public inbox for [email protected]help / color / mirror / Atom feed
[PATCH v7 3/4] restructure archive modules API 22+ messages / 7 participants [nested] [flat]
* [PATCH v7 3/4] restructure archive modules API @ 2023-01-31 22:58 Nathan Bossart <[email protected]> 0 siblings, 0 replies; 22+ messages in thread From: Nathan Bossart @ 2023-01-31 22:58 UTC (permalink / raw) --- contrib/basic_archive/basic_archive.c | 53 +++++++++++++++++++------ doc/src/sgml/archive-modules.sgml | 35 ++++++++++++---- src/backend/postmaster/pgarch.c | 23 ++++++----- src/backend/postmaster/shell_archive.c | 29 ++++++++------ src/include/postmaster/archive_module.h | 21 +++++++--- src/include/postmaster/shell_archive.h | 2 +- 6 files changed, 115 insertions(+), 48 deletions(-) diff --git a/contrib/basic_archive/basic_archive.c b/contrib/basic_archive/basic_archive.c index 87bbb2174d..74a51b9288 100644 --- a/contrib/basic_archive/basic_archive.c +++ b/contrib/basic_archive/basic_archive.c @@ -40,15 +40,27 @@ PG_MODULE_MAGIC; +typedef struct BasicArchiveData +{ + MemoryContext context; +} BasicArchiveData; + static char *archive_directory = NULL; -static MemoryContext basic_archive_context; -static bool basic_archive_configured(void); -static bool basic_archive_file(const char *file, const char *path); +static void basic_archive_startup(ArchiveModuleState *state); +static bool basic_archive_configured(ArchiveModuleState *state); +static bool basic_archive_file(ArchiveModuleState *state, const char *file, const char *path); static void basic_archive_file_internal(const char *file, const char *path); static bool check_archive_directory(char **newval, void **extra, GucSource source); static bool compare_files(const char *file1, const char *file2); +static const ArchiveModuleCallbacks basic_archive_callbacks = { + .startup_cb = basic_archive_startup, + .check_configured_cb = basic_archive_configured, + .archive_file_cb = basic_archive_file, + .shutdown_cb = NULL +}; + /* * _PG_init * @@ -67,10 +79,6 @@ _PG_init(void) check_archive_directory, NULL, NULL); MarkGUCPrefixReserved("basic_archive"); - - basic_archive_context = AllocSetContextCreate(TopMemoryContext, - "basic_archive", - ALLOCSET_DEFAULT_SIZES); } /* @@ -78,13 +86,30 @@ _PG_init(void) * * Returns the module's archiving callbacks. */ -void -_PG_archive_module_init(ArchiveModuleCallbacks *cb) +const ArchiveModuleCallbacks * +_PG_archive_module_init(void) { AssertVariableIsOfType(&_PG_archive_module_init, ArchiveModuleInit); - cb->check_configured_cb = basic_archive_configured; - cb->archive_file_cb = basic_archive_file; + return &basic_archive_callbacks; +} + +/* + * basic_archive_startup + * + * Creates the module's memory context. + */ +void +basic_archive_startup(ArchiveModuleState *state) +{ + BasicArchiveData *data; + + data = (BasicArchiveData *) MemoryContextAllocZero(TopMemoryContext, + sizeof(BasicArchiveData)); + data->context = AllocSetContextCreate(TopMemoryContext, + "basic_archive", + ALLOCSET_DEFAULT_SIZES); + state->private_data = (void *) data; } /* @@ -135,7 +160,7 @@ check_archive_directory(char **newval, void **extra, GucSource source) * Checks that archive_directory is not blank. */ static bool -basic_archive_configured(void) +basic_archive_configured(ArchiveModuleState *state) { return archive_directory != NULL && archive_directory[0] != '\0'; } @@ -146,10 +171,12 @@ basic_archive_configured(void) * Archives one file. */ static bool -basic_archive_file(const char *file, const char *path) +basic_archive_file(ArchiveModuleState *state, const char *file, const char *path) { sigjmp_buf local_sigjmp_buf; MemoryContext oldcontext; + BasicArchiveData *data = (BasicArchiveData *) (state->private_data); + MemoryContext basic_archive_context = data->context; /* * We run basic_archive_file_internal() in our own memory context so that diff --git a/doc/src/sgml/archive-modules.sgml b/doc/src/sgml/archive-modules.sgml index ef02051f7f..2db1b19216 100644 --- a/doc/src/sgml/archive-modules.sgml +++ b/doc/src/sgml/archive-modules.sgml @@ -47,23 +47,30 @@ normal library search path is used to locate the library. To provide the required archive module callbacks and to indicate that the library is actually an archive module, it needs to provide a function named - <function>_PG_archive_module_init</function>. This function is passed a - struct that needs to be filled with the callback function pointers for - individual actions. + <function>_PG_archive_module_init</function>. This function must return an + <structname>ArchiveModuleCallbacks</structname> struct filled with the + callback function pointers for individual actions. <programlisting> typedef struct ArchiveModuleCallbacks { + ArchiveStartupCB startup_cb; ArchiveCheckConfiguredCB check_configured_cb; ArchiveFileCB archive_file_cb; ArchiveShutdownCB shutdown_cb; } ArchiveModuleCallbacks; -typedef void (*ArchiveModuleInit) (struct ArchiveModuleCallbacks *cb); +typedef const ArchiveModuleCallbacks *(*ArchiveModuleInit) (void); </programlisting> Only the <function>archive_file_cb</function> callback is required. The others are optional. </para> + + <note> + <para> + <varname>archive_library</varname> is only loaded in the archiver process. + </para> + </note> </sect1> <sect1 id="archive-module-callbacks"> @@ -73,6 +80,20 @@ typedef void (*ArchiveModuleInit) (struct ArchiveModuleCallbacks *cb); The server will call them as required to process each individual WAL file. </para> + <sect2 id="archive-module-startup"> + <title>Startup Callback</title> + <para> + The <function>startup_cb</function> callback is called shortly after the + module is loaded. This callback can be used to perform any additional + initialization required. If the archive module needs to have a state, it + can use <structfield>state->private_data</structfield> to store it. + +<programlisting> +typedef void (*ArchiveStartupCB) (ArchiveModuleState *state); +</programlisting> + </para> + </sect2> + <sect2 id="archive-module-check"> <title>Check Callback</title> <para> @@ -83,7 +104,7 @@ typedef void (*ArchiveModuleInit) (struct ArchiveModuleCallbacks *cb); assumes the module is configured. <programlisting> -typedef bool (*ArchiveCheckConfiguredCB) (void); +typedef bool (*ArchiveCheckConfiguredCB) (ArchiveModuleState *state); </programlisting> If <literal>true</literal> is returned, the server will proceed with @@ -105,7 +126,7 @@ WARNING: archive_mode enabled, yet archiving is not configured single WAL file. <programlisting> -typedef bool (*ArchiveFileCB) (const char *file, const char *path); +typedef bool (*ArchiveFileCB) (ArchiveModuleState *state, const char *file, const char *path); </programlisting> If <literal>true</literal> is returned, the server proceeds as if the file @@ -128,7 +149,7 @@ typedef bool (*ArchiveFileCB) (const char *file, const char *path); these situations. <programlisting> -typedef void (*ArchiveShutdownCB) (void); +typedef void (*ArchiveShutdownCB) (ArchiveModuleState *state); </programlisting> </para> </sect2> diff --git a/src/backend/postmaster/pgarch.c b/src/backend/postmaster/pgarch.c index 281d9fd8b7..99a07996f7 100644 --- a/src/backend/postmaster/pgarch.c +++ b/src/backend/postmaster/pgarch.c @@ -99,7 +99,8 @@ char *XLogArchiveLibrary = ""; */ static time_t last_sigterm_time = 0; static PgArchData *PgArch = NULL; -static ArchiveModuleCallbacks ArchiveCallbacks; +static const ArchiveModuleCallbacks *ArchiveCallbacks; +static ArchiveModuleState *archive_module_state; /* @@ -408,8 +409,8 @@ pgarch_ArchiverCopyLoop(void) HandlePgArchInterrupts(); /* can't do anything if not configured ... */ - if (ArchiveCallbacks.check_configured_cb != NULL && - !ArchiveCallbacks.check_configured_cb()) + if (ArchiveCallbacks->check_configured_cb != NULL && + !ArchiveCallbacks->check_configured_cb(archive_module_state)) { ereport(WARNING, (errmsg("archive_mode enabled, yet archiving is not configured"))); @@ -510,7 +511,7 @@ pgarch_archiveXlog(char *xlog) snprintf(activitymsg, sizeof(activitymsg), "archiving %s", xlog); set_ps_display(activitymsg); - ret = ArchiveCallbacks.archive_file_cb(xlog, pathname); + ret = ArchiveCallbacks->archive_file_cb(archive_module_state, xlog, pathname); if (ret) snprintf(activitymsg, sizeof(activitymsg), "last was %s", xlog); else @@ -829,8 +830,6 @@ LoadArchiveLibrary(void) errmsg("both archive_command and archive_library set"), errdetail("Only one of archive_command, archive_library may be set."))); - memset(&ArchiveCallbacks, 0, sizeof(ArchiveModuleCallbacks)); - /* * If shell archiving is enabled, use our special initialization function. * Otherwise, load the library and call its _PG_archive_module_init(). @@ -846,12 +845,16 @@ LoadArchiveLibrary(void) ereport(ERROR, (errmsg("archive modules have to define the symbol %s", "_PG_archive_module_init"))); - (*archive_init) (&ArchiveCallbacks); + ArchiveCallbacks = (*archive_init) (); - if (ArchiveCallbacks.archive_file_cb == NULL) + if (ArchiveCallbacks->archive_file_cb == NULL) ereport(ERROR, (errmsg("archive modules must register an archive callback"))); + archive_module_state = (ArchiveModuleState *) palloc0(sizeof(ArchiveModuleState)); + if (ArchiveCallbacks->startup_cb != NULL) + ArchiveCallbacks->startup_cb(archive_module_state); + before_shmem_exit(pgarch_call_module_shutdown_cb, 0); } @@ -861,6 +864,6 @@ LoadArchiveLibrary(void) static void pgarch_call_module_shutdown_cb(int code, Datum arg) { - if (ArchiveCallbacks.shutdown_cb != NULL) - ArchiveCallbacks.shutdown_cb(); + if (ArchiveCallbacks->shutdown_cb != NULL) + ArchiveCallbacks->shutdown_cb(archive_module_state); } diff --git a/src/backend/postmaster/shell_archive.c b/src/backend/postmaster/shell_archive.c index 35f88c1222..cc2585218d 100644 --- a/src/backend/postmaster/shell_archive.c +++ b/src/backend/postmaster/shell_archive.c @@ -23,28 +23,33 @@ #include "postmaster/archive_module.h" #include "postmaster/shell_archive.h" -static bool shell_archive_configured(void); -static bool shell_archive_file(const char *file, const char *path); -static void shell_archive_shutdown(void); - -void -shell_archive_init(ArchiveModuleCallbacks *cb) +static bool shell_archive_configured(ArchiveModuleState *state); +static bool shell_archive_file(ArchiveModuleState *state, const char *file, const char *path); +static void shell_archive_shutdown(ArchiveModuleState *state); + +static const ArchiveModuleCallbacks shell_archive_callbacks = { + .startup_cb = NULL, + .check_configured_cb = shell_archive_configured, + .archive_file_cb = shell_archive_file, + .shutdown_cb = shell_archive_shutdown +}; + +const ArchiveModuleCallbacks * +shell_archive_init(void) { AssertVariableIsOfType(&shell_archive_init, ArchiveModuleInit); - cb->check_configured_cb = shell_archive_configured; - cb->archive_file_cb = shell_archive_file; - cb->shutdown_cb = shell_archive_shutdown; + return &shell_archive_callbacks; } static bool -shell_archive_configured(void) +shell_archive_configured(ArchiveModuleState *state) { return XLogArchiveCommand[0] != '\0'; } static bool -shell_archive_file(const char *file, const char *path) +shell_archive_file(ArchiveModuleState *state, const char *file, const char *path) { char *xlogarchcmd; char *nativePath = NULL; @@ -126,7 +131,7 @@ shell_archive_file(const char *file, const char *path) } static void -shell_archive_shutdown(void) +shell_archive_shutdown(ArchiveModuleState *state) { elog(DEBUG1, "archiver process shutting down"); } diff --git a/src/include/postmaster/archive_module.h b/src/include/postmaster/archive_module.h index f5015ff95d..79e9d53534 100644 --- a/src/include/postmaster/archive_module.h +++ b/src/include/postmaster/archive_module.h @@ -17,6 +17,15 @@ */ extern PGDLLIMPORT char *XLogArchiveLibrary; +typedef struct ArchiveModuleState +{ + /* + * Private data pointer for use by an archive module. This can be used to + * store state for the module that will be passed to each of its callbacks. + */ + void *private_data; +} ArchiveModuleState; + /* * Archive module callbacks * @@ -25,12 +34,14 @@ extern PGDLLIMPORT char *XLogArchiveLibrary; * For more information about the purpose of each callback, refer to the * archive modules documentation. */ -typedef bool (*ArchiveCheckConfiguredCB) (void); -typedef bool (*ArchiveFileCB) (const char *file, const char *path); -typedef void (*ArchiveShutdownCB) (void); +typedef void (*ArchiveStartupCB) (ArchiveModuleState *state); +typedef bool (*ArchiveCheckConfiguredCB) (ArchiveModuleState *state); +typedef bool (*ArchiveFileCB) (ArchiveModuleState *state, const char *file, const char *path); +typedef void (*ArchiveShutdownCB) (ArchiveModuleState *state); typedef struct ArchiveModuleCallbacks { + ArchiveStartupCB startup_cb; ArchiveCheckConfiguredCB check_configured_cb; ArchiveFileCB archive_file_cb; ArchiveShutdownCB shutdown_cb; @@ -40,8 +51,8 @@ typedef struct ArchiveModuleCallbacks * Type of the shared library symbol _PG_archive_module_init that is looked * up when loading an archive library. */ -typedef void (*ArchiveModuleInit) (ArchiveModuleCallbacks *cb); +typedef const ArchiveModuleCallbacks *(*ArchiveModuleInit) (void); -extern PGDLLEXPORT void _PG_archive_module_init(ArchiveModuleCallbacks *cb); +extern PGDLLEXPORT const ArchiveModuleCallbacks *_PG_archive_module_init(void); #endif /* _ARCHIVE_MODULE_H */ diff --git a/src/include/postmaster/shell_archive.h b/src/include/postmaster/shell_archive.h index 3cf737ce8e..be83c7715a 100644 --- a/src/include/postmaster/shell_archive.h +++ b/src/include/postmaster/shell_archive.h @@ -19,6 +19,6 @@ * and does not need to be loaded via a shared library, it has a special * initialization function. */ -extern void shell_archive_init(ArchiveModuleCallbacks *cb); +extern const ArchiveModuleCallbacks *shell_archive_init(void); #endif /* _SHELL_ARCHIVE_H */ -- 2.25.1 --/04w6evG8XlLl3ft Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="v7-0004-remove-unnecessary-assertions-for-functions-with-.patch" ^ permalink raw reply [nested|flat] 22+ messages in thread
* Re: trying again to get incremental backup @ 2023-06-14 20:40 Andres Freund <[email protected]> 0 siblings, 2 replies; 22+ messages in thread From: Andres Freund @ 2023-06-14 20:40 UTC (permalink / raw) To: Robert Haas <[email protected]>; +Cc: pgsql-hackers Hi, On 2023-06-14 16:10:38 -0400, Robert Haas wrote: > On Wed, Jun 14, 2023 at 3:47 PM Andres Freund <[email protected]> wrote: > > Could we just recompute the WAL summary for the [redo, end of chunk] for the > > relevant summary file? > > I'm not understanding how that would help. If we were going to compute > a WAL summary on the fly rather than waiting for one to show up on > disk, what we'd want is [end of last WAL summary that does exist on > disk, redo]. Oh, right. > But I'm not sure that's a great approach, because that LSN gap might be > large and then we're duplicating a lot of work that the summarizer has > probably already done most of. I guess that really depends on what the summary granularity is. If you create a separate summary every 32MB or so, recomputing just the required range shouldn't be too bad. > > FWIW, I like the idea of a special WAL record at that point, independent of > > this feature. It wouldn't be a meaningful overhead compared to the cost of a > > checkpoint, and it seems like it'd be quite useful for debugging. But I can > > see uses going beyond that - we occasionally have been discussing associating > > additional data with redo points, and that'd be a lot easier to deal with > > during recovery with such a record. > > > > I don't really see a performance and concurrency angle right now - what are > > you wondering about? > > I'm not really sure. I expect Dilip would be happy to post his patch, > and if you'd be willing to have a look at it and express your concerns > or lack thereof, that would be super valuable. Will do. Adding me to CC: might help, I have a backlog unfortunately :(. Greetings, Andres Freund ^ permalink raw reply [nested|flat] 22+ messages in thread
* Re: trying again to get incremental backup @ 2023-06-15 07:44 Dilip Kumar <[email protected]> parent: Andres Freund <[email protected]> 1 sibling, 0 replies; 22+ messages in thread From: Dilip Kumar @ 2023-06-15 07:44 UTC (permalink / raw) To: Andres Freund <[email protected]>; +Cc: Robert Haas <[email protected]>; pgsql-hackers On Thu, Jun 15, 2023 at 2:11 AM Andres Freund <[email protected]> wrote: > > > I'm not really sure. I expect Dilip would be happy to post his patch, > > and if you'd be willing to have a look at it and express your concerns > > or lack thereof, that would be super valuable. > > Will do. Adding me to CC: might help, I have a backlog unfortunately :(. Thanks, I have posted it here[1] [1] https://www.postgresql.org/message-id/CAFiTN-s-K%3DmVA%3DHPr_VoU-5bvyLQpNeuzjq1ebPJMEfCJZKFsg%40mail... -- Regards, Dilip Kumar EnterpriseDB: http://www.enterprisedb.com ^ permalink raw reply [nested|flat] 22+ messages in thread
* Re: trying again to get incremental backup @ 2023-06-19 13:46 Robert Haas <[email protected]> parent: Andres Freund <[email protected]> 1 sibling, 1 reply; 22+ messages in thread From: Robert Haas @ 2023-06-19 13:46 UTC (permalink / raw) To: Andres Freund <[email protected]>; +Cc: pgsql-hackers On Wed, Jun 14, 2023 at 4:40 PM Andres Freund <[email protected]> wrote: > > But I'm not sure that's a great approach, because that LSN gap might be > > large and then we're duplicating a lot of work that the summarizer has > > probably already done most of. > > I guess that really depends on what the summary granularity is. If you create > a separate summary every 32MB or so, recomputing just the required range > shouldn't be too bad. Yeah, but I don't think that's the right approach, for two reasons. First, one of the things I'm rather worried about is what happens when the WAL distance between the prior backup and the incremental backup is large. It could be a terabyte. If we have a WAL summary for every 32MB of WAL, that's 32k files we have to read, and I'm concerned that's too many. Maybe it isn't, but it's something that has really been weighing on my mind as I've been thinking through the design questions here. The files are really very small, and having to open a bazillion tiny little files to get the job done sounds lame. Second, I don't see what problem it actually solves. Why not just signal the summarizer to write out the accumulated data to a file instead of re-doing the work ourselves? Or else adopt the WAL-record-at-the-redo-pointer approach, and then the whole thing is moot? -- Robert Haas EDB: http://www.enterprisedb.com ^ permalink raw reply [nested|flat] 22+ messages in thread
* Re: trying again to get incremental backup @ 2023-06-19 15:51 Andres Freund <[email protected]> parent: Robert Haas <[email protected]> 0 siblings, 1 reply; 22+ messages in thread From: Andres Freund @ 2023-06-19 15:51 UTC (permalink / raw) To: Robert Haas <[email protected]>; +Cc: pgsql-hackers Hi, On 2023-06-19 09:46:12 -0400, Robert Haas wrote: > On Wed, Jun 14, 2023 at 4:40 PM Andres Freund <[email protected]> wrote: > > > But I'm not sure that's a great approach, because that LSN gap might be > > > large and then we're duplicating a lot of work that the summarizer has > > > probably already done most of. > > > > I guess that really depends on what the summary granularity is. If you create > > a separate summary every 32MB or so, recomputing just the required range > > shouldn't be too bad. > > Yeah, but I don't think that's the right approach, for two reasons. > First, one of the things I'm rather worried about is what happens when > the WAL distance between the prior backup and the incremental backup > is large. It could be a terabyte. If we have a WAL summary for every > 32MB of WAL, that's 32k files we have to read, and I'm concerned > that's too many. Maybe it isn't, but it's something that has really > been weighing on my mind as I've been thinking through the design > questions here. It doesn't have to be a separate file - you could easily summarize ranges at a higher granularity, storing multiple ranges into a single file with a coarser naming pattern. > The files are really very small, and having to open a bazillion tiny little > files to get the job done sounds lame. Second, I don't see what problem it > actually solves. Why not just signal the summarizer to write out the > accumulated data to a file instead of re-doing the work ourselves? Or else > adopt the WAL-record-at-the-redo-pointer approach, and then the whole thing > is moot? The one point for a relatively grainy summarization scheme that I see is that it would pave the way for using the WAL summary data for other purposes in the future. That could be done orthogonal to the other solutions to the redo pointer issues. Other potential use cases: - only restore parts of a base backup that aren't going to be overwritten by WAL replay - reconstructing database contents from WAL after data loss - more efficient pg_rewind - more efficient prefetching during WAL replay Greetings, Andres Freund ^ permalink raw reply [nested|flat] 22+ messages in thread
* Re: trying again to get incremental backup @ 2023-08-30 14:49 Robert Haas <[email protected]> parent: Andres Freund <[email protected]> 0 siblings, 3 replies; 22+ messages in thread From: Robert Haas @ 2023-08-30 14:49 UTC (permalink / raw) To: Andres Freund <[email protected]>; +Cc: pgsql-hackers Hi, In the limited time that I've had to work on this project lately, I've been trying to come up with a test case for this feature -- and since I've gotten completely stuck, I thought it might be time to post and see if anyone else has a better idea. I thought a reasonable test case would be: Do a full backup. Change some stuff. Do an incremental backup. Restore both backups and perform replay to the same LSN. Then compare the files on disk. But I cannot make this work. The first problem I ran into was that replay of the full backup does a restartpoint, while the replay of the incremental backup does not. That results in, for example, pg_subtrans having different contents. I'm not sure whether it can also result in data files having different contents: are changes that we replayed following the last restartpoint guaranteed to end up on disk when the server is shut down? It wasn't clear to me that this is the case. I thought maybe I could get both servers to perform a restartpoint at the same location by shutting down the primary and then replaying through the shutdown checkpoint, but that doesn't work because the primary doesn't finish archiving before shutting down. After some more fiddling I settled (at least for research purposes) on having the restored backups PITR and promote, instead of PITR and pause, so that we're guaranteed a checkpoint. But that just caused me to run into a far worse problem: replay on the standby doesn't actually create a state that is byte-for-byte identical to the one that exists on the primary. I quickly discovered that in my test case, I was ending up with different contents in the "hole" of a block wherein a tuple got updated. Replay doesn't think it's important to make the hole end up with the same contents on all machines that replay the WAL, so I end up with one server that has more junk in there than the other one and the tests fail. Unless someone has a brilliant idea that I lack, this suggests to me that this whole line of testing is a dead end. I can, of course, write tests that compare clusters *logically* -- do the correct relations exist, are they accessible, do they have the right contents? But I feel like it would be easy to have bugs that escape detection in such a test but would be detected by a physical comparison of the clusters. However, such a comparison can only be conducted if either (a) there's some way to set up the test so that byte-for-byte identical clusters can be expected or (b) there's some way to perform the comparison that can distinguish between expected, harmless differences and unexpected, problematic differences. And at the moment my conclusion is that neither (a) nor (b) exists. Does anyone think otherwise? Meanwhile, here's a rebased set of patches. The somewhat-primitive attempts at writing tests are in 0009, but they don't work, for the reasons explained above. I think I'd probably like to go ahead and commit 0001 and 0002 soon if there are no objections, since I think those are good refactorings independently of the rest of this. ...Robert Attachments: [application/octet-stream] 0004-Refactor-parse_filename_for_nontemp_relation-to-pars.patch (11.9K, ../../CA+TgmoZoHNqx=zconLLbV+8=gNhv7g5ocXd-Mee97wvTkKc-6A@mail.gmail.com/2-0004-Refactor-parse_filename_for_nontemp_relation-to-pars.patch) download | inline diff: From 816266cb17cbcea889a13de1e146fcb5ebfe2066 Mon Sep 17 00:00:00 2001 From: Robert Haas <[email protected]> Date: Mon, 5 Jun 2023 15:42:15 -0400 Subject: [PATCH 4/9] Refactor parse_filename_for_nontemp_relation to parse more. Instead of returning the number of characters in the RelFileNumber, return the RelFileNumber itself. Continue to return the fork number, as before, and additionally return the segment number. parse_filename_for_nontemp_relation now rejects a RelFileNumber or segment number that begins with a leading zero. Before, we accepted such cases as relation filenames, but if we continued to do so after this change, the function might return the same values for two different files (e.g. 1234.5 and 001234.5 or 1234.005) which could be annoying for callers. Since we don't actually ever generate filenames with leading zeroes in the names, any such files that we find must have been created by something other than PostgreSQL, and it is therefore reasonable to treat them as non-relation files. Along the way, change unlogged_relation_entry to store a RelFileNumber rather than an OID. This update should have been made in 851f4cc75cdd8c831f1baa9a7abf8c8248b65890, but it was overlooked. It could be done separately from the rest of this commit, but that would be more involved, whereas this way it's a 1-line change. --- src/backend/backup/basebackup.c | 15 ++-- src/backend/storage/file/reinit.c | 137 ++++++++++++++++++------------ src/include/storage/reinit.h | 5 +- 3 files changed, 93 insertions(+), 64 deletions(-) diff --git a/src/backend/backup/basebackup.c b/src/backend/backup/basebackup.c index cc3d2e0c41..24c038dfba 100644 --- a/src/backend/backup/basebackup.c +++ b/src/backend/backup/basebackup.c @@ -1198,9 +1198,9 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, { int excludeIdx; bool excludeFound; - ForkNumber relForkNum; /* Type of fork if file is a relation */ - int relnumchars; /* Chars in filename that are the - * relnumber */ + RelFileNumber relNumber; + ForkNumber relForkNum; + unsigned segno; /* Skip special stuff */ if (strcmp(de->d_name, ".") == 0 || strcmp(de->d_name, "..") == 0) @@ -1250,23 +1250,20 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, /* Exclude all forks for unlogged tables except the init fork */ if (isDbDir && - parse_filename_for_nontemp_relation(de->d_name, &relnumchars, - &relForkNum)) + parse_filename_for_nontemp_relation(de->d_name, &relNumber, + &relForkNum, &segno)) { /* Never exclude init forks */ if (relForkNum != INIT_FORKNUM) { char initForkFile[MAXPGPATH]; - char relNumber[OIDCHARS + 1]; /* * If any other type of fork, check if there is an init fork * with the same RelFileNumber. If so, the file can be * excluded. */ - memcpy(relNumber, de->d_name, relnumchars); - relNumber[relnumchars] = '\0'; - snprintf(initForkFile, sizeof(initForkFile), "%s/%s_init", + snprintf(initForkFile, sizeof(initForkFile), "%s/%u_init", path, relNumber); if (lstat(initForkFile, &statbuf) == 0) diff --git a/src/backend/storage/file/reinit.c b/src/backend/storage/file/reinit.c index fb55371b1b..31d6e01106 100644 --- a/src/backend/storage/file/reinit.c +++ b/src/backend/storage/file/reinit.c @@ -31,7 +31,7 @@ static void ResetUnloggedRelationsInDbspaceDir(const char *dbspacedirname, typedef struct { - Oid reloid; /* hash key */ + RelFileNumber relnumber; /* hash key */ } unlogged_relation_entry; /* @@ -195,12 +195,13 @@ ResetUnloggedRelationsInDbspaceDir(const char *dbspacedirname, int op) while ((de = ReadDir(dbspace_dir, dbspacedirname)) != NULL) { ForkNumber forkNum; - int relnumchars; + unsigned segno; unlogged_relation_entry ent; /* Skip anything that doesn't look like a relation data file. */ - if (!parse_filename_for_nontemp_relation(de->d_name, &relnumchars, - &forkNum)) + if (!parse_filename_for_nontemp_relation(de->d_name, + &ent.relnumber, + &forkNum, &segno)) continue; /* Also skip it unless this is the init fork. */ @@ -208,10 +209,8 @@ ResetUnloggedRelationsInDbspaceDir(const char *dbspacedirname, int op) continue; /* - * Put the OID portion of the name into the hash table, if it - * isn't already. + * Put the RelFileNumber into the hash table, if it isn't already. */ - ent.reloid = atooid(de->d_name); (void) hash_search(hash, &ent, HASH_ENTER, NULL); } @@ -235,12 +234,13 @@ ResetUnloggedRelationsInDbspaceDir(const char *dbspacedirname, int op) while ((de = ReadDir(dbspace_dir, dbspacedirname)) != NULL) { ForkNumber forkNum; - int relnumchars; + unsigned segno; unlogged_relation_entry ent; /* Skip anything that doesn't look like a relation data file. */ - if (!parse_filename_for_nontemp_relation(de->d_name, &relnumchars, - &forkNum)) + if (!parse_filename_for_nontemp_relation(de->d_name, + &ent.relnumber, + &forkNum, &segno)) continue; /* We never remove the init fork. */ @@ -251,7 +251,6 @@ ResetUnloggedRelationsInDbspaceDir(const char *dbspacedirname, int op) * See whether the OID portion of the name shows up in the hash * table. If so, nuke it! */ - ent.reloid = atooid(de->d_name); if (hash_search(hash, &ent, HASH_FIND, NULL)) { snprintf(rm_path, sizeof(rm_path), "%s/%s", @@ -285,14 +284,14 @@ ResetUnloggedRelationsInDbspaceDir(const char *dbspacedirname, int op) while ((de = ReadDir(dbspace_dir, dbspacedirname)) != NULL) { ForkNumber forkNum; - int relnumchars; - char relnumbuf[OIDCHARS + 1]; + RelFileNumber relNumber; + unsigned segno; char srcpath[MAXPGPATH * 2]; char dstpath[MAXPGPATH]; /* Skip anything that doesn't look like a relation data file. */ - if (!parse_filename_for_nontemp_relation(de->d_name, &relnumchars, - &forkNum)) + if (!parse_filename_for_nontemp_relation(de->d_name, &relNumber, + &forkNum, &segno)) continue; /* Also skip it unless this is the init fork. */ @@ -304,11 +303,12 @@ ResetUnloggedRelationsInDbspaceDir(const char *dbspacedirname, int op) dbspacedirname, de->d_name); /* Construct destination pathname. */ - memcpy(relnumbuf, de->d_name, relnumchars); - relnumbuf[relnumchars] = '\0'; - snprintf(dstpath, sizeof(dstpath), "%s/%s%s", - dbspacedirname, relnumbuf, de->d_name + relnumchars + 1 + - strlen(forkNames[INIT_FORKNUM])); + if (segno == 0) + snprintf(dstpath, sizeof(dstpath), "%s/%u", + dbspacedirname, relNumber); + else + snprintf(dstpath, sizeof(dstpath), "%s/%u.%u", + dbspacedirname, relNumber, segno); /* OK, we're ready to perform the actual copy. */ elog(DEBUG2, "copying %s to %s", srcpath, dstpath); @@ -327,14 +327,14 @@ ResetUnloggedRelationsInDbspaceDir(const char *dbspacedirname, int op) dbspace_dir = AllocateDir(dbspacedirname); while ((de = ReadDir(dbspace_dir, dbspacedirname)) != NULL) { + RelFileNumber relNumber; ForkNumber forkNum; - int relnumchars; - char relnumbuf[OIDCHARS + 1]; + unsigned segno; char mainpath[MAXPGPATH]; /* Skip anything that doesn't look like a relation data file. */ - if (!parse_filename_for_nontemp_relation(de->d_name, &relnumchars, - &forkNum)) + if (!parse_filename_for_nontemp_relation(de->d_name, &relNumber, + &forkNum, &segno)) continue; /* Also skip it unless this is the init fork. */ @@ -342,11 +342,12 @@ ResetUnloggedRelationsInDbspaceDir(const char *dbspacedirname, int op) continue; /* Construct main fork pathname. */ - memcpy(relnumbuf, de->d_name, relnumchars); - relnumbuf[relnumchars] = '\0'; - snprintf(mainpath, sizeof(mainpath), "%s/%s%s", - dbspacedirname, relnumbuf, de->d_name + relnumchars + 1 + - strlen(forkNames[INIT_FORKNUM])); + if (segno == 0) + snprintf(mainpath, sizeof(mainpath), "%s/%u", + dbspacedirname, relNumber); + else + snprintf(mainpath, sizeof(mainpath), "%s/%u.%u", + dbspacedirname, relNumber, segno); fsync_fname(mainpath, false); } @@ -371,52 +372,82 @@ ResetUnloggedRelationsInDbspaceDir(const char *dbspacedirname, int op) * This function returns true if the file appears to be in the correct format * for a non-temporary relation and false otherwise. * - * NB: If this function returns true, the caller is entitled to assume that - * *relnumchars has been set to a value no more than OIDCHARS, and thus - * that a buffer of OIDCHARS+1 characters is sufficient to hold the - * RelFileNumber portion of the filename. This is critical to protect against - * a possible buffer overrun. + * If it returns true, it sets *relnumber, *fork, and *segno to the values + * extracted from the filename. If it returns false, these values are set to + * InvalidRelFileNumber, InvalidForkNumber, and 0, respectively. */ bool -parse_filename_for_nontemp_relation(const char *name, int *relnumchars, - ForkNumber *fork) +parse_filename_for_nontemp_relation(const char *name, RelFileNumber *relnumber, + ForkNumber *fork, unsigned *segno) { - int pos; + unsigned long n, + s; + ForkNumber f; + char *endp; - /* Look for a non-empty string of digits (that isn't too long). */ - for (pos = 0; isdigit((unsigned char) name[pos]); ++pos) - ; - if (pos == 0 || pos > OIDCHARS) + *relnumber = InvalidRelFileNumber; + *fork = InvalidForkNumber; + *segno = 0; + + /* + * Relation filenames should begin with a digit that is not a zero. By + * rejecting cases involving leading zeroes, the caller can assume that + * there's only one possible string of characters that could have produced + * any given value for *relnumber. + * + * (To be clear, we don't expect files with names like 0017.3 to exist + * at all -- but if 0017.3 does exist, it's a non-relation file, not + * part of the main fork for relfilenode 17.) + */ + if (name[0] < '1' || name[0] > '9') + return false; + + /* + * Parse the leading digit string. If the value is out of range, we + * conclude that this isn't a relation file at all. + */ + errno = 0; + n = strtoul(name, &endp, 10); + if (errno || name == endp || n <= 0 || n > PG_UINT32_MAX) return false; - *relnumchars = pos; + name = endp; /* Check for a fork name. */ - if (name[pos] != '_') - *fork = MAIN_FORKNUM; + if (*name != '_') + f = MAIN_FORKNUM; else { int forkchar; - forkchar = forkname_chars(&name[pos + 1], fork); + forkchar = forkname_chars(name + 1, &f); if (forkchar <= 0) return false; - pos += forkchar + 1; + name += forkchar + 1; } /* Check for a segment number. */ - if (name[pos] == '.') + if (*name != '.') + s = 0; + else { - int segchar; + /* Reject leading zeroes, just like we do for RelFileNumber. */ + if (name[0] < '1' || name[0] > '9') + return false; - for (segchar = 1; isdigit((unsigned char) name[pos + segchar]); ++segchar) - ; - if (segchar <= 1) + errno = 0; + s = strtoul(name + 1, &endp, 10); + if (errno || name + 1 == endp || s <= 0 || s > PG_UINT32_MAX) return false; - pos += segchar; + name = endp; } /* Now we should be at the end. */ - if (name[pos] != '\0') + if (*name != '\0') return false; + + /* Set out parameters and return. */ + *relnumber = (RelFileNumber) n; + *fork = f; + *segno = (unsigned) s; return true; } diff --git a/src/include/storage/reinit.h b/src/include/storage/reinit.h index e2bbb5abe9..f8eb7ce234 100644 --- a/src/include/storage/reinit.h +++ b/src/include/storage/reinit.h @@ -20,8 +20,9 @@ extern void ResetUnloggedRelations(int op); extern bool parse_filename_for_nontemp_relation(const char *name, - int *relnumchars, - ForkNumber *fork); + RelFileNumber *relnumber, + ForkNumber *fork, + unsigned *segno); #define UNLOGGED_RELATION_CLEANUP 0x0001 #define UNLOGGED_RELATION_INIT 0x0002 -- 2.37.1 (Apple Git-137.1) [application/octet-stream] 0002-In-basebackup.c-refactor-to-create-read_file_data_in.patch (12.0K, ../../CA+TgmoZoHNqx=zconLLbV+8=gNhv7g5ocXd-Mee97wvTkKc-6A@mail.gmail.com/3-0002-In-basebackup.c-refactor-to-create-read_file_data_in.patch) download | inline diff: From f3aebf944f13080b108cbc1b0247e22dc0b8a187 Mon Sep 17 00:00:00 2001 From: Robert Haas <[email protected]> Date: Mon, 5 Jun 2023 15:40:14 -0400 Subject: [PATCH 2/9] In basebackup.c, refactor to create read_file_data_into_buffer. This further reduces the length and complexity of sendFile(), hopefully make it easier to understand and modify. In addition to moving some logic into a new function, I took this opportunity to make a few slight adjustments to sendFile() itself, including renaming the 'len' variable to 'bytes_done', since we use it to represent the number of bytes we've already handled so far, not the total length of the file. --- src/backend/backup/basebackup.c | 231 ++++++++++++++++++-------------- 1 file changed, 133 insertions(+), 98 deletions(-) diff --git a/src/backend/backup/basebackup.c b/src/backend/backup/basebackup.c index 0daf8257bc..f46f930329 100644 --- a/src/backend/backup/basebackup.c +++ b/src/backend/backup/basebackup.c @@ -83,6 +83,12 @@ static int64 sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeo static bool sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, struct stat *statbuf, bool missing_ok, Oid dboid, backup_manifest_info *manifest, const char *spcoid); +static off_t read_file_data_into_buffer(bbsink *sink, + const char *readfilename, int fd, + off_t offset, size_t length, + BlockNumber blkno, + bool verify_checksum, + int *checksum_failures); static bool verify_page_checksum(Page page, XLogRecPtr start_lsn, BlockNumber blkno, uint16 *expected_checksum); @@ -1490,9 +1496,7 @@ sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, BlockNumber blkno = 0; int checksum_failures = 0; off_t cnt; - int i; - pgoff_t len = 0; - char *page; + pgoff_t bytes_done = 0; int segmentno = 0; char *segmentpath; bool verify_checksum = false; @@ -1514,6 +1518,12 @@ sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, _tarWriteHeader(sink, tarfilename, NULL, statbuf, false); + /* + * Checksums are verified in multiples of BLCKSZ, so the buffer length + * should be a multiple of the block size as well. + */ + Assert((sink->bbs_buffer_length % BLCKSZ) == 0); + if (!noverify_checksums && DataChecksumsEnabled()) { char *filename; @@ -1551,23 +1561,21 @@ sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, * for a base backup we can ignore such extended data. It will be restored * from WAL. */ - while (len < statbuf->st_size) + while (bytes_done < statbuf->st_size) { - size_t remaining = statbuf->st_size - len; + size_t remaining = statbuf->st_size - bytes_done; /* Try to read some more data. */ - cnt = basebackup_read_file(fd, sink->bbs_buffer, - Min(sink->bbs_buffer_length, remaining), - len, readfilename, true); + cnt = read_file_data_into_buffer(sink, readfilename, fd, bytes_done, + remaining, + blkno + segmentno * RELSEG_SIZE, + verify_checksum, + &checksum_failures); /* - * The checksums are verified at block level, so we iterate over the - * buffer in chunks of BLCKSZ, after making sure that - * TAR_SEND_SIZE/buf is divisible by BLCKSZ and we read a multiple of - * BLCKSZ bytes. + * If the amount of data we were able to read was not a multiple of + * BLCKSZ, we cannot verify checksums, which are block-level. */ - Assert((sink->bbs_buffer_length % BLCKSZ) == 0); - if (verify_checksum && (cnt % BLCKSZ != 0)) { ereport(WARNING, @@ -1578,84 +1586,6 @@ sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, verify_checksum = false; } - if (verify_checksum) - { - for (i = 0; i < cnt / BLCKSZ; i++) - { - int reread_cnt; - uint16 expected_checksum; - - page = sink->bbs_buffer + BLCKSZ * i; - - /* If the page is OK, go on to the next one. */ - if (verify_page_checksum(page, sink->bbs_state->startptr, - blkno + i + segmentno * RELSEG_SIZE, - &expected_checksum)) - continue; - - /* - * Retry the block on the first failure. It's possible that - * we read the first 4K page of the block just before postgres - * updated the entire block so it ends up looking torn to us. - * If, before we retry the read, the concurrent write of the - * block finishes, the page LSN will be updated and we'll - * realize that we should ignore this block. - * - * There's no guarantee that this will actually happen, - * though: the torn write could take an arbitrarily long time - * to complete. Retrying multiple times wouldn't fix this - * problem, either, though it would reduce the chances of it - * happening in practice. The only real fix here seems to be - * to have some kind of interlock that allows us to wait until - * we can be certain that no write to the block is in - * progress. Since we don't have any such thing right now, we - * just do this and hope for the best. - */ - reread_cnt = - basebackup_read_file(fd, - sink->bbs_buffer + BLCKSZ * i, - BLCKSZ, len + BLCKSZ * i, - readfilename, - false); - if (reread_cnt == 0) - { - /* - * If we hit end-of-file, a concurrent truncation must - * have occurred, so break out of this loop just as if the - * initial fread() returned 0. We'll drop through to the - * same code that handles that case. (We must fix up cnt - * first, though.) - */ - cnt = BLCKSZ * i; - break; - } - - /* If the page now looks OK, go on to the next one. */ - if (verify_page_checksum(page, sink->bbs_state->startptr, - blkno + i + segmentno * RELSEG_SIZE, - &expected_checksum)) - continue; - - /* Handle checksum failure. */ - checksum_failures++; - if (checksum_failures <= 5) - ereport(WARNING, - (errmsg("checksum verification failed in " - "file \"%s\", block %u: calculated " - "%X but expected %X", - readfilename, blkno + i, expected_checksum, - ((PageHeader) page)->pd_checksum))); - if (checksum_failures == 5) - ereport(WARNING, - (errmsg("further checksum verification " - "failures in file \"%s\" will not " - "be reported", readfilename))); - } - - /* Update block number for next pass through the outer loop. */ - blkno += i; - } - /* * If we hit end-of-file, a concurrent truncation must have occurred. * That's not an error condition, because WAL replay will fix things @@ -1664,6 +1594,10 @@ sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, if (cnt == 0) break; + /* Update block number and # of bytes done for next loop iteration. */ + blkno += cnt / BLCKSZ; + bytes_done += cnt; + /* Archive the data we just read. */ bbsink_archive_contents(sink, cnt); @@ -1671,14 +1605,12 @@ sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, if (pg_checksum_update(&checksum_ctx, (uint8 *) sink->bbs_buffer, cnt) < 0) elog(ERROR, "could not update checksum of base backup"); - - len += cnt; } /* If the file was truncated while we were sending it, pad it with zeros */ - while (len < statbuf->st_size) + while (bytes_done < statbuf->st_size) { - size_t remaining = statbuf->st_size - len; + size_t remaining = statbuf->st_size - bytes_done; size_t nbytes = Min(sink->bbs_buffer_length, remaining); MemSet(sink->bbs_buffer, 0, nbytes); @@ -1687,7 +1619,7 @@ sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, nbytes) < 0) elog(ERROR, "could not update checksum of base backup"); bbsink_archive_contents(sink, nbytes); - len += nbytes; + bytes_done += nbytes; } /* @@ -1695,7 +1627,7 @@ sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, * of data is probably not worth throttling, and is not checksummed * because it's not actually part of the file.) */ - _tarWritePadding(sink, len); + _tarWritePadding(sink, bytes_done); CloseTransientFile(fd); @@ -1718,6 +1650,109 @@ sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, return true; } +/* + * Read some more data from the file into the bbsink's buffer, verifying + * checksums as required. + * + * 'offset' is the file offset from which we should begin to read, and + * 'length' is the amount of data that should be read. The actual amount + * of data read will be less than the requested amount if the bbsink's + * buffer isn't big enough to hold it all, or if the underlying file has + * been truncated. The return value is the number of bytes actually read. + * + * 'blkno' is the block number of the first page in the bbsink's buffer + * relative to the start of the relation. + * + * 'verify_checksum' indicates whether we should try to verify checksums + * for the blocks we read. If we do this, we'll update *checksum_failures + * and issue warnings as appropriate. + */ +static off_t +read_file_data_into_buffer(bbsink *sink, const char *readfilename, int fd, + off_t offset, size_t length, BlockNumber blkno, + bool verify_checksum, int *checksum_failures) +{ + off_t cnt; + int i; + char *page; + + /* Try to read some more data. */ + cnt = basebackup_read_file(fd, sink->bbs_buffer, + Min(sink->bbs_buffer_length, length), + offset, readfilename, true); + + /* Can't verify checksums if read length is not a multiple of BLCKSZ. */ + if (!verify_checksum || (cnt % BLCKSZ) != 0) + return cnt; + + /* Verify checksum for each block. */ + for (i = 0; i < cnt / BLCKSZ; i++) + { + int reread_cnt; + uint16 expected_checksum; + + page = sink->bbs_buffer + BLCKSZ * i; + + /* If the page is OK, go on to the next one. */ + if (verify_page_checksum(page, sink->bbs_state->startptr, blkno + i, + &expected_checksum)) + continue; + + /* + * Retry the block on the first failure. It's possible that we read + * the first 4K page of the block just before postgres updated the + * entire block so it ends up looking torn to us. If, before we retry + * the read, the concurrent write of the block finishes, the page LSN + * will be updated and we'll realize that we should ignore this block. + * + * There's no guarantee that this will actually happen, though: the + * torn write could take an arbitrarily long time to complete. + * Retrying multiple times wouldn't fix this problem, either, though + * it would reduce the chances of it happening in practice. The only + * real fix here seems to be to have some kind of interlock that + * allows us to wait until we can be certain that no write to the + * block is in progress. Since we don't have any such thing right now, + * we just do this and hope for the best. + */ + reread_cnt = + basebackup_read_file(fd, sink->bbs_buffer + BLCKSZ * i, + BLCKSZ, offset + BLCKSZ * i, + readfilename, false); + if (reread_cnt == 0) + { + /* + * If we hit end-of-file, a concurrent truncation must have + * occurred, so reduce cnt to reflect only the blocks already + * processed and break out of this loop. + */ + cnt = BLCKSZ * i; + break; + } + + /* If the page now looks OK, go on to the next one. */ + if (verify_page_checksum(page, sink->bbs_state->startptr, blkno + i, + &expected_checksum)) + continue; + + /* Handle checksum failure. */ + (*checksum_failures)++; + if (*checksum_failures <= 5) + ereport(WARNING, + (errmsg("checksum verification failed in " + "file \"%s\", block %u: calculated " + "%X but expected %X", + readfilename, blkno + i, expected_checksum, + ((PageHeader) page)->pd_checksum))); + if (*checksum_failures == 5) + ereport(WARNING, + (errmsg("further checksum verification " + "failures in file \"%s\" will not " + "be reported", readfilename))); + } + + return cnt; +} + /* * Try to verify the checksum for the provided page, if it seems appropriate * to do so. -- 2.37.1 (Apple Git-137.1) [application/octet-stream] 0001-In-basebackup.c-refactor-to-create-verify_page_check.patch (9.1K, ../../CA+TgmoZoHNqx=zconLLbV+8=gNhv7g5ocXd-Mee97wvTkKc-6A@mail.gmail.com/4-0001-In-basebackup.c-refactor-to-create-verify_page_check.patch) download | inline diff: From b65fb32ca1474a1158d5490970b9eb147fbe4f47 Mon Sep 17 00:00:00 2001 From: Robert Haas <[email protected]> Date: Mon, 5 Jun 2023 15:40:07 -0400 Subject: [PATCH 1/9] In basebackup.c, refactor to create verify_page_checksum. If checksum verification fails for a particular page, we reread the page and try one more time. The code that does this somewhat complex and difficult to follow. Move some of the logic into a new function and rearrange the code a bit to try to make it clearer. This way, we don't need the block_retry Boolean, a couple of other variables move from sendFile() into the new function, and some code is now less deeply indented. --- src/backend/backup/basebackup.c | 188 ++++++++++++++++++-------------- 1 file changed, 104 insertions(+), 84 deletions(-) diff --git a/src/backend/backup/basebackup.c b/src/backend/backup/basebackup.c index 45be21131c..0daf8257bc 100644 --- a/src/backend/backup/basebackup.c +++ b/src/backend/backup/basebackup.c @@ -83,6 +83,9 @@ static int64 sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeo static bool sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, struct stat *statbuf, bool missing_ok, Oid dboid, backup_manifest_info *manifest, const char *spcoid); +static bool verify_page_checksum(Page page, XLogRecPtr start_lsn, + BlockNumber blkno, + uint16 *expected_checksum); static void sendFileWithContent(bbsink *sink, const char *filename, const char *content, backup_manifest_info *manifest); @@ -1485,14 +1488,11 @@ sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, { int fd; BlockNumber blkno = 0; - bool block_retry = false; - uint16 checksum; int checksum_failures = 0; off_t cnt; int i; pgoff_t len = 0; char *page; - PageHeader phdr; int segmentno = 0; char *segmentpath; bool verify_checksum = false; @@ -1582,94 +1582,78 @@ sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, { for (i = 0; i < cnt / BLCKSZ; i++) { + int reread_cnt; + uint16 expected_checksum; + page = sink->bbs_buffer + BLCKSZ * i; + /* If the page is OK, go on to the next one. */ + if (verify_page_checksum(page, sink->bbs_state->startptr, + blkno + i + segmentno * RELSEG_SIZE, + &expected_checksum)) + continue; + /* - * Only check pages which have not been modified since the - * start of the base backup. Otherwise, they might have been - * written only halfway and the checksum would not be valid. - * However, replaying WAL would reinstate the correct page in - * this case. We also skip completely new pages, since they - * don't have a checksum yet. + * Retry the block on the first failure. It's possible that + * we read the first 4K page of the block just before postgres + * updated the entire block so it ends up looking torn to us. + * If, before we retry the read, the concurrent write of the + * block finishes, the page LSN will be updated and we'll + * realize that we should ignore this block. + * + * There's no guarantee that this will actually happen, + * though: the torn write could take an arbitrarily long time + * to complete. Retrying multiple times wouldn't fix this + * problem, either, though it would reduce the chances of it + * happening in practice. The only real fix here seems to be + * to have some kind of interlock that allows us to wait until + * we can be certain that no write to the block is in + * progress. Since we don't have any such thing right now, we + * just do this and hope for the best. */ - if (!PageIsNew(page) && PageGetLSN(page) < sink->bbs_state->startptr) + reread_cnt = + basebackup_read_file(fd, + sink->bbs_buffer + BLCKSZ * i, + BLCKSZ, len + BLCKSZ * i, + readfilename, + false); + if (reread_cnt == 0) { - checksum = pg_checksum_page((char *) page, blkno + segmentno * RELSEG_SIZE); - phdr = (PageHeader) page; - if (phdr->pd_checksum != checksum) - { - /* - * Retry the block on the first failure. It's - * possible that we read the first 4K page of the - * block just before postgres updated the entire block - * so it ends up looking torn to us. If, before we - * retry the read, the concurrent write of the block - * finishes, the page LSN will be updated and we'll - * realize that we should ignore this block. - * - * There's no guarantee that this will actually - * happen, though: the torn write could take an - * arbitrarily long time to complete. Retrying - * multiple times wouldn't fix this problem, either, - * though it would reduce the chances of it happening - * in practice. The only real fix here seems to be to - * have some kind of interlock that allows us to wait - * until we can be certain that no write to the block - * is in progress. Since we don't have any such thing - * right now, we just do this and hope for the best. - */ - if (block_retry == false) - { - int reread_cnt; - - /* Reread the failed block */ - reread_cnt = - basebackup_read_file(fd, - sink->bbs_buffer + BLCKSZ * i, - BLCKSZ, len + BLCKSZ * i, - readfilename, - false); - if (reread_cnt == 0) - { - /* - * If we hit end-of-file, a concurrent - * truncation must have occurred, so break out - * of this loop just as if the initial fread() - * returned 0. We'll drop through to the same - * code that handles that case. (We must fix - * up cnt first, though.) - */ - cnt = BLCKSZ * i; - break; - } - - /* Set flag so we know a retry was attempted */ - block_retry = true; - - /* Reset loop to validate the block again */ - i--; - continue; - } - - checksum_failures++; - - if (checksum_failures <= 5) - ereport(WARNING, - (errmsg("checksum verification failed in " - "file \"%s\", block %u: calculated " - "%X but expected %X", - readfilename, blkno, checksum, - phdr->pd_checksum))); - if (checksum_failures == 5) - ereport(WARNING, - (errmsg("further checksum verification " - "failures in file \"%s\" will not " - "be reported", readfilename))); - } + /* + * If we hit end-of-file, a concurrent truncation must + * have occurred, so break out of this loop just as if the + * initial fread() returned 0. We'll drop through to the + * same code that handles that case. (We must fix up cnt + * first, though.) + */ + cnt = BLCKSZ * i; + break; } - block_retry = false; - blkno++; + + /* If the page now looks OK, go on to the next one. */ + if (verify_page_checksum(page, sink->bbs_state->startptr, + blkno + i + segmentno * RELSEG_SIZE, + &expected_checksum)) + continue; + + /* Handle checksum failure. */ + checksum_failures++; + if (checksum_failures <= 5) + ereport(WARNING, + (errmsg("checksum verification failed in " + "file \"%s\", block %u: calculated " + "%X but expected %X", + readfilename, blkno + i, expected_checksum, + ((PageHeader) page)->pd_checksum))); + if (checksum_failures == 5) + ereport(WARNING, + (errmsg("further checksum verification " + "failures in file \"%s\" will not " + "be reported", readfilename))); } + + /* Update block number for next pass through the outer loop. */ + blkno += i; } /* @@ -1734,6 +1718,42 @@ sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, return true; } +/* + * Try to verify the checksum for the provided page, if it seems appropriate + * to do so. + * + * Returns true if verification succeeds or if we decide not to check it, + * and false if verification fails. When return false, it also sets + * *expected_checksum to the computed value. + */ +static bool +verify_page_checksum(Page page, XLogRecPtr start_lsn, BlockNumber blkno, + uint16 *expected_checksum) +{ + PageHeader phdr; + uint16 checksum; + + /* + * Only check pages which have not been modified since the start of the + * base backup. Otherwise, they might have been written only halfway and + * the checksum would not be valid. However, replaying WAL would + * reinstate the correct page in this case. We also skip completely new + * pages, since they don't have a checksum yet. + */ + if (PageIsNew(page) || PageGetLSN(page) >= start_lsn) + return true; + + /* Perform the actual checksum calculation. */ + checksum = pg_checksum_page(page, blkno); + + /* See whether it matches the value from the page. */ + phdr = (PageHeader) page; + if (phdr->pd_checksum == checksum) + return true; + *expected_checksum = checksum; + return false; +} + static int64 _tarWriteHeader(bbsink *sink, const char *filename, const char *linktarget, struct stat *statbuf, bool sizeonly) -- 2.37.1 (Apple Git-137.1) [application/octet-stream] 0005-Change-how-a-base-backup-decides-which-files-have-ch.patch (11.4K, ../../CA+TgmoZoHNqx=zconLLbV+8=gNhv7g5ocXd-Mee97wvTkKc-6A@mail.gmail.com/5-0005-Change-how-a-base-backup-decides-which-files-have-ch.patch) download | inline diff: From 23397800976a006b4280617003eec4413898f955 Mon Sep 17 00:00:00 2001 From: Robert Haas <[email protected]> Date: Mon, 5 Jun 2023 15:42:19 -0400 Subject: [PATCH 5/9] Change how a base backup decides which files have checksums. Previously, it thought that any plain file located under global, base, or a tablespace directory had checksums unless it was in a short list of excluded files. Now, it thinks that files in those directories have checksums if parse_filename_for_nontemp_relation says that they are relation files. (Temporary relation files don't matter because they're excluded from the backup anyway.) This changes the behavior if you have stray files not managed by PostgreSQL in the relevant directories. Previously, you'd get some kind of checksum-related complaint if such files existed, assuming that the cluster had checksums enabled and that the base backup wasn't run with NOVERIFY_CHECKSUMS. Now, you won't get those complaints any more. That seems like an improvement to me, because those files were presumably not created by PostgreSQL and so there is no reason to think that they would be checksummed like a PostgreSQL relation file. (If we want to complain about such files, we should complain about them existing at all, not just about their checksums.) The point of this change is to make the code more consistent. sendDir() was already calling parse_filename_for_nontemp_relation() as part of an effort to determine which files to include in the backup. So, it already had the information about whether a certain file was a relation file. sendFile() then used a separate method, embodied in is_checksummed_file(), to make what is essentially the same determination. It's better not to make the same decision using two different methods, especially in closely-related code. --- src/backend/backup/basebackup.c | 173 +++++++++++--------------------- 1 file changed, 56 insertions(+), 117 deletions(-) diff --git a/src/backend/backup/basebackup.c b/src/backend/backup/basebackup.c index 24c038dfba..64ab54fe06 100644 --- a/src/backend/backup/basebackup.c +++ b/src/backend/backup/basebackup.c @@ -82,7 +82,8 @@ static int64 sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeo backup_manifest_info *manifest, Oid spcoid); static bool sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, struct stat *statbuf, bool missing_ok, - Oid dboid, Oid spcoid, + Oid dboid, Oid spcoid, RelFileNumber relfilenumber, + unsigned segno, backup_manifest_info *manifest); static off_t read_file_data_into_buffer(bbsink *sink, const char *readfilename, int fd, @@ -104,7 +105,6 @@ static void convert_link_to_directory(const char *pathbuf, struct stat *statbuf) static void perform_base_backup(basebackup_options *opt, bbsink *sink); static void parse_basebackup_options(List *options, basebackup_options *opt); static int compareWalFileNames(const ListCell *a, const ListCell *b); -static bool is_checksummed_file(const char *fullpath, const char *filename); static int basebackup_read_file(int fd, char *buf, size_t nbytes, off_t offset, const char *filename, bool partial_read_ok); @@ -213,23 +213,6 @@ static const struct exclude_list_item excludeFiles[] = {NULL, false} }; -/* - * List of files excluded from checksum validation. - * - * Note: this list should be kept in sync with what pg_checksums.c - * includes. - */ -static const struct exclude_list_item noChecksumFiles[] = { - {"pg_control", false}, - {"pg_filenode.map", false}, - {"pg_internal.init", true}, - {"PG_VERSION", false}, -#ifdef EXEC_BACKEND - {"config_exec_params", true}, -#endif - {NULL, false} -}; - /* * Actually do a base backup for the specified tablespaces. * @@ -356,7 +339,8 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) errmsg("could not stat file \"%s\": %m", XLOG_CONTROL_FILE))); sendFile(sink, XLOG_CONTROL_FILE, XLOG_CONTROL_FILE, &statbuf, - false, InvalidOid, InvalidOid, &manifest); + false, InvalidOid, InvalidOid, + InvalidRelFileNumber, 0, &manifest); } else { @@ -625,7 +609,8 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) errmsg("could not stat file \"%s\": %m", pathbuf))); sendFile(sink, pathbuf, pathbuf, &statbuf, false, - InvalidOid, InvalidOid, &manifest); + InvalidOid, InvalidOid, InvalidRelFileNumber, 0, + &manifest); /* unconditionally mark file as archived */ StatusFilePath(pathbuf, fname, ".done"); @@ -1163,7 +1148,8 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, struct stat statbuf; int64 size = 0; const char *lastDir; /* Split last dir from parent path. */ - bool isDbDir = false; /* Does this directory contain relations? */ + bool isRelationDir = false; /* Does directory contain relations? */ + Oid dboid = InvalidOid; /* * Determine if the current path is a database directory that can contain @@ -1190,17 +1176,23 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, strncmp(lastDir - (sizeof(TABLESPACE_VERSION_DIRECTORY) - 1), TABLESPACE_VERSION_DIRECTORY, sizeof(TABLESPACE_VERSION_DIRECTORY) - 1) == 0)) - isDbDir = true; + { + isRelationDir = true; + dboid = atooid(lastDir + 1); + } } + else if (strcmp(path, "./global") == 0) + isRelationDir = true; dir = AllocateDir(path); while ((de = ReadDir(dir, path)) != NULL) { int excludeIdx; bool excludeFound; - RelFileNumber relNumber; - ForkNumber relForkNum; - unsigned segno; + RelFileNumber relfilenumber = InvalidRelFileNumber; + ForkNumber relForkNum = InvalidForkNumber; + unsigned segno = 0; + bool isRelationFile = false; /* Skip special stuff */ if (strcmp(de->d_name, ".") == 0 || strcmp(de->d_name, "..") == 0) @@ -1248,37 +1240,41 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, if (excludeFound) continue; + /* + * If there could be non-temporary relation files in this directory, + * try to parse the filename. + */ + if (isRelationDir) + isRelationFile = + parse_filename_for_nontemp_relation(de->d_name, + &relfilenumber, + &relForkNum, &segno); + /* Exclude all forks for unlogged tables except the init fork */ - if (isDbDir && - parse_filename_for_nontemp_relation(de->d_name, &relNumber, - &relForkNum, &segno)) + if (isRelationFile && relForkNum != INIT_FORKNUM) { - /* Never exclude init forks */ - if (relForkNum != INIT_FORKNUM) - { - char initForkFile[MAXPGPATH]; + char initForkFile[MAXPGPATH]; - /* - * If any other type of fork, check if there is an init fork - * with the same RelFileNumber. If so, the file can be - * excluded. - */ - snprintf(initForkFile, sizeof(initForkFile), "%s/%u_init", - path, relNumber); + /* + * If any other type of fork, check if there is an init fork + * with the same RelFileNumber. If so, the file can be + * excluded. + */ + snprintf(initForkFile, sizeof(initForkFile), "%s/%u_init", + path, relfilenumber); - if (lstat(initForkFile, &statbuf) == 0) - { - elog(DEBUG2, - "unlogged relation file \"%s\" excluded from backup", - de->d_name); + if (lstat(initForkFile, &statbuf) == 0) + { + elog(DEBUG2, + "unlogged relation file \"%s\" excluded from backup", + de->d_name); - continue; - } + continue; } } /* Exclude temporary relations */ - if (isDbDir && looks_like_temp_rel_name(de->d_name)) + if (OidIsValid(dboid) && looks_like_temp_rel_name(de->d_name)) { elog(DEBUG2, "temporary relation file \"%s\" excluded from backup", @@ -1417,8 +1413,8 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, if (!sizeonly) sent = sendFile(sink, pathbuf, pathbuf + basepathlen + 1, &statbuf, - true, isDbDir ? atooid(lastDir + 1) : InvalidOid, spcoid, - manifest); + true, dboid, spcoid, + relfilenumber, segno, manifest); if (sent || sizeonly) { @@ -1440,40 +1436,6 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, return size; } -/* - * Check if a file should have its checksum validated. - * We validate checksums on files in regular tablespaces - * (including global and default) only, and in those there - * are some files that are explicitly excluded. - */ -static bool -is_checksummed_file(const char *fullpath, const char *filename) -{ - /* Check that the file is in a tablespace */ - if (strncmp(fullpath, "./global/", 9) == 0 || - strncmp(fullpath, "./base/", 7) == 0 || - strncmp(fullpath, "/", 1) == 0) - { - int excludeIdx; - - /* Compare file against noChecksumFiles skip list */ - for (excludeIdx = 0; noChecksumFiles[excludeIdx].name != NULL; excludeIdx++) - { - int cmplen = strlen(noChecksumFiles[excludeIdx].name); - - if (!noChecksumFiles[excludeIdx].match_prefix) - cmplen++; - if (strncmp(filename, noChecksumFiles[excludeIdx].name, - cmplen) == 0) - return false; - } - - return true; - } - else - return false; -} - /* * Given the member, write the TAR header & send the file. * @@ -1488,6 +1450,7 @@ is_checksummed_file(const char *fullpath, const char *filename) static bool sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, struct stat *statbuf, bool missing_ok, Oid dboid, Oid spcoid, + RelFileNumber relfilenumber, unsigned segno, backup_manifest_info *manifest) { int fd; @@ -1495,8 +1458,6 @@ sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, int checksum_failures = 0; off_t cnt; pgoff_t bytes_done = 0; - int segmentno = 0; - char *segmentpath; bool verify_checksum = false; pg_checksum_context checksum_ctx; @@ -1522,36 +1483,14 @@ sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, */ Assert((sink->bbs_buffer_length % BLCKSZ) == 0); - if (!noverify_checksums && DataChecksumsEnabled()) - { - char *filename; - - /* - * Get the filename (excluding path). As last_dir_separator() - * includes the last directory separator, we chop that off by - * incrementing the pointer. - */ - filename = last_dir_separator(readfilename) + 1; - - if (is_checksummed_file(readfilename, filename)) - { - verify_checksum = true; - - /* - * Cut off at the segment boundary (".") to get the segment number - * in order to mix it into the checksum. - */ - segmentpath = strstr(filename, "."); - if (segmentpath != NULL) - { - segmentno = atoi(segmentpath + 1); - if (segmentno == 0) - ereport(ERROR, - (errmsg("invalid segment number %d in file \"%s\"", - segmentno, filename))); - } - } - } + /* + * If we weren't told not to verify checksums, and if checksums are + * enabled for this cluster, and if this is a relation file, then verify + * the checksum. + */ + if (!noverify_checksums && DataChecksumsEnabled() && + RelFileNumberIsValid(relfilenumber)) + verify_checksum = true; /* * Loop until we read the amount of data the caller told us to expect. The @@ -1566,7 +1505,7 @@ sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, /* Try to read some more data. */ cnt = read_file_data_into_buffer(sink, readfilename, fd, bytes_done, remaining, - blkno + segmentno * RELSEG_SIZE, + blkno + segno * RELSEG_SIZE, verify_checksum, &checksum_failures); -- 2.37.1 (Apple Git-137.1) [application/octet-stream] 0003-Change-struct-tablespaceinfo-s-oid-member-from-char-.patch (11.7K, ../../CA+TgmoZoHNqx=zconLLbV+8=gNhv7g5ocXd-Mee97wvTkKc-6A@mail.gmail.com/6-0003-Change-struct-tablespaceinfo-s-oid-member-from-char-.patch) download | inline diff: From de555bebd21507637a738c20d2933e05662404c4 Mon Sep 17 00:00:00 2001 From: Robert Haas <[email protected]> Date: Mon, 5 Jun 2023 15:42:00 -0400 Subject: [PATCH 3/9] Change struct tablespaceinfo's oid member from 'char *' to 'Oid' This shouldn't change behavior except in the unusual case where there are file in the tablespace directory that have entirely numeric names but are nevertheless not possible names for a tablespace directory, either because their names has leading zeroes that shouldn't be there, or the value is actually zero, or because the value is too large to represent as an OID. In those cases, the directory would previously have made it into the list of tablespaceinfo objects and no longer will. Thus, base backups will now ignore such directories, instead of treating them as legitimate tablespace directories. Similarly, if entries for such tablespaces occur in a tablespace_map file, they will now be rejected as erroneous, instead of being honored. This is infrastructure for future work that wants to be able to know the tablespace of each relation that is part of a backup *as an OID*. By strengthening the up-front validation, we don't have to worry about weird cases later, and can more easily avoid repeated string->integer conversions. --- src/backend/access/transam/xlog.c | 19 ++++++++++-- src/backend/access/transam/xlogrecovery.c | 12 ++++++-- src/backend/backup/backup_manifest.c | 6 ++-- src/backend/backup/basebackup.c | 35 ++++++++++++----------- src/backend/backup/basebackup_copy.c | 2 +- src/include/backup/backup_manifest.h | 2 +- src/include/backup/basebackup.h | 2 +- 7 files changed, 49 insertions(+), 29 deletions(-) diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f6f8adc72a..6f38d0eb9a 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -8484,9 +8484,22 @@ do_pg_backup_start(const char *backupidstr, bool fast, List **tablespaces, char *relpath = NULL; char *s; PGFileType de_type; + char *badp; + Oid tsoid; - /* Skip anything that doesn't look like a tablespace */ - if (strspn(de->d_name, "0123456789") != strlen(de->d_name)) + /* + * Try to parse the directory name as an unsigned integer. + * + * Tablespace directories should be positive integers that can + * be represented in 32 bits, with no leading zeroes or trailing + * garbage. If we come across a name that doesn't meet those + * criteria, skip it. + */ + if (de->d_name[0] < '1' || de->d_name[1] > '9') + continue; + errno = 0; + tsoid = strtoul(de->d_name, &badp, 10); + if (*badp != '\0' || errno == EINVAL || errno == ERANGE) continue; snprintf(fullpath, sizeof(fullpath), "pg_tblspc/%s", de->d_name); @@ -8561,7 +8574,7 @@ do_pg_backup_start(const char *backupidstr, bool fast, List **tablespaces, } ti = palloc(sizeof(tablespaceinfo)); - ti->oid = pstrdup(de->d_name); + ti->oid = tsoid; ti->path = pstrdup(linkpath); ti->rpath = relpath; ti->size = -1; diff --git a/src/backend/access/transam/xlogrecovery.c b/src/backend/access/transam/xlogrecovery.c index becc2bda62..4ff4430006 100644 --- a/src/backend/access/transam/xlogrecovery.c +++ b/src/backend/access/transam/xlogrecovery.c @@ -678,7 +678,7 @@ InitWalRecovery(ControlFileData *ControlFile, bool *wasShutdown_ptr, tablespaceinfo *ti = lfirst(lc); char *linkloc; - linkloc = psprintf("pg_tblspc/%s", ti->oid); + linkloc = psprintf("pg_tblspc/%u", ti->oid); /* * Remove the existing symlink if any and Create the symlink @@ -692,7 +692,6 @@ InitWalRecovery(ControlFileData *ControlFile, bool *wasShutdown_ptr, errmsg("could not create symbolic link \"%s\": %m", linkloc))); - pfree(ti->oid); pfree(ti->path); pfree(ti); } @@ -1341,6 +1340,8 @@ read_tablespace_map(List **tablespaces) { if (!was_backslash && (ch == '\n' || ch == '\r')) { + char *endp; + if (i == 0) continue; /* \r immediately followed by \n */ @@ -1360,7 +1361,12 @@ read_tablespace_map(List **tablespaces) str[n++] = '\0'; ti = palloc0(sizeof(tablespaceinfo)); - ti->oid = pstrdup(str); + errno = 0; + ti->oid = strtoul(str, &endp, 10); + if (*endp != '\0' || errno == EINVAL || errno == ERANGE) + ereport(FATAL, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("invalid data in file \"%s\"", TABLESPACE_MAP))); ti->path = pstrdup(str + n); *tablespaces = lappend(*tablespaces, ti); diff --git a/src/backend/backup/backup_manifest.c b/src/backend/backup/backup_manifest.c index cee6216524..aeed362a9a 100644 --- a/src/backend/backup/backup_manifest.c +++ b/src/backend/backup/backup_manifest.c @@ -97,7 +97,7 @@ FreeBackupManifest(backup_manifest_info *manifest) * Add an entry to the backup manifest for a file. */ void -AddFileToBackupManifest(backup_manifest_info *manifest, const char *spcoid, +AddFileToBackupManifest(backup_manifest_info *manifest, Oid spcoid, const char *pathname, size_t size, pg_time_t mtime, pg_checksum_context *checksum_ctx) { @@ -114,9 +114,9 @@ AddFileToBackupManifest(backup_manifest_info *manifest, const char *spcoid, * pathname relative to the data directory (ignoring the intermediate * symlink traversal). */ - if (spcoid != NULL) + if (OidIsValid(spcoid)) { - snprintf(pathbuf, sizeof(pathbuf), "pg_tblspc/%s/%s", spcoid, + snprintf(pathbuf, sizeof(pathbuf), "pg_tblspc/%u/%s", spcoid, pathname); pathname = pathbuf; } diff --git a/src/backend/backup/basebackup.c b/src/backend/backup/basebackup.c index f46f930329..cc3d2e0c41 100644 --- a/src/backend/backup/basebackup.c +++ b/src/backend/backup/basebackup.c @@ -75,14 +75,15 @@ typedef struct pg_checksum_type manifest_checksum_type; } basebackup_options; -static int64 sendTablespace(bbsink *sink, char *path, char *spcoid, bool sizeonly, +static int64 sendTablespace(bbsink *sink, char *path, Oid spcoid, bool sizeonly, struct backup_manifest_info *manifest); static int64 sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, List *tablespaces, bool sendtblspclinks, - backup_manifest_info *manifest, const char *spcoid); + backup_manifest_info *manifest, Oid spcoid); static bool sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, - struct stat *statbuf, bool missing_ok, Oid dboid, - backup_manifest_info *manifest, const char *spcoid); + struct stat *statbuf, bool missing_ok, + Oid dboid, Oid spcoid, + backup_manifest_info *manifest); static off_t read_file_data_into_buffer(bbsink *sink, const char *readfilename, int fd, off_t offset, size_t length, @@ -305,7 +306,7 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) if (tmp->path == NULL) tmp->size = sendDir(sink, ".", 1, true, state.tablespaces, - true, NULL, NULL); + true, NULL, InvalidOid); else tmp->size = sendTablespace(sink, tmp->path, tmp->oid, true, NULL); @@ -346,7 +347,7 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) /* Then the bulk of the files... */ sendDir(sink, ".", 1, false, state.tablespaces, - sendtblspclinks, &manifest, NULL); + sendtblspclinks, &manifest, InvalidOid); /* ... and pg_control after everything else. */ if (lstat(XLOG_CONTROL_FILE, &statbuf) != 0) @@ -355,11 +356,11 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) errmsg("could not stat file \"%s\": %m", XLOG_CONTROL_FILE))); sendFile(sink, XLOG_CONTROL_FILE, XLOG_CONTROL_FILE, &statbuf, - false, InvalidOid, &manifest, NULL); + false, InvalidOid, InvalidOid, &manifest); } else { - char *archive_name = psprintf("%s.tar", ti->oid); + char *archive_name = psprintf("%u.tar", ti->oid); bbsink_begin_archive(sink, archive_name); @@ -623,8 +624,8 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) (errcode_for_file_access(), errmsg("could not stat file \"%s\": %m", pathbuf))); - sendFile(sink, pathbuf, pathbuf, &statbuf, false, InvalidOid, - &manifest, NULL); + sendFile(sink, pathbuf, pathbuf, &statbuf, false, + InvalidOid, InvalidOid, &manifest); /* unconditionally mark file as archived */ StatusFilePath(pathbuf, fname, ".done"); @@ -1087,7 +1088,7 @@ sendFileWithContent(bbsink *sink, const char *filename, const char *content, _tarWritePadding(sink, len); - AddFileToBackupManifest(manifest, NULL, filename, len, + AddFileToBackupManifest(manifest, InvalidOid, filename, len, (pg_time_t) statbuf.st_mtime, &checksum_ctx); } @@ -1099,7 +1100,7 @@ sendFileWithContent(bbsink *sink, const char *filename, const char *content, * Only used to send auxiliary tablespaces, not PGDATA. */ static int64 -sendTablespace(bbsink *sink, char *path, char *spcoid, bool sizeonly, +sendTablespace(bbsink *sink, char *path, Oid spcoid, bool sizeonly, backup_manifest_info *manifest) { int64 size; @@ -1154,7 +1155,7 @@ sendTablespace(bbsink *sink, char *path, char *spcoid, bool sizeonly, static int64 sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, List *tablespaces, bool sendtblspclinks, backup_manifest_info *manifest, - const char *spcoid) + Oid spcoid) { DIR *dir; struct dirent *de; @@ -1419,8 +1420,8 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, if (!sizeonly) sent = sendFile(sink, pathbuf, pathbuf + basepathlen + 1, &statbuf, - true, isDbDir ? atooid(lastDir + 1) : InvalidOid, - manifest, spcoid); + true, isDbDir ? atooid(lastDir + 1) : InvalidOid, spcoid, + manifest); if (sent || sizeonly) { @@ -1489,8 +1490,8 @@ is_checksummed_file(const char *fullpath, const char *filename) */ static bool sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, - struct stat *statbuf, bool missing_ok, Oid dboid, - backup_manifest_info *manifest, const char *spcoid) + struct stat *statbuf, bool missing_ok, Oid dboid, Oid spcoid, + backup_manifest_info *manifest) { int fd; BlockNumber blkno = 0; diff --git a/src/backend/backup/basebackup_copy.c b/src/backend/backup/basebackup_copy.c index fee30c21e1..3bdbe1f989 100644 --- a/src/backend/backup/basebackup_copy.c +++ b/src/backend/backup/basebackup_copy.c @@ -407,7 +407,7 @@ SendTablespaceList(List *tablespaces) } else { - values[0] = ObjectIdGetDatum(strtoul(ti->oid, NULL, 10)); + values[0] = ObjectIdGetDatum(ti->oid); values[1] = CStringGetTextDatum(ti->path); } if (ti->size >= 0) diff --git a/src/include/backup/backup_manifest.h b/src/include/backup/backup_manifest.h index d41b439980..5a481dbcf5 100644 --- a/src/include/backup/backup_manifest.h +++ b/src/include/backup/backup_manifest.h @@ -39,7 +39,7 @@ extern void InitializeBackupManifest(backup_manifest_info *manifest, backup_manifest_option want_manifest, pg_checksum_type manifest_checksum_type); extern void AddFileToBackupManifest(backup_manifest_info *manifest, - const char *spcoid, + Oid spcoid, const char *pathname, size_t size, pg_time_t mtime, pg_checksum_context *checksum_ctx); diff --git a/src/include/backup/basebackup.h b/src/include/backup/basebackup.h index 3e68abc2bb..1432d9c206 100644 --- a/src/include/backup/basebackup.h +++ b/src/include/backup/basebackup.h @@ -27,7 +27,7 @@ */ typedef struct { - char *oid; /* tablespace's OID, as a decimal string */ + Oid oid; /* tablespace's OID */ char *path; /* full path to tablespace's directory */ char *rpath; /* relative path if it's within PGDATA, else * NULL */ -- 2.37.1 (Apple Git-137.1) [application/octet-stream] 0006-Move-src-bin-pg_verifybackup-parse_manifest.c-into-s.patch (4.3K, ../../CA+TgmoZoHNqx=zconLLbV+8=gNhv7g5ocXd-Mee97wvTkKc-6A@mail.gmail.com/7-0006-Move-src-bin-pg_verifybackup-parse_manifest.c-into-s.patch) download | inline diff: From d592bff5496ba8e11c2846532290a7b72e2e3b80 Mon Sep 17 00:00:00 2001 From: Robert Haas <[email protected]> Date: Mon, 5 Jun 2023 15:42:28 -0400 Subject: [PATCH 6/9] Move src/bin/pg_verifybackup/parse_manifest.c into src/common. This makes it possible for the code to be easily reused by other client-side tools, and/or by the server. --- src/bin/pg_verifybackup/Makefile | 1 - src/bin/pg_verifybackup/meson.build | 1 - src/bin/pg_verifybackup/pg_verifybackup.c | 2 +- src/common/Makefile | 1 + src/common/meson.build | 1 + src/{bin/pg_verifybackup => common}/parse_manifest.c | 4 ++-- src/{bin/pg_verifybackup => include/common}/parse_manifest.h | 2 +- 7 files changed, 6 insertions(+), 6 deletions(-) rename src/{bin/pg_verifybackup => common}/parse_manifest.c (99%) rename src/{bin/pg_verifybackup => include/common}/parse_manifest.h (97%) diff --git a/src/bin/pg_verifybackup/Makefile b/src/bin/pg_verifybackup/Makefile index 596df15118..8f04fa662c 100644 --- a/src/bin/pg_verifybackup/Makefile +++ b/src/bin/pg_verifybackup/Makefile @@ -21,7 +21,6 @@ LDFLAGS_INTERNAL += -L$(top_builddir)/src/fe_utils -lpgfeutils $(libpq_pgport) OBJS = \ $(WIN32RES) \ - parse_manifest.o \ pg_verifybackup.o all: pg_verifybackup diff --git a/src/bin/pg_verifybackup/meson.build b/src/bin/pg_verifybackup/meson.build index 9369da1bc6..58f780d1a6 100644 --- a/src/bin/pg_verifybackup/meson.build +++ b/src/bin/pg_verifybackup/meson.build @@ -1,7 +1,6 @@ # Copyright (c) 2022-2023, PostgreSQL Global Development Group pg_verifybackup_sources = files( - 'parse_manifest.c', 'pg_verifybackup.c' ) diff --git a/src/bin/pg_verifybackup/pg_verifybackup.c b/src/bin/pg_verifybackup/pg_verifybackup.c index 059836f0e6..ce423a03d4 100644 --- a/src/bin/pg_verifybackup/pg_verifybackup.c +++ b/src/bin/pg_verifybackup/pg_verifybackup.c @@ -20,9 +20,9 @@ #include "common/hashfn.h" #include "common/logging.h" +#include "common/parse_manifest.h" #include "fe_utils/simple_list.h" #include "getopt_long.h" -#include "parse_manifest.h" #include "pgtime.h" /* diff --git a/src/common/Makefile b/src/common/Makefile index 113029bf7b..e4cd26762b 100644 --- a/src/common/Makefile +++ b/src/common/Makefile @@ -65,6 +65,7 @@ OBJS_COMMON = \ kwlookup.o \ link-canary.o \ md5_common.o \ + parse_manifest.o \ percentrepl.o \ pg_get_line.o \ pg_lzcompress.o \ diff --git a/src/common/meson.build b/src/common/meson.build index 53942a9a61..a9ff7f9db8 100644 --- a/src/common/meson.build +++ b/src/common/meson.build @@ -17,6 +17,7 @@ common_sources = files( 'kwlookup.c', 'link-canary.c', 'md5_common.c', + 'parse_manifest.c', 'percentrepl.c', 'pg_get_line.c', 'pg_lzcompress.c', diff --git a/src/bin/pg_verifybackup/parse_manifest.c b/src/common/parse_manifest.c similarity index 99% rename from src/bin/pg_verifybackup/parse_manifest.c rename to src/common/parse_manifest.c index 2379f7be7b..672e8bcf25 100644 --- a/src/bin/pg_verifybackup/parse_manifest.c +++ b/src/common/parse_manifest.c @@ -6,15 +6,15 @@ * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group * Portions Copyright (c) 1994, Regents of the University of California * - * src/bin/pg_verifybackup/parse_manifest.c + * src/common/parse_manifest.c * *------------------------------------------------------------------------- */ #include "postgres_fe.h" -#include "parse_manifest.h" #include "common/jsonapi.h" +#include "common/parse_manifest.h" /* * Semantic states for JSON manifest parsing. diff --git a/src/bin/pg_verifybackup/parse_manifest.h b/src/include/common/parse_manifest.h similarity index 97% rename from src/bin/pg_verifybackup/parse_manifest.h rename to src/include/common/parse_manifest.h index 7387a917a2..7b24c5d785 100644 --- a/src/bin/pg_verifybackup/parse_manifest.h +++ b/src/include/common/parse_manifest.h @@ -6,7 +6,7 @@ * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group * Portions Copyright (c) 1994, Regents of the University of California * - * src/bin/pg_verifybackup/parse_manifest.h + * src/include/common/parse_manifest.h * *------------------------------------------------------------------------- */ -- 2.37.1 (Apple Git-137.1) [application/octet-stream] 0008-Add-new-pg_walsummary-tool.patch (11.4K, ../../CA+TgmoZoHNqx=zconLLbV+8=gNhv7g5ocXd-Mee97wvTkKc-6A@mail.gmail.com/8-0008-Add-new-pg_walsummary-tool.patch) download | inline diff: From d25598e3129566556eda0a161ef0763d115b6f25 Mon Sep 17 00:00:00 2001 From: Robert Haas <[email protected]> Date: Wed, 14 Jun 2023 11:31:39 -0400 Subject: [PATCH 8/9] Add new pg_walsummary tool. This can dump the contents of WAL summary files, either those in pg_wal/summaries, or the INCREMENTAL_BACKUP files that are part of an incremental backup proper. XXX. Needs documentation and tests. --- src/bin/Makefile | 1 + src/bin/meson.build | 1 + src/bin/pg_walsummary/.gitignore | 1 + src/bin/pg_walsummary/Makefile | 42 ++++ src/bin/pg_walsummary/meson.build | 24 +++ src/bin/pg_walsummary/pg_walsummary.c | 278 ++++++++++++++++++++++++++ 6 files changed, 347 insertions(+) create mode 100644 src/bin/pg_walsummary/.gitignore create mode 100644 src/bin/pg_walsummary/Makefile create mode 100644 src/bin/pg_walsummary/meson.build create mode 100644 src/bin/pg_walsummary/pg_walsummary.c diff --git a/src/bin/Makefile b/src/bin/Makefile index aa2210925e..f98f58d39e 100644 --- a/src/bin/Makefile +++ b/src/bin/Makefile @@ -31,6 +31,7 @@ SUBDIRS = \ pg_upgrade \ pg_verifybackup \ pg_waldump \ + pg_walsummary \ pgbench \ psql \ scripts diff --git a/src/bin/meson.build b/src/bin/meson.build index 4cb6fd59bb..d1e9ef4409 100644 --- a/src/bin/meson.build +++ b/src/bin/meson.build @@ -17,6 +17,7 @@ subdir('pg_test_timing') subdir('pg_upgrade') subdir('pg_verifybackup') subdir('pg_waldump') +subdir('pg_walsummary') subdir('pgbench') subdir('pgevent') subdir('psql') diff --git a/src/bin/pg_walsummary/.gitignore b/src/bin/pg_walsummary/.gitignore new file mode 100644 index 0000000000..d71ec192fa --- /dev/null +++ b/src/bin/pg_walsummary/.gitignore @@ -0,0 +1 @@ +pg_walsummary diff --git a/src/bin/pg_walsummary/Makefile b/src/bin/pg_walsummary/Makefile new file mode 100644 index 0000000000..852f7208f6 --- /dev/null +++ b/src/bin/pg_walsummary/Makefile @@ -0,0 +1,42 @@ +#------------------------------------------------------------------------- +# +# Makefile for src/bin/pg_walsummary +# +# Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group +# Portions Copyright (c) 1994, Regents of the University of California +# +# src/bin/pg_walsummary/Makefile +# +#------------------------------------------------------------------------- + +PGFILEDESC = "pg_walsummary - print contents of WAL summary files" +PGAPPICON=win32 + +subdir = src/bin/pg_walsummary +top_builddir = ../../.. +include $(top_builddir)/src/Makefile.global + +override CPPFLAGS := -I$(libpq_srcdir) $(CPPFLAGS) +LDFLAGS_INTERNAL += -L$(top_builddir)/src/fe_utils -lpgfeutils + +OBJS = \ + $(WIN32RES) \ + pg_walsummary.o + +all: pg_walsummary + +pg_walsummary: $(OBJS) | submake-libpgport submake-libpgfeutils + $(CC) $(CFLAGS) $^ $(LDFLAGS) $(LDFLAGS_EX) $(LIBS) -o $@$(X) + + +install: all installdirs + $(INSTALL_PROGRAM) pg_walsummary$(X) '$(DESTDIR)$(bindir)/pg_walsummary$(X)' + +installdirs: + $(MKDIR_P) '$(DESTDIR)$(bindir)' + +uninstall: + rm -f '$(DESTDIR)$(bindir)/pg_walsummary$(X)' + +clean distclean maintainer-clean: + rm -f pg_walsummary$(X) $(OBJS) diff --git a/src/bin/pg_walsummary/meson.build b/src/bin/pg_walsummary/meson.build new file mode 100644 index 0000000000..c2092960c6 --- /dev/null +++ b/src/bin/pg_walsummary/meson.build @@ -0,0 +1,24 @@ +# Copyright (c) 2022-2023, PostgreSQL Global Development Group + +pg_walsummary_sources = files( + 'pg_walsummary.c', +) + +if host_system == 'windows' + pg_walsummary_sources += rc_bin_gen.process(win32ver_rc, extra_args: [ + '--NAME', 'pg_walsummary', + '--FILEDESC', 'pg_walsummary - print contents of WAL summary files',]) +endif + +pg_walsummary = executable('pg_walsummary', + pg_walsummary_sources, + dependencies: [frontend_code], + kwargs: default_bin_args, +) +bin_targets += pg_walsummary + +tests += { + 'name': 'pg_walsummary', + 'sd': meson.current_source_dir(), + 'bd': meson.current_build_dir() +} diff --git a/src/bin/pg_walsummary/pg_walsummary.c b/src/bin/pg_walsummary/pg_walsummary.c new file mode 100644 index 0000000000..0304a42026 --- /dev/null +++ b/src/bin/pg_walsummary/pg_walsummary.c @@ -0,0 +1,278 @@ +/*------------------------------------------------------------------------- + * + * pg_walsummary.c + * Prints the contents of WAL summary files. + * + * Copyright (c) 2017-2023, PostgreSQL Global Development Group + * + * IDENTIFICATION + * src/bin/pg_walsummary/pg_walsummary.c + * + *------------------------------------------------------------------------- + */ +#include "postgres_fe.h" + +#include <fcntl.h> +#include <limits.h> + +#include "common/blkreftable.h" +#include "common/logging.h" +#include "fe_utils/option_utils.h" +#include "lib/stringinfo.h" +#include "getopt_long.h" + +typedef struct ws_options +{ + bool individual; + bool quiet; +} ws_options; + +typedef struct ws_file_info +{ + int fd; + char *filename; +} ws_file_info; + +static BlockNumber *block_buffer = NULL; +static unsigned block_buffer_size = 512; /* Initial size. */ + +static void dump_one_relation(ws_options *opt, RelFileLocator *rlocator, + ForkNumber forknum, BlockNumber limit_block, + BlockRefTableReader *reader); +static void help(const char *progname); +static int compare_block_numbers(const void *a, const void *b); +static int walsummary_read_callback(void *callback_arg, void *data, + int length); +static void walsummary_error_callback(void *callback_arg, char *fmt,...); + +/* + * Main program. + */ +int +main(int argc, char *argv[]) +{ + static struct option long_options[] = { + {"individual", no_argument, NULL, 'i'}, + {"quiet", no_argument, NULL, 'q'}, + {NULL, 0, NULL, 0} + }; + + const char *progname; + int optindex; + int c; + ws_options opt; + + pg_logging_init(argv[0]); + progname = get_progname(argv[0]); + handle_help_version_opts(argc, argv, progname, help); + + /* process command-line options */ + while ((c = getopt_long(argc, argv, "f:iqw:", + long_options, &optindex)) != -1) + { + switch (c) + { + case 'i': + opt.individual = true; + break; + case 'q': + opt.quiet = true; + break; + default: + /* getopt_long already emitted a complaint */ + pg_log_error_hint("Try \"%s --help\" for more information.", progname); + exit(1); + } + } + + if (optind >= argc) + { + pg_log_error("%s: no input files specified", progname); + pg_log_error_hint("Try \"%s --help\" for more information.", progname); + exit(1); + } + + while (optind < argc) + { + ws_file_info ws; + BlockRefTableReader *reader; + RelFileLocator rlocator; + ForkNumber forknum; + BlockNumber limit_block; + + ws.filename = argv[optind++]; + if ((ws.fd = open(ws.filename, O_RDONLY | PG_BINARY, 0)) < 0) + pg_fatal("could not open file \"%s\": %m", ws.filename); + + reader = CreateBlockRefTableReader(walsummary_read_callback, &ws, + ws.filename, + walsummary_error_callback, NULL); + while (BlockRefTableReaderNextRelation(reader, &rlocator, &forknum, + &limit_block)) + dump_one_relation(&opt, &rlocator, forknum, limit_block, reader); + + DestroyBlockRefTableReader(reader); + close(ws.fd); + } + + exit(0); +} + +/* + * Dump details for one relation. + */ +static void +dump_one_relation(ws_options *opt, RelFileLocator *rlocator, + ForkNumber forknum, BlockNumber limit_block, + BlockRefTableReader *reader) +{ + unsigned i = 0; + unsigned nblocks; + BlockNumber startblock = InvalidBlockNumber; + BlockNumber endblock = InvalidBlockNumber; + + /* Dump limit block, if any. */ + if (limit_block != InvalidBlockNumber) + printf("TS %u, DB %u, REL %u, FORK %s: limit %u\n", + rlocator->spcOid, rlocator->dbOid, rlocator->relNumber, + forkNames[forknum], limit_block); + + /* If we haven't allocated a block buffer yet, do that now. */ + if (block_buffer == NULL) + block_buffer = palloc_array(BlockNumber, block_buffer_size); + + /* Try to fill the block buffer. */ + nblocks = BlockRefTableReaderGetBlocks(reader, + block_buffer, + block_buffer_size); + + /* If we filled the block buffer completely, we must enlarge it. */ + while (nblocks >= block_buffer_size) + { + unsigned new_size; + + /* Double the size, being careful about overflow. */ + new_size = block_buffer_size * 2; + if (new_size < block_buffer_size) + new_size = PG_UINT32_MAX; + block_buffer = repalloc_array(block_buffer, BlockNumber, new_size); + + /* Try to fill the newly-allocated space. */ + nblocks += + BlockRefTableReaderGetBlocks(reader, + block_buffer + block_buffer_size, + new_size - block_buffer_size); + + /* Save the new size for later calls. */ + block_buffer_size = new_size; + } + + /* If we don't need to produce any output, skip the rest of this. */ + if (opt->quiet) + return; + + /* + * Sort the returned block numbers. If the block reference table was using + * the bitmap representation for a given chunk, the block numbers in that + * chunk will already be sorted, but when the array-of-offsets + * representation is used, we can receive block numbers here out of order. + */ + qsort(block_buffer, nblocks, sizeof(BlockNumber), compare_block_numbers); + + /* Dump block references. */ + while (i < nblocks) + { + /* + * Find the next range of blocks to print, but if --individual was + * specified, then consider each block a separate range. + */ + startblock = endblock = block_buffer[i++]; + if (!opt->individual) + { + while (i < nblocks && block_buffer[i] == endblock + 1) + { + endblock++; + i++; + } + } + + /* Format this range of block numbers as a string. */ + if (startblock == endblock) + printf("TS %u, DB %u, REL %u, FORK %s: block %u\n", + rlocator->spcOid, rlocator->dbOid, rlocator->relNumber, + forkNames[forknum], startblock); + else + printf("TS %u, DB %u, REL %u, FORK %s: blocks %u..%u\n", + rlocator->spcOid, rlocator->dbOid, rlocator->relNumber, + forkNames[forknum], startblock, endblock); + } +} + +/* + * Quicksort comparator for block numbers. + */ +static int +compare_block_numbers(const void *a, const void *b) +{ + BlockNumber aa = *(BlockNumber *) a; + BlockNumber bb = *(BlockNumber *) b; + + if (aa > bb) + return 1; + else if (aa == bb) + return 0; + else + return -1; +} + +/* + * Error callback. + */ +void +walsummary_error_callback(void *callback_arg, char *fmt,...) +{ + va_list ap; + + va_start(ap, fmt); + pg_log_generic_v(PG_LOG_ERROR, PG_LOG_PRIMARY, fmt, ap); + va_end(ap); + + exit(1); +} + +/* + * Read callback. + */ +int +walsummary_read_callback(void *callback_arg, void *data, int length) +{ + ws_file_info *ws = callback_arg; + int rc; + + if ((rc = read(ws->fd, data, length)) < 0) + pg_fatal("could not read file \"%s\": %m", ws->filename); + + return rc; +} + +/* + * help + * + * Prints help page for the program + * + * progname: the name of the executed program, such as "pg_walsummary" + */ +static void +help(const char *progname) +{ + printf(_("%s prints the contents of a WAL summary file.\n\n"), progname); + printf(_("Usage:\n")); + printf(_(" %s [OPTION]... FILE...\n"), progname); + printf(_("\nOptions:\n")); + printf(_(" -i, --individual list block numbers individually, not as ranges\n")); + printf(_(" -q, --quiet don't print anything, just parse the files\n")); + printf(_(" -?, --help show this help, then exit\n")); + + printf(_("\nReport bugs to <%s>.\n"), PACKAGE_BUGREPORT); + printf(_("%s home page: <%s>\n"), PACKAGE_NAME, PACKAGE_URL); +} -- 2.37.1 (Apple Git-137.1) [application/octet-stream] 0009-Add-TAP-tests-this-is-broken-doesn-t-work.patch (12.0K, ../../CA+TgmoZoHNqx=zconLLbV+8=gNhv7g5ocXd-Mee97wvTkKc-6A@mail.gmail.com/9-0009-Add-TAP-tests-this-is-broken-doesn-t-work.patch) download | inline diff: From c0cce1fc702d273d667a4356e2059837fc152b44 Mon Sep 17 00:00:00 2001 From: Robert Haas <[email protected]> Date: Thu, 17 Aug 2023 12:56:01 -0400 Subject: [PATCH 9/9] Add TAP tests (this is broken, doesn't work). --- src/bin/pg_combinebackup/Makefile | 6 + src/bin/pg_combinebackup/meson.build | 8 +- src/bin/pg_combinebackup/t/001_basic.pl | 23 ++ .../pg_combinebackup/t/002_compare_backups.pl | 276 ++++++++++++++++++ src/test/perl/PostgreSQL/Test/Cluster.pm | 21 +- 5 files changed, 332 insertions(+), 2 deletions(-) create mode 100644 src/bin/pg_combinebackup/t/001_basic.pl create mode 100644 src/bin/pg_combinebackup/t/002_compare_backups.pl diff --git a/src/bin/pg_combinebackup/Makefile b/src/bin/pg_combinebackup/Makefile index cb20480aae..78ba05e624 100644 --- a/src/bin/pg_combinebackup/Makefile +++ b/src/bin/pg_combinebackup/Makefile @@ -44,3 +44,9 @@ uninstall: clean distclean maintainer-clean: rm -f pg_combinebackup$(X) $(OBJS) + +check: + $(prove_check) + +installcheck: + $(prove_installcheck) diff --git a/src/bin/pg_combinebackup/meson.build b/src/bin/pg_combinebackup/meson.build index bea0db405e..a6036dea74 100644 --- a/src/bin/pg_combinebackup/meson.build +++ b/src/bin/pg_combinebackup/meson.build @@ -25,5 +25,11 @@ bin_targets += pg_combinebackup tests += { 'name': 'pg_combinebackup', 'sd': meson.current_source_dir(), - 'bd': meson.current_build_dir() + 'bd': meson.current_build_dir(), + 'tap': { + 'tests': [ + 't/001_basic.pl', + 't/002_compare_backups.pl', + ], + } } diff --git a/src/bin/pg_combinebackup/t/001_basic.pl b/src/bin/pg_combinebackup/t/001_basic.pl new file mode 100644 index 0000000000..fb66075d1a --- /dev/null +++ b/src/bin/pg_combinebackup/t/001_basic.pl @@ -0,0 +1,23 @@ +# Copyright (c) 2021-2023, PostgreSQL Global Development Group + +use strict; +use warnings; +use PostgreSQL::Test::Utils; +use Test::More; + +my $tempdir = PostgreSQL::Test::Utils::tempdir; + +program_help_ok('pg_combinebackup'); +program_version_ok('pg_combinebackup'); +program_options_handling_ok('pg_combinebackup'); + +command_fails_like( + ['pg_combinebackup'], + qr/no input directories specified/, + 'input directories must be specified'); +command_fails_like( + [ 'pg_combinebackup', $tempdir ], + qr/no output directory specified/, + 'output directory must be specified'); + +done_testing(); diff --git a/src/bin/pg_combinebackup/t/002_compare_backups.pl b/src/bin/pg_combinebackup/t/002_compare_backups.pl new file mode 100644 index 0000000000..c27f999a32 --- /dev/null +++ b/src/bin/pg_combinebackup/t/002_compare_backups.pl @@ -0,0 +1,276 @@ +# Copyright (c) 2021-2023, PostgreSQL Global Development Group + +use strict; +use warnings; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Set up a new database instance. +my $primary = PostgreSQL::Test::Cluster->new('primary'); +$primary->init(has_archiving => 1, allows_streaming => 1); +$primary->append_conf('postgresql.conf', 'autovacuum = off'); +$primary->start; + +# Create some test tables, each containing one row of data, plus a whole +# extra database. +$primary->safe_psql('postgres', <<EOM); +CREATE TABLE will_change (a int, b text); +INSERT INTO will_change VALUES (1, 'initial test row'); +CREATE TABLE will_grow (a int, b text); +INSERT INTO will_grow VALUES (1, 'initial test row'); +CREATE TABLE will_shrink (a int, b text); +INSERT INTO will_shrink VALUES (1, 'initial test row'); +CREATE TABLE will_get_vacuumed (a int, b text); +INSERT INTO will_get_vacuumed VALUES (1, 'initial test row'); +CREATE TABLE will_get_dropped (a int, b text); +INSERT INTO will_get_dropped VALUES (1, 'initial test row'); +CREATE TABLE will_get_rewritten (a int, b text); +INSERT INTO will_get_rewritten VALUES (1, 'initial test row'); +CREATE DATABASE db_will_get_dropped; +EOM + +# Take a full backup. +my $backup1path = $primary->backup_dir . '/backup1'; +$primary->command_ok( + [ 'pg_basebackup', '-D', $backup1path, '--no-sync', '-cfast' ], + "full backup"); + +# Now make some database changes. +$primary->safe_psql('postgres', <<EOM); +UPDATE will_change SET b = 'modified value' WHERE a = 1; +INSERT INTO will_grow + SELECT g, 'additional row' FROM generate_series(2, 5000) g; +TRUNCATE will_shrink; +VACUUM will_get_vacuumed; +DROP TABLE will_get_dropped; +CREATE TABLE newly_created (a int, b text); +INSERT INTO newly_created VALUES (1, 'row for new table'); +VACUUM FULL will_get_rewritten; +DROP DATABASE db_will_get_dropped; +CREATE DATABASE db_newly_created; +EOM + +# Take an incremental backup. +my $backup2path = $primary->backup_dir . '/backup2'; +$primary->command_ok( + [ 'pg_basebackup', '-D', $backup2path, '--no-sync', '-cfast', + '--incremental', $backup1path . '/backup_manifest' ], + "incremental backup"); + +# Find an LSN to which either backup can be recovered. +my $lsn = $primary->safe_psql('postgres', "SELECT pg_current_wal_lsn();"); + +# Make sure that the WAL segment containing that LSN has been archived. +# PostgreSQL won't issue two consecutive XLOG_SWITCH records, and the backup +# just issued one, so call txid_current() to generate some WAL activity +# before calling pg_switch_wal(). +$primary->safe_psql('postgres', 'SELECT txid_current();'); +$primary->safe_psql('postgres', 'SELECT pg_switch_wal()'); + +# Now wait for the LSN we chose above to be archived. +my $archive_wait_query = + "SELECT pg_walfile_name('$lsn') <= last_archived_wal FROM pg_stat_archiver;"; +$primary->poll_query_until('postgres', $archive_wait_query) + or die "Timed out while waiting for WAL segment to be archived"; + +# Perform PITR from the full backup. Disable archive_mode so that the archive +# doesn't find out about the new timeline; that way, the later PITR below will +# choose the same timeline. +my $pitr1 = PostgreSQL::Test::Cluster->new('pitr1'); +$pitr1->init_from_backup($primary, 'backup1', + standby => 1, has_restoring => 1); +$pitr1->append_conf('postgresql.conf', qq{ +recovery_target_lsn = '$lsn' +recovery_target_action = 'promote' +archive_mode = 'off' +}); +$pitr1->start(); + +# Wait until we exit recovery, then stop the server. +$pitr1->poll_query_until('postgres', + "SELECT NOT pg_is_in_recovery();") + or die "Timed out while waiting apply to reach LSN $lsn"; +$pitr1->stop; + +# Perform PITR to the same LSN from the incremental backup. Use the same +# basic configuration as before. +my $pitr2 = PostgreSQL::Test::Cluster->new('pitr2'); +$pitr2->init_from_backup($primary, 'backup2', + standby => 1, has_restoring => 1, + combine_with_prior => [ 'backup1' ]); +$pitr2->append_conf('postgresql.conf', qq{ +recovery_target_lsn = '$lsn' +recovery_target_action = 'promote' +archive_mode = 'off' +}); +$pitr2->start(); + +# Wait until we exit recovery, then stop the server. +$pitr2->poll_query_until('postgres', + "SELECT NOT pg_is_in_recovery();") + or die "Timed out while waiting apply to reach LSN $lsn"; +$pitr2->stop; + +my $cmp = compare_data_directories($pitr1->basedir . '/pgdata', + $pitr2->basedir . '/pgdata', ''); +is($cmp, 0, "directories are identical"); + +done_testing(); + +sub compare_data_directories +{ + my ($basedir1, $basedir2, $relpath) = @_; + my $result = 0; + + if ($relpath eq '/pg_wal') + { + # Since recovery started at different LSNs, pg_wal contents may not + # be identical. Ignore that. + return 0; + } + + my $dir1 = $basedir1 . $relpath; + my $dir2 = $basedir2 . $relpath; + + opendir(DIR1, $dir1) || die "$dir1: $!"; + my @files1 = grep { $_ ne '.' && $_ ne '..' } readdir(DIR1); + closedir(DIR1); + + opendir(DIR2, $dir2) || die "$dir2: $!"; + my %files2 = map { $_ => 'unmatched' } + grep { $_ ne '.' && $_ ne '..' } readdir(DIR2); + closedir(DIR2); + + for my $fname (@files1) + { + if (!exists $files2{$fname}) + { + warn "$dir1/$fname exists but $dir2/$fname does not"; + ++$result; + next; + } + + $files2{$fname} = 'matched'; + + if (-d "$dir1/$fname") + { + if (! -d "$dir2/$fname") + { + warn "$dir1/$fname is a directory but $dir2/$fname is not"; + ++$result; + } + else + { + $result += + compare_data_directories($basedir1, $basedir2, + "$relpath/$fname"); + } + } + elsif (-d "$dir2/$fname") + { + warn "$dir2/$fname is a directory but $dir1/$fname is not"; + ++$result; + } + else + { + # Both are plain files. + $result += compare_files($basedir1, $basedir2, "$relpath/$fname"); + } + } + + for my $fname (keys %files2) + { + if ($files2{$fname} eq 'unmatched') + { + warn "$dir2/$fname exists but $dir1/$fname does not"; + ++$result; + } + } + + return $result; +} + +sub compare_files +{ + my ($basedir1, $basedir2, $relpath) = @_; + my $file1 = $basedir1 . $relpath; + my $file2 = $basedir2 . $relpath; + + if ($relpath eq '/backup_manifest') + { + # We don't expect the backup manifest to be identical between two + # backups taken at different times, so just disregard it. + return 0; + } + + if ($relpath eq '/backup_label.old') + { + # We don't expect the backup label to be identical; the start WAL + # location and probably also the start time are expected to be + # different. + return 0; + } + + if ($relpath eq '/postgresql.conf') + { + # At least the port numbers are expected to be different, so + # disregard this file. + return 0; + } + + if ($relpath eq '/postmaster.opts') + { + # At least the cluster names are expected to be different, so + # disregard this file. + return 0; + } + + if ($relpath eq '/global/pg_control') + { + # At least the mock authentication nonce is expected to be different, + # so disregard this file. + return 0; + } + + if ($relpath eq '/pg_stat/pgstat.stat') + { + # Stats aren't stable enough to be compared here. + return 0; + } + + if ($relpath =~ m@/pg_internal\.init$@) + { + # relcache init files are rebuilt at startup, so they don't need to + # match. And because we write out the contents of data structures like + # RelationData that include pointers, they almost certainly won't. + return 0; + } + + # Check whether the lengths match. + my $length1 = -s $file1; + my $length2 = -s $file2; + if ($length1 != $length2) + { + warn "$file1 has length $length1, but $file2 has length $length2"; + return 1; + } + + # Compare contents. + my $contents1 = slurp_file($file1); + my $contents2 = slurp_file($file2); + if ($contents1 ne $contents2) + { + my $nchars = 1; + while (substr($contents1, 0, $nchars) eq substr($contents2, 0, $nchars)) + { + ++$nchars; + } + warn sprintf("%s and %s are both of length %s, but differ beginning at byte %d", + $file1, $file2, $length1, $nchars - 1); + return 1; + } + + # Files are identical. + return 0; +} diff --git a/src/test/perl/PostgreSQL/Test/Cluster.pm b/src/test/perl/PostgreSQL/Test/Cluster.pm index 2a478ba6ed..3b57379e13 100644 --- a/src/test/perl/PostgreSQL/Test/Cluster.pm +++ b/src/test/perl/PostgreSQL/Test/Cluster.pm @@ -779,6 +779,10 @@ a tar-format backup, pass the name of the tar program to use in the keyword parameter tar_program. Note that tablespace tar files aren't handled here. +To restore from an incremental backup, pass the parameter combine_with_prior +as a reference to an array of prior backup names with which this backup +is to be combined using pg_combinebackup. + Streaming replication can be enabled on this node by passing the keyword parameter has_streaming => 1. This is disabled by default. @@ -816,7 +820,22 @@ sub init_from_backup mkdir $self->archive_dir; my $data_path = $self->data_dir; - if (defined $params{tar_program}) + if (defined $params{combine_with_prior}) + { + my @prior_backups = @{$params{combine_with_prior}}; + my @prior_backup_path; + + for my $prior_backup_name (@prior_backups) + { + push @prior_backup_path, + $root_node->backup_dir . '/' . $prior_backup_name; + } + + local %ENV = $self->_get_env(); + PostgreSQL::Test::Utils::system_or_bail('pg_combinebackup', + @prior_backup_path, $backup_path, '-o', $data_path); + } + elsif (defined $params{tar_program}) { mkdir($data_path); PostgreSQL::Test::Utils::system_or_bail($params{tar_program}, 'xf', -- 2.37.1 (Apple Git-137.1) [application/octet-stream] 0007-Prototype-patch-for-incremental-and-differential-bac.patch (309.4K, ../../CA+TgmoZoHNqx=zconLLbV+8=gNhv7g5ocXd-Mee97wvTkKc-6A@mail.gmail.com/10-0007-Prototype-patch-for-incremental-and-differential-bac.patch) download | inline diff: From d76b71f72567cfb32340f870cf768e31ea3461c6 Mon Sep 17 00:00:00 2001 From: Robert Haas <[email protected]> Date: Wed, 14 Jun 2023 11:31:29 -0400 Subject: [PATCH 7/9] Prototype patch for incremental and differential backup. We don't differentiate between incremental and differential backups; the term "incremental" as used herein means "either incremental or differential". This adds a new background process, the WAL summarizer, whose behavor is governed by new GUCs wal_summarize_mb and wal_summarize_keep_time. This writes out WAL summary files to $PGDATA/pg_wal/summaries. Each summary file contains information for a certain range of LSNs on a certain TLI. For each relation, it stores a "limit block" which is 0 if a relation is created or destroyed within a certain range of WAL records, or otherwise the shortest length to which the relation was truncated during that range of WAL records, or otherwise InvalidBlockNumber. In addition, it stores any blocks which have been modified during that range of WAL records, but excluding blocks which were removed by truncation after they were modified and which were never modified thereafter. In other words, it tells us which blocks need to copied in case of an incremental backup covering that range of WAL records. To take an incremental backup, you use the new replication command UPLOAD_MANIFEST to upload the manifest for the prior backup. This prior backup could either be a full backup or another incremental backup. You then use BASE_BACKUP with the INCREMENTAL option to take the backup. pg_basebackup now has an --incremental=PATH_TO_MANIFEST option to trigger this behavior. An incremental backup is like a regular full backup except that some relation files are replaced with files with names like INCREMENTAL.${ORIGINAL_NAME}, and the backup_label file contains additional lines identifying it as an incremental backup. The new pg_combinebackup tool can be used to reconstruct a data directory from a full backup and a series of incremental backups. XXX. It would be nice if we could do something about incremental JSON parsing. XXX. This needs a lot of work on documentation and tests. Patch by me. Thanks to Dilip Kumar and Andres Freund for some helpful design discussions. --- src/backend/access/transam/xlog.c | 97 +- src/backend/access/transam/xlogbackup.c | 10 + src/backend/access/transam/xlogrecovery.c | 10 +- src/backend/backup/Makefile | 5 +- src/backend/backup/basebackup.c | 340 +++- src/backend/backup/basebackup_incremental.c | 867 ++++++++++ src/backend/backup/meson.build | 3 + src/backend/backup/walsummary.c | 356 +++++ src/backend/backup/walsummaryfuncs.c | 169 ++ src/backend/postmaster/Makefile | 1 + src/backend/postmaster/auxprocess.c | 8 + src/backend/postmaster/meson.build | 1 + src/backend/postmaster/postmaster.c | 53 + src/backend/postmaster/walsummarizer.c | 1414 +++++++++++++++++ src/backend/replication/repl_gram.y | 14 +- src/backend/replication/repl_scanner.l | 2 + src/backend/replication/walsender.c | 162 +- src/backend/storage/ipc/ipci.c | 3 + src/backend/storage/lmgr/lwlocknames.txt | 1 + src/backend/utils/activity/pgstat_io.c | 4 +- .../utils/activity/wait_event_names.txt | 5 + src/backend/utils/init/miscinit.c | 3 + src/backend/utils/misc/guc_tables.c | 29 + src/backend/utils/misc/postgresql.conf.sample | 5 + src/bin/Makefile | 1 + src/bin/initdb/initdb.c | 1 + src/bin/meson.build | 1 + src/bin/pg_basebackup/bbstreamer_file.c | 1 + src/bin/pg_basebackup/pg_basebackup.c | 108 +- src/bin/pg_basebackup/t/010_pg_basebackup.pl | 4 +- src/bin/pg_combinebackup/.gitignore | 1 + src/bin/pg_combinebackup/Makefile | 46 + src/bin/pg_combinebackup/backup_label.c | 281 ++++ src/bin/pg_combinebackup/backup_label.h | 29 + src/bin/pg_combinebackup/copy_file.c | 169 ++ src/bin/pg_combinebackup/copy_file.h | 19 + src/bin/pg_combinebackup/load_manifest.c | 245 +++ src/bin/pg_combinebackup/load_manifest.h | 67 + src/bin/pg_combinebackup/meson.build | 29 + src/bin/pg_combinebackup/pg_combinebackup.c | 1268 +++++++++++++++ src/bin/pg_combinebackup/reconstruct.c | 618 +++++++ src/bin/pg_combinebackup/reconstruct.h | 32 + src/bin/pg_combinebackup/write_manifest.c | 293 ++++ src/bin/pg_combinebackup/write_manifest.h | 33 + src/bin/pg_resetwal/pg_resetwal.c | 36 + src/common/Makefile | 1 + src/common/blkreftable.c | 1309 +++++++++++++++ src/common/meson.build | 1 + src/include/access/xlog.h | 1 + src/include/access/xlogbackup.h | 2 + src/include/backup/basebackup.h | 5 +- src/include/backup/basebackup_incremental.h | 56 + src/include/backup/walsummary.h | 49 + src/include/catalog/pg_proc.dat | 19 + src/include/common/blkreftable.h | 120 ++ src/include/miscadmin.h | 3 + src/include/nodes/replnodes.h | 9 + src/include/postmaster/walsummarizer.h | 31 + src/include/storage/proc.h | 9 +- src/include/utils/guc_tables.h | 1 + src/test/recovery/t/001_stream_rep.pl | 2 + src/test/recovery/t/019_replslot_limit.pl | 3 + .../t/035_standby_logical_decoding.pl | 1 + src/tools/pgindent/typedefs.list | 24 + 64 files changed, 8421 insertions(+), 69 deletions(-) create mode 100644 src/backend/backup/basebackup_incremental.c create mode 100644 src/backend/backup/walsummary.c create mode 100644 src/backend/backup/walsummaryfuncs.c create mode 100644 src/backend/postmaster/walsummarizer.c create mode 100644 src/bin/pg_combinebackup/.gitignore create mode 100644 src/bin/pg_combinebackup/Makefile create mode 100644 src/bin/pg_combinebackup/backup_label.c create mode 100644 src/bin/pg_combinebackup/backup_label.h create mode 100644 src/bin/pg_combinebackup/copy_file.c create mode 100644 src/bin/pg_combinebackup/copy_file.h create mode 100644 src/bin/pg_combinebackup/load_manifest.c create mode 100644 src/bin/pg_combinebackup/load_manifest.h create mode 100644 src/bin/pg_combinebackup/meson.build create mode 100644 src/bin/pg_combinebackup/pg_combinebackup.c create mode 100644 src/bin/pg_combinebackup/reconstruct.c create mode 100644 src/bin/pg_combinebackup/reconstruct.h create mode 100644 src/bin/pg_combinebackup/write_manifest.c create mode 100644 src/bin/pg_combinebackup/write_manifest.h create mode 100644 src/common/blkreftable.c create mode 100644 src/include/backup/basebackup_incremental.h create mode 100644 src/include/backup/walsummary.h create mode 100644 src/include/common/blkreftable.h create mode 100644 src/include/postmaster/walsummarizer.h diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 6f38d0eb9a..3e19ec9ad1 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -77,6 +77,7 @@ #include "port/pg_iovec.h" #include "postmaster/bgwriter.h" #include "postmaster/startup.h" +#include "postmaster/walsummarizer.h" #include "postmaster/walwriter.h" #include "replication/logical.h" #include "replication/origin.h" @@ -3500,6 +3501,43 @@ XLogGetLastRemovedSegno(void) return lastRemovedSegNo; } +/* + * Return the oldest WAL segment on the given TLI that still exists in + * XLOGDIR, or 0 if none. + */ +XLogSegNo +XLogGetOldestSegno(TimeLineID tli) +{ + DIR *xldir; + struct dirent *xlde; + XLogSegNo oldest_segno = 0; + + xldir = AllocateDir(XLOGDIR); + while ((xlde = ReadDir(xldir, XLOGDIR)) != NULL) + { + TimeLineID file_tli; + XLogSegNo file_segno; + + /* Ignore files that are not XLOG segments */ + if (!IsXLogFileName(xlde->d_name)) + continue; + + /* Parse filename to get TLI and segno. */ + XLogFromFileName(xlde->d_name, &file_tli, &file_segno, + wal_segment_size); + + /* Ignore anything that's not from the TLI of interest. */ + if (tli != file_tli) + continue; + + /* If it's the oldest so far, update oldest_segno. */ + if (oldest_segno == 0 || file_segno < oldest_segno) + oldest_segno = file_segno; + } + + FreeDir(xldir); + return oldest_segno; +} /* * Update the last removed segno pointer in shared memory, to reflect that the @@ -3779,8 +3817,8 @@ RemoveXlogFile(const struct dirent *segment_de, } /* - * Verify whether pg_wal and pg_wal/archive_status exist. - * If the latter does not exist, recreate it. + * Verify whether pg_wal, pg_wal/archive_status, and pg_wal/summaries exist. + * If the latter do not exist, recreate them. * * It is not the goal of this function to verify the contents of these * directories, but to help in cases where someone has performed a cluster @@ -3823,6 +3861,26 @@ ValidateXLOGDirectoryStructure(void) (errmsg("could not create missing directory \"%s\": %m", path))); } + + /* Check for summaries */ + snprintf(path, MAXPGPATH, XLOGDIR "/summaries"); + if (stat(path, &stat_buf) == 0) + { + /* Check for weird cases where it exists but isn't a directory */ + if (!S_ISDIR(stat_buf.st_mode)) + ereport(FATAL, + (errmsg("required WAL directory \"%s\" does not exist", + path))); + } + else + { + ereport(LOG, + (errmsg("creating missing WAL directory \"%s\"", path))); + if (MakePGDirectory(path) < 0) + ereport(FATAL, + (errmsg("could not create missing directory \"%s\": %m", + path))); + } } /* @@ -5147,9 +5205,9 @@ StartupXLOG(void) #endif /* - * Verify that pg_wal and pg_wal/archive_status exist. In cases where - * someone has performed a copy for PITR, these directories may have been - * excluded and need to be re-created. + * Verify that pg_wal, pg_wal/archive_status, and pg_wal/summaries exist. + * In cases where someone has performed a copy for PITR, these directories + * may have been excluded and need to be re-created. */ ValidateXLOGDirectoryStructure(); @@ -6830,6 +6888,17 @@ CreateCheckPoint(int flags) */ END_CRIT_SECTION(); + /* + * If there hasn't been much system activity in a while, the WAL + * summarizer may be sleeping for relatively long periods, which could + * delay an incremental backup that has started concurrently. In the hopes + * of avoiding that, poke the WAL summarizer here. + * + * Possibly this should instead be done at some earlier point in this + * function, but it's not clear that it matters much. + */ + SetWalSummarizerLatch(); + /* * Let smgr do post-checkpoint cleanup (eg, deleting old files). */ @@ -7504,6 +7573,20 @@ KeepLogSeg(XLogRecPtr recptr, XLogSegNo *logSegNo) } } + /* + * If WAL summarization is in use, don't remove WAL that has yet to be + * summarized. + */ + keep = GetOldestUnsummarizedLSN(NULL, NULL); + if (keep != InvalidXLogRecPtr) + { + XLogSegNo unsummarized_segno; + + XLByteToSeg(keep, unsummarized_segno, wal_segment_size); + if (unsummarized_segno < segno) + segno = unsummarized_segno; + } + /* but, keep at least wal_keep_size if that's set */ if (wal_keep_size_mb > 0) { @@ -8490,8 +8573,8 @@ do_pg_backup_start(const char *backupidstr, bool fast, List **tablespaces, /* * Try to parse the directory name as an unsigned integer. * - * Tablespace directories should be positive integers that can - * be represented in 32 bits, with no leading zeroes or trailing + * Tablespace directories should be positive integers that can be + * represented in 32 bits, with no leading zeroes or trailing * garbage. If we come across a name that doesn't meet those * criteria, skip it. */ diff --git a/src/backend/access/transam/xlogbackup.c b/src/backend/access/transam/xlogbackup.c index 23461c9d2c..3ad6b679d5 100644 --- a/src/backend/access/transam/xlogbackup.c +++ b/src/backend/access/transam/xlogbackup.c @@ -77,6 +77,16 @@ build_backup_content(BackupState *state, bool ishistoryfile) appendStringInfo(result, "STOP TIMELINE: %u\n", state->stoptli); } + /* either both istartpoint and istarttli should be set, or neither */ + Assert(XLogRecPtrIsInvalid(state->istartpoint) == (state->istarttli == 0)); + if (!XLogRecPtrIsInvalid(state->istartpoint)) + { + appendStringInfo(result, "INCREMENTAL FROM LSN: %X/%X\n", + LSN_FORMAT_ARGS(state->istartpoint)); + appendStringInfo(result, "INCREMENTAL FROM TLI: %u\n", + state->istarttli); + } + data = result->data; pfree(result); diff --git a/src/backend/access/transam/xlogrecovery.c b/src/backend/access/transam/xlogrecovery.c index 4ff4430006..89ddec5bf9 100644 --- a/src/backend/access/transam/xlogrecovery.c +++ b/src/backend/access/transam/xlogrecovery.c @@ -1284,6 +1284,12 @@ read_backup_label(XLogRecPtr *checkPointLoc, TimeLineID *backupLabelTLI, tli_from_file, BACKUP_LABEL_FILE))); } + if (fscanf(lfp, "INCREMENTAL FROM LSN: %X/%X\n", &hi, &lo) > 0) + ereport(FATAL, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("this is an incremental backup, not a data directory"), + errhint("Use pg_combinebackup to reconstruct a valid data directory."))); + if (ferror(lfp) || FreeFile(lfp)) ereport(FATAL, (errcode_for_file_access(), @@ -1340,7 +1346,7 @@ read_tablespace_map(List **tablespaces) { if (!was_backslash && (ch == '\n' || ch == '\r')) { - char *endp; + char *endp; if (i == 0) continue; /* \r immediately followed by \n */ @@ -1363,7 +1369,7 @@ read_tablespace_map(List **tablespaces) ti = palloc0(sizeof(tablespaceinfo)); errno = 0; ti->oid = strtoul(str, &endp, 10); - if (*endp != '\0' || errno == EINVAL || errno == ERANGE) + if (*endp != '\0' || errno == EINVAL || errno == ERANGE) ereport(FATAL, (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("invalid data in file \"%s\"", TABLESPACE_MAP))); diff --git a/src/backend/backup/Makefile b/src/backend/backup/Makefile index b21bd8ff43..751e6d3d5e 100644 --- a/src/backend/backup/Makefile +++ b/src/backend/backup/Makefile @@ -19,12 +19,15 @@ OBJS = \ basebackup.o \ basebackup_copy.o \ basebackup_gzip.o \ + basebackup_incremental.o \ basebackup_lz4.o \ basebackup_zstd.o \ basebackup_progress.o \ basebackup_server.o \ basebackup_sink.o \ basebackup_target.o \ - basebackup_throttle.o + basebackup_throttle.o \ + walsummary.o \ + walsummaryfuncs.o include $(top_srcdir)/src/backend/common.mk diff --git a/src/backend/backup/basebackup.c b/src/backend/backup/basebackup.c index 64ab54fe06..8aea2a4a76 100644 --- a/src/backend/backup/basebackup.c +++ b/src/backend/backup/basebackup.c @@ -20,8 +20,10 @@ #include "access/xlogbackup.h" #include "backup/backup_manifest.h" #include "backup/basebackup.h" +#include "backup/basebackup_incremental.h" #include "backup/basebackup_sink.h" #include "backup/basebackup_target.h" +#include "catalog/pg_tablespace_d.h" #include "commands/defrem.h" #include "common/compression.h" #include "common/file_perm.h" @@ -64,6 +66,7 @@ typedef struct bool fastcheckpoint; bool nowait; bool includewal; + bool incremental; uint32 maxrate; bool sendtblspcmapfile; bool send_to_client; @@ -75,22 +78,37 @@ typedef struct pg_checksum_type manifest_checksum_type; } basebackup_options; +typedef struct +{ + const char *filename; + pg_checksum_context *checksum_ctx; + bbsink *sink; + size_t bytes_sent; +} FileChunkContext; + static int64 sendTablespace(bbsink *sink, char *path, Oid spcoid, bool sizeonly, - struct backup_manifest_info *manifest); + struct backup_manifest_info *manifest, + IncrementalBackupInfo *ib); static int64 sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, List *tablespaces, bool sendtblspclinks, - backup_manifest_info *manifest, Oid spcoid); + backup_manifest_info *manifest, Oid spcoid, + IncrementalBackupInfo *ib); static bool sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, struct stat *statbuf, bool missing_ok, Oid dboid, Oid spcoid, RelFileNumber relfilenumber, unsigned segno, - backup_manifest_info *manifest); + backup_manifest_info *manifest, + unsigned num_incremental_blocks, + BlockNumber *incremental_blocks, + unsigned truncation_block_length); static off_t read_file_data_into_buffer(bbsink *sink, const char *readfilename, int fd, off_t offset, size_t length, BlockNumber blkno, bool verify_checksum, int *checksum_failures); +static void push_to_sink(bbsink *sink, pg_checksum_context *checksum_ctx, + size_t *bytes_done, void *data, size_t length); static bool verify_page_checksum(Page page, XLogRecPtr start_lsn, BlockNumber blkno, uint16 *expected_checksum); @@ -102,7 +120,8 @@ static int64 _tarWriteHeader(bbsink *sink, const char *filename, bool sizeonly); static void _tarWritePadding(bbsink *sink, int len); static void convert_link_to_directory(const char *pathbuf, struct stat *statbuf); -static void perform_base_backup(basebackup_options *opt, bbsink *sink); +static void perform_base_backup(basebackup_options *opt, bbsink *sink, + IncrementalBackupInfo *ib); static void parse_basebackup_options(List *options, basebackup_options *opt); static int compareWalFileNames(const ListCell *a, const ListCell *b); static int basebackup_read_file(int fd, char *buf, size_t nbytes, off_t offset, @@ -220,7 +239,8 @@ static const struct exclude_list_item excludeFiles[] = * clobbered by longjmp" from stupider versions of gcc. */ static void -perform_base_backup(basebackup_options *opt, bbsink *sink) +perform_base_backup(basebackup_options *opt, bbsink *sink, + IncrementalBackupInfo *ib) { bbsink_state state; XLogRecPtr endptr; @@ -270,6 +290,10 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) ListCell *lc; tablespaceinfo *newti; + /* If this is an incremental backup, execute preparatory steps. */ + if (ib != NULL) + PrepareForIncrementalBackup(ib, backup_state); + /* Add a node for the base directory at the end */ newti = palloc0(sizeof(tablespaceinfo)); newti->size = -1; @@ -289,10 +313,10 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) if (tmp->path == NULL) tmp->size = sendDir(sink, ".", 1, true, state.tablespaces, - true, NULL, InvalidOid); + true, NULL, InvalidOid, NULL); else tmp->size = sendTablespace(sink, tmp->path, tmp->oid, true, - NULL); + NULL, NULL); state.bytes_total += tmp->size; } state.bytes_total_is_valid = true; @@ -330,7 +354,7 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) /* Then the bulk of the files... */ sendDir(sink, ".", 1, false, state.tablespaces, - sendtblspclinks, &manifest, InvalidOid); + sendtblspclinks, &manifest, InvalidOid, ib); /* ... and pg_control after everything else. */ if (lstat(XLOG_CONTROL_FILE, &statbuf) != 0) @@ -340,7 +364,7 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) XLOG_CONTROL_FILE))); sendFile(sink, XLOG_CONTROL_FILE, XLOG_CONTROL_FILE, &statbuf, false, InvalidOid, InvalidOid, - InvalidRelFileNumber, 0, &manifest); + InvalidRelFileNumber, 0, &manifest, 0, NULL, 0); } else { @@ -348,7 +372,7 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) bbsink_begin_archive(sink, archive_name); - sendTablespace(sink, ti->path, ti->oid, false, &manifest); + sendTablespace(sink, ti->path, ti->oid, false, &manifest, ib); } /* @@ -610,7 +634,7 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) sendFile(sink, pathbuf, pathbuf, &statbuf, false, InvalidOid, InvalidOid, InvalidRelFileNumber, 0, - &manifest); + &manifest, 0, NULL, 0); /* unconditionally mark file as archived */ StatusFilePath(pathbuf, fname, ".done"); @@ -686,6 +710,7 @@ parse_basebackup_options(List *options, basebackup_options *opt) bool o_checkpoint = false; bool o_nowait = false; bool o_wal = false; + bool o_incremental = false; bool o_maxrate = false; bool o_tablespace_map = false; bool o_noverify_checksums = false; @@ -764,6 +789,15 @@ parse_basebackup_options(List *options, basebackup_options *opt) opt->includewal = defGetBoolean(defel); o_wal = true; } + else if (strcmp(defel->defname, "incremental") == 0) + { + if (o_incremental) + ereport(ERROR, + (errcode(ERRCODE_SYNTAX_ERROR), + errmsg("duplicate option \"%s\"", defel->defname))); + opt->incremental = defGetBoolean(defel); + o_incremental = true; + } else if (strcmp(defel->defname, "max_rate") == 0) { int64 maxrate; @@ -956,7 +990,7 @@ parse_basebackup_options(List *options, basebackup_options *opt) * the filesystem, bypassing the buffer cache. */ void -SendBaseBackup(BaseBackupCmd *cmd) +SendBaseBackup(BaseBackupCmd *cmd, IncrementalBackupInfo *ib) { basebackup_options opt; bbsink *sink; @@ -980,6 +1014,20 @@ SendBaseBackup(BaseBackupCmd *cmd) set_ps_display(activitymsg); } + /* + * If we're asked to perform an incremental backup and the user has not + * supplied a manifest, that's an ERROR. + * + * If we're asked to perform a full backup and the user did supply a + * manifest, just ignore it. + */ + if (!opt.incremental) + ib = NULL; + else if (ib == NULL) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("must UPLOAD_MANIFEST before performing an incremental BASE_BACKUP"))); + /* * If the target is specifically 'client' then set up to stream the backup * to the client; otherwise, it's being sent someplace else and should not @@ -1011,7 +1059,7 @@ SendBaseBackup(BaseBackupCmd *cmd) */ PG_TRY(); { - perform_base_backup(&opt, sink); + perform_base_backup(&opt, sink, ib); } PG_FINALLY(); { @@ -1086,7 +1134,7 @@ sendFileWithContent(bbsink *sink, const char *filename, const char *content, */ static int64 sendTablespace(bbsink *sink, char *path, Oid spcoid, bool sizeonly, - backup_manifest_info *manifest) + backup_manifest_info *manifest, IncrementalBackupInfo *ib) { int64 size; char pathbuf[MAXPGPATH]; @@ -1120,7 +1168,7 @@ sendTablespace(bbsink *sink, char *path, Oid spcoid, bool sizeonly, /* Send all the files in the tablespace version directory */ size += sendDir(sink, pathbuf, strlen(path), sizeonly, NIL, true, manifest, - spcoid); + spcoid, ib); return size; } @@ -1140,7 +1188,7 @@ sendTablespace(bbsink *sink, char *path, Oid spcoid, bool sizeonly, static int64 sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, List *tablespaces, bool sendtblspclinks, backup_manifest_info *manifest, - Oid spcoid) + Oid spcoid, IncrementalBackupInfo *ib) { DIR *dir; struct dirent *de; @@ -1148,7 +1196,8 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, struct stat statbuf; int64 size = 0; const char *lastDir; /* Split last dir from parent path. */ - bool isRelationDir = false; /* Does directory contain relations? */ + bool isRelationDir = false; /* Does directory contain relations? */ + bool isGlobalDir = false; Oid dboid = InvalidOid; /* @@ -1182,14 +1231,17 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, } } else if (strcmp(path, "./global") == 0) + { isRelationDir = true; + isGlobalDir = true; + } dir = AllocateDir(path); while ((de = ReadDir(dir, path)) != NULL) { int excludeIdx; bool excludeFound; - RelFileNumber relfilenumber = InvalidRelFileNumber; + RelFileNumber relfilenumber = InvalidRelFileNumber; ForkNumber relForkNum = InvalidForkNumber; unsigned segno = 0; bool isRelationFile = false; @@ -1256,9 +1308,8 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, char initForkFile[MAXPGPATH]; /* - * If any other type of fork, check if there is an init fork - * with the same RelFileNumber. If so, the file can be - * excluded. + * If any other type of fork, check if there is an init fork with + * the same RelFileNumber. If so, the file can be excluded. */ snprintf(initForkFile, sizeof(initForkFile), "%s/%u_init", path, relfilenumber); @@ -1332,11 +1383,13 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, &statbuf, sizeonly); /* - * Also send archive_status directory (by hackishly reusing - * statbuf from above ...). + * Also send archive_status and summaries directories (by + * hackishly reusing statbuf from above ...). */ size += _tarWriteHeader(sink, "./pg_wal/archive_status", NULL, &statbuf, sizeonly); + size += _tarWriteHeader(sink, "./pg_wal/summaries", NULL, + &statbuf, sizeonly); continue; /* don't recurse into pg_wal */ } @@ -1405,27 +1458,79 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, if (!skip_this_dir) size += sendDir(sink, pathbuf, basepathlen, sizeonly, tablespaces, - sendtblspclinks, manifest, spcoid); + sendtblspclinks, manifest, spcoid, ib); } else if (S_ISREG(statbuf.st_mode)) { bool sent = false; + unsigned num_blocks_required = 0; + unsigned truncation_block_length = 0; + BlockNumber relative_block_numbers[RELSEG_SIZE]; + char tarfilenamebuf[MAXPGPATH * 2]; + char *tarfilename = pathbuf + basepathlen + 1; + FileBackupMethod method = BACK_UP_FILE_FULLY; + + if (ib != NULL && isRelationFile) + { + Oid relspcoid; + char *lookup_path; + + if (OidIsValid(spcoid)) + { + relspcoid = spcoid; + lookup_path = psprintf("pg_tblspc/%u/%s", spcoid, + pathbuf + basepathlen + 1); + } + else + { + if (isGlobalDir) + relspcoid = GLOBALTABLESPACE_OID; + else + relspcoid = DEFAULTTABLESPACE_OID; + lookup_path = pstrdup(pathbuf + basepathlen + 1); + } - if (!sizeonly) - sent = sendFile(sink, pathbuf, pathbuf + basepathlen + 1, &statbuf, - true, dboid, spcoid, - relfilenumber, segno, manifest); + method = GetFileBackupMethod(ib, lookup_path, dboid, relspcoid, + relfilenumber, relForkNum, + segno, statbuf.st_size, + &num_blocks_required, + relative_block_numbers, + &truncation_block_length); + if (method == BACK_UP_FILE_INCREMENTALLY) + { + statbuf.st_size = + GetIncrementalFileSize(num_blocks_required); + snprintf(tarfilenamebuf, sizeof(tarfilenamebuf), + "%s/INCREMENTAL.%s", + path + basepathlen + 1, + de->d_name); + tarfilename = tarfilenamebuf; + } + + pfree(lookup_path); + } - if (sent || sizeonly) + if (method != DO_NOT_BACK_UP_FILE) { - /* Add size. */ - size += statbuf.st_size; + if (!sizeonly) + sent = sendFile(sink, pathbuf, tarfilename, &statbuf, + true, dboid, spcoid, + relfilenumber, segno, manifest, + num_blocks_required, + method == BACK_UP_FILE_INCREMENTALLY ? relative_block_numbers : NULL, + truncation_block_length); + + if (sent || sizeonly) + { + /* Add size. */ + size += statbuf.st_size; - /* Pad to a multiple of the tar block size. */ - size += tarPaddingBytesRequired(statbuf.st_size); + /* Pad to a multiple of the tar block size. */ + size += tarPaddingBytesRequired(statbuf.st_size); - /* Size of the header for the file. */ - size += TAR_BLOCK_SIZE; + /* Size of the header for the file. */ + size += TAR_BLOCK_SIZE; + } } } else @@ -1444,6 +1549,12 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, * If dboid is anything other than InvalidOid then any checksum failures * detected will get reported to the cumulative stats system. * + * If the file is to be set incrementally, then num_incremental_blocks + * should be the number of blocks to be sent, and incremental_blocks + * an array of block numbers relative to the start of the current segment. + * If the whole file is to be sent, then incremental_blocks should be NULL, + * and num_incremental_blocks can have any value, as it will be ignored. + * * Returns true if the file was successfully sent, false if 'missing_ok', * and the file did not exist. */ @@ -1451,7 +1562,8 @@ static bool sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, struct stat *statbuf, bool missing_ok, Oid dboid, Oid spcoid, RelFileNumber relfilenumber, unsigned segno, - backup_manifest_info *manifest) + backup_manifest_info *manifest, unsigned num_incremental_blocks, + BlockNumber *incremental_blocks, unsigned truncation_block_length) { int fd; BlockNumber blkno = 0; @@ -1460,6 +1572,7 @@ sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, pgoff_t bytes_done = 0; bool verify_checksum = false; pg_checksum_context checksum_ctx; + int ibindex = 0; if (pg_checksum_init(&checksum_ctx, manifest->checksum_type) < 0) elog(ERROR, "could not initialize checksum of file \"%s\"", @@ -1492,22 +1605,111 @@ sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, RelFileNumberIsValid(relfilenumber)) verify_checksum = true; + /* + * If we're sending an incremental file, write the file header. + */ + if (incremental_blocks != NULL) + { + unsigned magic = INCREMENTAL_MAGIC; + size_t header_bytes_done = 0; + + /* Emit header data. */ + push_to_sink(sink, &checksum_ctx, &header_bytes_done, + &magic, sizeof(magic)); + push_to_sink(sink, &checksum_ctx, &header_bytes_done, + &num_incremental_blocks, sizeof(num_incremental_blocks)); + push_to_sink(sink, &checksum_ctx, &header_bytes_done, + &truncation_block_length, sizeof(truncation_block_length)); + push_to_sink(sink, &checksum_ctx, &header_bytes_done, + incremental_blocks, + sizeof(BlockNumber) * num_incremental_blocks); + + /* Flush out any data still in the buffer so it's again empty. */ + if (header_bytes_done > 0) + { + bbsink_archive_contents(sink, header_bytes_done); + if (pg_checksum_update(&checksum_ctx, + (uint8 *) sink->bbs_buffer, + header_bytes_done) < 0) + elog(ERROR, "could not update checksum of base backup"); + } + + /* Update our notion of file position. */ + bytes_done += sizeof(magic); + bytes_done += sizeof(num_incremental_blocks); + bytes_done += sizeof(truncation_block_length); + bytes_done += sizeof(BlockNumber) * num_incremental_blocks; + } + /* * Loop until we read the amount of data the caller told us to expect. The * file could be longer, if it was extended while we were sending it, but * for a base backup we can ignore such extended data. It will be restored * from WAL. */ - while (bytes_done < statbuf->st_size) + while (1) { - size_t remaining = statbuf->st_size - bytes_done; + /* + * Determine whether we've read all the data that we need, and if not, + * read some more. + */ + if (incremental_blocks == NULL) + { + size_t remaining = statbuf->st_size - bytes_done; + + /* + * If we've read the required number of bytes, then it's time to + * stop. + */ + if (bytes_done >= statbuf->st_size) + break; + + /* + * Read as many bytes as will fit in the buffer, or however many + * are left to read, whichever is less. + */ + cnt = read_file_data_into_buffer(sink, readfilename, fd, + bytes_done, remaining, + blkno + segno * RELSEG_SIZE, + verify_checksum, + &checksum_failures); + } + else + { + BlockNumber relative_blkno; + + /* + * If we've read all the blocks, then it's time to stop. + */ + if (ibindex >= num_incremental_blocks) + break; + + /* + * Read just one block, whichever one is the next that we're + * supposed to include. + */ + relative_blkno = incremental_blocks[ibindex++]; + cnt = read_file_data_into_buffer(sink, readfilename, fd, + relative_blkno * BLCKSZ, + BLCKSZ, + relative_blkno + segno * RELSEG_SIZE, + verify_checksum, + &checksum_failures); - /* Try to read some more data. */ - cnt = read_file_data_into_buffer(sink, readfilename, fd, bytes_done, - remaining, - blkno + segno * RELSEG_SIZE, - verify_checksum, - &checksum_failures); + /* + * If we get a partial read, that must mean that the relation is + * being truncated. Ultimately, it should be truncated to a + * multiple of BLCKSZ, since this path should only be reached for + * relation files, but we might transiently observe an + * intermediate value. + * + * It should be fine to treat this just as if the entire block had + * been truncated away - i.e. fill this and all later blocks with + * zeroes. WAL replay will fix things up. + */ + if (cnt < BLCKSZ) + break; + } /* * If the amount of data we were able to read was not a multiple of @@ -1690,6 +1892,56 @@ read_file_data_into_buffer(bbsink *sink, const char *readfilename, int fd, return cnt; } +/* + * Push data into a bbsink. + * + * It's better, when possible, to read data directly into the bbsink's buffer, + * rather than using this function to copy it into the buffer; this function is + * for cases where that approach is not practical. + * + * bytes_done should point to a count of the number of bytes that are + * currently used in the bbsink's buffer. Upon return, the bytes identified by + * data and length will have been copied into the bbsink's buffer, flushing + * as required, and *bytes_done will have been updated accordingly. If the + * buffer was flushed, the previous contents will also have been fed to + * checksum_ctx. + * + * Note that after one or more calls to this function it is the caller's + * responsibility to perform any required final flush. + */ +static void +push_to_sink(bbsink *sink, pg_checksum_context *checksum_ctx, + size_t *bytes_done, void *data, size_t length) +{ + while (length > 0) + { + size_t bytes_to_copy; + + /* + * We use < here rather than <= so that if the data exactly fills the + * remaining buffer space, we trigger a flush now. + */ + if (length < sink->bbs_buffer_length - *bytes_done) + { + /* Append remaining data to buffer. */ + memcpy(sink->bbs_buffer + *bytes_done, data, length); + *bytes_done += length; + return; + } + + /* Copy until buffer is full and flush it. */ + bytes_to_copy = sink->bbs_buffer_length - *bytes_done; + memcpy(sink->bbs_buffer + *bytes_done, data, bytes_to_copy); + data = ((char *) data) + bytes_to_copy; + length -= bytes_to_copy; + bbsink_archive_contents(sink, sink->bbs_buffer_length); + if (pg_checksum_update(checksum_ctx, (uint8 *) sink->bbs_buffer, + sink->bbs_buffer_length) < 0) + elog(ERROR, "could not update checksum"); + *bytes_done = 0; + } +} + /* * Try to verify the checksum for the provided page, if it seems appropriate * to do so. diff --git a/src/backend/backup/basebackup_incremental.c b/src/backend/backup/basebackup_incremental.c new file mode 100644 index 0000000000..b70eeb0282 --- /dev/null +++ b/src/backend/backup/basebackup_incremental.c @@ -0,0 +1,867 @@ +/*------------------------------------------------------------------------- + * + * basebackup_incremental.c + * code for incremental backup support + * + * This code isn't actually in charge of taking an incremental backup; + * the actual construction of the incremental backup happens in + * basebackup.c. Here, we're concerned with providing the necessary + * supports for that operation. In particular, we need to parse the + * backup manifest supplied by the user taking the incremental backup + * and extract the required information from it. + * + * Portions Copyright (c) 2010-2022, PostgreSQL Global Development Group + * + * IDENTIFICATION + * src/backend/backup/basebackup_incremental.c + * + *------------------------------------------------------------------------- + */ +#include "postgres.h" + +#include "access/timeline.h" +#include "access/xlog.h" +#include "access/xlogrecovery.h" +#include "backup/basebackup_incremental.h" +#include "backup/walsummary.h" +#include "common/blkreftable.h" +#include "common/parse_manifest.h" +#include "common/hashfn.h" +#include "postmaster/walsummarizer.h" + +#define BLOCKS_PER_READ 512 + +typedef struct +{ + TimeLineID tli; + XLogRecPtr start_lsn; + XLogRecPtr end_lsn; +} backup_wal_range; + +typedef struct +{ + uint32 status; + const char *path; + size_t size; +} backup_file_entry; + +static uint32 hash_string_pointer(const char *s); +#define SH_PREFIX backup_file +#define SH_ELEMENT_TYPE backup_file_entry +#define SH_KEY_TYPE const char * +#define SH_KEY path +#define SH_HASH_KEY(tb, key) hash_string_pointer(key) +#define SH_EQUAL(tb, a, b) (strcmp(a, b) == 0) +#define SH_SCOPE static inline +#define SH_DECLARE +#define SH_DEFINE +#include "lib/simplehash.h" + +struct IncrementalBackupInfo +{ + /* Memory context for this object and its subsidiary objects. */ + MemoryContext mcxt; + + /* Temporary buffer for storing the manifest while parsing it. */ + StringInfoData buf; + + /* WAL ranges extracted from the backup manifest. */ + List *manifest_wal_ranges; + + /* + * Files extracted from the backup manifest. + * + * We don't really need this information, because we use WAL summaries to + * figure what's changed. It would be unsafe to just rely on the list of + * files that existed before, because it's possible for a file to be + * removed and a new one created with the same name and different + * contents. In such cases, the whole file must still be sent. We can tell + * from the WAL summaries whether that happened, but not from the file + * list. + * + * Nonetheless, this data is useful for sanity checking. If a file that we + * think we shouldn't need to send is not present in the manifest for the + * prior backup, something has gone terribly wrong. We retain the file + * names and sizes, but not the checksums or last modified times, for + * which we have no use. + * + * One significant downside of storing this data is that it consumes + * memory. If that turns out to be a problem, we might have to decide not + * to retain this information, or to make it optional. + */ + backup_file_hash *manifest_files; + + /* + * Block-reference table for the incremental backup. + * + * It's possible that storing the entire block-reference table in memory + * will be a problem for some users. The in-memory format that we're using + * here is pretty efficient, converging to little more than 1 bit per + * block for relation forks with large numbers of modified blocks. It's + * possible, however, that if you try to perform an incremental backup of + * a database with a sufficiently large number of relations on a + * sufficiently small machine, you could run out of memory here. If that + * turns out to be a problem in practice, we'll need to be more clever. + */ + BlockRefTable *brtab; +}; + +static void manifest_process_file(JsonManifestParseContext *, + char *pathname, + size_t size, + pg_checksum_type checksum_type, + int checksum_length, + uint8 *checksum_payload); +static void manifest_process_wal_range(JsonManifestParseContext *, + TimeLineID tli, + XLogRecPtr start_lsn, + XLogRecPtr end_lsn); +static void manifest_report_error(JsonManifestParseContext *ib, + const char *fmt,...) + pg_attribute_printf(2, 3) pg_attribute_noreturn(); + +/* + * Create a new object for storing information extracted from the manifest + * supplied when creating an incremental backup. + */ +IncrementalBackupInfo * +CreateIncrementalBackupInfo(MemoryContext mcxt) +{ + IncrementalBackupInfo *ib; + MemoryContext oldcontext; + + oldcontext = MemoryContextSwitchTo(mcxt); + + ib = palloc0(sizeof(IncrementalBackupInfo)); + ib->mcxt = mcxt; + initStringInfo(&ib->buf); + + /* + * It's hard to guess how many files a "typical" installation will have in + * the data directory, but a fresh initdb creates almost 1000 files as of + * this writing, so it seems to make sense for our estimate to + * substantially higher. + */ + ib->manifest_files = backup_file_create(mcxt, 10000, NULL); + + MemoryContextSwitchTo(oldcontext); + + return ib; +} + +/* + * Before taking an incremental backup, the caller must supply the backup + * manifest from a prior backup. Each chunk of manifest data recieved + * from the client should be passed to this function. + */ +void +AppendIncrementalManifestData(IncrementalBackupInfo *ib, const char *data, + int len) +{ + MemoryContext oldcontext; + + /* Switch to our memory context. */ + oldcontext = MemoryContextSwitchTo(ib->mcxt); + + /* + * XXX. Our json parser is at present incapable of parsing json blobs + * incrementally, so we have to accumulate the entire backup manifest + * before we can do anything with it. This should really be fixed, since + * some users might have very large numbers of files in the data + * directory. + */ + appendBinaryStringInfo(&ib->buf, data, len); + + /* Switch back to previous memory context. */ + MemoryContextSwitchTo(oldcontext); +} + +/* + * Finalize an IncrementalBackupInfo object after all manifest data has + * been supplied via calls to AppendIncrementalManifestData. + */ +void +FinalizeIncrementalManifest(IncrementalBackupInfo *ib) +{ + JsonManifestParseContext context; + MemoryContext oldcontext; + + /* Switch to our memory context. */ + oldcontext = MemoryContextSwitchTo(ib->mcxt); + + /* Parse the manifest. */ + context.private_data = ib; + context.perfile_cb = manifest_process_file; + context.perwalrange_cb = manifest_process_wal_range; + context.error_cb = manifest_report_error; + json_parse_manifest(&context, ib->buf.data, ib->buf.len); + + /* Done with the buffer, so release memory. */ + pfree(ib->buf.data); + ib->buf.data = NULL; + + /* Switch back to previous memory context. */ + MemoryContextSwitchTo(oldcontext); +} + +/* + * Prepare to take an incremental backup. + * + * Before this function is called, AppendIncrementalManifestData and + * FinalizeIncrementalManifest should have already been called to pass all + * the manifest data to this object. + * + * This function performs sanity checks on the data extracted from the + * manifest and figures out for which WAL ranges we need summaries, and + * whether those summaries are available. Then, it reads and combines the + * data from those summary files. It also updates the backup_state with the + * reference TLI and LSN for the prior backup. + */ +void +PrepareForIncrementalBackup(IncrementalBackupInfo *ib, + BackupState *backup_state) +{ + MemoryContext oldcontext; + List *expectedTLEs; + List *all_wslist, + *required_wslist = NIL; + ListCell *lc; + TimeLineHistoryEntry **tlep; + int num_wal_ranges; + int i; + bool found_backup_start_tli = false; + TimeLineID earliest_wal_range_tli = 0; + XLogRecPtr earliest_wal_range_start_lsn; + TimeLineID latest_wal_range_tli = 0; + XLogRecPtr summarized_lsn; + + Assert(ib->buf.data == NULL); + + /* Switch to our memory context. */ + oldcontext = MemoryContextSwitchTo(ib->mcxt); + + /* + * Match up the TLIs that appear in the WAL ranges of the backup manifest + * with those that appear in this server's timeline history. We expect + * every backup_wal_range to match to a TimeLineHistoryEntry; if it does + * not, that's an error. + * + * This loop also decides which of the WAL ranges is the manifest is most + * ancient and which one is the newest, according to the timeline history + * of this server, and stores TLIs of those WAL ranges into + * earliest_wal_range_tli and latest_wal_range_tli. It also updates + * earliest_wal_range_start_lsn to the start LSN of the WAL range for + * earliest_wal_range_tli. + * + * Note that the return value of readTimeLineHistory puts the latest + * timeline at the beginning of the list, not the end. Hence, the earliest + * TLI is the one that occurs nearest the end of the list returned by + * readTimeLineHistory, and the latest TLI is the one that occurs closest + * to the beginning. + */ + expectedTLEs = readTimeLineHistory(backup_state->starttli); + num_wal_ranges = list_length(ib->manifest_wal_ranges); + tlep = palloc0(num_wal_ranges * sizeof(TimeLineHistoryEntry *)); + for (i = 0; i < num_wal_ranges; ++i) + { + backup_wal_range *range = list_nth(ib->manifest_wal_ranges, i); + bool saw_earliest_wal_range_tli = false; + bool saw_latest_wal_range_tli = false; + + /* Search this server's history for this WAL range's TLI. */ + foreach(lc, expectedTLEs) + { + TimeLineHistoryEntry *tle = lfirst(lc); + + if (tle->tli == range->tli) + { + tlep[i] = tle; + break; + } + + if (tle->tli == earliest_wal_range_tli) + saw_earliest_wal_range_tli = true; + if (tle->tli == latest_wal_range_tli) + saw_latest_wal_range_tli = true; + } + + /* + * An incremental backup can only be taken relative to a backup that + * represents a previous state of this server. If the backup requires + * WAL from a timeline that's not in our history, that definitely + * isn't the case. + */ + if (tlep[i] == NULL) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("timeline %u found in manifest, but not in this server's history", + range->tli))); + + /* + * If we found this TLI in the server's history before encountering + * the latest TLI seen so far in the server's history, then this TLI + * is the latest one seen so far. + * + * If on the other hand we saw the earliest TLI seen so far before + * finding this TLI, this TLI is earlier than the earliest one seen so + * far. And if this is the first TLI for which we've searched, it's + * also the earliest one seen so far. + * + * On the first loop iteration, both things should necessarily be + * true. + */ + if (!saw_latest_wal_range_tli) + latest_wal_range_tli = range->tli; + if (earliest_wal_range_tli == 0 || saw_earliest_wal_range_tli) + { + earliest_wal_range_tli = range->tli; + earliest_wal_range_start_lsn = range->start_lsn; + } + } + + /* + * Propagate information about the prior backup into the backup_label that + * will be generated for this backup. + */ + backup_state->istartpoint = earliest_wal_range_start_lsn; + backup_state->istarttli = earliest_wal_range_tli; + + /* + * Sanity check start and end LSNs for the WAL ranges in the manifest. + * + * Commonly, there won't be any timeline switches during the prior backup + * at all, but if there are, they should happen at the same LSNs that this + * server switched timelines. + * + * Whether there are any timeline switches during the prior backup or not, + * the prior backup shouldn't require any WAL from a timeline prior to the + * start of that timeline. It also shouldn't require any WAL from later + * than the start of this backup. + * + * If any of these sanity checks fail, one possible explanation is that + * the user has generated WAL on the same timeline with the same LSNs more + * than once. For instance, if two standbys running on timeline 1 were + * both promoted and (due to a broken archiving setup) both selected new + * timeline ID 2, then it's possible that one of these checks might trip. + * + * Note that there are lots of ways for the user to do something very bad + * without tripping any of these checks, and they are not intended to be + * comprehensive. It's pretty hard to see how we could be certain of + * anything here. However, if there's a problem staring us right in the + * face, it's best to report it, so we do. + */ + for (i = 0; i < num_wal_ranges; ++i) + { + backup_wal_range *range = list_nth(ib->manifest_wal_ranges, i); + + if (range->tli == earliest_wal_range_tli) + { + if (range->start_lsn < tlep[i]->begin) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("manifest requires WAL from initial timeline %u starting at %X/%X, but that timeline begins at %X/%X", + range->tli, + LSN_FORMAT_ARGS(range->start_lsn), + LSN_FORMAT_ARGS(tlep[i]->begin)))); + } + else + { + if (range->start_lsn != tlep[i]->begin) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("manifest requires WAL from continuation timeline %u starting at %X/%X, but that timeline begins at %X/%X", + range->tli, + LSN_FORMAT_ARGS(range->start_lsn), + LSN_FORMAT_ARGS(tlep[i]->begin)))); + } + + if (range->tli == latest_wal_range_tli) + { + if (range->end_lsn > backup_state->startpoint) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("manifest requires WAL from final timeline %u ending at %X/%X, but this backup starts at %X/%X", + range->tli, + LSN_FORMAT_ARGS(range->end_lsn), + LSN_FORMAT_ARGS(backup_state->startpoint)))); + } + else + { + if (range->end_lsn != tlep[i]->end) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("manifest requires WAL from non-final timeline %u ending at %X/%X, but this server switched timelines at %X/%X", + range->tli, + LSN_FORMAT_ARGS(range->end_lsn), + LSN_FORMAT_ARGS(tlep[i]->end)))); + } + + } + + /* + * Wait for WAL summarization to catch up to the backup start LSN (but + * time out if it doesn't do so quickly enough). + */ + /* XXX make timeout configurable */ + summarized_lsn = WaitForWalSummarization(backup_state->startpoint, 60000); + if (summarized_lsn < backup_state->startpoint) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("timeout waiting for WAL summarization"), + errdetail("This backup requires WAL to be summarized up to %X/%X, but summarizer has only reached %X/%X.", + LSN_FORMAT_ARGS(backup_state->startpoint), + LSN_FORMAT_ARGS(summarized_lsn)))); + + /* + * Retrieve a list of all WAL summaries on any timeline that overlap with + * the LSN range of interest. We could instead call GetWalSummaries() once + * per timeline in the loop that follows, but that would involve reading + * the directory multiple times. It should be mildly faster - and perhaps + * a bit safer - to do it just once. + */ + all_wslist = GetWalSummaries(0, earliest_wal_range_start_lsn, + backup_state->startpoint); + + /* + * We need WAL summaries for everything that happened during the prior + * backup and everything that happened afterward up until the point where + * the current backup started. + */ + foreach(lc, expectedTLEs) + { + TimeLineHistoryEntry *tle = lfirst(lc); + XLogRecPtr tli_start_lsn = tle->begin; + XLogRecPtr tli_end_lsn = tle->end; + XLogRecPtr tli_missing_lsn = InvalidXLogRecPtr; + List *tli_wslist; + + /* + * Working through the history of this server from the current + * timeline backwards, we skip everything until we find the timeline + * where this backup started. Most of the time, this means we won't + * skip anything at all, as it's unlikely that the timeline has + * changed since the beginning of the backup moments ago. + */ + if (tle->tli == backup_state->starttli) + { + found_backup_start_tli = true; + tli_end_lsn = backup_state->startpoint; + } + else if (!found_backup_start_tli) + continue; + + /* + * Find the summaries that overlap the LSN range of interest for this + * timeline. If this is the earliest timeline involved, the range of + * interest begins with the start LSN of the prior backup; otherwise, + * it begins at the LSN at which this timeline came into existence. If + * this is the latest TLI involved, the range of interest ends at the + * start LSN of the current backup; otherwise, it ends at the point + * where we switched from this timeline to the next one. + */ + if (tle->tli == earliest_wal_range_tli) + tli_start_lsn = earliest_wal_range_start_lsn; + tli_wslist = FilterWalSummaries(all_wslist, tle->tli, + tli_start_lsn, tli_end_lsn); + + /* + * There is no guarantee that the WAL summaries we found cover the + * entire range of LSNs for which summaries are required, or indeed + * that we found any WAL summaries at all. Check whether we have a + * problem of that sort. + */ + if (!WalSummariesAreComplete(tli_wslist, tli_start_lsn, tli_end_lsn, + &tli_missing_lsn)) + { + if (XLogRecPtrIsInvalid(tli_missing_lsn)) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("WAL summaries are required on timeline %u from %X/%X to %X/%X, but no summaries for that timeline and LSN range exist", + tle->tli, + LSN_FORMAT_ARGS(tli_start_lsn), + LSN_FORMAT_ARGS(tli_end_lsn)))); + else + ereport(ERROR, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("WAL summaries are required on timeline %u from %X/%X to %X/%X, but the summaries for that timeline and LSN range are incomplete", + tle->tli, + LSN_FORMAT_ARGS(tli_start_lsn), + LSN_FORMAT_ARGS(tli_end_lsn)), + errdetail("The first unsummarized LSN is this range is %X/%X.", + LSN_FORMAT_ARGS(tli_missing_lsn)))); + } + + /* + * Remember that we need to read these summaries. + * + * Technically, it's possible that this could read more files than + * required, since tli_wslist in theory could contain redundant + * summaries. For instance, if we have a summary from 0/10000000 to + * 0/20000000 and also one from 0/00000000 to 0/30000000, then the + * latter subsumes the former and the former could be ignored. + * + * We ignore this possibility because the WAL summarizer only tries to + * generate summaries that do not overlap. If somehow they exist, + * we'll do a bit of extra work but the results should still be + * correct. + */ + required_wslist = list_concat(required_wslist, tli_wslist); + + /* + * Timelines earlier than the one in which the prior backup began are + * not relevant. + */ + if (tle->tli == earliest_wal_range_tli) + break; + } + + /* + * Read all of the required block reference table files and merge all of + * the data into a single in-memory block reference table. + * + * See the comments for struct IncrementalBackupInfo for some thoughts on + * memory usage. + */ + ib->brtab = CreateEmptyBlockRefTable(); + foreach(lc, required_wslist) + { + WalSummaryFile *ws = lfirst(lc); + WalSummaryIO wsio; + BlockRefTableReader *reader; + RelFileLocator rlocator; + ForkNumber forknum; + BlockNumber limit_block; + BlockNumber blocks[BLOCKS_PER_READ]; + + wsio.file = OpenWalSummaryFile(ws, false); + wsio.filepos = 0; + ereport(DEBUG1, + (errmsg_internal("reading WAL summary file \"%s\"", + FilePathName(wsio.file)))); + reader = CreateBlockRefTableReader(ReadWalSummary, &wsio, + FilePathName(wsio.file), + ReportWalSummaryError, NULL); + while (BlockRefTableReaderNextRelation(reader, &rlocator, &forknum, + &limit_block)) + { + BlockRefTableSetLimitBlock(ib->brtab, &rlocator, + forknum, limit_block); + + while (1) + { + unsigned nblocks; + unsigned i; + + nblocks = BlockRefTableReaderGetBlocks(reader, blocks, + BLOCKS_PER_READ); + if (nblocks == 0) + break; + + for (i = 0; i < nblocks; ++i) + BlockRefTableMarkBlockModified(ib->brtab, &rlocator, + forknum, blocks[i]); + } + } + DestroyBlockRefTableReader(reader); + FileClose(wsio.file); + } + + /* Switch back to previous memory context. */ + MemoryContextSwitchTo(oldcontext); +} + +/* + * Get the pathname that should be used when a file is sent incrementally. + * + * The result is a palloc'd string. + */ +char * +GetIncrementalFilePath(Oid dboid, Oid spcoid, RelFileNumber relfilenumber, + ForkNumber forknum, unsigned segno) +{ + char *path; + char *lastslash; + char *ipath; + + path = GetRelationPath(dboid, spcoid, relfilenumber, InvalidBackendId, + forknum); + + lastslash = strrchr(path, '/'); + Assert(lastslash != NULL); + *lastslash = '\0'; + + if (segno > 0) + ipath = psprintf("%s/INCREMENTAL.%s.%u", path, lastslash + 1, segno); + else + ipath = psprintf("%s/INCREMENTAL.%s", path, lastslash + 1); + + pfree(path); + + return ipath; +} + +/* + * How should we back up a particular file as part of an incremental backup? + * + * If the return value is BACK_UP_FILE_FULLY, caller should back up the whole + * file just as if this were not an incremental backup. + * + * If the return value is BACK_UP_FILE_INCREMENTALLY, caller should include + * an incremental file in the backup instead of the entire file. On return, + * *num_blocks_required will be set to the number of blocks that need to be + * sent, and the actual block numbers will have been stored in + * relative_block_numbers, which should be an array of at least RELSEG_SIZE. + * In addition, *truncation_block_length will be set to the value that should + * be included in the incremental file. + * + * If the return value is DO_NOT_BACK_UP_FILE, the caller should not include + * the file in the backup at all. + */ +FileBackupMethod +GetFileBackupMethod(IncrementalBackupInfo *ib, char *path, + Oid dboid, Oid spcoid, + RelFileNumber relfilenumber, ForkNumber forknum, + unsigned segno, size_t size, + unsigned *num_blocks_required, + BlockNumber *relative_block_numbers, + unsigned *truncation_block_length) +{ + BlockNumber absolute_block_numbers[RELSEG_SIZE]; + BlockNumber limit_block; + BlockNumber start_blkno; + BlockNumber stop_blkno; + RelFileLocator rlocator; + BlockRefTableEntry *brtentry; + unsigned i; + unsigned nblocks; + + /* Should only be called after PrepareForIncrementalBackup. */ + Assert(ib->buf.data == NULL); + + /* + * dboid could be InvalidOid if shared rel, but spcoid and relfilenumber + * should have legal values. + */ + Assert(OidIsValid(spcoid)); + Assert(RelFileNumberIsValid(relfilenumber)); + + /* + * If the file size is too large or not a multiple of BLCKSZ, then + * something weird is happening, so give up and send the whole file. + */ + if ((size % BLCKSZ) != 0 || size / BLCKSZ > RELSEG_SIZE) + return BACK_UP_FILE_FULLY; + + /* + * The free-space map fork is not properly WAL-logged, so we need to + * backup the entire file every time. + */ + if (forknum == FSM_FORKNUM) + return BACK_UP_FILE_FULLY; + + /* + * Check whether this file is part of the prior backup. If it isn't, back + * up the whole file. + */ + if (backup_file_lookup(ib->manifest_files, path) == NULL) + { + char *ipath; + + ipath = GetIncrementalFilePath(dboid, spcoid, relfilenumber, + forknum, segno); + if (backup_file_lookup(ib->manifest_files, ipath) == NULL) + return BACK_UP_FILE_FULLY; + } + + /* Look up the block reference table entry. */ + rlocator.spcOid = spcoid; + rlocator.dbOid = dboid; + rlocator.relNumber = relfilenumber; + brtentry = BlockRefTableGetEntry(ib->brtab, &rlocator, forknum, + &limit_block); + + /* + * If there is no entry, then there have been no WAL-logged changes to the + * relation since the predecessor backup was taken, so we can back it up + * incrementally and need not include any modified blocks. + * + * However, if the file is zero-length, we should do a full backup, + * because an incremental file is always more than zero length, and it's + * silly to take an incremental backup when a full backup would be + * smaller. + */ + if (brtentry == NULL) + { + *num_blocks_required = 0; + *truncation_block_length = size / BLCKSZ; + if (size == 0) + return BACK_UP_FILE_FULLY; + return BACK_UP_FILE_INCREMENTALLY; + } + + /* + * If the limit_block is less than or equal to the point where this + * segment starts, send the whole file. + */ + if (limit_block <= segno * RELSEG_SIZE) + return BACK_UP_FILE_FULLY; + + /* + * Get relevant entries from the block reference table entry. + * + * We shouldn't overflow computing the start or stop block numbers, but if + * it manages to happen somehow, detect it and throw an error. + */ + start_blkno = segno * RELSEG_SIZE; + stop_blkno = start_blkno + (size / BLCKSZ); + if (start_blkno / RELSEG_SIZE != segno || stop_blkno < start_blkno) + ereport(ERROR, + errcode(ERRCODE_INTERNAL_ERROR), + errmsg_internal("overflow computing block number bounds for segment %u with size %lu", + segno, size)); + nblocks = BlockRefTableEntryGetBlocks(brtentry, start_blkno, stop_blkno, + absolute_block_numbers, RELSEG_SIZE); + Assert(nblocks <= RELSEG_SIZE); + + /* + * If we're going to have to send nearly all of the blocks, then just send + * the whole file, because that won't require much extra storage or + * transfer and will speed up and simplify backup restoration. It's not + * clear what threshold is most appropriate here and perhaps it ought to + * be configurable, but for now we're just going to say that if we'd need + * to send 90% of the blocks anyway, give up and send the whole file. + * + * NB: If you change the threshold here, at least make sure to back up the + * file fully when every single block must be sent, because there's + * nothing good about sending an incremental file in that case. + */ + if (nblocks * BLCKSZ > size * 0.9) + return BACK_UP_FILE_FULLY; + + /* + * Looks like we can send an incremental file. + * + * Return the relevant details to the caller, transposing absolute block + * numbers to relative block numbers. + * + * The truncation block length is the minimum length of the reconstructed + * file. Any block numbers below this threshold that are not present in + * the backup need to be fetched from the prior backup. At or above this + * threshold, blocks should only be included in the result if they are + * present in the backup. (This may require inserting zero blocks if the + * blocks included in the backup are non-consecutive.) + */ + for (i = 0; i < nblocks; ++i) + relative_block_numbers[i] = absolute_block_numbers[i] - start_blkno; + *num_blocks_required = nblocks; + *truncation_block_length = + Min(size / BLCKSZ, limit_block - segno * RELSEG_SIZE); + return BACK_UP_FILE_INCREMENTALLY; +} + +/* + * Compute the size for an incremental file containing a given number of blocks. + */ +extern size_t +GetIncrementalFileSize(unsigned num_blocks_required) +{ + size_t result; + + /* Make sure we're not going to overflow. */ + Assert(num_blocks_required <= RELSEG_SIZE); + + /* + * Three four byte quantities (magic number, truncation block length, + * block count) followed by block numbers followed by block contents. + */ + result = 3 * sizeof(uint32); + result += (BLCKSZ + sizeof(BlockNumber)) * num_blocks_required; + + return result; +} + +/* + * Helper function for filemap hash table. + */ +static uint32 +hash_string_pointer(const char *s) +{ + unsigned char *ss = (unsigned char *) s; + + return hash_bytes(ss, strlen(s)); +} + +/* + * This callback is invoked for each file mentioned in the backup manifest. + * + * We store the path to each file and the size of each file for sanity-checking + * purposes. For further details, see comments for IncrementalBackupInfo. + */ +static void +manifest_process_file(JsonManifestParseContext *context, + char *pathname, size_t size, + pg_checksum_type checksum_type, + int checksum_length, + uint8 *checksum_payload) +{ + IncrementalBackupInfo *ib = context->private_data; + backup_file_entry *entry; + bool found; + + entry = backup_file_insert(ib->manifest_files, pathname, &found); + if (!found) + { + entry->path = MemoryContextStrdup(ib->manifest_files->ctx, + pathname); + entry->size = size; + } +} + +/* + * This callback is invoked for each WAL range mentioned in the backup + * manifest. + * + * We're just interested in learning the oldest LSN and the corresponding TLI + * that appear in any WAL range. + */ +static void +manifest_process_wal_range(JsonManifestParseContext *context, + TimeLineID tli, XLogRecPtr start_lsn, + XLogRecPtr end_lsn) +{ + IncrementalBackupInfo *ib = context->private_data; + backup_wal_range *range = palloc(sizeof(backup_wal_range)); + + range->tli = tli; + range->start_lsn = start_lsn; + range->end_lsn = end_lsn; + ib->manifest_wal_ranges = lappend(ib->manifest_wal_ranges, range); +} + +/* + * This callback is invoked if an error occurs while parsing the backup + * manifest. + */ +static void +manifest_report_error(JsonManifestParseContext *context, const char *fmt,...) +{ + StringInfoData errbuf; + + initStringInfo(&errbuf); + + for (;;) + { + va_list ap; + int needed; + + va_start(ap, fmt); + needed = appendStringInfoVA(&errbuf, fmt, ap); + va_end(ap); + if (needed == 0) + break; + enlargeStringInfo(&errbuf, needed); + } + + ereport(ERROR, + errmsg_internal("%s", errbuf.data)); +} diff --git a/src/backend/backup/meson.build b/src/backend/backup/meson.build index 11a79bbf80..19c355ceca 100644 --- a/src/backend/backup/meson.build +++ b/src/backend/backup/meson.build @@ -5,6 +5,7 @@ backend_sources += files( 'basebackup.c', 'basebackup_copy.c', 'basebackup_gzip.c', + 'basebackup_incremental.c', 'basebackup_lz4.c', 'basebackup_progress.c', 'basebackup_server.c', @@ -12,4 +13,6 @@ backend_sources += files( 'basebackup_target.c', 'basebackup_throttle.c', 'basebackup_zstd.c', + 'walsummary.c', + 'walsummaryfuncs.c' ) diff --git a/src/backend/backup/walsummary.c b/src/backend/backup/walsummary.c new file mode 100644 index 0000000000..ebf4ea038d --- /dev/null +++ b/src/backend/backup/walsummary.c @@ -0,0 +1,356 @@ +/*------------------------------------------------------------------------- + * + * walsummary.c + * Functions for accessing and managing WAL summary data. + * + * Portions Copyright (c) 2010-2022, PostgreSQL Global Development Group + * + * src/backend/backup/walsummary.c + * + *------------------------------------------------------------------------- + */ + +#include "postgres.h" + +#include <sys/stat.h> +#include <unistd.h> + +#include "access/xlog_internal.h" +#include "backup/walsummary.h" +#include "utils/wait_event.h" + +static bool IsWalSummaryFilename(char *filename); +static int ListComparatorForWalSummaryFiles(const ListCell *a, + const ListCell *b); + +/* + * Get a list of WAL summaries. + * + * If tli != 0, only WAL summaries with the indicated TLI will be included. + * + * If start_lsn != InvalidXLogRecPtr, only summaries that end before the + * indicated LSN will be included. + * + * If end_lsn != InvalidXLogRecPtr, only summaries that start before the + * indicated LSN will be included. + * + * The intent is that you can call GetWalSummaries(tli, start_lsn, end_lsn) + * to get all WAL summaries on the indicated timeline that overlap the + * specified LSN range. + */ +List * +GetWalSummaries(TimeLineID tli, XLogRecPtr start_lsn, XLogRecPtr end_lsn) +{ + DIR *sdir; + struct dirent *dent; + List *result = NIL; + + sdir = AllocateDir(XLOGDIR "/summaries"); + while ((dent = ReadDir(sdir, XLOGDIR "/summaries")) != NULL) + { + WalSummaryFile *ws; + uint32 tmp[5]; + TimeLineID file_tli; + XLogRecPtr file_start_lsn; + XLogRecPtr file_end_lsn; + + /* Decode filename, or skip if it's not in the expected format. */ + if (!IsWalSummaryFilename(dent->d_name)) + continue; + sscanf(dent->d_name, "%08X%08X%08X%08X%08X", + &tmp[0], &tmp[1], &tmp[2], &tmp[3], &tmp[4]); + file_tli = tmp[0]; + file_start_lsn = ((uint64) tmp[1]) << 32 | tmp[2]; + file_end_lsn = ((uint64) tmp[3]) << 32 | tmp[4]; + + /* Skip if it doesn't match the filter criteria. */ + if (tli != 0 && tli != file_tli) + continue; + if (!XLogRecPtrIsInvalid(start_lsn) && start_lsn > file_end_lsn) + continue; + if (!XLogRecPtrIsInvalid(end_lsn) && end_lsn < file_start_lsn) + continue; + + /* Add it to the list. */ + ws = palloc(sizeof(WalSummaryFile)); + ws->tli = file_tli; + ws->start_lsn = file_start_lsn; + ws->end_lsn = file_end_lsn; + result = lappend(result, ws); + } + FreeDir(sdir); + + return result; +} + +/* + * Build a new list of WAL summaries based on an existing list, but filtering + * out summaries that don't match the search parameters. + * + * If tli != 0, only WAL summaries with the indicated TLI will be included. + * + * If start_lsn != InvalidXLogRecPtr, only summaries that end before the + * indicated LSN will be included. + * + * If end_lsn != InvalidXLogRecPtr, only summaries that start before the + * indicated LSN will be included. + */ +List * +FilterWalSummaries(List *wslist, TimeLineID tli, + XLogRecPtr start_lsn, XLogRecPtr end_lsn) +{ + List *result = NIL; + ListCell *lc; + + /* Loop over input. */ + foreach(lc, wslist) + { + WalSummaryFile *ws = lfirst(lc); + + /* Skip if it doesn't match the filter criteria. */ + if (tli != 0 && tli != ws->tli) + continue; + if (!XLogRecPtrIsInvalid(start_lsn) && start_lsn > ws->end_lsn) + continue; + if (!XLogRecPtrIsInvalid(end_lsn) && end_lsn < ws->start_lsn) + continue; + + /* Add it to the result list. */ + result = lappend(result, ws); + } + + return result; +} + +/* + * Check whether the supplied list of WalSummaryFile objects covers the + * whole range of LSNs from start_lsn to end_lsn. This function ignores + * timelines, so the caller should probably filter using the appropriate + * timeline before calling this. + * + * If the whole range of LSNs is covered, returns true, otherwise false. + * If false is returned, *missing_lsn is set either to InvalidXLogRecPtr + * if there are no WAL summary files in the input list, or to the first LSN + * in the range that is not covered by a WAL summary file in the input list. + */ +bool +WalSummariesAreComplete(List *wslist, XLogRecPtr start_lsn, + XLogRecPtr end_lsn, XLogRecPtr *missing_lsn) +{ + XLogRecPtr current_lsn = start_lsn; + ListCell *lc; + + /* Special case for empty list. */ + if (wslist == NIL) + { + *missing_lsn = InvalidXLogRecPtr; + return false; + } + + /* Make a private copy of the list and sort it by start LSN. */ + wslist = list_copy(wslist); + list_sort(wslist, ListComparatorForWalSummaryFiles); + + /* + * Consider summary files in order of increasing start_lsn, advancing the + * known-summarized range from start_lsn toward end_lsn. + * + * Normally, the summary files should cover non-overlapping WAL ranges, + * but this algorithm is intended to be correct even in case of overlap. + */ + foreach(lc, wslist) + { + WalSummaryFile *ws = lfirst(lc); + + if (ws->start_lsn > current_lsn) + { + /* We found a gap. */ + break; + } + if (ws->end_lsn > current_lsn) + { + /* + * Next summary extends beyond end of previous summary, so extend + * the end of the range known to be summarized. + */ + current_lsn = ws->end_lsn; + + /* + * If the range we know to be summarized has reached the required + * end LSN, we have proved completeness. + */ + if (current_lsn >= end_lsn) + return true; + } + } + + /* + * We either ran out of summary files without reaching the end LSN, or we + * hit a gap in the sequence that resulted in us bailing out of the loop + * above. + */ + *missing_lsn = current_lsn; + return false; +} + +/* + * Open a WAL summary file. + * + * This will throw an error in case of trouble. As an exception, if + * missing_ok = true and the trouble is specifically that the file does + * not exist, it will not throw an error and will return a value less than 0. + */ +File +OpenWalSummaryFile(WalSummaryFile *ws, bool missing_ok) +{ + char path[MAXPGPATH]; + File file; + + snprintf(path, MAXPGPATH, + XLOGDIR "/summaries/%08X%08X%08X%08X%08X.summary", + ws->tli, + LSN_FORMAT_ARGS(ws->start_lsn), + LSN_FORMAT_ARGS(ws->end_lsn)); + + file = PathNameOpenFile(path, O_RDONLY); + if (file < 0 && (errno != EEXIST || !missing_ok)) + ereport(ERROR, + (errcode_for_file_access(), + errmsg("could not open file \"%s\": %m", path))); + + return file; +} + +/* + * Remove a WAL summary file if the last modification time precedes the + * cutoff time. + */ +void +RemoveWalSummaryIfOlderThan(WalSummaryFile *ws, time_t cutoff_time) +{ + char path[MAXPGPATH]; + struct stat statbuf; + + snprintf(path, MAXPGPATH, + XLOGDIR "/summaries/%08X%08X%08X%08X%08X.summary", + ws->tli, + LSN_FORMAT_ARGS(ws->start_lsn), + LSN_FORMAT_ARGS(ws->end_lsn)); + + if (lstat(path, &statbuf) != 0) + { + if (errno == ENOENT) + return; + ereport(ERROR, + (errcode_for_file_access(), + errmsg("could not stat file \"%s\": %m", path))); + } + if (statbuf.st_mtime >= cutoff_time) + return; + if (unlink(path) != 0) + ereport(ERROR, + (errcode_for_file_access(), + errmsg("could not stat file \"%s\": %m", path))); + ereport(DEBUG2, + (errmsg_internal("removing file \"%s\"", path))); +} + +/* + * Test whether a filename looks like a WAL summary file. + */ +static bool +IsWalSummaryFilename(char *filename) +{ + return strspn(filename, "0123456789ABCDEF") == 40 && + strcmp(filename + 40, ".summary") == 0; +} + +/* + * Data read callback for use with CreateBlockRefTableReader. + */ +int +ReadWalSummary(void *wal_summary_io, void *data, int length) +{ + WalSummaryIO *io = wal_summary_io; + int nbytes; + + nbytes = FileRead(io->file, data, length, io->filepos, + WAIT_EVENT_WAL_SUMMARY_READ); + if (nbytes < 0) + ereport(ERROR, + (errcode_for_file_access(), + errmsg("could not write file \"%s\": %m", + FilePathName(io->file)))); + + io->filepos += nbytes; + return nbytes; +} + +/* + * Data write callback for use with WriteBlockRefTable. + */ +int +WriteWalSummary(void *wal_summary_io, void *data, int length) +{ + WalSummaryIO *io = wal_summary_io; + int nbytes; + + nbytes = FileWrite(io->file, data, length, io->filepos, + WAIT_EVENT_WAL_SUMMARY_WRITE); + if (nbytes < 0) + ereport(ERROR, + (errcode_for_file_access(), + errmsg("could not write file \"%s\": %m", + FilePathName(io->file)))); + if (nbytes != length) + ereport(ERROR, + (errcode_for_file_access(), + errmsg("could not write file \"%s\": wrote only %d of %d bytes at offset %u", + FilePathName(io->file), nbytes, + length, (unsigned) io->filepos), + errhint("Check free disk space."))); + + io->filepos += nbytes; + return nbytes; +} + +/* + * Error-reporting callback for use with CreateBlockRefTableReader. + */ +void +ReportWalSummaryError(void *callback_arg, char *fmt,...) +{ + StringInfoData buf; + va_list ap; + int needed; + + initStringInfo(&buf); + for (;;) + { + va_start(ap, fmt); + needed = appendStringInfoVA(&buf, fmt, ap); + va_end(ap); + if (needed == 0) + break; + enlargeStringInfo(&buf, needed); + } + ereport(ERROR, + errcode(ERRCODE_DATA_CORRUPTED), + errmsg_internal("%s", buf.data)); +} + +/* + * Comparator to sort a List of WalSummaryFile objects by start_lsn. + */ +static int +ListComparatorForWalSummaryFiles(const ListCell *a, const ListCell *b) +{ + WalSummaryFile *ws1 = lfirst(a); + WalSummaryFile *ws2 = lfirst(b); + + if (ws1->start_lsn < ws2->start_lsn) + return -1; + if (ws1->start_lsn > ws2->start_lsn) + return 1; + return 0; +} diff --git a/src/backend/backup/walsummaryfuncs.c b/src/backend/backup/walsummaryfuncs.c new file mode 100644 index 0000000000..2e77d38b4a --- /dev/null +++ b/src/backend/backup/walsummaryfuncs.c @@ -0,0 +1,169 @@ +/*------------------------------------------------------------------------- + * + * walsummaryfuncs.c + * SQL-callable functions for accessing WAL summary data. + * + * Portions Copyright (c) 2010-2022, PostgreSQL Global Development Group + * + * src/backend/backup/walsummaryfuncs.c + * + *------------------------------------------------------------------------- + */ + +#include "postgres.h" + +#include "backup/walsummary.h" +#include "common/blkreftable.h" +#include "funcapi.h" +#include "miscadmin.h" +#include "utils/fmgrprotos.h" +#include "utils/pg_lsn.h" + +#define NUM_WS_ATTS 3 +#define NUM_SUMMARY_ATTS 6 +#define MAX_BLOCKS_PER_CALL 256 + +/* + * List the WAL summary files available in pg_wal/summaries. + */ +Datum +pg_available_wal_summaries(PG_FUNCTION_ARGS) +{ + ReturnSetInfo *rsi; + List *wslist; + ListCell *lc; + Datum values[NUM_WS_ATTS]; + bool nulls[NUM_WS_ATTS]; + + InitMaterializedSRF(fcinfo, 0); + rsi = (ReturnSetInfo *) fcinfo->resultinfo; + + memset(nulls, 0, sizeof(nulls)); + + wslist = GetWalSummaries(0, InvalidXLogRecPtr, InvalidXLogRecPtr); + foreach(lc, wslist) + { + WalSummaryFile *ws = (WalSummaryFile *) lfirst(lc); + HeapTuple tuple; + + CHECK_FOR_INTERRUPTS(); + + values[0] = Int64GetDatum((int64) ws->tli); + values[1] = LSNGetDatum(ws->start_lsn); + values[2] = LSNGetDatum(ws->end_lsn); + + tuple = heap_form_tuple(rsi->setDesc, values, nulls); + tuplestore_puttuple(rsi->setResult, tuple); + } + + return (Datum) 0; +} + +/* + * List the contents of a WAL summary file identified by TLI, start LSN, + * and end LSN. + */ +Datum +pg_wal_summary_contents(PG_FUNCTION_ARGS) +{ + ReturnSetInfo *rsi; + Datum values[NUM_SUMMARY_ATTS]; + bool nulls[NUM_SUMMARY_ATTS]; + WalSummaryFile ws; + WalSummaryIO io; + BlockRefTableReader *reader; + int64 raw_tli; + RelFileLocator rlocator; + ForkNumber forknum; + BlockNumber limit_block; + + InitMaterializedSRF(fcinfo, 0); + rsi = (ReturnSetInfo *) fcinfo->resultinfo; + memset(nulls, 0, sizeof(nulls)); + + /* + * Since the timeline could at least in theory be more than 2^31, and + * since we don't have unsigned types at the SQL level, it is passed as a + * 64-bit integer. Test whether it's out of range. + */ + raw_tli = PG_GETARG_INT64(0); + if (raw_tli < 1 || raw_tli > PG_INT32_MAX) + ereport(ERROR, + errcode(ERRCODE_INVALID_PARAMETER_VALUE), + errmsg("invalid timeline %lld", (long long) raw_tli)); + + /* Prepare to read the specified WAL summry file. */ + ws.tli = (TimeLineID) raw_tli; + ws.start_lsn = PG_GETARG_LSN(1); + ws.end_lsn = PG_GETARG_LSN(2); + io.filepos = 0; + io.file = OpenWalSummaryFile(&ws, false); + reader = CreateBlockRefTableReader(ReadWalSummary, &io, + FilePathName(io.file), + ReportWalSummaryError, NULL); + + /* Loop over relation forks. */ + while (BlockRefTableReaderNextRelation(reader, &rlocator, &forknum, + &limit_block)) + { + BlockNumber blocks[MAX_BLOCKS_PER_CALL]; + HeapTuple tuple; + + CHECK_FOR_INTERRUPTS(); + + values[0] = ObjectIdGetDatum(rlocator.relNumber); + values[1] = ObjectIdGetDatum(rlocator.spcOid); + values[2] = ObjectIdGetDatum(rlocator.dbOid); + values[3] = Int16GetDatum((int16) forknum); + + /* Loop over blocks within the current relation fork. */ + while (true) + { + unsigned nblocks; + unsigned i; + + CHECK_FOR_INTERRUPTS(); + + nblocks = BlockRefTableReaderGetBlocks(reader, blocks, + MAX_BLOCKS_PER_CALL); + if (nblocks == 0) + break; + + /* + * For each block that we specifically know to have been modified, + * emit a row with that block number and limit_block = false. + */ + values[5] = BoolGetDatum(false); + for (i = 0; i < nblocks; ++i) + { + values[4] = Int64GetDatum((int64) blocks[i]); + + tuple = heap_form_tuple(rsi->setDesc, values, nulls); + tuplestore_puttuple(rsi->setResult, tuple); + } + + /* + * If the limit block is not InvalidBlockNumber, emit an exta row + * with that block number and limit_block = true. + * + * There is no point in doing this when the limit_block is + * InvalidBlockNumber, because no block with that number or any + * higher number can ever exist. + */ + if (BlockNumberIsValid(limit_block)) + { + values[4] = Int64GetDatum((int64) limit_block); + values[5] = BoolGetDatum(true); + + tuple = heap_form_tuple(rsi->setDesc, values, nulls); + tuplestore_puttuple(rsi->setResult, tuple); + } + } + } + + /* Cleanup */ + DestroyBlockRefTableReader(reader); + FileClose(io.file); + + return (Datum) 0; +} diff --git a/src/backend/postmaster/Makefile b/src/backend/postmaster/Makefile index 047448b34e..367a46c617 100644 --- a/src/backend/postmaster/Makefile +++ b/src/backend/postmaster/Makefile @@ -24,6 +24,7 @@ OBJS = \ postmaster.o \ startup.o \ syslogger.o \ + walsummarizer.o \ walwriter.o include $(top_srcdir)/src/backend/common.mk diff --git a/src/backend/postmaster/auxprocess.c b/src/backend/postmaster/auxprocess.c index cae6feb356..0c15c1777d 100644 --- a/src/backend/postmaster/auxprocess.c +++ b/src/backend/postmaster/auxprocess.c @@ -21,6 +21,7 @@ #include "postmaster/auxprocess.h" #include "postmaster/bgwriter.h" #include "postmaster/startup.h" +#include "postmaster/walsummarizer.h" #include "postmaster/walwriter.h" #include "replication/walreceiver.h" #include "storage/bufmgr.h" @@ -80,6 +81,9 @@ AuxiliaryProcessMain(AuxProcType auxtype) case WalReceiverProcess: MyBackendType = B_WAL_RECEIVER; break; + case WalSummarizerProcess: + MyBackendType = B_WAL_SUMMARIZER; + break; default: elog(PANIC, "unrecognized process type: %d", (int) MyAuxProcType); MyBackendType = B_INVALID; @@ -161,6 +165,10 @@ AuxiliaryProcessMain(AuxProcType auxtype) WalReceiverMain(); proc_exit(1); + case WalSummarizerProcess: + WalSummarizerMain(); + proc_exit(1); + default: elog(PANIC, "unrecognized process type: %d", (int) MyAuxProcType); proc_exit(1); diff --git a/src/backend/postmaster/meson.build b/src/backend/postmaster/meson.build index cda921fd10..a30eb6692f 100644 --- a/src/backend/postmaster/meson.build +++ b/src/backend/postmaster/meson.build @@ -12,5 +12,6 @@ backend_sources += files( 'postmaster.c', 'startup.c', 'syslogger.c', + 'walsummarizer.c', 'walwriter.c', ) diff --git a/src/backend/postmaster/postmaster.c b/src/backend/postmaster/postmaster.c index d7bfb28ff3..8ae8291a3b 100644 --- a/src/backend/postmaster/postmaster.c +++ b/src/backend/postmaster/postmaster.c @@ -114,6 +114,7 @@ #include "postmaster/pgarch.h" #include "postmaster/postmaster.h" #include "postmaster/syslogger.h" +#include "postmaster/walsummarizer.h" #include "replication/logicallauncher.h" #include "replication/walsender.h" #include "storage/fd.h" @@ -250,6 +251,7 @@ static pid_t StartupPID = 0, CheckpointerPID = 0, WalWriterPID = 0, WalReceiverPID = 0, + WalSummarizerPID = 0, AutoVacPID = 0, PgArchPID = 0, SysLoggerPID = 0; @@ -441,6 +443,7 @@ static bool CreateOptsFile(int argc, char *argv[], char *fullprogname); static pid_t StartChildProcess(AuxProcType type); static void StartAutovacuumWorker(void); static void MaybeStartWalReceiver(void); +static void MaybeStartWalSummarizer(void); static void InitPostmasterDeathWatchHandle(void); /* @@ -560,6 +563,7 @@ static void ShmemBackendArrayRemove(Backend *bn); #define StartCheckpointer() StartChildProcess(CheckpointerProcess) #define StartWalWriter() StartChildProcess(WalWriterProcess) #define StartWalReceiver() StartChildProcess(WalReceiverProcess) +#define StartWalSummarizer() StartChildProcess(WalSummarizerProcess) /* Macros to check exit status of a child process */ #define EXIT_STATUS_0(st) ((st) == 0) @@ -1846,6 +1850,9 @@ ServerLoop(void) if (WalReceiverRequested) MaybeStartWalReceiver(); + /* If we need to start a WAL summarizer, try to do that now */ + MaybeStartWalSummarizer(); + /* Get other worker processes running, if needed */ if (StartWorkerNeeded || HaveCrashedWorker) maybe_start_bgworkers(); @@ -2713,6 +2720,8 @@ process_pm_reload_request(void) signal_child(WalWriterPID, SIGHUP); if (WalReceiverPID != 0) signal_child(WalReceiverPID, SIGHUP); + if (WalSummarizerPID != 0) + signal_child(WalSummarizerPID, SIGHUP); if (AutoVacPID != 0) signal_child(AutoVacPID, SIGHUP); if (PgArchPID != 0) @@ -3066,6 +3075,7 @@ process_pm_child_exit(void) BgWriterPID = StartBackgroundWriter(); if (WalWriterPID == 0) WalWriterPID = StartWalWriter(); + MaybeStartWalSummarizer(); /* * Likewise, start other special children as needed. In a restart @@ -3184,6 +3194,20 @@ process_pm_child_exit(void) continue; } + /* + * Was it the wal summarizer? Normal exit can be ignored; we'll start + * a new one at the next iteration of the postmaster's main loop, if + * necessary. Any other exit condition is treated as a crash. + */ + if (pid == WalSummarizerPID) + { + WalSummarizerPID = 0; + if (!EXIT_STATUS_0(exitstatus)) + HandleChildCrash(pid, exitstatus, + _("WAL summarizer process")); + continue; + } + /* * Was it the autovacuum launcher? Normal exit can be ignored; we'll * start a new one at the next iteration of the postmaster's main @@ -3579,6 +3603,12 @@ HandleChildCrash(int pid, int exitstatus, const char *procname) else if (WalReceiverPID != 0 && take_action) sigquit_child(WalReceiverPID); + /* Take care of the walsummarizer too */ + if (pid == WalSummarizerPID) + WalSummarizerPID = 0; + else if (WalSummarizerPID != 0 && take_action) + sigquit_child(WalSummarizerPID); + /* Take care of the autovacuum launcher too */ if (pid == AutoVacPID) AutoVacPID = 0; @@ -3729,6 +3759,8 @@ PostmasterStateMachine(void) signal_child(StartupPID, SIGTERM); if (WalReceiverPID != 0) signal_child(WalReceiverPID, SIGTERM); + if (WalSummarizerPID != 0) + signal_child(WalSummarizerPID, SIGTERM); /* checkpointer, archiver, stats, and syslogger may continue for now */ /* Now transition to PM_WAIT_BACKENDS state to wait for them to die */ @@ -3755,6 +3787,7 @@ PostmasterStateMachine(void) if (CountChildren(BACKEND_TYPE_ALL - BACKEND_TYPE_WALSND) == 0 && StartupPID == 0 && WalReceiverPID == 0 && + WalSummarizerPID == 0 && BgWriterPID == 0 && (CheckpointerPID == 0 || (!FatalError && Shutdown < ImmediateShutdown)) && @@ -3852,6 +3885,7 @@ PostmasterStateMachine(void) /* These other guys should be dead already */ Assert(StartupPID == 0); Assert(WalReceiverPID == 0); + Assert(WalSummarizerPID == 0); Assert(BgWriterPID == 0); Assert(CheckpointerPID == 0); Assert(WalWriterPID == 0); @@ -4073,6 +4107,8 @@ TerminateChildren(int signal) signal_child(WalWriterPID, signal); if (WalReceiverPID != 0) signal_child(WalReceiverPID, signal); + if (WalSummarizerPID != 0) + signal_child(WalSummarizerPID, signal); if (AutoVacPID != 0) signal_child(AutoVacPID, signal); if (PgArchPID != 0) @@ -5379,6 +5415,10 @@ StartChildProcess(AuxProcType type) ereport(LOG, (errmsg("could not fork WAL receiver process: %m"))); break; + case WalSummarizerProcess: + ereport(LOG, + (errmsg("could not fork WAL summarizer process: %m"))); + break; default: ereport(LOG, (errmsg("could not fork process: %m"))); @@ -5515,6 +5555,19 @@ MaybeStartWalReceiver(void) } } +/* + * MaybeStartWalSummarizer + * Start the WAL summarizer process, if not running and our state allows. + */ +static void +MaybeStartWalSummarizer(void) +{ + if (wal_summarize_mb != 0 && WalSummarizerPID == 0 && + (pmState == PM_RUN || pmState == PM_HOT_STANDBY) && + Shutdown <= SmartShutdown) + WalSummarizerPID = StartWalSummarizer(); +} + /* * Create the opts file diff --git a/src/backend/postmaster/walsummarizer.c b/src/backend/postmaster/walsummarizer.c new file mode 100644 index 0000000000..926b6c6ae4 --- /dev/null +++ b/src/backend/postmaster/walsummarizer.c @@ -0,0 +1,1414 @@ +/*------------------------------------------------------------------------- + * + * walsummarizer.c + * + * Background process to perform WAL summarization, if it is enabled. + * It continuously scans the write-ahead log and periodically emits a + * summary file which indicates which blocks in which relation forks + * were modified by WAL records in the LSN range covered by the summary + * file. See walsummary.c and blkreftable.c for more details on the + * naming and contents of WAL summary files. + * + * If configured to do, this background process will also remove WAL + * summary files when the file timestamp is older than a configurable + * threshold (but only if the WAL has been removed first). + * + * Portions Copyright (c) 1996-2022, PostgreSQL Global Development Group + * + * IDENTIFICATION + * src/backend/postmaster/walsummarizer.c + * + *------------------------------------------------------------------------- + */ +#include "postgres.h" + +#include "access/timeline.h" +#include "access/xlog.h" +#include "access/xlog_internal.h" +#include "access/xlogrecovery.h" +#include "access/xlogutils.h" +#include "backup/walsummary.h" +#include "catalog/storage_xlog.h" +#include "common/blkreftable.h" +#include "libpq/pqsignal.h" +#include "miscadmin.h" +#include "postmaster/bgwriter.h" +#include "postmaster/interrupt.h" +#include "postmaster/walsummarizer.h" +#include "replication/walreceiver.h" +#include "storage/fd.h" +#include "storage/ipc.h" +#include "storage/lwlock.h" +#include "storage/latch.h" +#include "storage/proc.h" +#include "storage/procsignal.h" +#include "storage/shmem.h" +#include "storage/spin.h" +#include "utils/guc.h" +#include "utils/memutils.h" +#include "utils/wait_event.h" + +/* + * Data in shared memory related to WAL summarization. + */ +typedef struct +{ + /* + * These fields are protected by WALSummarizerLock. + * + * Until we've discovered what summary files already exist on disk and + * stored that information in shared memory, initialized is false and the + * other fields here contain no meaningful information. After that has + * been done, initialized is true. + * + * summarized_tli and summarized_lsn indicate the last LSN and TLI at + * which the next summary file will start. Normally, these are the LSN + * and TLI at which the last file ended; in such case, lsn_is_exact is + * true. If, however, the LSN is just an approximation, then lsn_is_exact + * is false. This can happen if, for example, there are no existing WAL + * summary files at startup. In that case, we have to derive the position + * at which to start summarizing from the WAL files that exist on disk, + * and so the LSN might point to the start of the next file even though + * that might happen to be in the middle of a WAL record. + * + * summarizer_pgprocno is the pgprocno value for the summarizer process, + * if one is running, or else INVALID_PGPROCNO. + * + * pending_lsn is used by the summarizer to advertise the ending LSN of + * a record it has recently read. It shouldn't ever be less than + * summarized_lsn, but might be greater, because the summarizer buffers + * data for a range of LSNs in memory before writing out a new file. + * + * switch_requested can be set to true to notify the summarizer that a + * new WAL summary file should be written as soon as possible, without + * trying to read more WAL first. + */ + bool initialized; + TimeLineID summarized_tli; + XLogRecPtr summarized_lsn; + bool lsn_is_exact; + int summarizer_pgprocno; + XLogRecPtr pending_lsn; + bool switch_requested; + + /* + * This field handles its own synchronizaton. + */ + ConditionVariable summary_file_cv; +} WalSummarizerData; + +/* + * Private data for our xlogreader's page read callback. + */ +typedef struct +{ + TimeLineID tli; + bool historic; + XLogRecPtr read_upto; + bool end_of_wal; + bool waited; + XLogRecPtr redo_pointer; + bool redo_pointer_reached; + XLogRecPtr redo_pointer_refresh_lsn; +} SummarizerReadLocalXLogPrivate; + +/* Pointer to shared memory state. */ +static WalSummarizerData *WalSummarizerCtl; + +/* + * When we reach end of WAL and need to read more, we sleep for a number of + * milliseconds that is a integer multiple of MS_PER_SLEEP_QUANTUM. This is + * the multiplier. It should vary between 1 and MAX_SLEEP_QUANTA, depending + * on system activity. See summarizer_wait_for_wal() for how we adjust this. + */ +static long sleep_quanta = 1; + +/* + * The sleep time will always be a multiple of 200ms and will not exceed + * one minute (300 * 200 = 60 * 1000). + */ +#define MAX_SLEEP_QUANTA 300 +#define MS_PER_SLEEP_QUANTUM 200 + +/* + * This is a count of the number of pages of WAL that we've read since the + * last time we waited for more WAL to appear. + */ +static long pages_read_since_last_sleep = 0; + +/* + * Most recent RedoRecPtr value observed by MaybeRemoveOldWalSummaries. + */ +static XLogRecPtr redo_pointer_at_last_summary_removal = InvalidXLogRecPtr; + +/* + * GUC parameters + */ +int wal_summarize_mb = 256; +int wal_summarize_keep_time = 7 * 24 * 60; + +static XLogRecPtr GetLatestLSN(TimeLineID *tli); +static void HandleWalSummarizerInterrupts(void); +static XLogRecPtr SummarizeWAL(TimeLineID tli, bool historic, + XLogRecPtr start_lsn, bool exact, + XLogRecPtr cutoff_lsn, XLogRecPtr maximum_lsn); +static void SummarizeSmgrRecord(XLogReaderState *xlogreader, + BlockRefTable *brtab); +static void SummarizeXactRecord(XLogReaderState *xlogreader, + BlockRefTable *brtab); +static int summarizer_read_local_xlog_page(XLogReaderState *state, + XLogRecPtr targetPagePtr, + int reqLen, + XLogRecPtr targetRecPtr, + char *cur_page); +static void summarizer_wait_for_wal(void); +static void MaybeRemoveOldWalSummaries(void); + +/* + * Amount of shared memory required for this module. + */ +Size +WalSummarizerShmemSize(void) +{ + return sizeof(WalSummarizerData); +} + +/* + * Create or attach to shared memory segment for this module. + */ +void +WalSummarizerShmemInit(void) +{ + bool found; + + WalSummarizerCtl = (WalSummarizerData *) + ShmemInitStruct("Wal Summarizer Ctl", WalSummarizerShmemSize(), + &found); + + if (!found) + { + /* + * First time through, so initialize. + * + * We're just filling in dummy values here -- the real initialization + * will happen when GetOldestUnsummarizedLSN() is called for the first + * time. + */ + WalSummarizerCtl->initialized = false; + WalSummarizerCtl->summarized_tli = 0; + WalSummarizerCtl->summarized_lsn = InvalidXLogRecPtr; + WalSummarizerCtl->lsn_is_exact = false; + WalSummarizerCtl->summarizer_pgprocno = INVALID_PGPROCNO; + WalSummarizerCtl->pending_lsn = InvalidXLogRecPtr; + WalSummarizerCtl->switch_requested = false; + ConditionVariableInit(&WalSummarizerCtl->summary_file_cv); + } +} + +/* + * Entry point for walsummarizer process. + */ +void +WalSummarizerMain(void) +{ + sigjmp_buf local_sigjmp_buf; + MemoryContext context; + + /* + * Within this function, 'current_lsn' and 'current_tli' refer to the + * point from which the next WAL summary file should start. 'exact' is + * true if 'current_lsn' is known to be the start of a WAL recod or WAL + * segment, and false if it might be in the middle of a record someplace. + * + * 'switch_lsn' and 'switch_tli', if set, are the LSN at which we need to + * switch to a new timeline and the timeline to which we need to switch. + * If not set, we either haven't figured out the answers yet or we're + * already on the latest timeline. + */ + XLogRecPtr current_lsn; + TimeLineID current_tli; + bool exact; + XLogRecPtr switch_lsn = InvalidXLogRecPtr; + TimeLineID switch_tli = 0; + + ereport(DEBUG1, + (errmsg_internal("WAL summarizer started"))); + + /* + * Properly accept or ignore signals the postmaster might send us + * + * We have no particular use for SIGINT at the moment, but seems + * reasonable to treat like SIGTERM. + */ + pqsignal(SIGHUP, SignalHandlerForConfigReload); + pqsignal(SIGINT, SignalHandlerForShutdownRequest); + pqsignal(SIGTERM, SignalHandlerForShutdownRequest); + /* SIGQUIT handler was already set up by InitPostmasterChild */ + pqsignal(SIGALRM, SIG_IGN); + pqsignal(SIGPIPE, SIG_IGN); + pqsignal(SIGUSR1, procsignal_sigusr1_handler); + pqsignal(SIGUSR2, SIG_IGN); /* not used */ + + /* Advertise ourselves. */ + LWLockAcquire(WALSummarizerLock, LW_EXCLUSIVE); + WalSummarizerCtl->summarizer_pgprocno = MyProc->pgprocno; + LWLockRelease(WALSummarizerLock); + + /* Create and switch to a memory context that we can reset on error. */ + context = AllocSetContextCreate(TopMemoryContext, + "Wal Summarizer", + ALLOCSET_DEFAULT_SIZES); + MemoryContextSwitchTo(context); + + /* + * Reset some signals that are accepted by postmaster but not here + */ + pqsignal(SIGCHLD, SIG_DFL); + + /* + * If an exception is encountered, processing resumes here. + */ + if (sigsetjmp(local_sigjmp_buf, 1) != 0) + { + /* Since not using PG_TRY, must reset error stack by hand */ + error_context_stack = NULL; + + /* Prevent interrupts while cleaning up */ + HOLD_INTERRUPTS(); + + /* Report the error to the server log */ + EmitErrorReport(); + + /* Release resources we might have acquired. */ + LWLockReleaseAll(); + ConditionVariableCancelSleep(); + pgstat_report_wait_end(); + ReleaseAuxProcessResources(false); + AtEOXact_Files(false); + AtEOXact_HashTables(false); + + /* + * Now return to normal top-level context and clear ErrorContext for + * next time. + */ + MemoryContextSwitchTo(context); + FlushErrorState(); + + /* Flush any leaked data in the top-level context */ + MemoryContextResetAndDeleteChildren(context); + + /* Now we can allow interrupts again */ + RESUME_INTERRUPTS(); + + /* + * Sleep for 10 seconds before attempting to resume operations in + * order to avoid excessing logging. + * + * Many of the likely error conditions are things that will repeat + * every time. For example, if the WAL can't be read or the summary + * can't be written, only administrator action will cure the problem. + * So a really fast retry time doesn't seem to be especially + * beneficial, and it will clutter the logs. + */ + (void) WaitLatch(MyLatch, + WL_TIMEOUT | WL_EXIT_ON_PM_DEATH, + 10000, + WAIT_EVENT_WAL_SUMMARIZER_ERROR); + } + + /* We can now handle ereport(ERROR) */ + PG_exception_stack = &local_sigjmp_buf; + + /* + * Unblock signals (they were blocked when the postmaster forked us) + */ + sigprocmask(SIG_SETMASK, &UnBlockSig, NULL); + + /* + * Fetch information about previous progress from shared memory. + * + * If we discover that WAL summarization is not enabled, just exit. + */ + current_lsn = GetOldestUnsummarizedLSN(¤t_tli, &exact); + if (XLogRecPtrIsInvalid(current_lsn)) + proc_exit(0); + + /* + * Loop forever + */ + for (;;) + { + XLogRecPtr latest_lsn; + TimeLineID latest_tli; + XLogRecPtr cutoff_lsn; + XLogRecPtr end_of_summary_lsn; + + /* Flush any leaked data in the top-level context */ + MemoryContextResetAndDeleteChildren(context); + + /* Process any signals received recently. */ + HandleWalSummarizerInterrupts(); + + /* If it's time to remove any old WAL summaries, do that now. */ + MaybeRemoveOldWalSummaries(); + + /* Find the LSN and TLI up to which we can safely summarize. */ + latest_lsn = GetLatestLSN(&latest_tli); + + /* + * If we're summarizing a historic timeline and we haven't yet + * computed the point at which to switch to the next timeline, do that + * now. + * + * Note that if this is a standby, what was previously the current + * timeline could become historic at any time. + * + * We could try to make this more efficient by caching the results of + * readTimeLineHistory when latest_tli has not changed, but since we + * only have to do this once per timeline switch, we probably wouldn't + * save any significant amount of work in practice. + */ + if (current_tli != latest_tli && XLogRecPtrIsInvalid(switch_lsn)) + { + List *tles = readTimeLineHistory(latest_tli); + + switch_lsn = tliSwitchPoint(current_tli, tles, &switch_tli); + elog(DEBUG2, + "switch point from TLI %u to TLI %u is at %X/%X", + current_tli, switch_tli, LSN_FORMAT_ARGS(switch_lsn)); + } + + /* + * wal_summarize_mb sets a soft limit on the amont of WAL covered + * by a single summary file. If we read a WAL record that ends after + * the cutoff LSN computed here, we'll stop the summary. In most cases, + * it will actually stop earlier than that, but this is here as a + * backstop. + */ + cutoff_lsn = current_lsn + wal_summarize_mb * 1024 * 1024; + if (!XLogRecPtrIsInvalid(switch_lsn) && cutoff_lsn > switch_lsn) + cutoff_lsn = switch_lsn; + elog(DEBUG2, + "WAL summarization cutoff is TLI %d @ %X/%X, flush position is %X/%X", + current_tli, LSN_FORMAT_ARGS(cutoff_lsn), LSN_FORMAT_ARGS(latest_lsn)); + + /* Summarize WAL. */ + end_of_summary_lsn = SummarizeWAL(current_tli, + current_tli != latest_tli, + current_lsn, exact, + cutoff_lsn, latest_lsn); + Assert(!XLogRecPtrIsInvalid(end_of_summary_lsn)); + Assert(end_of_summary_lsn >= current_lsn); + + /* + * Update state for next loop iteration. + * + * Next summary file should start from exactly where this one ended. + * Timeline remains unchanged unless a switch LSN was computed and we + * have reached it. + */ + current_lsn = end_of_summary_lsn; + exact = true; + if (!XLogRecPtrIsInvalid(switch_lsn) && cutoff_lsn >= switch_lsn) + { + current_tli = switch_tli; + switch_lsn = InvalidXLogRecPtr; + switch_tli = 0; + } + + /* Update state in shared memory. */ + LWLockAcquire(WALSummarizerLock, LW_EXCLUSIVE); + Assert(WalSummarizerCtl->pending_lsn <= end_of_summary_lsn); + WalSummarizerCtl->summarized_lsn = end_of_summary_lsn; + WalSummarizerCtl->summarized_tli = current_tli; + WalSummarizerCtl->lsn_is_exact = true; + WalSummarizerCtl->pending_lsn = end_of_summary_lsn; + WalSummarizerCtl->switch_requested = false; + LWLockRelease(WALSummarizerLock); + + /* Wake up anyone waiting for more summary files to be written. */ + ConditionVariableBroadcast(&WalSummarizerCtl->summary_file_cv); + } +} + +/* + * Get the oldest LSN in this server's timeline history that has not yet been + * summarized. + * + * If *tli != NULL, it will be set to the TLI for the LSN that is returned. + * + * If *lsn_is_exact != NULL, it will be set to true if the returned LSN is + * necessarily the start of a WAL record and false if it's just the beginning + * of a WAL segment. + */ +XLogRecPtr +GetOldestUnsummarizedLSN(TimeLineID *tli, bool *lsn_is_exact) +{ + TimeLineID latest_tli; + LWLockMode mode = LW_SHARED; + int n; + List *tles; + XLogRecPtr unsummarized_lsn; + TimeLineID unsummarized_tli = 0; + bool should_make_exact = false; + List *existing_summaries; + ListCell *lc; + + /* If not summarizing WAL, do nothing. */ + if (wal_summarize_mb == 0) + return InvalidXLogRecPtr; + + /* + * Initially, we acquire the lock in shared mode and try to fetch the + * required information. If the data structure hasn't been initialized, we + * reacquire the lock in shared mode so that we can initialize it. + * However, if someone else does that first before we get the lock, then + * we can just return the requested information after all. + */ + while (true) + { + LWLockAcquire(WALSummarizerLock, mode); + + if (WalSummarizerCtl->initialized) + { + unsummarized_lsn = WalSummarizerCtl->summarized_lsn; + if (tli != NULL) + *tli = WalSummarizerCtl->summarized_tli; + if (lsn_is_exact != NULL) + *lsn_is_exact = WalSummarizerCtl->lsn_is_exact; + LWLockRelease(WALSummarizerLock); + return unsummarized_lsn; + } + + if (mode == LW_EXCLUSIVE) + break; + + LWLockRelease(WALSummarizerLock); + mode = LW_EXCLUSIVE; + } + + /* + * The data structure needs to be initialized, and we are the first to + * obtain the lock in exclusive mode, so it's our job to do that + * initialization. + * + * So, find the oldest timeline on which WAL still exists, and the + * earliest segment for which it exists. + */ + (void) GetLatestLSN(&latest_tli); + tles = readTimeLineHistory(latest_tli); + for (n = list_length(tles) - 1; n >= 0; --n) + { + TimeLineHistoryEntry *tle = list_nth(tles, n); + XLogSegNo oldest_segno; + + oldest_segno = XLogGetOldestSegno(tle->tli); + if (oldest_segno != 0) + { + /* Compute oldest LSN that still exists on disk. */ + XLogSegNoOffsetToRecPtr(oldest_segno, 0, wal_segment_size, + unsummarized_lsn); + + unsummarized_tli = tle->tli; + break; + } + } + + /* It really should not be possible for us to find no WAL. */ + if (unsummarized_tli == 0) + ereport(ERROR, + errcode(ERRCODE_INTERNAL_ERROR), + errmsg_internal("no WAL found on timeline %d", latest_tli)); + + /* + * Don't try to summarize anything older than the end LSN of the newest + * summary file that exists for this timeline. + */ + existing_summaries = + GetWalSummaries(unsummarized_tli, + InvalidXLogRecPtr, InvalidXLogRecPtr); + foreach(lc, existing_summaries) + { + WalSummaryFile *ws = lfirst(lc); + + if (ws->end_lsn > unsummarized_lsn) + { + unsummarized_lsn = ws->end_lsn; + should_make_exact = true; + } + } + + /* Update shared memory with the discovered values. */ + WalSummarizerCtl->initialized = true; + WalSummarizerCtl->summarized_lsn = unsummarized_lsn; + WalSummarizerCtl->summarized_tli = unsummarized_tli; + WalSummarizerCtl->lsn_is_exact = should_make_exact; + WalSummarizerCtl->pending_lsn = unsummarized_lsn; + + /* Also return the to the caller as required. */ + if (tli != NULL) + *tli = WalSummarizerCtl->summarized_tli; + if (lsn_is_exact != NULL) + *lsn_is_exact = WalSummarizerCtl->lsn_is_exact; + LWLockRelease(WALSummarizerLock); + + return unsummarized_lsn; +} + +/* + * Attempt to set the WAL summarizer's latch. + * + * This might not work, because there's no guarantee that the WAL summarizer + * process was successfully started, and it also might have started but + * subsequently terminated. So, under normal circumstances, this will get the + * latch set, but there's no guarantee. + */ +void +SetWalSummarizerLatch(void) +{ + int pgprocno; + + LWLockAcquire(WALSummarizerLock, LW_EXCLUSIVE); + pgprocno = WalSummarizerCtl->summarizer_pgprocno; + LWLockRelease(WALSummarizerLock); + + if (pgprocno != INVALID_PGPROCNO) + SetLatch(&ProcGlobal->allProcs[pgprocno].procLatch); +} + +/* + * Wait until WAL summarization reaches the given LSN, but not longer than + * the given timeout. + * + * The return value is the first still-unsummarized LSN. If it's greater than + * or equal to the passed LSN, then that LSN was reached. If not, we timed out. + */ +XLogRecPtr +WaitForWalSummarization(XLogRecPtr lsn, long timeout) +{ + TimestampTz start_time = GetCurrentTimestamp(); + TimestampTz deadline = TimestampTzPlusMilliseconds(start_time, timeout); + XLogRecPtr summarized_lsn; + + Assert(!XLogRecPtrIsInvalid(lsn)); + Assert(timeout > 0); + + while (1) + { + TimestampTz now; + long remaining_timeout; + + /* + * If the LSN summarized on disk has reached the target value, stop. + * If it hasn't, but the in-memory value has reached the target value, + * request that a file be written as soon as possible. + */ + LWLockAcquire(WALSummarizerLock, LW_EXCLUSIVE); + summarized_lsn = WalSummarizerCtl->summarized_lsn; + if (summarized_lsn < lsn && + WalSummarizerCtl->pending_lsn >= lsn) + WalSummarizerCtl->switch_requested = true; + LWLockRelease(WALSummarizerLock); + if (summarized_lsn >= lsn) + break; + + /* Timeout reached? If yes, stop. */ + now = GetCurrentTimestamp(); + remaining_timeout = TimestampDifferenceMilliseconds(now, deadline); + if (remaining_timeout <= 0) + break; + + /* + * Limit the sleep to 1 second, because we may need to request a + * switch. + */ + if (remaining_timeout > 1000) + remaining_timeout = 1000; + + /* Wait and see. */ + ConditionVariableTimedSleep(&WalSummarizerCtl->summary_file_cv, + remaining_timeout, + WAIT_EVENT_WAL_SUMMARY_READY); + } + + return summarized_lsn; +} + +/* + * Get the latest LSN that is eligible to be summarized, and set *tli to the + * corresponding timeline. + */ +static XLogRecPtr +GetLatestLSN(TimeLineID *tli) +{ + if (!RecoveryInProgress()) + { + /* Don't summarize WAL before it's flushed. */ + return GetFlushRecPtr(tli); + } + else + { + XLogRecPtr flush_lsn; + TimeLineID flush_tli; + XLogRecPtr replay_lsn; + TimeLineID replay_tli; + + /* + * What we really want to know is how much WAL has been flushed to + * disk, but the only flush position available is the one provided by + * the walreceiver, which may not be running, because this could be + * crash recovery or recovery via restore_command. So use either the + * WAL receiver's flush position or the replay position, whichever is + * further ahead, on the theory that if the WAL has been replayed then + * it must also have been flushed to disk. + */ + flush_lsn = GetWalRcvFlushRecPtr(NULL, &flush_tli); + replay_lsn = GetXLogReplayRecPtr(&replay_tli); + if (flush_lsn > replay_lsn) + { + *tli = flush_tli; + return flush_lsn; + } + else + { + *tli = replay_tli; + return replay_lsn; + } + } +} + +/* + * Interrupt handler for main loop of WAL summarizer process. + */ +static void +HandleWalSummarizerInterrupts(void) +{ + if (ProcSignalBarrierPending) + ProcessProcSignalBarrier(); + + if (ConfigReloadPending) + { + ConfigReloadPending = false; + ProcessConfigFile(PGC_SIGHUP); + } + + if (ShutdownRequestPending || wal_summarize_mb == 0) + { + ereport(DEBUG1, + errmsg_internal("WAL summarizer shutting down")); + proc_exit(0); + } + + /* Perform logging of memory contexts of this process */ + if (LogMemoryContextPending) + ProcessLogMemoryContextInterrupt(); +} + +/* + * Summarize a range of WAL records on a single timeline. + * + * 'tli' is the timeline to be summarized. 'historic' should be false if the + * timeline in question is the latest one and true otherwise. + * + * 'start_lsn' is the point at which we should start summarizing. If this + * value comes from the end LSN of the previous record as returned by the + * xlograder machinery, 'exact' should be true; otherwise, 'exact' should + * be false, and this function will search forward for the start of a valid + * WAL record. + * + * 'cutoff_lsn' is the point at which we should stop summarizing. The first + * record that ends at or after cutoff_lsn will be the last one included + * in the summary. + * + * 'maximum_lsn' identifies the point beyond which we can't count on being + * able to read any more WAL. It should be the switch point when reading a + * historic timeline, or the most-recently-measured end of WAL when reading + * the current timeline. + * + * The return value is the LSN at which the WAL summary actually ends. Most + * often, a summary file ends because we notice that a checkpoint has + * occurred and reach the redo pointer of that checkpoint, but sometimes + * we stop for other reasons, such as a timeline switch, or reading a record + * that ends after the cutoff_lsn. + */ +static XLogRecPtr +SummarizeWAL(TimeLineID tli, bool historic, + XLogRecPtr start_lsn, bool exact, + XLogRecPtr cutoff_lsn, XLogRecPtr maximum_lsn) +{ + SummarizerReadLocalXLogPrivate *private_data; + XLogReaderState *xlogreader; + XLogRecPtr summary_start_lsn; + XLogRecPtr summary_end_lsn = cutoff_lsn; + char temp_path[MAXPGPATH]; + char final_path[MAXPGPATH]; + WalSummaryIO io; + BlockRefTable *brtab = CreateEmptyBlockRefTable(); + + /* Initialize private data for xlogreader. */ + private_data = (SummarizerReadLocalXLogPrivate *) + palloc0(sizeof(SummarizerReadLocalXLogPrivate)); + private_data->tli = tli; + private_data->historic = historic; + private_data->read_upto = maximum_lsn; + private_data->redo_pointer = GetRedoRecPtr(); + private_data->redo_pointer_refresh_lsn = start_lsn; + private_data->redo_pointer_reached = + (start_lsn >= private_data->redo_pointer); + + /* Create xlogreader. */ + xlogreader = XLogReaderAllocate(wal_segment_size, NULL, + XL_ROUTINE(.page_read = &summarizer_read_local_xlog_page, + .segment_open = &wal_segment_open, + .segment_close = &wal_segment_close), + private_data); + if (xlogreader == NULL) + ereport(ERROR, + (errcode(ERRCODE_OUT_OF_MEMORY), + errmsg("out of memory"), + errdetail("Failed while allocating a WAL reading processor."))); + + /* + * When exact = false, we're starting from an arbitrary point in the WAL + * and must search forward for the start of the next record. + * + * When exact = true, start_lsn should be either the LSN where a record + * begins, or the LSN of a page where the page header is immediately + * followed by the start of a new record. XLogBeginRead should tolerate + * either case. + * + * We need to allow for both cases because the behavior of xlogreader + * varies. When a record spans two or more xlog pages, the ending LSN + * reported by xlogreader will be the starting LSN of the following + * record, but when an xlog page boundary falls between two records, the + * end LSN for the first will be reported as the first byte of the + * following page. We can't know until we read that page how large the + * header will be, but we'll have to skip over it to find the next record. + */ + if (exact) + { + /* + * Even if start_lsn is the beginning of a page rather than the + * beginning of the first record on that page, we should still use it + * as the start LSN for the summary file. That's because we detect + * missing summary files by looking for cases where the end LSN of one + * file is less than the start LSN of the next file. When only a page + * header is skipped, nothing has been missed. + */ + XLogBeginRead(xlogreader, start_lsn); + summary_start_lsn = start_lsn; + } + else + { + summary_start_lsn = XLogFindNextRecord(xlogreader, start_lsn); + if (XLogRecPtrIsInvalid(summary_start_lsn)) + { + /* + * If we hit end-of-WAL while trying to find the next valid + * record, we must be on a historic timeline that has no valid + * records that begin after start_lsn and before end of WAL. + */ + if (private_data->end_of_wal) + { + ereport(DEBUG1, + errmsg_internal("could not read WAL from timeline %d at %X/%X: end of WAL at %X/%X", + tli, + LSN_FORMAT_ARGS(start_lsn), + LSN_FORMAT_ARGS(private_data->read_upto))); + + /* + * The timeline ends at or after start_lsn, without containing + * any records. Thus, we must make sure the main loop does not + * iterate. If start_lsn is the end of the timeline, then we + * won't actually emit an empty summary file, but otherwise, + * we must, to capture the fact that the LSN range in question + * contains no interesting WAL records. + */ + summary_start_lsn = start_lsn; + summary_end_lsn = private_data->read_upto; + cutoff_lsn = xlogreader->EndRecPtr; + } + else + ereport(ERROR, + (errmsg("could not find a valid record after %X/%X", + LSN_FORMAT_ARGS(start_lsn)))); + } + + /* We shouldn't go backward. */ + Assert(summary_start_lsn >= start_lsn); + } + + /* + * Main loop: read xlog records one by one. + */ + while (xlogreader->EndRecPtr < cutoff_lsn) + { + int block_id; + char *errormsg; + XLogRecord *record; + bool switch_requested; + + /* We shouldn't go backward. */ + Assert(summary_start_lsn <= xlogreader->EndRecPtr); + + /* + * This flag tracks whether the read of a particular record had to + * wait for more WAL to arrive, so reset it before reading the next + * record. + */ + private_data->waited = false; + + /* Now read the next record. */ + record = XLogReadRecord(xlogreader, &errormsg); + if (record == NULL) + { + SummarizerReadLocalXLogPrivate *private_data; + + private_data = (SummarizerReadLocalXLogPrivate *) + xlogreader->private_data; + if (private_data->end_of_wal) + { + /* + * This timeline must be historic and must end before we were + * able to read a complete record. + */ + ereport(DEBUG1, + errmsg_internal("could not read WAL from timeline %d at %X/%X: end of WAL at %X/%X", + tli, + LSN_FORMAT_ARGS(xlogreader->EndRecPtr), + LSN_FORMAT_ARGS(private_data->read_upto))); + /* Summary ends at end of WAL. */ + summary_end_lsn = private_data->read_upto; + break; + } + if (errormsg) + ereport(ERROR, + (errcode_for_file_access(), + errmsg("could not read WAL at %X/%X: %s", + LSN_FORMAT_ARGS(xlogreader->EndRecPtr), errormsg))); + else + ereport(ERROR, + (errcode_for_file_access(), + errmsg("could not read WAL at %X/%X", + LSN_FORMAT_ARGS(xlogreader->EndRecPtr)))); + } + + /* We shouldn't go backward. */ + Assert(summary_start_lsn <= xlogreader->EndRecPtr); + + if (xlogreader->ReadRecPtr >= cutoff_lsn) + { + /* + * Woops! We've read a record that *starts* after the cutoff LSN, + * contrary to our goal of reading only until we hit the first + * record that ends at or after the cutoff LSN. Pretend we didn't + * read it after all by bailing out of this loop right here, + * before we do anything with this record. + * + * This can happen because the last record before the cutoff LSN + * might be continued across multiple pages, and then we might + * come to a page with XLP_FIRST_IS_OVERWRITE_CONTRECORD set. In + * that case, the record that was continued across multiple pages + * is incomplete and will be disregarded, and the read will + * restart from the beginning of the page that is flagged + * XLP_FIRST_IS_OVERWRITE_CONTRECORD. + * + * If this case occurs, we can fairly say that the current summary + * file ends at the cutoff LSN exactly. The first record on the + * page marked XLP_FIRST_IS_OVERWRITE_CONTRECORD will be + * discovered when generating the next summary file. + */ + summary_end_lsn = cutoff_lsn; + break; + } + + /* + * We attempt, on a best effort basis only, to make WAL summary file + * boundaries line up with checkpoint cycles. So, if the last redo + * pointer we've seen was in the future, and this record starts at + * that redo pointer, stop before processing and let it be included in + * the next summary file. + * + * Note that in the case of a checkpoint triggered by a backup, the + * redo pointer is likely to be pointing to the first record on a + * page. Before reading the record, xlogreader->EndRecPtr will have + * pointed to the start of the page, which precedes the redo LSN. But + * after reading the next record, we'll advance over the page header + * and realize that the next record starts at the redo LSN exactly, + * making this the first point at which we can realize that it's time + * to stop. + */ + if (!private_data->redo_pointer_reached && + xlogreader->ReadRecPtr >= private_data->redo_pointer) + { + summary_end_lsn = xlogreader->ReadRecPtr; + break; + } + + /* Special handling for particular types of WAL records. */ + switch (XLogRecGetRmid(xlogreader)) + { + case RM_SMGR_ID: + SummarizeSmgrRecord(xlogreader, brtab); + break; + case RM_XACT_ID: + SummarizeXactRecord(xlogreader, brtab); + break; + default: + break; + } + + /* Feed block references from xlog record to block reference table. */ + for (block_id = 0; block_id <= XLogRecMaxBlockId(xlogreader); + block_id++) + { + RelFileLocator rlocator; + ForkNumber forknum; + BlockNumber blocknum; + + if (!XLogRecGetBlockTagExtended(xlogreader, block_id, &rlocator, + &forknum, &blocknum, NULL)) + continue; + + BlockRefTableMarkBlockModified(brtab, &rlocator, forknum, + blocknum); + } + + /* Update our notion of where this summary file ends. */ + summary_end_lsn = xlogreader->EndRecPtr; + + /* + * Also update shared memory, and handle any request for a + * WAL summary file switch. + */ + LWLockAcquire(WALSummarizerLock, LW_EXCLUSIVE); + Assert(summary_end_lsn >= WalSummarizerCtl->pending_lsn); + Assert(summary_end_lsn >= WalSummarizerCtl->summarized_lsn); + WalSummarizerCtl->pending_lsn = summary_end_lsn; + switch_requested = WalSummarizerCtl->switch_requested; + LWLockRelease(WALSummarizerLock); + if (switch_requested) + break; + + /* + * Periodically update our notion of the redo pointer, because it + * might be changing concurrently. There's no interlocking here: we + * might race past the new redo pointer before we learn about it. + * That's OK; we only use the redo pointer as a heuristic for where to + * stop summarizing. + * + * It would be nice if we could just fetch the updated redo pointer on + * every pass through this loop, but that seems a bit too expensive: + * GetRedoRecPtr acquires a heavily-contended spinlock. So, instead, + * just fetch the updated value if we've just had to sleep, or if + * we've read more than a segment's worth of WAL without sleeping. + */ + if (private_data->waited || xlogreader->EndRecPtr > + private_data->redo_pointer_refresh_lsn + wal_segment_size) + { + private_data->redo_pointer = GetRedoRecPtr(); + private_data->redo_pointer_refresh_lsn = xlogreader->EndRecPtr; + private_data->redo_pointer_reached = + (xlogreader->EndRecPtr >= private_data->redo_pointer); + } + + /* + * Recheck whether we've just caught up with the redo pointer, and + * if so, stop. This has the same purpose as the earlier check for + * the same condition above, but there we've just read a record and + * might decide against including it in the current summary file, + * whereas here we've already included it and might decide against + * reading the next one. Note that we may have just refreshed our + * notion of the redo pointer, so it's smart to check here before we + * do any more work. + */ + if (!private_data->redo_pointer_reached && + xlogreader->EndRecPtr >= private_data->redo_pointer) + break; + } + + /* Destroy xlogreader. */ + pfree(xlogreader->private_data); + XLogReaderFree(xlogreader); + + /* + * If a timeline switch occurs, we may fail to make any progress at all + * before exiting the loop above. If that happens, we don't write a WAL + * summary file at all. + */ + if (summary_end_lsn > summary_start_lsn) + { + /* Generate temporary and final path name. */ + snprintf(temp_path, MAXPGPATH, + XLOGDIR "/summaries/temp.summary"); + snprintf(final_path, MAXPGPATH, + XLOGDIR "/summaries/%08X%08X%08X%08X%08X.summary", + tli, + LSN_FORMAT_ARGS(summary_start_lsn), + LSN_FORMAT_ARGS(summary_end_lsn)); + + /* Open the temporary file for writing. */ + io.filepos = 0; + io.file = PathNameOpenFile(temp_path, O_WRONLY | O_CREAT | O_TRUNC); + if (io.file < 0) + ereport(ERROR, + (errcode_for_file_access(), + errmsg("could not create file \"%s\": %m", temp_path))); + + /* Write the data. */ + WriteBlockRefTable(brtab, WriteWalSummary, &io); + + /* Close temporary file and shut down xlogreader. */ + FileClose(io.file); + + /* Tell the user what we did. */ + ereport(LOG, + errmsg("summarized WAL on TLI %d from %X/%X to %X/%X", + tli, + LSN_FORMAT_ARGS(summary_start_lsn), + LSN_FORMAT_ARGS(summary_end_lsn))); + + /* Durably rename the new summary into place. */ + durable_rename(temp_path, final_path, ERROR); + } + + return summary_end_lsn; +} + +/* + * Special handling for WAL records with RM_SMGR_ID. + */ +static void +SummarizeSmgrRecord(XLogReaderState *xlogreader, BlockRefTable *brtab) +{ + uint8 info = XLogRecGetInfo(xlogreader) & ~XLR_INFO_MASK; + + if (info == XLOG_SMGR_CREATE) + { + xl_smgr_create *xlrec; + + /* + * If a new relation fork is created on disk, there is no point + * tracking anything about which blocks have been modified, because + * the whole thing will be new. Hence, set the limit block for this + * fork to 0. + * + * Ignore the FSM fork, which is not fully WAL-logged. + */ + xlrec = (xl_smgr_create *) XLogRecGetData(xlogreader); + + if (xlrec->forkNum != FSM_FORKNUM) + BlockRefTableSetLimitBlock(brtab, &xlrec->rlocator, + xlrec->forkNum, 0); + } + else if (info == XLOG_SMGR_TRUNCATE) + { + xl_smgr_truncate *xlrec; + + xlrec = (xl_smgr_truncate *) XLogRecGetData(xlogreader); + + /* + * If a relation fork is truncated on disk, there is in point in + * tracking anything about block modifications beyond the truncation + * point. + * + * We ignore SMGR_TRUNCATE_FSM here because the FSM isn't fully + * WAL-logged and thus we can't track modified blocks for it anyway. + */ + if ((xlrec->flags & SMGR_TRUNCATE_HEAP) != 0) + BlockRefTableSetLimitBlock(brtab, &xlrec->rlocator, + MAIN_FORKNUM, xlrec->blkno); + if ((xlrec->flags & SMGR_TRUNCATE_VM) != 0) + BlockRefTableSetLimitBlock(brtab, &xlrec->rlocator, + VISIBILITYMAP_FORKNUM, xlrec->blkno); + } +} + +/* + * Special handling for WAL recods with RM_XACT_ID. + */ +static void +SummarizeXactRecord(XLogReaderState *xlogreader, BlockRefTable *brtab) +{ + uint8 info = XLogRecGetInfo(xlogreader) & ~XLR_INFO_MASK; + uint8 xact_info = info & XLOG_XACT_OPMASK; + + if (xact_info == XLOG_XACT_COMMIT || + xact_info == XLOG_XACT_COMMIT_PREPARED) + { + xl_xact_commit *xlrec = (xl_xact_commit *) XLogRecGetData(xlogreader); + xl_xact_parsed_commit parsed; + int i; + + ParseCommitRecord(XLogRecGetInfo(xlogreader), xlrec, &parsed); + for (i = 0; i < parsed.nrels; ++i) + { + ForkNumber forknum; + + for (forknum = 0; forknum <= MAX_FORKNUM; ++forknum) + if (forknum != FSM_FORKNUM) + BlockRefTableSetLimitBlock(brtab, &parsed.xlocators[i], + forknum, 0); + } + } + else if (xact_info == XLOG_XACT_ABORT || + xact_info == XLOG_XACT_ABORT_PREPARED) + { + xl_xact_abort *xlrec = (xl_xact_abort *) XLogRecGetData(xlogreader); + xl_xact_parsed_abort parsed; + int i; + + ParseAbortRecord(XLogRecGetInfo(xlogreader), xlrec, &parsed); + for (i = 0; i < parsed.nrels; ++i) + { + ForkNumber forknum; + + for (forknum = 0; forknum <= MAX_FORKNUM; ++forknum) + if (forknum != FSM_FORKNUM) + BlockRefTableSetLimitBlock(brtab, &parsed.xlocators[i], + forknum, 0); + } + } +} + +/* + * Similar to read_local_xlog_page, but limited to read from one particular + * timeline. If the end of WAL is reached, it will wait for more if reading + * from the current timeline, or give up if reading from a historic timeline. + * In the latter case, it will also set private_data->end_of_wal = true. + * + * Caller must set private_data->tli to the TLI of interest, + * private_data->read_upto to the lowest LSN that is not known to be safe + * to read on that timeline, and private_data->historic to true if and only + * if the timeline is not the current timeline. This function will update + * private_data->read_upto and private_data->historic if more WAL appears + * on the current timeline or if the current timeline becomes historic. + */ +static int +summarizer_read_local_xlog_page(XLogReaderState *state, + XLogRecPtr targetPagePtr, int reqLen, + XLogRecPtr targetRecPtr, char *cur_page) +{ + int count; + WALReadError errinfo; + SummarizerReadLocalXLogPrivate *private_data; + + private_data = (SummarizerReadLocalXLogPrivate *) + state->private_data; + + while (true) + { + if (targetPagePtr + XLOG_BLCKSZ <= private_data->read_upto) + { + /* + * more than one block available; read only that block, have + * caller come back if they need more. + */ + count = XLOG_BLCKSZ; + break; + } + else if (targetPagePtr + reqLen > private_data->read_upto) + { + /* We don't seem to have enough data. */ + if (private_data->historic) + { + /* + * This is a historic timeline, so there will never be any + * more data than we have currently. + */ + private_data->end_of_wal = true; + return -1; + } + else + { + XLogRecPtr latest_lsn; + TimeLineID latest_tli; + + /* + * This is - or at least was up until very recently - the + * current timeline, so more data might show up. Delay here + * so we don't tight-loop. + */ + HandleWalSummarizerInterrupts(); + summarizer_wait_for_wal(); + private_data->waited = true; + + /* Recheck end-of-WAL. */ + latest_lsn = GetLatestLSN(&latest_tli); + if (private_data->tli == latest_tli) + { + /* Still the current timeline, update max LSN. */ + Assert(latest_lsn >= private_data->read_upto); + private_data->read_upto = latest_lsn; + } + else + { + List *tles = readTimeLineHistory(latest_tli); + XLogRecPtr switchpoint; + + /* + * The timeline we're scanning is no longer the latest + * one. Figure out when it ended and allow reads up to + * exactly that point. + */ + private_data->historic = true; + switchpoint = tliSwitchPoint(private_data->tli, tles, + NULL); + Assert(switchpoint >= private_data->read_upto); + private_data->read_upto = switchpoint; + } + + /* Go around and try again. */ + } + } + else + { + /* enough bytes available to satisfy the request */ + count = private_data->read_upto - targetPagePtr; + break; + } + } + + /* + * Even though we just determined how much of the page can be validly read + * as 'count', read the whole page anyway. It's guaranteed to be + * zero-padded up to the page boundary if it's incomplete. + */ + if (!WALRead(state, cur_page, targetPagePtr, XLOG_BLCKSZ, + private_data->tli, &errinfo)) + WALReadRaiseError(&errinfo); + + /* Track that we read a page, for sleep time calculation. */ + ++pages_read_since_last_sleep; + + /* number of valid bytes in the buffer */ + return count; +} + +/* + * Sleep for long enough that we believe it's likely that more WAL will + * be available afterwards. + */ +static void +summarizer_wait_for_wal(void) +{ + if (pages_read_since_last_sleep == 0) + { + /* + * No pages were read since the last sleep, so double the sleep time, + * but not beyond the maximum allowable value. + */ + sleep_quanta = Min(sleep_quanta * 2, MAX_SLEEP_QUANTA); + } + else if (pages_read_since_last_sleep > 1) + { + /* + * Multiple pages were read since the last sleep, so reduce the sleep + * time. + * + * A large burst of activity should be able to quickly reduce the + * sleep time to the minimum, but we don't want a handful of extra WAL + * records to provoke a strong reaction. We choose to reduce the sleep + * time by 1 quantum for each page read beyond the first, which is a + * fairly arbitrary way of trying to be reactive without + * overrreacting. + */ + if (pages_read_since_last_sleep > sleep_quanta - 1) + sleep_quanta = 1; + else + sleep_quanta -= pages_read_since_last_sleep; + } + + /* OK, now sleep. */ + (void) WaitLatch(MyLatch, + WL_LATCH_SET | WL_TIMEOUT | WL_EXIT_ON_PM_DEATH, + sleep_quanta * MS_PER_SLEEP_QUANTUM, + WAIT_EVENT_WAL_SUMMARIZER_WAL); + ResetLatch(MyLatch); + + /* Reset count of pages read. */ + pages_read_since_last_sleep = 0; +} + +/* + * Most recent RedoRecPtr value observed by RemoveOldWalSummaries. + */ +static void +MaybeRemoveOldWalSummaries(void) +{ + XLogRecPtr redo_pointer = GetRedoRecPtr(); + List *wslist; + time_t cutoff_time; + + /* If WAL summary removal is disabled, don't do anything. */ + if (wal_summarize_keep_time == 0) + return; + + /* + * If the redo pointer has not advanced, don't do anything. + * + * This has the effect that we only try to remove old WAL summary files + * once per checkpoint cycle. + */ + if (redo_pointer == redo_pointer_at_last_summary_removal) + return; + redo_pointer_at_last_summary_removal = redo_pointer; + + /* + * Files should only be removed if the last modification time precedes the + * cutoff time we compute here. + */ + cutoff_time = time(NULL) - 60 * wal_summarize_keep_time; + + /* Get all the summaries that currently exist. */ + wslist = GetWalSummaries(0, InvalidXLogRecPtr, InvalidXLogRecPtr); + + /* Loop until all summaries have been considered for removal. */ + while (wslist != NIL) + { + ListCell *lc; + XLogSegNo oldest_segno; + XLogRecPtr oldest_lsn = InvalidXLogRecPtr; + TimeLineID selected_tli; + + CHECK_FOR_INTERRUPTS(); + + /* + * Pick a timeline for which some summary files still exist on disk, + * and find the oldest LSN that still exists on disk for that + * timeline. + */ + selected_tli = ((WalSummaryFile *) linitial(wslist))->tli; + oldest_segno = XLogGetOldestSegno(selected_tli); + if (oldest_segno != 0) + XLogSegNoOffsetToRecPtr(oldest_segno, 0, wal_segment_size, + oldest_lsn); + + + /* Consider each WAL file on the selected timeline in turn. */ + foreach(lc, wslist) + { + WalSummaryFile *ws = lfirst(lc); + + CHECK_FOR_INTERRUPTS(); + + /* If it's not on this timeline, it's not time to consider it. */ + if (selected_tli != ws->tli) + continue; + + /* + * If the WAL doesn't exist any more, we can remove it if the file + * modification time is old enough. + */ + if (XLogRecPtrIsInvalid(oldest_lsn) || ws->end_lsn <= oldest_lsn) + RemoveWalSummaryIfOlderThan(ws, cutoff_time); + + /* + * Whether we we removed the file or not, we need not consider it + * again. + */ + wslist = foreach_delete_current(wslist, lc); + pfree(ws); + } + } +} diff --git a/src/backend/replication/repl_gram.y b/src/backend/replication/repl_gram.y index 0c874e33cf..a5d118ed68 100644 --- a/src/backend/replication/repl_gram.y +++ b/src/backend/replication/repl_gram.y @@ -76,11 +76,12 @@ Node *replication_parse_result; %token K_EXPORT_SNAPSHOT %token K_NOEXPORT_SNAPSHOT %token K_USE_SNAPSHOT +%token K_UPLOAD_MANIFEST %type <node> command %type <node> base_backup start_replication start_logical_replication create_replication_slot drop_replication_slot identify_system - read_replication_slot timeline_history show + read_replication_slot timeline_history show upload_manifest %type <list> generic_option_list %type <defelt> generic_option %type <uintval> opt_timeline @@ -114,6 +115,7 @@ command: | read_replication_slot | timeline_history | show + | upload_manifest ; /* @@ -307,6 +309,15 @@ timeline_history: } ; +/* UPLOAD_MANIFEST doesn't currently accept any arguments */ +upload_manifest: + K_UPLOAD_MANIFEST + { + UploadManifestCmd *cmd = makeNode(UploadManifestCmd); + + $$ = (Node *) cmd; + } + opt_physical: K_PHYSICAL | /* EMPTY */ @@ -411,6 +422,7 @@ ident_or_keyword: | K_EXPORT_SNAPSHOT { $$ = "export_snapshot"; } | K_NOEXPORT_SNAPSHOT { $$ = "noexport_snapshot"; } | K_USE_SNAPSHOT { $$ = "use_snapshot"; } + | K_UPLOAD_MANIFEST { $$ = "upload_manifest"; } ; %% diff --git a/src/backend/replication/repl_scanner.l b/src/backend/replication/repl_scanner.l index 1cc7fb858c..4805da08ee 100644 --- a/src/backend/replication/repl_scanner.l +++ b/src/backend/replication/repl_scanner.l @@ -136,6 +136,7 @@ EXPORT_SNAPSHOT { return K_EXPORT_SNAPSHOT; } NOEXPORT_SNAPSHOT { return K_NOEXPORT_SNAPSHOT; } USE_SNAPSHOT { return K_USE_SNAPSHOT; } WAIT { return K_WAIT; } +UPLOAD_MANIFEST { return K_UPLOAD_MANIFEST; } {space}+ { /* do nothing */ } @@ -303,6 +304,7 @@ replication_scanner_is_replication_command(void) case K_DROP_REPLICATION_SLOT: case K_READ_REPLICATION_SLOT: case K_TIMELINE_HISTORY: + case K_UPLOAD_MANIFEST: case K_SHOW: /* Yes; push back the first token so we can parse later. */ repl_pushed_back_token = first_token; diff --git a/src/backend/replication/walsender.c b/src/backend/replication/walsender.c index 80374c55be..a73d15fdd5 100644 --- a/src/backend/replication/walsender.c +++ b/src/backend/replication/walsender.c @@ -58,6 +58,7 @@ #include "access/xlogrecovery.h" #include "access/xlogutils.h" #include "backup/basebackup.h" +#include "backup/basebackup_incremental.h" #include "catalog/pg_authid.h" #include "catalog/pg_type.h" #include "commands/dbcommands.h" @@ -137,6 +138,17 @@ bool wake_wal_senders = false; */ static XLogReaderState *xlogreader = NULL; +/* + * If the UPLOAD_MANIFEST command is used to provide a backup manifest in + * preparation for an incremental backup, uploaded_manifest will be point + * to an object containing information about its contexts, and + * uploaded_manifest_mcxt will point to the memory context that contains + * that object and all of its subordinate data. Otherwise, both values will + * be NULL. + */ +static IncrementalBackupInfo *uploaded_manifest = NULL; +static MemoryContext uploaded_manifest_mcxt = NULL; + /* * These variables keep track of the state of the timeline we're currently * sending. sendTimeLine identifies the timeline. If sendTimeLineIsHistoric, @@ -233,6 +245,9 @@ static void XLogSendLogical(void); static void WalSndDone(WalSndSendDataCallback send_data); static XLogRecPtr GetStandbyFlushRecPtr(TimeLineID *tli); static void IdentifySystem(void); +static void UploadManifest(void); +static bool HandleUploadManifestPacket(StringInfo buf, off_t *offset, + IncrementalBackupInfo *ib); static void ReadReplicationSlot(ReadReplicationSlotCmd *cmd); static void CreateReplicationSlot(CreateReplicationSlotCmd *cmd); static void DropReplicationSlot(DropReplicationSlotCmd *cmd); @@ -660,6 +675,143 @@ SendTimeLineHistory(TimeLineHistoryCmd *cmd) pq_endmessage(&buf); } +/* + * Handle UPLOAD_MANIFEST command. + */ +static void +UploadManifest(void) +{ + MemoryContext mcxt; + IncrementalBackupInfo *ib; + off_t offset = 0; + StringInfoData buf; + + /* + * parsing the manifest will use the cryptohash stuff, which requires a + * resource owner + */ + Assert(CurrentResourceOwner == NULL); + CurrentResourceOwner = ResourceOwnerCreate(NULL, "base backup"); + + /* Prepare to read manifest data into a temporary context. */ + mcxt = AllocSetContextCreate(CurrentMemoryContext, + "incremental backup information", + ALLOCSET_DEFAULT_SIZES); + ib = CreateIncrementalBackupInfo(mcxt); + + /* Send a CopyInResponse message */ + pq_beginmessage(&buf, 'G'); + pq_sendbyte(&buf, 0); + pq_sendint16(&buf, 0); + pq_endmessage_reuse(&buf); + pq_flush(); + + /* Recieve packets from client until done. */ + while (HandleUploadManifestPacket(&buf, &offset, ib)) + ; + + /* Finish up manifest processing. */ + FinalizeIncrementalManifest(ib); + + /* + * Discard any old manifest information and arrange to preserve the new + * information we just got. + * + * We assume that MemoryContextDelete and MemoryContextSetParent won't + * fail, and thus we shouldn't end up bailing out of here in such a way as + * to leave dangling pointrs. + */ + if (uploaded_manifest_mcxt != NULL) + MemoryContextDelete(uploaded_manifest_mcxt); + MemoryContextSetParent(mcxt, CacheMemoryContext); + uploaded_manifest = ib; + uploaded_manifest_mcxt = mcxt; + + /* clean up the resource owner we created */ + WalSndResourceCleanup(true); +} + +/* + * Process one packet received during the handling of an UPLOAD_MANIFEST + * operation. + * + * 'buf' is scratch space. This function expects it to be initialized, doesn't + * care what the current contents are, and may override them with completely + * new contents. + * + * The return value is true if the caller should continue processing + * additional packets and false if the UPLOAD_MANIFEST operation is complete. + */ +static bool +HandleUploadManifestPacket(StringInfo buf, off_t *offset, + IncrementalBackupInfo *ib) +{ + int mtype; + int maxmsglen; + + HOLD_CANCEL_INTERRUPTS(); + + pq_startmsgread(); + mtype = pq_getbyte(); + if (mtype == EOF) + ereport(ERROR, + (errcode(ERRCODE_CONNECTION_FAILURE), + errmsg("unexpected EOF on client connection with an open transaction"))); + + switch (mtype) + { + case 'd': /* CopyData */ + maxmsglen = PQ_LARGE_MESSAGE_LIMIT; + break; + case 'c': /* CopyDone */ + case 'f': /* CopyFail */ + case 'H': /* Flush */ + case 'S': /* Sync */ + maxmsglen = PQ_SMALL_MESSAGE_LIMIT; + break; + default: + ereport(ERROR, + (errcode(ERRCODE_PROTOCOL_VIOLATION), + errmsg("unexpected message type 0x%02X during COPY from stdin", + mtype))); + maxmsglen = 0; /* keep compiler quiet */ + break; + } + + /* Now collect the message body */ + if (pq_getmessage(buf, maxmsglen)) + ereport(ERROR, + (errcode(ERRCODE_CONNECTION_FAILURE), + errmsg("unexpected EOF on client connection with an open transaction"))); + RESUME_CANCEL_INTERRUPTS(); + + /* Process the message */ + switch (mtype) + { + case 'd': /* CopyData */ + AppendIncrementalManifestData(ib, buf->data, buf->len); + return true; + + case 'c': /* CopyDone */ + return false; + + case 'H': /* Sync */ + case 'S': /* Flush */ + /* Ignore these while in CopyOut mode as we do elsewhere. */ + return true; + + case 'f': + ereport(ERROR, + (errcode(ERRCODE_QUERY_CANCELED), + errmsg("COPY from stdin failed: %s", + pq_getmsgstring(buf)))); + } + + /* Not reached. */ + Assert(false); + return false; +} + /* * Handle START_REPLICATION command. * @@ -1801,7 +1953,7 @@ exec_replication_command(const char *cmd_string) cmdtag = "BASE_BACKUP"; set_ps_display(cmdtag); PreventInTransactionBlock(true, cmdtag); - SendBaseBackup((BaseBackupCmd *) cmd_node); + SendBaseBackup((BaseBackupCmd *) cmd_node, uploaded_manifest); EndReplicationCommand(cmdtag); break; @@ -1863,6 +2015,14 @@ exec_replication_command(const char *cmd_string) } break; + case T_UploadManifestCmd: + cmdtag = "UPLOAD_MANIFEST"; + set_ps_display(cmdtag); + PreventInTransactionBlock(true, cmdtag); + UploadManifest(); + EndReplicationCommand(cmdtag); + break; + default: elog(ERROR, "unrecognized replication command node tag: %u", cmd_node->type); diff --git a/src/backend/storage/ipc/ipci.c b/src/backend/storage/ipc/ipci.c index 5551afffc0..ff0660656c 100644 --- a/src/backend/storage/ipc/ipci.c +++ b/src/backend/storage/ipc/ipci.c @@ -31,6 +31,7 @@ #include "postmaster/bgworker_internals.h" #include "postmaster/bgwriter.h" #include "postmaster/postmaster.h" +#include "postmaster/walsummarizer.h" #include "replication/logicallauncher.h" #include "replication/origin.h" #include "replication/slot.h" @@ -136,6 +137,7 @@ CalculateShmemSize(int *num_semaphores) size = add_size(size, ReplicationOriginShmemSize()); size = add_size(size, WalSndShmemSize()); size = add_size(size, WalRcvShmemSize()); + size = add_size(size, WalSummarizerShmemSize()); size = add_size(size, PgArchShmemSize()); size = add_size(size, ApplyLauncherShmemSize()); size = add_size(size, SnapMgrShmemSize()); @@ -292,6 +294,7 @@ CreateSharedMemoryAndSemaphores(void) ReplicationOriginShmemInit(); WalSndShmemInit(); WalRcvShmemInit(); + WalSummarizerShmemInit(); PgArchShmemInit(); ApplyLauncherShmemInit(); diff --git a/src/backend/storage/lmgr/lwlocknames.txt b/src/backend/storage/lmgr/lwlocknames.txt index 811ad94742..4a315bfe93 100644 --- a/src/backend/storage/lmgr/lwlocknames.txt +++ b/src/backend/storage/lmgr/lwlocknames.txt @@ -54,3 +54,4 @@ XactTruncationLock 44 WrapLimitsVacuumLock 46 NotifyQueueTailLock 47 WaitEventExtensionLock 48 +WALSummarizerLock 49 diff --git a/src/backend/utils/activity/pgstat_io.c b/src/backend/utils/activity/pgstat_io.c index eb7d35d422..bd0a921a3e 100644 --- a/src/backend/utils/activity/pgstat_io.c +++ b/src/backend/utils/activity/pgstat_io.c @@ -292,7 +292,8 @@ pgstat_io_snapshot_cb(void) * - Syslogger because it is not connected to shared memory * - Archiver because most relevant archiving IO is delegated to a * specialized command or module -* - WAL Receiver and WAL Writer IO is not tracked in pg_stat_io for now +* - WAL Receiver, WAL Writer, and WAL Summarizer IO are not tracked in +* pg_stat_io for now * * Function returns true if BackendType participates in the cumulative stats * subsystem for IO and false if it does not. @@ -314,6 +315,7 @@ pgstat_tracks_io_bktype(BackendType bktype) case B_LOGGER: case B_WAL_RECEIVER: case B_WAL_WRITER: + case B_WAL_SUMMARIZER: return false; case B_AUTOVAC_LAUNCHER: diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 13774254d2..2ea39fd824 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -56,6 +56,7 @@ WAIT_EVENT_RECOVERY_WAL_STREAM RecoveryWalStream "Waiting in main loop of startu WAIT_EVENT_SYSLOGGER_MAIN SysLoggerMain "Waiting in main loop of syslogger process." WAIT_EVENT_WAL_RECEIVER_MAIN WalReceiverMain "Waiting in main loop of WAL receiver process." WAIT_EVENT_WAL_SENDER_MAIN WalSenderMain "Waiting in main loop of WAL sender process." +WAIT_EVENT_WAL_SUMMARIZER_WAL WalSummarizerError "Waiting in WAL summarizer for more WAL to be generated." WAIT_EVENT_WAL_WRITER_MAIN WalWriterMain "Waiting in main loop of WAL writer process." @@ -140,6 +141,7 @@ WAIT_EVENT_SAFE_SNAPSHOT SafeSnapshot "Waiting to obtain a valid snapshot for a WAIT_EVENT_SYNC_REP SyncRep "Waiting for confirmation from a remote server during synchronous replication." WAIT_EVENT_WAL_RECEIVER_EXIT WalReceiverExit "Waiting for the WAL receiver to exit." WAIT_EVENT_WAL_RECEIVER_WAIT_START WalReceiverWaitStart "Waiting for startup process to send initial data for streaming replication." +WAIT_EVENT_WAL_SUMMARY_READY WalSummaryReady "Waiting for a new WAL summary to be generated." WAIT_EVENT_XACT_GROUP_UPDATE XactGroupUpdate "Waiting for the group leader to update transaction status at end of a parallel operation." @@ -160,6 +162,7 @@ WAIT_EVENT_REGISTER_SYNC_REQUEST RegisterSyncRequest "Waiting while sending sync WAIT_EVENT_SPIN_DELAY SpinDelay "Waiting while acquiring a contended spinlock." WAIT_EVENT_VACUUM_DELAY VacuumDelay "Waiting in a cost-based vacuum delay point." WAIT_EVENT_VACUUM_TRUNCATE VacuumTruncate "Waiting to acquire an exclusive lock to truncate off any empty pages at the end of a table vacuumed." +WAIT_EVENT_WAL_SUMMARIZER_ERROR WalSummarizerError "Waiting after a WAL summarizer error." # @@ -241,6 +244,8 @@ WAIT_EVENT_WAL_COPY_WRITE WALCopyWrite "Waiting for a write when creating a new WAIT_EVENT_WAL_INIT_SYNC WALInitSync "Waiting for a newly initialized WAL file to reach durable storage." WAIT_EVENT_WAL_INIT_WRITE WALInitWrite "Waiting for a write while initializing a new WAL file." WAIT_EVENT_WAL_READ WALRead "Waiting for a read from a WAL file." +WAIT_EVENT_WAL_SUMMARY_READ WALSummaryRead "Waiting for a read from a WAL summary file." +WAIT_EVENT_WAL_SUMMARY_WRITE WALSummaryWrite "Waiting for a write to a WAL summary file." WAIT_EVENT_WAL_SYNC WALSync "Waiting for a WAL file to reach durable storage." WAIT_EVENT_WAL_SYNC_METHOD_ASSIGN WALSyncMethodAssign "Waiting for data to reach durable storage while assigning a new WAL sync method." WAIT_EVENT_WAL_WRITE WALWrite "Waiting for a write to a WAL file." diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c index 1e671c560c..037111b89f 100644 --- a/src/backend/utils/init/miscinit.c +++ b/src/backend/utils/init/miscinit.c @@ -306,6 +306,9 @@ GetBackendTypeDesc(BackendType backendType) case B_WAL_SENDER: backendDesc = "walsender"; break; + case B_WAL_SUMMARIZER: + backendDesc = "walsummarizer"; + break; case B_WAL_WRITER: backendDesc = "walwriter"; break; diff --git a/src/backend/utils/misc/guc_tables.c b/src/backend/utils/misc/guc_tables.c index e565a3092f..91dc345d8b 100644 --- a/src/backend/utils/misc/guc_tables.c +++ b/src/backend/utils/misc/guc_tables.c @@ -61,6 +61,7 @@ #include "postmaster/postmaster.h" #include "postmaster/startup.h" #include "postmaster/syslogger.h" +#include "postmaster/walsummarizer.h" #include "postmaster/walwriter.h" #include "replication/logicallauncher.h" #include "replication/slot.h" @@ -702,6 +703,8 @@ const char *const config_group_names[] = gettext_noop("Write-Ahead Log / Archive Recovery"), /* WAL_RECOVERY_TARGET */ gettext_noop("Write-Ahead Log / Recovery Target"), + /* WAL_SUMMARIZATION */ + gettext_noop("Write-Ahead Log / Summarization"), /* REPLICATION_SENDING */ gettext_noop("Replication / Sending Servers"), /* REPLICATION_PRIMARY */ @@ -3169,6 +3172,32 @@ struct config_int ConfigureNamesInt[] = check_wal_segment_size, NULL, NULL }, + { + {"wal_summarize_mb", PGC_SIGHUP, WAL_SUMMARIZATION, + gettext_noop("Number of bytes of WAL per summary file."), + gettext_noop("Smaller values minimize extra work performed by incremental backup, but increase the number of files on disk."), + GUC_UNIT_MB, + }, + &wal_summarize_mb, + 256, + 0, + INT_MAX, + NULL, NULL, NULL + }, + + { + {"wal_summarize_keep_time", PGC_SIGHUP, WAL_SUMMARIZATION, + gettext_noop("Time for which WAL summary files should be kept."), + NULL, + GUC_UNIT_MIN, + }, + &wal_summarize_keep_time, + 7 * 24 * 60, /* 1 week */ + 0, + INT_MAX, + NULL, NULL, NULL + }, + { {"autovacuum_naptime", PGC_SIGHUP, AUTOVACUUM, gettext_noop("Time to sleep between autovacuum runs."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index c768af9a73..1211de5ea3 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -301,6 +301,11 @@ #recovery_target_action = 'pause' # 'pause', 'promote', 'shutdown' # (change requires restart) +# - WAL Summarization - + +#wal_summarize_mb = 256 # MB of WAL per summary file, 0 disables +#wal_summarize_keep_time = '7d' # when to remove old summary files, 0 = never + #------------------------------------------------------------------------------ # REPLICATION diff --git a/src/bin/Makefile b/src/bin/Makefile index 373077bf52..aa2210925e 100644 --- a/src/bin/Makefile +++ b/src/bin/Makefile @@ -19,6 +19,7 @@ SUBDIRS = \ pg_archivecleanup \ pg_basebackup \ pg_checksums \ + pg_combinebackup \ pg_config \ pg_controldata \ pg_ctl \ diff --git a/src/bin/initdb/initdb.c b/src/bin/initdb/initdb.c index 905b979947..09d153ed88 100644 --- a/src/bin/initdb/initdb.c +++ b/src/bin/initdb/initdb.c @@ -226,6 +226,7 @@ static char *extra_options = ""; static const char *const subdirs[] = { "global", "pg_wal/archive_status", + "pg_wal/summaries", "pg_commit_ts", "pg_dynshmem", "pg_notify", diff --git a/src/bin/meson.build b/src/bin/meson.build index 67cb50630c..4cb6fd59bb 100644 --- a/src/bin/meson.build +++ b/src/bin/meson.build @@ -5,6 +5,7 @@ subdir('pg_amcheck') subdir('pg_archivecleanup') subdir('pg_basebackup') subdir('pg_checksums') +subdir('pg_combinebackup') subdir('pg_config') subdir('pg_controldata') subdir('pg_ctl') diff --git a/src/bin/pg_basebackup/bbstreamer_file.c b/src/bin/pg_basebackup/bbstreamer_file.c index 45f32974ff..6b78ee283d 100644 --- a/src/bin/pg_basebackup/bbstreamer_file.c +++ b/src/bin/pg_basebackup/bbstreamer_file.c @@ -296,6 +296,7 @@ should_allow_existing_directory(const char *pathname) if (strcmp(filename, "pg_wal") == 0 || strcmp(filename, "pg_xlog") == 0 || strcmp(filename, "archive_status") == 0 || + strcmp(filename, "summaries") == 0 || strcmp(filename, "pg_tblspc") == 0) return true; diff --git a/src/bin/pg_basebackup/pg_basebackup.c b/src/bin/pg_basebackup/pg_basebackup.c index 1dc8efe0cb..3ffe15ac74 100644 --- a/src/bin/pg_basebackup/pg_basebackup.c +++ b/src/bin/pg_basebackup/pg_basebackup.c @@ -101,6 +101,11 @@ typedef void (*WriteDataCallback) (size_t nbytes, char *buf, */ #define MINIMUM_VERSION_FOR_TERMINATED_TARFILE 150000 +/* + * pg_wal/summaries exists beginning with v16. + */ +#define MINIMUM_VERSION_FOR_WAL_SUMMARIES 160000 + /* * Different ways to include WAL */ @@ -216,7 +221,8 @@ static void ReceiveBackupManifestInMemoryChunk(size_t r, char *copybuf, void *callback_data); static void BaseBackup(char *compression_algorithm, char *compression_detail, CompressionLocation compressloc, - pg_compress_specification *client_compress); + pg_compress_specification *client_compress, + char *incremental_manifest); static bool reached_end_position(XLogRecPtr segendpos, uint32 timeline, bool segment_finished); @@ -684,6 +690,23 @@ StartLogStreamer(char *startpos, uint32 timeline, char *sysidentifier, if (pg_mkdir_p(statusdir, pg_dir_create_mode) != 0 && errno != EEXIST) pg_fatal("could not create directory \"%s\": %m", statusdir); + + /* + * For newer server versions, likewise create pg_wal/summaries + */ + if (PQserverVersion(conn) < MINIMUM_VERSION_FOR_WAL_SUMMARIES) + { + char summarydir[MAXPGPATH]; + + snprintf(summarydir, sizeof(summarydir), "%s/%s/summaries", + basedir, + PQserverVersion(conn) < MINIMUM_VERSION_FOR_PG_WAL ? + "pg_xlog" : "pg_wal"); + + if (pg_mkdir_p(statusdir, pg_dir_create_mode) != 0 && + errno != EEXIST) + pg_fatal("could not create directory \"%s\": %m", summarydir); + } } /* @@ -1724,7 +1747,9 @@ ReceiveBackupManifestInMemoryChunk(size_t r, char *copybuf, static void BaseBackup(char *compression_algorithm, char *compression_detail, - CompressionLocation compressloc, pg_compress_specification *client_compress) + CompressionLocation compressloc, + pg_compress_specification *client_compress, + char *incremental_manifest) { PGresult *res; char *sysidentifier; @@ -1790,7 +1815,74 @@ BaseBackup(char *compression_algorithm, char *compression_detail, exit(1); /* - * Start the actual backup + * If the user wants an incremental backup, we must upload the manifest + * for the previous backup upon which it is to be based. + */ + if (incremental_manifest != NULL) + { + int fd; + char mbuf[65536]; + int nbytes; + + /* XXX add a server version check here */ + + /* Open the file. */ + fd = open(incremental_manifest, O_RDONLY | PG_BINARY, 0); + if (fd < 0) + pg_fatal("could not open file \"%s\": %m", incremental_manifest); + + /* Tell the server what we want to do. */ + if (PQsendQuery(conn, "UPLOAD_MANIFEST") == 0) + pg_fatal("could not send replication command \"%s\": %s", + "UPLOAD_MANIFEST", PQerrorMessage(conn)); + res = PQgetResult(conn); + if (PQresultStatus(res) != PGRES_COPY_IN) + { + if (PQresultStatus(res) == PGRES_FATAL_ERROR) + pg_fatal("could not upload manifest: %s", + PQerrorMessage(conn)); + else + pg_fatal("could not upload manifest: unexpected status %s", + PQresStatus(PQresultStatus(res))); + } + + /* Loop, reading from the file and sending the data to the server. */ + while ((nbytes = read(fd, mbuf, sizeof mbuf)) > 0) + { + if (PQputCopyData(conn, mbuf, nbytes) < 0) + pg_fatal("could not send COPY data: %s", + PQerrorMessage(conn)); + } + + /* Bail out if we exited the loop due to an error. */ + if (nbytes < 0) + pg_fatal("could not read file \"%s\": %m", incremental_manifest); + + /* End the COPY operation. */ + if (PQputCopyEnd(conn, NULL) < 0) + pg_fatal("could not send end-of-COPY: %s", + PQerrorMessage(conn)); + + /* See whether the server is happy with what we sent. */ + res = PQgetResult(conn); + if (PQresultStatus(res) == PGRES_FATAL_ERROR) + pg_fatal("could not upload manifest: %s", + PQerrorMessage(conn)); + else if (PQresultStatus(res) != PGRES_COMMAND_OK) + pg_fatal("could not upload manifest: unexpected status %s", + PQresStatus(PQresultStatus(res))); + + /* Consume ReadyForQuery message from server. */ + res = PQgetResult(conn); + if (res != NULL) + pg_fatal("unexpected extra result while sending manifest"); + + /* Add INCREMENTAL option to BASE_BACKUP command. */ + AppendPlainCommandOption(&buf, use_new_option_syntax, "INCREMENTAL"); + } + + /* + * Continue building up the options list for the BASE_BACKUP command. */ AppendStringCommandOption(&buf, use_new_option_syntax, "LABEL", label); if (estimatesize) @@ -1897,6 +1989,7 @@ BaseBackup(char *compression_algorithm, char *compression_detail, else basebkp = psprintf("BASE_BACKUP %s", buf.data); + /* OK, try to start the backup. */ if (PQsendQuery(conn, basebkp) == 0) pg_fatal("could not send replication command \"%s\": %s", "BASE_BACKUP", PQerrorMessage(conn)); @@ -2252,6 +2345,7 @@ main(int argc, char **argv) {"version", no_argument, NULL, 'V'}, {"pgdata", required_argument, NULL, 'D'}, {"format", required_argument, NULL, 'F'}, + {"incremental", required_argument, NULL, 'i'}, {"checkpoint", required_argument, NULL, 'c'}, {"create-slot", no_argument, NULL, 'C'}, {"max-rate", required_argument, NULL, 'r'}, @@ -2288,6 +2382,7 @@ main(int argc, char **argv) int option_index; char *compression_algorithm = "none"; char *compression_detail = NULL; + char *incremental_manifest = NULL; CompressionLocation compressloc = COMPRESS_LOCATION_UNSPECIFIED; pg_compress_specification client_compress; @@ -2312,7 +2407,7 @@ main(int argc, char **argv) atexit(cleanup_directories_atexit); - while ((c = getopt_long(argc, argv, "c:Cd:D:F:h:l:nNp:Pr:Rs:S:t:T:U:vwWX:zZ:", + while ((c = getopt_long(argc, argv, "c:Cd:D:F:h:i:l:nNp:Pr:Rs:S:t:T:U:vwWX:zZ:", long_options, &option_index)) != -1) { switch (c) @@ -2347,6 +2442,9 @@ main(int argc, char **argv) case 'h': dbhost = pg_strdup(optarg); break; + case 'i': + incremental_manifest = pg_strdup(optarg); + break; case 'l': label = pg_strdup(optarg); break; @@ -2756,7 +2854,7 @@ main(int argc, char **argv) } BaseBackup(compression_algorithm, compression_detail, compressloc, - &client_compress); + &client_compress, incremental_manifest); success = true; return 0; diff --git a/src/bin/pg_basebackup/t/010_pg_basebackup.pl b/src/bin/pg_basebackup/t/010_pg_basebackup.pl index b9f5e1266b..bf765291e7 100644 --- a/src/bin/pg_basebackup/t/010_pg_basebackup.pl +++ b/src/bin/pg_basebackup/t/010_pg_basebackup.pl @@ -223,10 +223,10 @@ SKIP: "check backup dir permissions"); } -# Only archive_status directory should be copied in pg_wal/. +# Only archive_status and summaries directories should be copied in pg_wal/. is_deeply( [ sort(slurp_dir("$tempdir/backup/pg_wal/")) ], - [ sort qw(. .. archive_status) ], + [ sort qw(. .. archive_status summaries) ], 'no WAL files copied'); # Contents of these directories should not be copied. diff --git a/src/bin/pg_combinebackup/.gitignore b/src/bin/pg_combinebackup/.gitignore new file mode 100644 index 0000000000..d7e617438c --- /dev/null +++ b/src/bin/pg_combinebackup/.gitignore @@ -0,0 +1 @@ +pg_combinebackup diff --git a/src/bin/pg_combinebackup/Makefile b/src/bin/pg_combinebackup/Makefile new file mode 100644 index 0000000000..cb20480aae --- /dev/null +++ b/src/bin/pg_combinebackup/Makefile @@ -0,0 +1,46 @@ +#------------------------------------------------------------------------- +# +# Makefile for src/bin/pg_combinebackup +# +# Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group +# Portions Copyright (c) 1994, Regents of the University of California +# +# src/bin/pg_combinebackup/Makefile +# +#------------------------------------------------------------------------- + +PGFILEDESC = "pg_combinebackup - combine incremental backups" +PGAPPICON=win32 + +subdir = src/bin/pg_combinebackup +top_builddir = ../../.. +include $(top_builddir)/src/Makefile.global + +override CPPFLAGS := -I$(libpq_srcdir) $(CPPFLAGS) +LDFLAGS_INTERNAL += -L$(top_builddir)/src/fe_utils -lpgfeutils + +OBJS = \ + $(WIN32RES) \ + pg_combinebackup.o \ + backup_label.o \ + copy_file.o \ + load_manifest.o \ + reconstruct.o \ + write_manifest.o + +all: pg_combinebackup + +pg_combinebackup: $(OBJS) | submake-libpgport submake-libpgfeutils + $(CC) $(CFLAGS) $^ $(LDFLAGS) $(LDFLAGS_EX) $(LIBS) -o $@$(X) + +install: all installdirs + $(INSTALL_PROGRAM) pg_combinebackup$(X) '$(DESTDIR)$(bindir)/pg_combinebackup$(X)' + +installdirs: + $(MKDIR_P) '$(DESTDIR)$(bindir)' + +uninstall: + rm -f '$(DESTDIR)$(bindir)/pg_combinebackup$(X)' + +clean distclean maintainer-clean: + rm -f pg_combinebackup$(X) $(OBJS) diff --git a/src/bin/pg_combinebackup/backup_label.c b/src/bin/pg_combinebackup/backup_label.c new file mode 100644 index 0000000000..2a62aa6fad --- /dev/null +++ b/src/bin/pg_combinebackup/backup_label.c @@ -0,0 +1,281 @@ +/*------------------------------------------------------------------------- + * + * Read and manipulate backup label files + * + * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group + * Portions Copyright (c) 1994, Regents of the University of California + * + * src/bin/pg_combinebackup/backup_label.c + * + *------------------------------------------------------------------------- + */ +#include "postgres_fe.h" + +#include <unistd.h> + +#include "access/xlogdefs.h" +#include "backup_label.h" +#include "common/logging.h" +#include "common/file_perm.h" +#include "write_manifest.h" + +static int get_eol_offset(StringInfo buf); +static bool line_starts_with(char *s, char *e, char *match, char **sout); +static bool parse_lsn(char *s, char *e, XLogRecPtr *lsn, char **c); +static bool parse_tli(char *s, char *e, TimeLineID *tli); + +/* + * Parse a backup label file, starting at buf->cursor. + * + * We expect to find a START WAL LOCATION line, followed by a LSN, followed + * by a space; the resulting LSN is stored into *start_lsn. + * + * We expect to find a START TIMELINE line, followed by a TLI, followed by + * a newline; the resulting TLI is stored into *start_tli. + * + * We expect to find either both INCREMENTAL FROM LSN and INCREMENTAL FROM TLI + * or neither. If these are found, they should be followed by an LSN or TLI + * respectively and then by a newline, and the values will be stored into + * *previous_lsn and *previous_tli, respectively. + * + * Other lines in the provided backup_label data are ignored. filename is used + * for error reporting; errors are fatal. + */ +void +parse_backup_label(char *filename, StringInfo buf, + TimeLineID *start_tli, XLogRecPtr *start_lsn, + TimeLineID *previous_tli, XLogRecPtr *previous_lsn) +{ + int found = 0; + + *start_tli = 0; + *start_lsn = InvalidXLogRecPtr; + *previous_tli = 0; + *previous_lsn = InvalidXLogRecPtr; + + while (buf->cursor < buf->len) + { + char *s = &buf->data[buf->cursor]; + int eo = get_eol_offset(buf); + char *e = &buf->data[eo]; + char *c; + + if (line_starts_with(s, e, "START WAL LOCATION: ", &s)) + { + if (!parse_lsn(s, e, start_lsn, &c)) + pg_fatal("%s: could not parse START WAL LOCATION", + filename); + if (c >= e || *c != ' ') + pg_fatal("%s: improper terminator for START WAL LOCATION", + filename); + found |= 1; + } + else if (line_starts_with(s, e, "START TIMELINE: ", &s)) + { + if (!parse_tli(s, e, start_tli)) + pg_fatal("%s: could not parse TLI for START TIMELINE", + filename); + if (*start_tli == 0) + pg_fatal("%s: invalid TLI", filename); + found |= 2; + } + else if (line_starts_with(s, e, "INCREMENTAL FROM LSN: ", &s)) + { + if (!parse_lsn(s, e, previous_lsn, &c)) + pg_fatal("%s: could not parse INCREMENTAL FROM LSN", + filename); + if (c >= e || *c != '\n') + pg_fatal("%s: improper terminator for INCREMENTAL FROM LSN", + filename); + found |= 4; + } + else if (line_starts_with(s, e, "INCREMENTAL FROM TLI: ", &s)) + { + if (!parse_tli(s, e, previous_tli)) + pg_fatal("%s: could not parse INCREMENTAL FROM TLI", + filename); + if (*previous_tli == 0) + pg_fatal("%s: invalid TLI", filename); + found |= 8; + } + + buf->cursor = eo; + } + + if ((found & 1) == 0) + pg_fatal("%s: could not find START WAL LOCATION", filename); + if ((found & 2) == 0) + pg_fatal("%s: could not find START TIMELINE", filename); + if ((found & 4) != 0 && (found & 8) == 0) + pg_fatal("%s: INCREMENTAL FROM LSN requires INCREMENTAL FROM TLI", filename); + if ((found & 8) != 0 && (found & 4) == 0) + pg_fatal("%s: INCREMENTAL FROM TLI requires INCREMENTAL FROM LSN", filename); +} + +/* + * Write a backup label file to the output directory. + * + * This will be identical to the provided backup_label file, except that the + * INCREMENTAL FROM LSN and INCREMENTAL FROM TLI lines will be omitted. + * + * The new file will be checksummed using the specified algorithm. If + * mwriter != NULL, it will be added to the manifest. + */ +void +write_backup_label(char *output_directory, StringInfo buf, + pg_checksum_type checksum_type, manifest_writer *mwriter) +{ + char output_filename[MAXPGPATH]; + int output_fd; + pg_checksum_context checksum_ctx; + uint8 checksum_payload[PG_CHECKSUM_MAX_LENGTH]; + int checksum_length; + + pg_checksum_init(&checksum_ctx, checksum_type); + + snprintf(output_filename, MAXPGPATH, "%s/backup_label", output_directory); + + if ((output_fd = open(output_filename, + O_WRONLY | O_CREAT | O_EXCL | PG_BINARY, + pg_file_create_mode)) < 0) + pg_fatal("could not open file \"%s\": %m", output_filename); + + while (buf->cursor < buf->len) + { + char *s = &buf->data[buf->cursor]; + int eo = get_eol_offset(buf); + char *e = &buf->data[eo]; + + if (!line_starts_with(s, e, "INCREMENTAL FROM LSN: ", NULL) && + !line_starts_with(s, e, "INCREMENTAL FROM TLI: ", NULL)) + { + ssize_t wb; + + wb = write(output_fd, s, e - s); + if (wb != e - s) + { + if (wb < 0) + pg_fatal("could not write file \"%s\": %m", output_filename); + else + pg_fatal("could not write file \"%s\": wrote only %d of %d bytes", + output_filename, (int) wb, (int) (e - s)); + } + if (pg_checksum_update(&checksum_ctx, (uint8 *) s, e - s) < 0) + pg_fatal("could not update checksum of file \"%s\"", + output_filename); + } + + buf->cursor = eo; + } + + if (close(output_fd) != 0) + pg_fatal("could not close \"%s\": %m", output_filename); + + checksum_length = pg_checksum_final(&checksum_ctx, checksum_payload); + + if (mwriter != NULL) + { + struct stat sb; + + /* + * We could track the length ourselves, but must stat() to get the + * mtime. + */ + if (stat(output_filename, &sb) < 0) + pg_fatal("could not stat file \"%s\": %m", output_filename); + add_file_to_manifest(mwriter, "backup_label", sb.st_size, + sb.st_mtime, checksum_type, + checksum_length, checksum_payload); + } +} + +/* + * Return the offset at which the next line in the buffer starts, or there + * is none, the offset at which the buffer ends. + * + * The search begins at buf->cursor. + */ +static int +get_eol_offset(StringInfo buf) +{ + int eo = buf->cursor; + + while (eo < buf->len) + { + if (buf->data[eo] == '\n') + return eo + 1; + ++eo; + } + + return eo; +} + +/* + * Test whether the line that runs from s to e (inclusive of *s, but not + * inclusive of *e) starts with the match string provided, and return true + * or false according to whether or not this is the case. + * + * If the function returns true and if *sout != NULL, stores a pointer to the + * byte following the match into *sout. + */ +static bool +line_starts_with(char *s, char *e, char *match, char **sout) +{ + while (s < e && *match != '\0' && *s == *match) + ++s, ++match; + + if (*match == '\0' && sout != NULL) + *sout = s; + + return (*match == '\0'); +} + +/* + * Parse an LSN starting at s and not stopping at or before e. The return value + * is true on success and otherwise false. On success, stores the result into + * *lsn and sets *c to the first character that is not part of the LSN. + */ +static bool +parse_lsn(char *s, char *e, XLogRecPtr *lsn, char **c) +{ + char save = *e; + int nchars; + bool success; + unsigned hi; + unsigned lo; + + *e = '\0'; + success = (sscanf(s, "%X/%X%n", &hi, &lo, &nchars) == 2); + *e = save; + + if (success) + { + *lsn = ((XLogRecPtr) hi) << 32 | (XLogRecPtr) lo; + *c = s + nchars; + } + + return success; +} + +/* + * Parse a TLI starting at s and stopping at or before e. The return value is + * true on success and otherwise false. On success, stores the result into + * *tli. If the first character that is not part of the TLI is anything other + * than a newline, that is deemed a failure. + */ +static bool +parse_tli(char *s, char *e, TimeLineID *tli) +{ + char save = *e; + int nchars; + bool success; + + *e = '\0'; + success = (sscanf(s, "%u%n", tli, &nchars) == 1); + *e = save; + + if (success && s[nchars] != '\n') + success = false; + + return success; +} diff --git a/src/bin/pg_combinebackup/backup_label.h b/src/bin/pg_combinebackup/backup_label.h new file mode 100644 index 0000000000..08d6ed67a9 --- /dev/null +++ b/src/bin/pg_combinebackup/backup_label.h @@ -0,0 +1,29 @@ +/*------------------------------------------------------------------------- + * + * Read and manipulate backup label files + * + * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group + * Portions Copyright (c) 1994, Regents of the University of California + * + * src/bin/pg_combinebackup/backup_label.h + * + *------------------------------------------------------------------------- + */ +#ifndef BACKUP_LABEL_H +#define BACKUP_LABEL_H + +#include "common/checksum_helper.h" +#include "lib/stringinfo.h" + +struct manifest_writer; + +extern void parse_backup_label(char *filename, StringInfo buf, + TimeLineID *start_tli, + XLogRecPtr *start_lsn, + TimeLineID *previous_tli, + XLogRecPtr *previous_lsn); +extern void write_backup_label(char *output_directory, StringInfo buf, + pg_checksum_type checksum_type, + struct manifest_writer *mwriter); + +#endif /* BACKUP_LABEL_H */ diff --git a/src/bin/pg_combinebackup/copy_file.c b/src/bin/pg_combinebackup/copy_file.c new file mode 100644 index 0000000000..8ba6cc09e4 --- /dev/null +++ b/src/bin/pg_combinebackup/copy_file.c @@ -0,0 +1,169 @@ +/* + * Copy entire files. + * + * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group + * Portions Copyright (c) 1994, Regents of the University of California + * + * src/bin/pg_combinebackup/copy_file.h + * + *------------------------------------------------------------------------- + */ +#include "postgres_fe.h" + +#ifdef HAVE_COPYFILE_H +#include <copyfile.h> +#endif +#include <fcntl.h> +#include <sys/stat.h> +#include <unistd.h> + +#include "common/file_perm.h" +#include "common/logging.h" +#include "copy_file.h" + +static void copy_file_blocks(const char *src, const char *dst, + pg_checksum_context *checksum_ctx); + +#ifdef WIN32 +static void copy_file_copyfile(const char *src, const char *dst); +#endif + +/* + * Copy a regular file, optionally computing a checksum, and emitting + * appropriate debug messages. But if we're in dry-run mode, then just emit + * the messages and don't copy anything. + */ +void +copy_file(const char *src, const char *dst, + pg_checksum_context *checksum_ctx, bool dry_run) +{ + /* + * In dry-run mode, we don't actually copy anything, nor do we read any + * data from the source file, but we do verify that we can open it. + */ + if (dry_run) + { + int fd; + + if ((fd = open(src, O_RDONLY | PG_BINARY)) < 0) + pg_fatal("could not open \"%s\": %m", src); + if (close(fd) < 0) + pg_fatal("could not close \"%s\": %m", src); + } + + /* + * If we don't need to compute a checksum, then we can use any special + * operating system primitives that we know about to copy the file; this + * may be quicker than a naive block copy. + */ + if (checksum_ctx->type != CHECKSUM_TYPE_NONE) + { + char *strategy_name = NULL; + void (*strategy_implementation) (const char *, const char *) = NULL; + +#ifdef WIN32 + strategy_name = "CopyFile"; + strategy_implementation = copy_file_copyfile; +#endif + + if (strategy_name != NULL) + { + if (dry_run) + pg_log_debug("would copy \"%s\" to \"%s\" using strategy %s", + src, dst, strategy_name); + else + { + pg_log_debug("copying \"%s\" to \"%s\" using strategy %s", + src, dst, strategy_name); + (*strategy_implementation) (src, dst); + } + return; + } + } + + /* + * Fall back to the simple approach of reading and writing all the blocks, + * feeding them into the checksum context as we go. + */ + if (dry_run) + { + if (checksum_ctx->type == CHECKSUM_TYPE_NONE) + pg_log_debug("would copy \"%s\" to \"%s\"", + src, dst); + else + pg_log_debug("would copy \"%s\" to \"%s\" and checksum with %s", + src, dst, pg_checksum_type_name(checksum_ctx->type)); + } + else + { + if (checksum_ctx->type == CHECKSUM_TYPE_NONE) + pg_log_debug("copying \"%s\" to \"%s\"", + src, dst); + else + pg_log_debug("copying \"%s\" to \"%s\" and checksumming with %s", + src, dst, pg_checksum_type_name(checksum_ctx->type)); + copy_file_blocks(src, dst, checksum_ctx); + } +} + +/* + * Copy a file block by block, and optionally compute a checksum as we go. + */ +static void +copy_file_blocks(const char *src, const char *dst, + pg_checksum_context *checksum_ctx) +{ + int src_fd; + int dest_fd; + uint8 *buffer; + const int buffer_size = 50 * BLCKSZ; + ssize_t rb; + unsigned offset = 0; + + if ((src_fd = open(src, O_RDONLY | PG_BINARY, 0)) < 0) + pg_fatal("could not open file \"%s\": %m", src); + + if ((dest_fd = open(dst, O_WRONLY | O_CREAT | O_EXCL | PG_BINARY, + pg_file_create_mode)) < 0) + pg_fatal("could not open file \"%s\": %m", dst); + + buffer = pg_malloc(buffer_size); + + while ((rb = read(src_fd, buffer, buffer_size)) > 0) + { + ssize_t wb; + + if ((wb = write(dest_fd, buffer, rb)) != rb) + { + if (wb < 0) + pg_fatal("could not write file \"%s\": %m", dst); + else + pg_fatal("could not write file \"%s\": wrote only %d of %d bytes at offset %u", + dst, (int) wb, (int) rb, offset); + } + + if (pg_checksum_update(checksum_ctx, buffer, rb) < 0) + pg_fatal("could not update checksum of file \"%s\"", dst); + + offset += rb; + } + + if (rb < 0) + pg_fatal("could not read file \"%s\": %m", dst); + + pg_free(buffer); + close(src_fd); + close(dest_fd); +} + +#ifdef WIN32 +static void +copy_file_copyfile(const char *src, const char *dst) +{ + if (CopyFile(src, dst, true) == 0) + { + _dosmaperr(GetLastError()); + pg_fatal("could not copy \"%s\" to \"%s\": %m", src, dst); + } +} +#endif /* WIN32 */ diff --git a/src/bin/pg_combinebackup/copy_file.h b/src/bin/pg_combinebackup/copy_file.h new file mode 100644 index 0000000000..031030bacb --- /dev/null +++ b/src/bin/pg_combinebackup/copy_file.h @@ -0,0 +1,19 @@ +/* + * Copy entire files. + * + * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group + * Portions Copyright (c) 1994, Regents of the University of California + * + * src/bin/pg_combinebackup/copy_file.h + * + *------------------------------------------------------------------------- + */ +#ifndef COPY_FILE_H +#define COPY_FILE_H + +#include "common/checksum_helper.h" + +extern void copy_file(const char *src, const char *dst, + pg_checksum_context *checksum_ctx, bool dry_run); + +#endif /* COPY_FILE_H */ diff --git a/src/bin/pg_combinebackup/load_manifest.c b/src/bin/pg_combinebackup/load_manifest.c new file mode 100644 index 0000000000..d0b8de7912 --- /dev/null +++ b/src/bin/pg_combinebackup/load_manifest.c @@ -0,0 +1,245 @@ +/*------------------------------------------------------------------------- + * + * Load data from a backup manifest into memory. + * + * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group + * Portions Copyright (c) 1994, Regents of the University of California + * + * src/bin/pg_combinebackup/load_manifest.c + * + *------------------------------------------------------------------------- + */ + +#include "postgres_fe.h" + +#include <sys/stat.h> +#include <unistd.h> + +#include "common/hashfn.h" +#include "common/logging.h" +#include "common/parse_manifest.h" +#include "load_manifest.h" + +/* + * For efficiency, we'd like our hash table containing information about the + * manifest to start out with approximately the correct number of entries. + * There's no way to know the exact number of entries without reading the whole + * file, but we can get an estimate by dividing the file size by the estimated + * number of bytes per line. + * + * This could be off by about a factor of two in either direction, because the + * checksum algorithm has a big impact on the line lengths; e.g. a SHA512 + * checksum is 128 hex bytes, whereas a CRC-32C value is only 8, and there + * might be no checksum at all. + */ +#define ESTIMATED_BYTES_PER_MANIFEST_LINE 100 + +/* + * Define a hash table which we can use to store information about the files + * mentioned in the backup manifest. + */ +static uint32 hash_string_pointer(char *s); +#define SH_PREFIX manifest_files +#define SH_ELEMENT_TYPE manifest_file +#define SH_KEY_TYPE char * +#define SH_KEY pathname +#define SH_HASH_KEY(tb, key) hash_string_pointer(key) +#define SH_EQUAL(tb, a, b) (strcmp(a, b) == 0) +#define SH_SCOPE extern +#define SH_RAW_ALLOCATOR pg_malloc0 +#define SH_DEFINE +#include "lib/simplehash.h" + +static void record_manifest_details_for_file(JsonManifestParseContext *context, + char *pathname, size_t size, + pg_checksum_type checksum_type, + int checksum_length, + uint8 *checksum_payload); +static void record_manifest_details_for_wal_range(JsonManifestParseContext *context, + TimeLineID tli, + XLogRecPtr start_lsn, + XLogRecPtr end_lsn); +static void report_manifest_error(JsonManifestParseContext *context, + const char *fmt,...) + pg_attribute_printf(2, 3) pg_attribute_noreturn(); + +/* + * Load backup_manifest files from an array of backups and produces an array + * of manifest_data objects. + * + * NB: Since load_backup_manifest() can return NULL, the resulting array could + * contain NULL entries. + */ +manifest_data ** +load_backup_manifests(int n_backups, char **backup_directories) +{ + manifest_data **result; + int i; + + result = pg_malloc(sizeof(manifest_data *) * n_backups); + for (i = 0; i < n_backups; ++i) + result[i] = load_backup_manifest(backup_directories[i]); + + return result; +} + +/* + * Parse the backup_manifest file in the named backup directory. Construct a + * hash table with information about all the files it mentions, and a linked + * list of all the WAL ranges it mentions. + * + * If the backup_manifest file simply doesn't exist, logs a warning and returns + * NULL. Any other error, or any error parsing the contents of the file, is + * fatal. + */ +manifest_data * +load_backup_manifest(char *backup_directory) +{ + char pathname[MAXPGPATH]; + int fd; + struct stat statbuf; + off_t estimate; + uint32 initial_size; + manifest_files_hash *ht; + char *buffer; + int rc; + JsonManifestParseContext context; + manifest_data *result; + + /* Open the manifest file. */ + snprintf(pathname, MAXPGPATH, "%s/backup_manifest", backup_directory); + if ((fd = open(pathname, O_RDONLY | PG_BINARY, 0)) < 0) + { + if (errno == EEXIST) + { + pg_log_warning("\"%s\" does not exist", pathname); + return NULL; + } + pg_fatal("could not open file \"%s\": %m", pathname); + } + + /* Figure out how big the manifest is. */ + if (fstat(fd, &statbuf) != 0) + pg_fatal("could not stat file \"%s\": %m", pathname); + + /* Guess how large to make the hash table based on the manifest size. */ + estimate = statbuf.st_size / ESTIMATED_BYTES_PER_MANIFEST_LINE; + initial_size = Min(PG_UINT32_MAX, Max(estimate, 256)); + + /* Create the hash table. */ + ht = manifest_files_create(initial_size, NULL); + + /* + * Slurp in the whole file. + * + * This is not ideal, but there's currently no way to get pg_parse_json() + * to perform incremental parsing. + */ + buffer = pg_malloc(statbuf.st_size); + rc = read(fd, buffer, statbuf.st_size); + if (rc != statbuf.st_size) + { + if (rc < 0) + pg_fatal("could not read file \"%s\": %m", pathname); + else + pg_fatal("could not read file \"%s\": read %d of %lld", + pathname, rc, (long long int) statbuf.st_size); + } + + /* Close the manifest file. */ + close(fd); + + /* Parse the manifest. */ + result = pg_malloc0(sizeof(manifest_data)); + result->files = ht; + context.private_data = result; + context.perfile_cb = record_manifest_details_for_file; + context.perwalrange_cb = record_manifest_details_for_wal_range; + context.error_cb = report_manifest_error; + json_parse_manifest(&context, buffer, statbuf.st_size); + + /* All done. */ + pfree(buffer); + return result; +} + +/* + * Report an error while parsing the manifest. + * + * We consider all such errors to be fatal errors. The manifest parser + * expects this function not to return. + */ +static void +report_manifest_error(JsonManifestParseContext *context, const char *fmt,...) +{ + va_list ap; + + va_start(ap, fmt); + pg_log_generic_v(PG_LOG_ERROR, PG_LOG_PRIMARY, gettext(fmt), ap); + va_end(ap); + + exit(1); +} + +/* + * Record details extracted from the backup manifest for one file. + */ +static void +record_manifest_details_for_file(JsonManifestParseContext *context, + char *pathname, size_t size, + pg_checksum_type checksum_type, + int checksum_length, uint8 *checksum_payload) +{ + manifest_data *manifest = context->private_data; + manifest_file *m; + bool found; + + /* Make a new entry in the hash table for this file. */ + m = manifest_files_insert(manifest->files, pathname, &found); + if (found) + pg_fatal("duplicate path name in backup manifest: \"%s\"", pathname); + + /* Initialize the entry. */ + m->size = size; + m->checksum_type = checksum_type; + m->checksum_length = checksum_length; + m->checksum_payload = checksum_payload; +} + +/* + * Record details extracted from the backup manifest for one WAL range. + */ +static void +record_manifest_details_for_wal_range(JsonManifestParseContext *context, + TimeLineID tli, + XLogRecPtr start_lsn, XLogRecPtr end_lsn) +{ + manifest_data *manifest = context->private_data; + manifest_wal_range *range; + + /* Allocate and initialize a struct describing this WAL range. */ + range = palloc(sizeof(manifest_wal_range)); + range->tli = tli; + range->start_lsn = start_lsn; + range->end_lsn = end_lsn; + range->prev = manifest->last_wal_range; + range->next = NULL; + + /* Add it to the end of the list. */ + if (manifest->first_wal_range == NULL) + manifest->first_wal_range = range; + else + manifest->last_wal_range->next = range; + manifest->last_wal_range = range; +} + +/* + * Helper function for manifest_files hash table. + */ +static uint32 +hash_string_pointer(char *s) +{ + unsigned char *ss = (unsigned char *) s; + + return hash_bytes(ss, strlen(s)); +} diff --git a/src/bin/pg_combinebackup/load_manifest.h b/src/bin/pg_combinebackup/load_manifest.h new file mode 100644 index 0000000000..2bfeeff156 --- /dev/null +++ b/src/bin/pg_combinebackup/load_manifest.h @@ -0,0 +1,67 @@ +/*------------------------------------------------------------------------- + * + * Load data from a backup manifest into memory. + * + * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group + * Portions Copyright (c) 1994, Regents of the University of California + * + * src/bin/pg_combinebackup/load_manifest.h + * + *------------------------------------------------------------------------- + */ +#ifndef LOAD_MANIFEST_H +#define LOAD_MANIFEST_H + +#include "access/xlogdefs.h" +#include "common/checksum_helper.h" + +/* + * Each file described by the manifest file is parsed to produce an object + * like this. + */ +typedef struct manifest_file +{ + uint32 status; /* hash status */ + char *pathname; + size_t size; + pg_checksum_type checksum_type; + int checksum_length; + uint8 *checksum_payload; +} manifest_file; + +#define SH_PREFIX manifest_files +#define SH_ELEMENT_TYPE manifest_file +#define SH_KEY_TYPE char * +#define SH_SCOPE extern +#define SH_RAW_ALLOCATOR pg_malloc0 +#define SH_DECLARE +#include "lib/simplehash.h" + +/* + * Each WAL range described by the manifest file is parsed to produce an + * object like this. + */ +typedef struct manifest_wal_range +{ + TimeLineID tli; + XLogRecPtr start_lsn; + XLogRecPtr end_lsn; + struct manifest_wal_range *next; + struct manifest_wal_range *prev; +} manifest_wal_range; + +/* + * All the data parsed from a backup_manifest file. + */ +typedef struct manifest_data +{ + manifest_files_hash *files; + manifest_wal_range *first_wal_range; + manifest_wal_range *last_wal_range; +} manifest_data; + +extern manifest_data *load_backup_manifest(char *backup_directory); +extern manifest_data **load_backup_manifests(int n_backups, + char **backup_directories); + +#endif /* LOAD_MANIFEST_H */ diff --git a/src/bin/pg_combinebackup/meson.build b/src/bin/pg_combinebackup/meson.build new file mode 100644 index 0000000000..bea0db405e --- /dev/null +++ b/src/bin/pg_combinebackup/meson.build @@ -0,0 +1,29 @@ +# Copyright (c) 2022-2023, PostgreSQL Global Development Group + +pg_combinebackup_sources = files( + 'pg_combinebackup.c', + 'backup_label.c', + 'copy_file.c', + 'load_manifest.c', + 'reconstruct.c', + 'write_manifest.c', +) + +if host_system == 'windows' + pg_combinebackup_sources += rc_bin_gen.process(win32ver_rc, extra_args: [ + '--NAME', 'pg_combinebackup', + '--FILEDESC', 'pg_combinebackup - combine incremental backups',]) +endif + +pg_combinebackup = executable('pg_combinebackup', + pg_combinebackup_sources, + dependencies: [frontend_code], + kwargs: default_bin_args, +) +bin_targets += pg_combinebackup + +tests += { + 'name': 'pg_combinebackup', + 'sd': meson.current_source_dir(), + 'bd': meson.current_build_dir() +} diff --git a/src/bin/pg_combinebackup/pg_combinebackup.c b/src/bin/pg_combinebackup/pg_combinebackup.c new file mode 100644 index 0000000000..6c7fd3290e --- /dev/null +++ b/src/bin/pg_combinebackup/pg_combinebackup.c @@ -0,0 +1,1268 @@ +/*------------------------------------------------------------------------- + * + * pg_combinebackup.c + * Combine incremental backups with prior backups. + * + * Copyright (c) 2017-2023, PostgreSQL Global Development Group + * + * IDENTIFICATION + * src/bin/pg_combinebackup/pg_combinebackup.c + * + *------------------------------------------------------------------------- + */ +#include "postgres_fe.h" + +#include <dirent.h> +#include <fcntl.h> +#include <limits.h> + +#include "access/xlogdefs.h" +#include "backup_label.h" +#include "common/blkreftable.h" +#include "common/checksum_helper.h" +#include "common/controldata_utils.h" +#include "common/file_perm.h" +#include "common/file_utils.h" +#include "common/logging.h" +#include "copy_file.h" +#include "fe_utils/option_utils.h" +#include "lib/stringinfo.h" +#include "load_manifest.h" +#include "getopt_long.h" +#include "reconstruct.h" +#include "write_manifest.h" + +/* Incremental file naming convention. */ +#define INCREMENTAL_PREFIX "INCREMENTAL." +#define INCREMENTAL_PREFIX_LENGTH 12 + +/* + * Tracking for directories that need to be removed, or have their contents + * removed, if the operation fails. + */ +typedef struct cb_cleanup_dir +{ + char *target_path; + bool rmtopdir; + struct cb_cleanup_dir *next; +} cb_cleanup_dir; + +/* + * Stores a tablespace mapping provided using -T, --tablespace-mapping. + */ +typedef struct cb_tablespace_mapping +{ + char old_dir[MAXPGPATH]; + char new_dir[MAXPGPATH]; + struct cb_tablespace_mapping *next; +} cb_tablespace_mapping; + +/* + * Stores data parsed from all command-line options. + */ +typedef struct cb_options +{ + bool debug; + char *output; + bool dry_run; + bool no_sync; + bool progress; + cb_tablespace_mapping *tsmappings; + pg_checksum_type manifest_checksums; + bool no_manifest; +} cb_options; + +/* + * Data about a tablespace. + * + * Every normal tablespace needs a tablespace mapping, but in-place tablespaces + * don't, so the list of tablespaces can contain more entries than the list of + * tablespace mappings. + */ +typedef struct cb_tablespace +{ + Oid oid; + bool in_place; + char old_dir[MAXPGPATH]; + char new_dir[MAXPGPATH]; + struct cb_tablespace *next; +} cb_tablespace; + +/* Directories to be removed if we exit uncleanly. */ +cb_cleanup_dir *cleanup_dir_list = NULL; + +static void add_tablespace_mapping(cb_options *opt, char *arg); +static StringInfo check_backup_label_files(int n_backups, char **backup_dirs); +static void check_control_files(int n_backups, char **backup_dirs); +static void check_input_dir_permissions(char *dir); +static void cleanup_directories_atexit(void); +static void create_output_directory(char *dirname, cb_options *opt); +static void help(const char *progname); +static bool parse_oid(char *s, Oid *result); +static void process_directory_recursively(Oid tsoid, + char *input_directory, + char *output_directory, + char *relative_path, + int n_prior_backups, + char **prior_backup_dirs, + manifest_data **manifests, + manifest_writer *mwriter, + cb_options *opt); +static int read_pg_version_file(char *directory); +static void remember_to_cleanup_directory(char *target_path, bool rmtopdir); +static void reset_directory_cleanup_list(void); +static cb_tablespace *scan_for_existing_tablespaces(char *pathname, + cb_options *opt); +static void slurp_file(int fd, char *filename, StringInfo buf, int maxlen); + +/* + * Main program. + */ +int +main(int argc, char *argv[]) +{ + static struct option long_options[] = { + {"debug", no_argument, NULL, 'd'}, + {"output", required_argument, NULL, 'o'}, + {"dry-run", no_argument, NULL, 'n'}, + {"no-sync", no_argument, NULL, 'N'}, + {"progress", no_argument, NULL, 'P'}, + {"tablespace-mapping", no_argument, NULL, 'T'}, + {"manifest-checksums", required_argument, NULL, 1}, + {"no-manifest", no_argument, NULL, 2}, + {NULL, 0, NULL, 0} + }; + + const char *progname; + char *last_input_dir; + int optindex; + int c; + int n_backups; + int n_prior_backups; + int version; + char **prior_backup_dirs; + cb_options opt; + cb_tablespace *tablespaces; + cb_tablespace *ts; + StringInfo last_backup_label; + manifest_data **manifests; + manifest_writer *mwriter; + + pg_logging_init(argv[0]); + progname = get_progname(argv[0]); + handle_help_version_opts(argc, argv, progname, help); + + memset(&opt, 0, sizeof(opt)); + opt.manifest_checksums = CHECKSUM_TYPE_CRC32C; + + /* process command-line options */ + while ((c = getopt_long(argc, argv, "do:nNPT:", + long_options, &optindex)) != -1) + { + switch (c) + { + case 'd': + opt.debug = true; + pg_logging_increase_verbosity(); + break; + case 'o': + opt.output = optarg; + break; + case 'n': + opt.dry_run = true; + break; + case 'N': + opt.no_sync = true; + break; + case 'P': + opt.progress = true; + break; + case 'T': + add_tablespace_mapping(&opt, optarg); + break; + case 1: + if (!pg_checksum_parse_type(optarg, + &opt.manifest_checksums)) + pg_fatal("unrecognized checksum algorithm: \"%s\"", + optarg); + break; + case 2: + opt.no_manifest = true; + break; + default: + /* getopt_long already emitted a complaint */ + pg_log_error_hint("Try \"%s --help\" for more information.", progname); + exit(1); + } + } + + if (optind >= argc) + { + pg_log_error("%s: no input directories specified", progname); + pg_log_error_hint("Try \"%s --help\" for more information.", progname); + exit(1); + } + + if (opt.output == NULL) + pg_fatal("no output directory specified"); + + /* If no manifest is needed, no checksums are needed, either. */ + if (opt.no_manifest) + opt.manifest_checksums = CHECKSUM_TYPE_NONE; + + /* Read the server version from the final backup. */ + version = read_pg_version_file(argv[argc - 1]); + + /* Sanity-check control files. */ + n_backups = argc - optind; + check_control_files(n_backups, argv + optind); + + /* Sanity-check backup_label files, and get the contents of the last one. */ + last_backup_label = check_backup_label_files(n_backups, argv + optind); + + /* Load backup manifests. */ + manifests = load_backup_manifests(n_backups, argv + optind); + + /* Figure out which tablespaces are going to be included in the output. */ + last_input_dir = argv[argc - 1]; + check_input_dir_permissions(last_input_dir); + tablespaces = scan_for_existing_tablespaces(last_input_dir, &opt); + + /* + * Create output directories. + * + * We create one output directory for the main data directory plus one for + * each non-in-place tablespace. create_output_directory() will arrange + * for those directories to be cleaned up on failure. In-place tablespaces + * aren't handled at this stage because they're located beneath the main + * output directory, and thus the cleanup of that directory will get rid + * of them. Plus, the pg_tblspc directory that needs to contain them + * doesn't exist yet. + */ + atexit(cleanup_directories_atexit); + create_output_directory(opt.output, &opt); + for (ts = tablespaces; ts != NULL; ts = ts->next) + if (!ts->in_place) + create_output_directory(ts->new_dir, &opt); + + /* If we need to write a backup_manifest, prepare to do so. */ + if (!opt.dry_run && !opt.no_manifest) + mwriter = create_manifest_writer(opt.output); + else + mwriter = NULL; + + /* Write backup label into output directory. */ + if (opt.dry_run) + pg_log_debug("would generate \"%s/backup_label\"", opt.output); + else + { + pg_log_debug("generating \"%s/backup_label\"", opt.output); + last_backup_label->cursor = 0; + write_backup_label(opt.output, last_backup_label, + opt.manifest_checksums, mwriter); + } + + /* + * We'll need the pathnames to the prior backups. By "prior" we mean all + * but the last one listed on the command line. + */ + n_prior_backups = argc - optind - 1; + prior_backup_dirs = argv + optind; + + /* Process everything that's not part of a user-defined tablespace. */ + pg_log_debug("processing backup directory \"%s\"", last_input_dir); + process_directory_recursively(InvalidOid, last_input_dir, opt.output, + NULL, n_prior_backups, prior_backup_dirs, + manifests, mwriter, &opt); + + /* Process user-defined tablespaces. */ + for (ts = tablespaces; ts != NULL; ts = ts->next) + { + pg_log_debug("processing tablespace directory \"%s\"", ts->old_dir); + + /* + * If it's a normal tablespace, we need to set up a symbolic link from + * pg_tblspc/${OID} to the target directory; if it's an in-place + * tablespace, we need to create a directory at pg_tblspc/${OID}. + */ + if (!ts->in_place) + { + char linkpath[MAXPGPATH]; + + snprintf(linkpath, MAXPGPATH, "%s/pg_tblspc/%u", opt.output, + ts->oid); + + if (opt.dry_run) + pg_log_debug("would create symbolic link from \"%s\" to \"%s\"", + linkpath, ts->new_dir); + else + { + pg_log_debug("creating symbolic link from \"%s\" to \"%s\"", + linkpath, ts->new_dir); + if (symlink(ts->new_dir, linkpath) != 0) + pg_fatal("could not create symbolic link from \"%s\" to \"%s\": %m", + linkpath, ts->new_dir); + } + } + else + { + if (opt.dry_run) + pg_log_debug("would create directory \"%s\"", ts->new_dir); + else + { + pg_log_debug("creating directory \"%s\"", ts->new_dir); + if (pg_mkdir_p(ts->new_dir, pg_dir_create_mode) == -1) + pg_fatal("could not create directory \"%s\": %m", + ts->new_dir); + } + } + + /* OK, now handle the directory contents. */ + process_directory_recursively(ts->oid, ts->old_dir, ts->new_dir, + NULL, n_prior_backups, prior_backup_dirs, + manifests, mwriter, &opt); + } + + /* Finalize the backup_manifest, if we're generating one. */ + if (mwriter != NULL) + finalize_manifest(mwriter, + manifests[n_prior_backups]->first_wal_range); + + /* fsync that output directory unless we've been told not to do so */ + if (!opt.no_sync) + { + if (opt.dry_run) + pg_log_debug("would recursively fsync \"%s\"", opt.output); + else + { + pg_log_debug("recursively fsyncing \"%s\"", opt.output); + fsync_pgdata(opt.output, version * 10000); + } + } + + /* It's a success, so don't remove the output directories. */ + reset_directory_cleanup_list(); + exit(0); +} + +/* + * Process the option argument for the -T, --tablespace-mapping switch. + */ +static void +add_tablespace_mapping(cb_options *opt, char *arg) +{ + cb_tablespace_mapping *tsmap = pg_malloc0(sizeof(cb_tablespace_mapping)); + char *dst; + char *dst_ptr; + char *arg_ptr; + + /* + * Basically, we just want to copy everything before the equals sign to + * tsmap->old_dir and everything afterwards to tsmap->new_dir, but if + * there's more or less than one equals sign, that's an error, and if + * there's an equals sign preceded by a backslash, don't treat it as a + * field separator but instead copy a literal equals sign. + */ + dst_ptr = dst = tsmap->old_dir; + for (arg_ptr = arg; *arg_ptr != '\0'; arg_ptr++) + { + if (dst_ptr - dst >= MAXPGPATH) + pg_fatal("directory name too long"); + + if (*arg_ptr == '\\' && *(arg_ptr + 1) == '=') + ; /* skip backslash escaping = */ + else if (*arg_ptr == '=' && (arg_ptr == arg || *(arg_ptr - 1) != '\\')) + { + if (tsmap->new_dir[0] != '\0') + pg_fatal("multiple \"=\" signs in tablespace mapping"); + else + dst = dst_ptr = tsmap->new_dir; + } + else + *dst_ptr++ = *arg_ptr; + } + if (!tsmap->old_dir[0] || !tsmap->new_dir[0]) + pg_fatal("invalid tablespace mapping format \"%s\", must be \"OLDDIR=NEWDIR\"", arg); + + /* + * All tablespaces are created with absolute directories, so specifying a + * non-absolute path here would never match, possibly confusing users. + * + * In contrast to pg_basebackup, both the old and new directories are on + * the local machine, so the local machine's definition of an absolute + * path is the only relevant one. + */ + if (!is_absolute_path(tsmap->old_dir)) + pg_fatal("old directory is not an absolute path in tablespace mapping: %s", + tsmap->old_dir); + + if (!is_absolute_path(tsmap->new_dir)) + pg_fatal("old directory is not an absolute path in tablespace mapping: %s", + tsmap->new_dir); + + /* Canonicalize paths to avoid spurious failures when comparing. */ + canonicalize_path(tsmap->old_dir); + canonicalize_path(tsmap->new_dir); + + /* Add it to the list. */ + tsmap->next = opt->tsmappings; + opt->tsmappings = tsmap; +} + +/* + * Check that the backup_label files form a coherent backup chain, and return + * the contents of the backup_label file from the latest backup. + */ +static StringInfo +check_backup_label_files(int n_backups, char **backup_dirs) +{ + StringInfo buf = makeStringInfo(); + StringInfo lastbuf = buf; + int i; + TimeLineID check_tli = 0; + XLogRecPtr check_lsn = InvalidXLogRecPtr; + + /* Try to read each backup_label file in turn, last to first. */ + for (i = n_backups - 1; i >= 0; --i) + { + char pathbuf[MAXPGPATH]; + int fd; + TimeLineID start_tli; + TimeLineID previous_tli; + XLogRecPtr start_lsn; + XLogRecPtr previous_lsn; + + /* Open the backup_label file. */ + snprintf(pathbuf, MAXPGPATH, "%s/backup_label", backup_dirs[i]); + pg_log_debug("reading \"%s\"", pathbuf); + if ((fd = open(pathbuf, O_RDONLY, 0)) < 0) + pg_fatal("could not open file \"%s\": %m", pathbuf); + + /* + * Slurp the whole file into memory. + * + * The exact size limit that we impose here doesn't really matter -- + * most of what's supposed to be in the file is fixed size and quite + * short. However, the length of the backup_label is limited (at least + * by some parts of the code) to MAXGPATH, so include that value in + * the maximum length that we tolerate. + */ + slurp_file(fd, pathbuf, buf, 10000 + MAXPGPATH); + + /* Close the file. */ + if (close(fd) != 0) + pg_fatal("could not close \"%s\": %m", pathbuf); + + /* Parse the file contents. */ + parse_backup_label(pathbuf, buf, &start_tli, &start_lsn, + &previous_tli, &previous_lsn); + + /* + * Sanity checks. + * + * XXX. It's actually not required that start_lsn == check_lsn. It + * would be OK if start_lsn > check_lsn provided that start_lsn is + * less than or equal to the relevant switchpoint. But at the moment + * we don't have that information. + */ + if (i > 0 && previous_tli == 0) + pg_fatal("backup at \"%s\" is a full backup, but only the first backup should be a full backup", + backup_dirs[i]); + if (i == 0 && previous_tli != 0) + pg_fatal("backup at \"%s\" is an incremental backup, but the first backup should be a full backup", + backup_dirs[i]); + if (i < n_backups - 1 && start_tli != check_tli) + pg_fatal("backup at \"%s\" starts on timeline %u, but expected %u", + backup_dirs[i], start_tli, check_tli); + if (i < n_backups - 1 && start_lsn != check_lsn) + pg_fatal("backup at \"%s\" starts at LSN %X/%X, but expected %X/%X", + backup_dirs[i], + LSN_FORMAT_ARGS(start_lsn), + LSN_FORMAT_ARGS(check_lsn)); + check_tli = previous_tli; + check_lsn = previous_lsn; + + /* + * The last backup label in the chain needs to be saved for later use, + * while the others are only needed within this loop. + */ + if (lastbuf == buf) + buf = makeStringInfo(); + else + resetStringInfo(buf); + } + + /* Free memory that we don't need any more. */ + if (lastbuf != buf) + { + pfree(buf->data); + pfree(buf); + } + + /* + * Return the data from the first backup_info that we read (which is the + * backup_label from the last directory specified on the command line). + */ + return lastbuf; +} + +/* + * Sanity check control files. + */ +static void +check_control_files(int n_backups, char **backup_dirs) +{ + int i; + uint64 system_identifier; + + /* Try to read each control file in turn, last to first. */ + for (i = n_backups - 1; i >= 0; --i) + { + ControlFileData *control_file; + bool crc_ok; + + pg_log_debug("reading \"%s/global/pg_control\"", backup_dirs[i]); + control_file = get_controlfile(backup_dirs[i], &crc_ok); + + /* Control file contents not meaningful if CRC is bad. */ + if (!crc_ok) + pg_fatal("%s/global/pg_control: crc is incorrect", backup_dirs[i]); + + /* Can't interpret control file if not current version. */ + if (control_file->pg_control_version != PG_CONTROL_VERSION) + pg_fatal("%s/global/pg_control: unexpected control file version", + backup_dirs[i]); + + /* System identifiers should all match. */ + if (i == n_backups - 1) + system_identifier = control_file->system_identifier; + else if (system_identifier != control_file->system_identifier) + pg_fatal("%s/global/pg_control: expected system identifier %llu, but found %llu", + backup_dirs[i], (unsigned long long) system_identifier, + (unsigned long long) control_file->system_identifier); + + /* Release memory. */ + pfree(control_file); + } + + /* + * If debug output is enabled, make a note of the system identifier that + * we found in all of the relevant control files. + */ + pg_log_debug("system identifier is %llu", + (unsigned long long) system_identifier); +} + +/* + * Set default permissions for new files and directories based on the + * permissions of the given directory. The intent here is that the output + * directory should use the same permissions scheme as the final input + * directory. + */ +static void +check_input_dir_permissions(char *dir) +{ + struct stat st; + + if (stat(dir, &st) != 0) + pg_fatal("could not stat \"%s\": %m", dir); + + SetDataDirectoryCreatePerm(st.st_mode); +} + +/* + * Clean up output directories before exiting. + */ +static void +cleanup_directories_atexit(void) +{ + while (cleanup_dir_list != NULL) + { + cb_cleanup_dir *dir = cleanup_dir_list; + + if (dir->rmtopdir) + { + pg_log_info("removing output directory \"%s\"", dir->target_path); + if (!rmtree(dir->target_path, dir->rmtopdir)) + pg_log_error("failed to remove output directory"); + } + else + { + pg_log_info("removing contents of output directory \"%s\"", + dir->target_path); + if (!rmtree(dir->target_path, dir->rmtopdir)) + pg_log_error("failed to remove contents of output directory"); + } + + cleanup_dir_list = cleanup_dir_list->next; + pfree(dir); + } +} + +/* + * Create the named output directory, unless it already exists or we're in + * dry-run mode. If it already exists but is not empty, that's a fatal error. + * + * Adds the created directory to the list of directories to be cleaned up + * at process exit. + */ +static void +create_output_directory(char *dirname, cb_options *opt) +{ + switch (pg_check_dir(dirname)) + { + case 0: + if (opt->dry_run) + { + pg_log_debug("would create directory \"%s\"", dirname); + return; + } + pg_log_debug("creating directory \"%s\"", dirname); + if (pg_mkdir_p(dirname, pg_dir_create_mode) == -1) + pg_fatal("could not create directory \"%s\": %m", dirname); + remember_to_cleanup_directory(dirname, true); + break; + + case 1: + pg_log_debug("using existing directory \"%s\"", dirname); + remember_to_cleanup_directory(dirname, false); + break; + + case 2: + case 3: + case 4: + pg_fatal("directory \"%s\" exists but is not empty", dirname); + + case -1: + pg_fatal("could not access directory \"%s\": %m", dirname); + } +} + +/* + * help + * + * Prints help page for the program + * + * progname: the name of the executed program, such as "pg_combinebackup" + */ +static void +help(const char *progname) +{ + printf(_("%s combines incremental backups.\n\n"), progname); + printf(_("Usage:\n")); + printf(_(" %s [OPTION]... DIRECTORY...\n"), progname); + printf(_("\nOptions:\n")); + printf(_(" -d, --debug generate lots of debugging output\n")); + printf(_(" -o, --output output directory\n")); + printf(_(" -n, --dry-run don't actually do anything\n")); + printf(_(" -N, --no-sync do not wait for changes to be written safely to disk\n")); + printf(_(" -P, --progress show progress information\n")); + printf(_(" -T, --tablespace-mapping=OLDDIR=NEWDIR\n")); + printf(_(" relocate tablespace in OLDDIR to NEWDIR\n")); + printf(_(" --manifest-checksums=SHA{224,256,384,512}|CRC32C|NONE\n" + " use algorithm for manifest checksums\n")); + printf(_(" --no-manifest suppress generation of backup manifest\n")); + printf(_(" -?, --help show this help, then exit\n")); + + printf(_("\nReport bugs to <%s>.\n"), PACKAGE_BUGREPORT); + printf(_("%s home page: <%s>\n"), PACKAGE_NAME, PACKAGE_URL); +} + +/* + * Try to parse a string as a non-zero OID without leading zeroes. + * + * If it works, return true and set *result to the answer, else return false. + */ +static bool +parse_oid(char *s, Oid *result) +{ + Oid oid; + char *ep; + + errno = 0; + oid = strtoul(s, &ep, 10); + if (errno != 0 || *ep != '\0' || oid < 1 || oid > PG_UINT32_MAX) + return false; + + *result = oid; + return true; +} + +/* + * Copy files from the input directory to the output directory, reconstructing + * full files from incremental files as required. + * + * If processing is a user-defined tablespace, the tsoid should be the OID + * of that tablespace and input_directory and output_directory should be the + * toplevel input and output directories for that tablespace. Otherwise, + * tsoid should be InvalidOid and input_directory and output_directory should + * be the main input and output directories. + * + * relative_path is the path beneath the given input and output directories + * that we are currently processing. If NULL, it indicates that we're + * processing the input and output directories themselves. + * + * n_prior_backups is the number of prior backups that we have available. + * This doesn't count the very last backup, which is referenced by + * output_directory, just the older ones. prior_backup_dirs is an array of + * the locations of those previous backups. + */ +static void +process_directory_recursively(Oid tsoid, + char *input_directory, + char *output_directory, + char *relative_path, + int n_prior_backups, + char **prior_backup_dirs, + manifest_data **manifests, + manifest_writer *mwriter, + cb_options *opt) +{ + char ifulldir[MAXPGPATH]; + char ofulldir[MAXPGPATH]; + char manifest_prefix[MAXPGPATH]; + DIR *dir; + struct dirent *de; + bool is_pg_tblspc; + bool is_pg_wal; + manifest_data *latest_manifest = manifests[n_prior_backups]; + pg_checksum_type checksum_type; + + StaticAssertStmt(strlen(INCREMENTAL_PREFIX) == INCREMENTAL_PREFIX_LENGTH, + "INCREMENTAL_PREFIX_LENGTH is incorrect"); + + /* + * pg_tblspc and pg_wal are special cases, so detect those here. + * + * pg_tblspc is only special at the top level, but subdirectories of + * pg_wal are just as special as the top level directory. + * + * Since incremental backup does not exist in pre-v10 versions, we don't + * have to worry about the old pg_xlog naming. + */ + is_pg_tblspc = !OidIsValid(tsoid) && relative_path != NULL && + strcmp(relative_path, "pg_tblspc") == 0; + is_pg_wal = !OidIsValid(tsoid) && relative_path != NULL && + (strcmp(relative_path, "pg_wal") == 0 || + strncmp(relative_path, "pg_wal/", 7) == 0); + + /* + * If we're under pg_wal, then we don't need checksums, because these + * files aren't included in the backup manifest. Otherwise use whatever + * type of checksum is configured. + */ + if (!is_pg_wal) + checksum_type = opt->manifest_checksums; + else + checksum_type = CHECKSUM_TYPE_NONE; + + /* + * Append the relative path to the input and output directories, and + * figure out the appropriate prefix to add to files in this directory + * when looking them up in a backup manifest. + */ + if (relative_path == NULL) + { + strncpy(ifulldir, input_directory, MAXPGPATH); + strncpy(ofulldir, output_directory, MAXPGPATH); + if (OidIsValid(tsoid)) + snprintf(manifest_prefix, MAXPGPATH, "pg_tblspc/%u/", tsoid); + else + manifest_prefix[0] = '\0'; + } + else + { + snprintf(ifulldir, MAXPGPATH, "%s/%s", input_directory, + relative_path); + snprintf(ofulldir, MAXPGPATH, "%s/%s", output_directory, + relative_path); + if (OidIsValid(tsoid)) + snprintf(manifest_prefix, MAXPGPATH, "pg_tblspc/%u/%s/", + tsoid, relative_path); + else + snprintf(manifest_prefix, MAXPGPATH, "%s/", relative_path); + } + + /* + * Toplevel output directories have already been created by the time this + * function is called, but any subdirectories are our responsibility. + */ + if (relative_path != NULL) + { + if (opt->dry_run) + pg_log_debug("would create directory \"%s\"", ofulldir); + else + { + pg_log_debug("creating directory \"%s\"", ofulldir); + if (mkdir(ofulldir, pg_dir_create_mode) == -1) + pg_fatal("could not create directory \"%s\": %m", ofulldir); + } + } + + /* It's time to scan the directory. */ + if ((dir = opendir(ifulldir)) == NULL) + pg_fatal("could not open directory \"%s\": %m", ifulldir); + while (errno = 0, (de = readdir(dir)) != NULL) + { + PGFileType type; + char ifullpath[MAXPGPATH]; + char ofullpath[MAXPGPATH]; + char manifest_path[MAXPGPATH]; + Oid oid = InvalidOid; + int checksum_length = 0; + uint8 *checksum_payload = NULL; + pg_checksum_context checksum_ctx; + + /* Ignore "." and ".." entries. */ + if (strcmp(de->d_name, ".") == 0 || + strcmp(de->d_name, "..") == 0) + continue; + + /* Construct input path. */ + snprintf(ifullpath, MAXPGPATH, "%s/%s", ifulldir, de->d_name); + + /* Figure out what kind of directory entry this is. */ + type = get_dirent_type(ifullpath, de, false, PG_LOG_ERROR); + if (type == PGFILETYPE_ERROR) + exit(1); + + /* + * If we're processing pg_tblspc, then check whether the filename + * looks like it could be a tablespace OID. If so, and if the + * directory entry is a symbolic link or a directory, skip it. + * + * Our goal here is to ignore anything that would have been considered + * by scan_for_existing_tablespaces to be a tablespace. + */ + if (is_pg_tblspc && parse_oid(de->d_name, &oid) && + (type == PGFILETYPE_LNK || type == PGFILETYPE_DIR)) + continue; + + /* If it's a directory, recurse. */ + if (type == PGFILETYPE_DIR) + { + char new_relative_path[MAXPGPATH]; + + /* Append new pathname component to relative path. */ + if (relative_path == NULL) + strncpy(new_relative_path, de->d_name, MAXPGPATH); + else + snprintf(new_relative_path, MAXPGPATH, "%s/%s", relative_path, + de->d_name); + + /* And recurse. */ + process_directory_recursively(tsoid, + input_directory, output_directory, + new_relative_path, + n_prior_backups, prior_backup_dirs, + manifests, mwriter, opt); + continue; + } + + /* Skip anything that's not a regular file. */ + if (type != PGFILETYPE_REG) + { + if (type == PGFILETYPE_LNK) + pg_log_warning("skipping symbolic link \"%s\"", ifullpath); + else + pg_log_warning("skipping special file \"%s\"", ifullpath); + continue; + } + + /* + * Skip the backup_label and backup_manifest files; they require + * special handling and are handled elsewhere. + */ + if (relative_path == NULL && + (strcmp(de->d_name, "backup_label") == 0 || + strcmp(de->d_name, "backup_manifest") == 0)) + continue; + + /* + * If it's an incremental file, hand it off to the reconstruction + * code, which will figure out what to do. + */ + if (strncmp(de->d_name, INCREMENTAL_PREFIX, + INCREMENTAL_PREFIX_LENGTH) == 0) + { + /* Output path should not include "INCREMENTAL." prefix. */ + snprintf(ofullpath, MAXPGPATH, "%s/%s", ofulldir, + de->d_name + INCREMENTAL_PREFIX_LENGTH); + + + /* Manifest path likewise omits incremental prefix. */ + snprintf(manifest_path, MAXPGPATH, "%s%s", manifest_prefix, + de->d_name + INCREMENTAL_PREFIX_LENGTH); + + /* Reconstruction logic will do the rest. */ + reconstruct_from_incremental_file(ifullpath, ofullpath, + relative_path, + de->d_name + INCREMENTAL_PREFIX_LENGTH, + n_prior_backups, + prior_backup_dirs, + manifests, + manifest_path, + checksum_type, + &checksum_length, + &checksum_payload, + opt->dry_run); + } + else + { + /* Construct the path that the backup_manifest will use. */ + snprintf(manifest_path, MAXPGPATH, "%s%s", manifest_prefix, + de->d_name); + + /* + * It's not an incremental file, so we need to copy the entire + * file to the output directory. + * + * If a checksum of the required type already exists in the + * backup_manifest for the final input directory, we can save some + * work by reusing that checksum instead of computing a new one. + */ + if (checksum_type != CHECKSUM_TYPE_NONE && + latest_manifest != NULL) + { + manifest_file *mfile; + + mfile = manifest_files_lookup(latest_manifest->files, + manifest_path); + if (mfile == NULL) + { + /* + * The directory is out of sync with the backup_manifest, + * so emit a warning. + */ + pg_log_warning("\"%s/backup_manifest\" contains no entry for \"%s\"", + input_directory, manifest_path); + } + else if (mfile->checksum_type == checksum_type) + { + checksum_length = mfile->checksum_length; + checksum_payload = mfile->checksum_payload; + } + } + + /* + * If we're reusing a checksum, then we don't need copy_file() to + * compute one for us, but otherwise, it needs to compute whatever + * type of checksum we need. + */ + if (checksum_length != 0) + pg_checksum_init(&checksum_ctx, CHECKSUM_TYPE_NONE); + else + pg_checksum_init(&checksum_ctx, checksum_type); + + /* Actually copy the file. */ + snprintf(ofullpath, MAXPGPATH, "%s/%s", ofulldir, de->d_name); + copy_file(ifullpath, ofullpath, &checksum_ctx, opt->dry_run); + + /* + * If copy_file() performed a checksum calculation for us, then + * save the results (except in dry-run mode, when there's no + * point). + */ + if (checksum_ctx.type != CHECKSUM_TYPE_NONE && !opt->dry_run) + { + checksum_payload = pg_malloc(PG_CHECKSUM_MAX_LENGTH); + checksum_length = pg_checksum_final(&checksum_ctx, + checksum_payload); + } + } + + /* Generate manifest entry, if needed. */ + if (mwriter != NULL) + { + struct stat sb; + + /* + * In order to generate a manifest entry, we need the file size + * and mtime. We have no way to know the correct mtime except to + * stat() the file, so just do that and get the size as well. + * + * If we didn't need the mtime here, we could try to obtain the + * file size from the reconstruction or file copy process above, + * although that is actually not convenient in all cases. If we + * write the file ourselves then clearly we can keep a count of + * bytes, but if we use something like CopyFile() then it's + * trickier. Since we have to stat() anyway to get the mtime, + * there's no point in worrying about it. + */ + if (stat(ofullpath, &sb) < 0) + pg_fatal("could not stat file \"%s\": %m", ofullpath); + + /* OK, now do the work. */ + add_file_to_manifest(mwriter, manifest_path, + sb.st_size, sb.st_mtime, + checksum_type, checksum_length, + checksum_payload); + } + + /* Avoid leaking memory. */ + if (checksum_payload != NULL) + pfree(checksum_payload); + } + + closedir(dir); +} + +/* + * Read the version number from PG_VERSION and convert it to the usual server + * version number format. (e.g. If PG_VERSION contains "14\n" this function + * will return 140000) + */ +static int +read_pg_version_file(char *directory) +{ + char filename[MAXPGPATH]; + StringInfoData buf; + int fd; + int version; + char *ep; + + /* Construct pathname. */ + snprintf(filename, MAXPGPATH, "%s/PG_VERSION", directory); + + /* Open file. */ + if ((fd = open(filename, O_RDONLY, 0)) < 0) + pg_fatal("could not open file \"%s\": %m", filename); + + /* Read into memory. Length limit of 128 should be more than generous. */ + initStringInfo(&buf); + slurp_file(fd, filename, &buf, 128); + + /* Close the file. */ + if (close(fd) != 0) + pg_fatal("could not close \"%s\": %m", filename); + + /* Convert to integer. */ + errno = 0; + version = strtoul(buf.data, &ep, 10); + if (errno != 0 || *ep != '\n') + { + /* + * Incremental backup is not relevant to very old server versions that + * used multi-part version number (e.g. 9.6, or 8.4). So if we see + * what looks like the beginning of such a version number, just bail + * out. + */ + if (version < 10 && *ep == '.') + pg_fatal("%s: server version too old\n", filename); + pg_fatal("%s: could not parse version number\n", filename); + } + + /* Debugging output. */ + pg_log_debug("read server version %d from \"%s\"", version, filename); + + /* Release memory and return result. */ + pfree(buf.data); + return version * 10000; +} + +/* + * Add a directory to the list of output directories to clean up. + */ +static void +remember_to_cleanup_directory(char *target_path, bool rmtopdir) +{ + cb_cleanup_dir *dir = pg_malloc(sizeof(cb_cleanup_dir)); + + dir->target_path = target_path; + dir->rmtopdir = rmtopdir; + dir->next = cleanup_dir_list; + cleanup_dir_list = dir; +} + +/* + * Empty out the list of directories scheduled for cleanup a exit. + * + * We want to remove the output directories only on a failure, so call this + * function when we know that the operation has succeeded. + * + * Since we only expect this to be called when we're about to exit, we could + * just set cleanup_dir_list to NULL and be done with it, but we free the + * memory to be tidy. + */ +static void +reset_directory_cleanup_list(void) +{ + while (cleanup_dir_list != NULL) + { + cb_cleanup_dir *dir = cleanup_dir_list; + + cleanup_dir_list = cleanup_dir_list->next; + pfree(dir); + } +} + +/* + * Scan the pg_tblspc directory of the final input backup to get a canonical + * list of what tablespaces are part of the backup. + * + * 'pathname' should be the path to the toplevel backup directory for the + * final backup in the backup chain. + */ +static cb_tablespace * +scan_for_existing_tablespaces(char *pathname, cb_options *opt) +{ + char pg_tblspc[MAXPGPATH]; + DIR *dir; + struct dirent *de; + cb_tablespace *tslist = NULL; + + snprintf(pg_tblspc, MAXPGPATH, "%s/pg_tblspc", pathname); + pg_log_debug("scanning \"%s\"", pg_tblspc); + + if ((dir = opendir(pg_tblspc)) == NULL) + pg_fatal("could not open directory \"%s\": %m", pathname); + + while (errno = 0, (de = readdir(dir)) != NULL) + { + Oid oid; + char tblspcdir[MAXPGPATH]; + char link_target[MAXPGPATH]; + int link_length; + cb_tablespace *ts; + cb_tablespace *otherts; + PGFileType type; + + /* Silently ignore "." and ".." entries. */ + if (strcmp(de->d_name, ".") == 0 || strcmp(de->d_name, "..") == 0) + continue; + + /* Construct full pathname. */ + snprintf(tblspcdir, MAXPGPATH, "%s/%s", pg_tblspc, de->d_name); + + /* Ignore any file name that doesn't look like a proper OID. */ + if (!parse_oid(de->d_name, &oid)) + { + pg_log_debug("skipping \"%s\" because the filename is not a legal tablespace OID", + tblspcdir); + continue; + } + + /* Only symbolic links and directories are tablespaces. */ + type = get_dirent_type(tblspcdir, de, false, PG_LOG_ERROR); + if (type == PGFILETYPE_ERROR) + exit(1); + if (type != PGFILETYPE_LNK && type != PGFILETYPE_DIR) + { + pg_log_debug("skipping \"%s\" because it is neither a symbolic link nor a directory", + tblspcdir); + continue; + } + + /* Create a new tablespace object. */ + ts = pg_malloc0(sizeof(cb_tablespace)); + ts->oid = oid; + + /* + * If it's a link, it's not an in-place tablespace. Otherwise, it must + * be a directory, and thus an in-place tablespace. + */ + if (type == PGFILETYPE_LNK) + { + cb_tablespace_mapping *tsmap; + + /* Read the link target. */ + link_length = readlink(tblspcdir, link_target, sizeof(link_target)); + if (link_length < 0) + pg_fatal("could not read symbolic link \"%s\": %m", + tblspcdir); + if (link_length >= sizeof(link_target)) + pg_fatal("symbolic link \"%s\" is too long", tblspcdir); + link_target[link_length] = '\0'; + if (!is_absolute_path(link_target)) + pg_fatal("symbolic link \"%s\" is relative", tblspcdir); + + /* Caonicalize the link target. */ + canonicalize_path(link_target); + + /* + * Find the corresponding tablespace mapping and copy the relevant + * details into the new tablespace entry. + */ + for (tsmap = opt->tsmappings; tsmap != NULL; tsmap = tsmap->next) + { + if (strcmp(tsmap->old_dir, link_target) == 0) + { + strncpy(ts->old_dir, tsmap->old_dir, MAXPGPATH); + strncpy(ts->new_dir, tsmap->new_dir, MAXPGPATH); + ts->in_place = false; + break; + } + } + + /* Every non-in-place tablespace must be mapped. */ + if (tsmap == NULL) + pg_fatal("tablespace at \"%s\" has no tablespace mapping", + link_target); + } + else + { + /* + * For an in-place tablespace, there's no separate directory, so + * we just record the paths within the data directories. + */ + snprintf(ts->old_dir, MAXPGPATH, "%s/%s", pg_tblspc, de->d_name); + snprintf(ts->new_dir, MAXPGPATH, "%s/pg_tblpc/%s", opt->output, + de->d_name); + ts->in_place = true; + } + + /* Tablespaces should not share a directory. */ + for (otherts = tslist; otherts != NULL; otherts = otherts->next) + if (strcmp(ts->new_dir, otherts->new_dir) == 0) + pg_fatal("tablespaces with OIDs %u and %u both point at \"%s\"", + otherts->oid, oid, ts->new_dir); + + /* Add this tablespace to the list. */ + ts->next = tslist; + tslist = ts; + } + + return tslist; +} + +/* + * Read a file into a StringInfo. + * + * fd is used for the actual file I/O, filename for error reporting purposes. + * A file longer than maxlen is a fatal error. + */ +static void +slurp_file(int fd, char *filename, StringInfo buf, int maxlen) +{ + struct stat st; + ssize_t rb; + + /* Check file size, and complain if it's too large. */ + if (fstat(fd, &st) != 0) + pg_fatal("could not stat \"%s\": %m", filename); + if (st.st_size > maxlen) + pg_fatal("file \"%s\" is too large", filename); + + /* Make sure we have enough space. */ + enlargeStringInfo(buf, st.st_size); + + /* Read the data. */ + rb = read(fd, &buf->data[buf->len], st.st_size); + + /* + * We don't expect any concurrent changes, so we should read exactly the + * expected number of bytes. + */ + if (rb != st.st_size) + { + if (rb < 0) + pg_fatal("could not read file \"%s\": %m", filename); + else + pg_fatal("could not read file \"%s\": read only %d of %d bytes", + filename, (int) rb, (int) st.st_size); + } + + /* Adjust buffer length for new data and restore trailing-\0 invariant */ + buf->len += rb; + buf->data[buf->len] = '\0'; +} diff --git a/src/bin/pg_combinebackup/reconstruct.c b/src/bin/pg_combinebackup/reconstruct.c new file mode 100644 index 0000000000..c774bf1842 --- /dev/null +++ b/src/bin/pg_combinebackup/reconstruct.c @@ -0,0 +1,618 @@ +/*------------------------------------------------------------------------- + * + * reconstruct.c + * Reconstruct full file from incremental file and backup chain. + * + * Copyright (c) 2017-2023, PostgreSQL Global Development Group + * + * IDENTIFICATION + * src/bin/pg_combinebackup/reconstruct.c + * + *------------------------------------------------------------------------- + */ +#include "postgres_fe.h" + +#include <unistd.h> + +#include "backup/basebackup_incremental.h" +#include "common/logging.h" +#include "common/file_perm.h" +#include "copy_file.h" +#include "reconstruct.h" +#include "storage/block.h" + +/* + * An rfile stores the data that we need in order to be able to use some file + * on disk for reconstruction. For any given output file, we create one rfile + * per backup that we need to consult when we constructing that output file. + * + * If we find a full version of the file in the backup chain, then only + * filename and fd are initialized; the remaining fields are 0 or NULL. + * For an incremental file, header_length, num_blocks, relative_block_numbers, + * and truncation_block_length are also set. + * + * num_blocks_read and highest_offset_read always start out as 0. + */ +typedef struct rfile +{ + char *filename; + int fd; + size_t header_length; + unsigned num_blocks; + BlockNumber *relative_block_numbers; + unsigned truncation_block_length; + unsigned num_blocks_read; + off_t highest_offset_read; +} rfile; + +static void debug_reconstruction(int n_source, + rfile **sources, + bool dry_run); +static unsigned find_reconstructed_block_length(rfile *s); +static rfile *make_incremental_rfile(char *filename); +static rfile *make_rfile(char *filename, bool missing_ok); +static void write_reconstructed_file(char *input_filename, + char *output_filename, + unsigned block_length, + rfile **sourcemap, + off_t *offsetmap, + pg_checksum_context *checksum_ctx, + bool dry_run); +static void read_bytes(rfile *rf, void *buffer, unsigned length); + +/* + * Reconstruct a full file from an incremental file and a chain of prior + * backups. + * + * input_filename should be the path to the incremental file, and + * output_filename should be the path where the reconstructed file is to be + * written. + * + * relative_path should be the relative path to the directory containing this + * file. bare_file_name should be the name of the file within that directory, + * without "INCREMENTAL.". + * + * n_prior_backups is the number of prior backups, and prior_backup_dirs is + * an array of pathnames where those backups can be found. + */ +void +reconstruct_from_incremental_file(char *input_filename, + char *output_filename, + char *relative_path, + char *bare_file_name, + int n_prior_backups, + char **prior_backup_dirs, + manifest_data **manifests, + char *manifest_path, + pg_checksum_type checksum_type, + int *checksum_length, + uint8 **checksum_payload, + bool dry_run) +{ + rfile **source; + rfile *latest_source = NULL; + rfile **sourcemap; + off_t *offsetmap; + unsigned block_length; + unsigned num_missing_blocks; + unsigned i; + unsigned sidx = n_prior_backups; + bool full_copy_possible = true; + int copy_source_index = -1; + rfile *copy_source = NULL; + pg_checksum_context checksum_ctx; + + /* + * Every block must come either from the latest version of the file or + * from one of the prior backups. + */ + source = pg_malloc0(sizeof(rfile *) * (1 + n_prior_backups)); + + /* + * Use the information from the latest incremental file to figure out how + * long the reconstructed file should be. + */ + latest_source = make_incremental_rfile(input_filename); + source[n_prior_backups] = latest_source; + block_length = find_reconstructed_block_length(latest_source); + + /* + * For each block in the output file, we need to know from which file we + * need to obtain it and at what offset in that file it's stored. + * sourcemap gives us the first of these things, and offsetmap the latter. + */ + sourcemap = pg_malloc0(sizeof(rfile *) * block_length); + offsetmap = pg_malloc0(sizeof(off_t) * block_length); + + /* + * Blocks prior to the truncation_block_length threshold must be obtained + * from some prior backup, while those after that threshold are left as + * zeroes if not present in the newest incremental file. + * num_missing_blocks counts the number of blocks that we must be found + * somewhere in the backup chain, and is thus initially equal to + * truncation_block_length. + */ + num_missing_blocks = latest_source->truncation_block_length; + + /* + * Every block that is present in the newest incremental file should be + * sourced from that file. If it precedes the truncation_block_length, + * it's a block that we would otherwise have had to find in an older + * backup and thus reduces the number of blocks remaining to be found by + * one; otherwise, it's an extra block that needs to be included in the + * output but would not have needed to be found in an older backup if it + * had not been present. + */ + for (i = 0; i < latest_source->num_blocks; ++i) + { + BlockNumber b = latest_source->relative_block_numbers[i]; + + Assert(b < block_length); + sourcemap[b] = latest_source; + offsetmap[b] = latest_source->header_length + (i * BLCKSZ); + if (b < latest_source->truncation_block_length) + num_missing_blocks--; + + /* + * A full copy of a file from an earlier backup is only possible if no + * blocks are needed from any later incremental file. + */ + full_copy_possible = false; + } + + while (num_missing_blocks > 0) + { + char source_filename[MAXPGPATH]; + rfile *s; + + /* + * Move to the next backup in the chain. If there are no more, then + * something has gone wrong and reconstruction has failed. + */ + if (sidx == 0) + pg_fatal("reconstruction for file \"%s\" failed to find %u required blocks", + output_filename, num_missing_blocks); + --sidx; + + /* + * Look for the full file in the previous backup. If not found, then + * look for an incremental file instead. + */ + snprintf(source_filename, MAXPGPATH, "%s/%s/%s", + prior_backup_dirs[sidx], relative_path, bare_file_name); + if ((s = make_rfile(source_filename, true)) == NULL) + { + snprintf(source_filename, MAXPGPATH, "%s/%s/INCREMENTAL.%s", + prior_backup_dirs[sidx], relative_path, bare_file_name); + s = make_incremental_rfile(source_filename); + } + source[sidx] = s; + + /* + * If s->header_length == 0, then this is a full file; otherwise, it's + * an incremental file. + */ + if (s->header_length != 0) + { + /* + * Since we found another incremental file, source all blocks from + * it that we need but don't yet have. + */ + for (i = 0; i < s->num_blocks; ++i) + { + BlockNumber b = s->relative_block_numbers[i]; + + if (b < latest_source->truncation_block_length && + sourcemap[b] == NULL) + { + sourcemap[b] = s; + offsetmap[b] = s->header_length + (i * BLCKSZ); + + Assert(num_missing_blocks > 0); + --num_missing_blocks; + + /* + * A full copy of a file from an earlier backup is only + * possible if no blocks are needed from any later + * incremental file. + */ + full_copy_possible = false; + } + } + } + else + { + BlockNumber b; + + /* + * Since we found a full file, source all remaining required + * blocks from it. + */ + for (b = 0; b < latest_source->truncation_block_length; ++b) + { + if (sourcemap[b] == NULL) + { + sourcemap[b] = s; + offsetmap[b] = b * BLCKSZ; + + Assert(num_missing_blocks > 0); + --num_missing_blocks; + } + } + Assert(num_missing_blocks == 0); + + /* + * If a full copy looks possible, check whether the resulting file + * should be exactly as long as the source file is. If so, a full + * copy is acceptable, otherwise not. + */ + if (full_copy_possible) + { + struct stat sb; + uint64 expected_length; + + if (fstat(s->fd, &sb) < 0) + pg_fatal("could not stat \"%s\": %m", s->filename); + expected_length = + (uint64) latest_source->truncation_block_length; + expected_length *= BLCKSZ; + if (expected_length == sb.st_size) + { + copy_source = s; + copy_source_index = sidx; + } + } + } + } + + /* + * If a checksum of the required type already exists in the + * backup_manifest for the relevant input directory, we can save some work + * by reusing that checksum instead of computing a new one. + */ + if (copy_source_index >= 0 && manifests[copy_source_index] != NULL && + checksum_type != CHECKSUM_TYPE_NONE) + { + manifest_file *mfile; + + mfile = manifest_files_lookup(manifests[copy_source_index]->files, + manifest_path); + if (mfile == NULL) + { + /* + * The directory is out of sync with the backup_manifest, so emit + * a warning. + */ + pg_log_warning("\"%s/backup_manifest\" contains no entry for \"%s\"", + prior_backup_dirs[copy_source_index], + manifest_path); + } + else if (mfile->checksum_type == checksum_type) + { + *checksum_length = mfile->checksum_length; + *checksum_payload = pg_malloc(*checksum_length); + memcpy(*checksum_payload, mfile->checksum_payload, + *checksum_length); + checksum_type = CHECKSUM_TYPE_NONE; + } + } + + /* Prepare for checksum calculation, if required. */ + pg_checksum_init(&checksum_ctx, checksum_type); + + /* + * If the full file can be created by copying a file from an older backup + * in the chain without needing to overwrite any blocks or truncate the + * result, then forget about performing reconstruction and just copy that + * file in its entirety. + * + * Otherwise, reconstruct. + */ + if (copy_source != NULL) + copy_file(copy_source->filename, output_filename, + &checksum_ctx, dry_run); + else + { + write_reconstructed_file(input_filename, output_filename, + block_length, sourcemap, offsetmap, + &checksum_ctx, dry_run); + debug_reconstruction(n_prior_backups + 1, source, dry_run); + } + + /* Save results of checksum calculation. */ + if (checksum_type != CHECKSUM_TYPE_NONE) + { + *checksum_payload = pg_malloc(PG_CHECKSUM_MAX_LENGTH); + *checksum_length = pg_checksum_final(&checksum_ctx, + *checksum_payload); + } + + /* + * Close files and release memory. + */ + for (i = 0; i <= n_prior_backups; ++i) + { + rfile *s = source[i]; + + if (s == NULL) + continue; + if (close(s->fd) != 0) + pg_fatal("could not close \"%s\": %m", s->filename); + if (s->relative_block_numbers != NULL) + pfree(s->relative_block_numbers); + pg_free(s->filename); + } + pfree(sourcemap); + pfree(offsetmap); + pfree(source); +} + +/* + * Perform post-reconstruction logging and sanity checks. + */ +static void +debug_reconstruction(int n_source, rfile **sources, bool dry_run) +{ + unsigned i; + + for (i = 0; i < n_source; ++i) + { + rfile *s = sources[i]; + + /* Ignore source if not used. */ + if (s == NULL) + continue; + + /* If no data is needed from this file, we can ignore it. */ + if (s->num_blocks_read == 0) + continue; + + /* Debug logging. */ + if (dry_run) + pg_log_debug("would have read %u blocks from \"%s\"", + s->num_blocks_read, s->filename); + else + pg_log_debug("read %u blocks from \"%s\"", + s->num_blocks_read, s->filename); + + /* + * In dry-run mode, we don't actually try to read data from the file, + * but we do try to verify that the file is long enough that we could + * have read the data if we'd tried. + * + * If this fails, then it means that a non-dry-run attempt would fail, + * complaining of not being able to read the required bytes from the + * file. + */ + if (dry_run) + { + struct stat sb; + + if (fstat(s->fd, &sb) < 0) + pg_fatal("could not stat \"%s\": %m", s->filename); + if (sb.st_size < s->highest_offset_read) + pg_fatal("file \"%s\" is too short: expected %llu, found %llu", + s->filename, + (unsigned long long) s->highest_offset_read, + (unsigned long long) sb.st_size); + } + } +} + +/* + * When we perform reconstruction using an incremental file, the output file + * should be at least as long as the truncation_block_length. Any blocks + * present in the incremental file increase the output length as far as is + * necessary to include those blocks. + */ +static unsigned +find_reconstructed_block_length(rfile *s) +{ + unsigned block_length = s->truncation_block_length; + unsigned i; + + for (i = 0; i < s->num_blocks; ++i) + if (s->relative_block_numbers[i] >= block_length) + block_length = s->relative_block_numbers[i] + 1; + + return block_length; +} + +/* + * Initialize an incremental rfile, reading the header so that we know which + * blocks it contains. + */ +static rfile * +make_incremental_rfile(char *filename) +{ + rfile *rf; + unsigned magic; + + rf = make_rfile(filename, false); + + /* Read and validate magic number. */ + read_bytes(rf, &magic, sizeof(magic)); + if (magic != INCREMENTAL_MAGIC) + pg_fatal("file \"%s\" has bad incremental magic number (0x%x not 0x%x)", + filename, magic, INCREMENTAL_MAGIC); + + /* Read block count. */ + read_bytes(rf, &rf->num_blocks, sizeof(rf->num_blocks)); + if (rf->num_blocks > RELSEG_SIZE) + pg_fatal("file \"%s\" has block count %u in excess of segment size %u", + filename, rf->num_blocks, RELSEG_SIZE); + + /* Read truncation block length. */ + read_bytes(rf, &rf->truncation_block_length, + sizeof(rf->truncation_block_length)); + if (rf->truncation_block_length > RELSEG_SIZE) + pg_fatal("file \"%s\" has truncation block length %u in excess of segment size %u", + filename, rf->truncation_block_length, RELSEG_SIZE); + + /* Read block numbers if there are any. */ + if (rf->num_blocks > 0) + { + rf->relative_block_numbers = + pg_malloc0(sizeof(BlockNumber) * rf->num_blocks); + read_bytes(rf, rf->relative_block_numbers, + sizeof(BlockNumber) * rf->num_blocks); + } + + /* Remember length of header. */ + rf->header_length = sizeof(magic) + sizeof(rf->num_blocks) + + sizeof(rf->truncation_block_length) + + sizeof(BlockNumber) * rf->num_blocks; + + return rf; +} + +/* + * Allocate and perform basic initialization of an rfile. + */ +static rfile * +make_rfile(char *filename, bool missing_ok) +{ + rfile *rf; + + rf = pg_malloc0(sizeof(rfile)); + rf->filename = pstrdup(filename); + if ((rf->fd = open(filename, O_RDONLY | PG_BINARY, 0)) < 0) + { + if (missing_ok && errno == ENOENT) + { + pg_free(rf); + return NULL; + } + pg_fatal("could not open file \"%s\": %m", filename); + } + + return rf; +} + +/* + * Read the indicated number of bytes from an rfile into the buffer. + */ +static void +read_bytes(rfile *rf, void *buffer, unsigned length) +{ + unsigned rb = read(rf->fd, buffer, length); + + if (rb != length) + { + if (rb < 0) + pg_fatal("could not read file \"%s\": %m", rf->filename); + else + pg_fatal("could not read file \"%s\": read only %d of %d bytes", + rf->filename, (int) rb, length); + } +} + +/* + * Write out a reconstructed file. + */ +static void +write_reconstructed_file(char *input_filename, + char *output_filename, + unsigned block_length, + rfile **sourcemap, + off_t *offsetmap, + pg_checksum_context *checksum_ctx, + bool dry_run) +{ + int wfd = -1; + unsigned i; + unsigned zero_blocks = 0; + + /* Debugging output. */ + if (dry_run) + pg_log_debug("would reconstruct \"%s\" (%u blocks, checksum %s)", + output_filename, block_length, + pg_checksum_type_name(checksum_ctx->type)); + else + pg_log_debug("reconstructing \"%s\" (%u blocks, checksum %s)", + output_filename, block_length, + pg_checksum_type_name(checksum_ctx->type)); + + /* Open the output file, except in dry_run mode. */ + if (!dry_run && + (wfd = open(output_filename, + O_RDWR | PG_BINARY | O_CREAT | O_EXCL, + pg_file_create_mode)) < 0) + pg_fatal("could not open file \"%s\": %m", output_filename); + + /* Read and write the blocks as required. */ + for (i = 0; i < block_length; ++i) + { + uint8 buffer[BLCKSZ]; + rfile *s = sourcemap[i]; + unsigned wb; + + /* Update accounting information. */ + if (s == NULL) + ++zero_blocks; + else + { + s->num_blocks_read++; + s->highest_offset_read = Max(s->highest_offset_read, + offsetmap[i] + BLCKSZ); + } + + /* Skip the rest of this in dry-run mode. */ + if (dry_run) + continue; + + /* Read or zero-fill the block as appropriate. */ + if (s == NULL) + { + /* + * New block not mentioned in the WAL summary. Should have been an + * uninitialized block, so just zero-fill it. + */ + memset(buffer, 0, BLCKSZ); + } + else + { + unsigned rb; + + /* Read the block from the correct source, except if dry-run. */ + rb = pg_pread(s->fd, buffer, BLCKSZ, offsetmap[i]); + if (rb != BLCKSZ) + { + if (rb < 0) + pg_fatal("could not read file \"%s\": %m", s->filename); + else + pg_fatal("could not read file \"%s\": read only %d of %d bytes at offset %u", + s->filename, (int) rb, BLCKSZ, + (unsigned) offsetmap[i]); + } + } + + /* Write out the block. */ + if ((wb = write(wfd, buffer, BLCKSZ)) != BLCKSZ) + { + if (wb < 0) + pg_fatal("could not write file \"%s\": %m", output_filename); + else + pg_fatal("could not write file \"%s\": wrote only %d of %d bytes", + output_filename, (int) wb, BLCKSZ); + } + + /* Update the checksum computation. */ + if (pg_checksum_update(checksum_ctx, buffer, BLCKSZ) < 0) + pg_fatal("could not update checksum of file \"%s\"", + output_filename); + } + + /* Debugging output. */ + if (zero_blocks > 0) + { + if (dry_run) + pg_log_debug("would have zero-filled %u blocks", zero_blocks); + else + pg_log_debug("zero-filled %u blocks", zero_blocks); + } + + /* Close the output file. */ + if (wfd >= 0 && close(wfd) != 0) + pg_fatal("could not close \"%s\": %m", output_filename); +} diff --git a/src/bin/pg_combinebackup/reconstruct.h b/src/bin/pg_combinebackup/reconstruct.h new file mode 100644 index 0000000000..c599a70d42 --- /dev/null +++ b/src/bin/pg_combinebackup/reconstruct.h @@ -0,0 +1,32 @@ +/*------------------------------------------------------------------------- + * + * reconstruct.h + * Reconstruct full file from incremental file and backup chain. + * + * Copyright (c) 2017-2023, PostgreSQL Global Development Group + * + * IDENTIFICATION + * src/bin/pg_combinebackup/reconstruct.h + * + *------------------------------------------------------------------------- + */ +#ifndef RECONSTRUCT_H +#define RECONSTRUCT_H + +#include "common/checksum_helper.h" +#include "load_manifest.h" + +extern void reconstruct_from_incremental_file(char *input_filename, + char *output_filename, + char *relative_path, + char *bare_file_name, + int n_prior_backups, + char **prior_backup_dirs, + manifest_data **manifests, + char *manifest_path, + pg_checksum_type checksum_type, + int *checksum_length, + uint8 **checksum_payload, + bool dry_run); + +#endif diff --git a/src/bin/pg_combinebackup/write_manifest.c b/src/bin/pg_combinebackup/write_manifest.c new file mode 100644 index 0000000000..82160134d8 --- /dev/null +++ b/src/bin/pg_combinebackup/write_manifest.c @@ -0,0 +1,293 @@ +/*------------------------------------------------------------------------- + * + * Write a new backup manifest. + * + * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group + * Portions Copyright (c) 1994, Regents of the University of California + * + * src/bin/pg_combinebackup/write_manifest.c + * + *------------------------------------------------------------------------- + */ + +#include "postgres_fe.h" + +#include <fcntl.h> +#include <time.h> +#include <unistd.h> + +#include "common/checksum_helper.h" +#include "common/file_perm.h" +#include "common/logging.h" +#include "lib/stringinfo.h" +#include "load_manifest.h" +#include "mb/pg_wchar.h" +#include "write_manifest.h" + +struct manifest_writer +{ + char pathname[MAXPGPATH]; + int fd; + StringInfoData buf; + bool first_file; + bool still_checksumming; + pg_checksum_context manifest_ctx; +}; + +static void escape_json(StringInfo buf, const char *str); +static void flush_manifest(manifest_writer *mwriter); +static size_t hex_encode(const uint8 *src, size_t len, char *dst); + +/* + * Create a new backup manifest writer. + * + * The backup manifest will be written into a file named backup_manifest + * in the specified directory. + */ +manifest_writer * +create_manifest_writer(char *directory) +{ + manifest_writer *mwriter = pg_malloc(sizeof(manifest_writer)); + + snprintf(mwriter->pathname, MAXPGPATH, "%s/backup_manifest", directory); + mwriter->fd = -1; + initStringInfo(&mwriter->buf); + mwriter->first_file = true; + mwriter->still_checksumming = true; + pg_checksum_init(&mwriter->manifest_ctx, CHECKSUM_TYPE_SHA256); + + appendStringInfo(&mwriter->buf, + "{ \"PostgreSQL-Backup-Manifest-Version\": 1,\n" + "\"Files\": ["); + + return mwriter; +} + +/* + * Add an entry for a file to a backup manifest. + * + * This is very similar to the backend's AddFileToBackupManifest, but + * various adjustments are required due to frontend/backend differences + * and other details. + */ +void +add_file_to_manifest(manifest_writer *mwriter, const char *manifest_path, + size_t size, pg_time_t mtime, + pg_checksum_type checksum_type, + int checksum_length, + uint8 *checksum_payload) +{ + int pathlen = strlen(manifest_path); + + if (mwriter->first_file) + { + appendStringInfoChar(&mwriter->buf, '\n'); + mwriter->first_file = false; + } + else + appendStringInfoString(&mwriter->buf, ",\n"); + + if (pg_encoding_verifymbstr(PG_UTF8, manifest_path, pathlen) == pathlen) + { + appendStringInfoString(&mwriter->buf, "{ \"Path\": "); + escape_json(&mwriter->buf, manifest_path); + appendStringInfoString(&mwriter->buf, ", "); + } + else + { + appendStringInfoString(&mwriter->buf, "{ \"Encoded-Path\": \""); + enlargeStringInfo(&mwriter->buf, 2 * pathlen); + mwriter->buf.len += hex_encode((const uint8 *) manifest_path, pathlen, + &mwriter->buf.data[mwriter->buf.len]); + appendStringInfoString(&mwriter->buf, "\", "); + } + + appendStringInfo(&mwriter->buf, "\"Size\": %zu, ", size); + + appendStringInfoString(&mwriter->buf, "\"Last-Modified\": \""); + enlargeStringInfo(&mwriter->buf, 128); + mwriter->buf.len += strftime(&mwriter->buf.data[mwriter->buf.len], 128, + "%Y-%m-%d %H:%M:%S %Z", + gmtime(&mtime)); + appendStringInfoChar(&mwriter->buf, '"'); + + if (mwriter->buf.len > 128 * 1024) + flush_manifest(mwriter); + + if (checksum_length > 0) + { + appendStringInfo(&mwriter->buf, + ", \"Checksum-Algorithm\": \"%s\", \"Checksum\": \"", + pg_checksum_type_name(checksum_type)); + + enlargeStringInfo(&mwriter->buf, 2 * checksum_length); + mwriter->buf.len += hex_encode(checksum_payload, checksum_length, + &mwriter->buf.data[mwriter->buf.len]); + + appendStringInfoChar(&mwriter->buf, '"'); + } + + appendStringInfoString(&mwriter->buf, " }"); + + if (mwriter->buf.len > 128 * 1024) + flush_manifest(mwriter); +} + +/* + * Finalize the backup_manifest. + */ +void +finalize_manifest(manifest_writer *mwriter, + manifest_wal_range *first_wal_range) +{ + uint8 checksumbuf[PG_SHA256_DIGEST_LENGTH]; + int len; + manifest_wal_range *wal_range; + + /* Terminate the list of files. */ + appendStringInfoString(&mwriter->buf, "\n],\n"); + + /* Start a list of LSN ranges. */ + appendStringInfoString(&mwriter->buf, "\"WAL-Ranges\": [\n"); + + for (wal_range = first_wal_range; wal_range != NULL; + wal_range = wal_range->next) + appendStringInfo(&mwriter->buf, + "%s{ \"Timeline\": %u, \"Start-LSN\": \"%X/%X\", \"End-LSN\": \"%X/%X\" }", + wal_range == first_wal_range ? "" : ",\n", + wal_range->tli, + LSN_FORMAT_ARGS(wal_range->start_lsn), + LSN_FORMAT_ARGS(wal_range->end_lsn)); + + /* Terminate the list of WAL ranges. */ + appendStringInfoString(&mwriter->buf, "\n],\n"); + + /* Flush accumulated data and update checksum calculation. */ + flush_manifest(mwriter); + + /* Checksum only includes data up to this point. */ + mwriter->still_checksumming = false; + + /* Compute and insert manifest checksum. */ + appendStringInfoString(&mwriter->buf, "\"Manifest-Checksum\": \""); + enlargeStringInfo(&mwriter->buf, 2 * PG_SHA256_DIGEST_STRING_LENGTH); + len = pg_checksum_final(&mwriter->manifest_ctx, checksumbuf); + Assert(len == PG_SHA256_DIGEST_LENGTH); + mwriter->buf.len += + hex_encode(checksumbuf, len, &mwriter->buf.data[mwriter->buf.len]); + appendStringInfoString(&mwriter->buf, "\"}\n"); + + /* Flush the last manifest checksum itself. */ + flush_manifest(mwriter); + + /* Close the file. */ + if (close(mwriter->fd) != 0) + pg_fatal("could not close \"%s\": %m", mwriter->pathname); + mwriter->fd = -1; +} + +/* + * Produce a JSON string literal, properly escaping characters in the text. + */ +static void +escape_json(StringInfo buf, const char *str) +{ + const char *p; + + appendStringInfoCharMacro(buf, '"'); + for (p = str; *p; p++) + { + switch (*p) + { + case '\b': + appendStringInfoString(buf, "\\b"); + break; + case '\f': + appendStringInfoString(buf, "\\f"); + break; + case '\n': + appendStringInfoString(buf, "\\n"); + break; + case '\r': + appendStringInfoString(buf, "\\r"); + break; + case '\t': + appendStringInfoString(buf, "\\t"); + break; + case '"': + appendStringInfoString(buf, "\\\""); + break; + case '\\': + appendStringInfoString(buf, "\\\\"); + break; + default: + if ((unsigned char) *p < ' ') + appendStringInfo(buf, "\\u%04x", (int) *p); + else + appendStringInfoCharMacro(buf, *p); + break; + } + } + appendStringInfoCharMacro(buf, '"'); +} + +/* + * Flush whatever portion of the backup manifest we have generated and + * buffered in memory out to a file on disk. + * + * The first call to this function will create the file. After that, we + * keep it open and just append more data. + */ +static void +flush_manifest(manifest_writer *mwriter) +{ + char pathname[MAXPGPATH]; + + if (mwriter->fd == -1 && + (mwriter->fd = open(mwriter->pathname, + O_WRONLY | O_CREAT | O_EXCL | PG_BINARY, + pg_file_create_mode)) < 0) + pg_fatal("could not open file \"%s\": %m", mwriter->pathname); + + if (mwriter->buf.len > 0) + { + ssize_t wb; + + wb = write(mwriter->fd, mwriter->buf.data, mwriter->buf.len); + if (wb != mwriter->buf.len) + { + if (wb < 0) + pg_fatal("could not write \"%s\": %m", mwriter->pathname); + else + pg_fatal("could not write file \"%s\": wrote only %d of %d bytes", + pathname, (int) wb, mwriter->buf.len); + } + + if (mwriter->still_checksumming) + pg_checksum_update(&mwriter->manifest_ctx, + (uint8 *) mwriter->buf.data, + mwriter->buf.len); + resetStringInfo(&mwriter->buf); + } +} + +/* + * Encode bytes using two hexademical digits for each one. + */ +static size_t +hex_encode(const uint8 *src, size_t len, char *dst) +{ + const uint8 *end = src + len; + + while (src < end) + { + unsigned n1 = (*src >> 4) & 0xF; + unsigned n2 = *src & 0xF; + + *dst++ = n1 < 10 ? '0' + n1 : 'a' + n1 - 10; + *dst++ = n2 < 10 ? '0' + n2 : 'a' + n2 - 10; + ++src; + } + + return len * 2; +} diff --git a/src/bin/pg_combinebackup/write_manifest.h b/src/bin/pg_combinebackup/write_manifest.h new file mode 100644 index 0000000000..8fd7fe02c8 --- /dev/null +++ b/src/bin/pg_combinebackup/write_manifest.h @@ -0,0 +1,33 @@ +/*------------------------------------------------------------------------- + * + * Write a new backup manifest. + * + * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group + * Portions Copyright (c) 1994, Regents of the University of California + * + * src/bin/pg_combinebackup/write_manifest.h + * + *------------------------------------------------------------------------- + */ +#ifndef WRITE_MANIFEST_H +#define WRITE_MANIFEST_H + +#include "common/checksum_helper.h" +#include "pgtime.h" + +struct manifest_wal_range; + +struct manifest_writer; +typedef struct manifest_writer manifest_writer; + +extern manifest_writer *create_manifest_writer(char *directory); +extern void add_file_to_manifest(manifest_writer *mwriter, + const char *manifest_path, + size_t size, pg_time_t mtime, + pg_checksum_type checksum_type, + int checksum_length, + uint8 *checksum_payload); +extern void finalize_manifest(manifest_writer *mwriter, + struct manifest_wal_range *first_wal_range); + +#endif /* WRITE_MANIFEST_H */ diff --git a/src/bin/pg_resetwal/pg_resetwal.c b/src/bin/pg_resetwal/pg_resetwal.c index 25ecdaaa15..15a40cd17e 100644 --- a/src/bin/pg_resetwal/pg_resetwal.c +++ b/src/bin/pg_resetwal/pg_resetwal.c @@ -86,6 +86,7 @@ static void RewriteControlFile(void); static void FindEndOfXLOG(void); static void KillExistingXLOG(void); static void KillExistingArchiveStatus(void); +static void KillExistingWALSummaries(void); static void WriteEmptyXLOG(void); static void usage(void); @@ -492,6 +493,7 @@ main(int argc, char *argv[]) RewriteControlFile(); KillExistingXLOG(); KillExistingArchiveStatus(); + KillExistingWALSummaries(); WriteEmptyXLOG(); printf(_("Write-ahead log reset\n")); @@ -1033,6 +1035,40 @@ KillExistingArchiveStatus(void) pg_fatal("could not close directory \"%s\": %m", ARCHSTATDIR); } +/* + * Remove existing WAL summary files + */ +static void +KillExistingWALSummaries(void) +{ +#define WALSUMMARYDIR XLOGDIR "/summaries" +#define WALSUMMARY_NHEXCHARS 40 + + DIR *xldir; + struct dirent *xlde; + char path[MAXPGPATH + sizeof(WALSUMMARYDIR)]; + + xldir = opendir(WALSUMMARYDIR); + if (xldir == NULL) + pg_fatal("could not open directory \"%s\": %m", WALSUMMARYDIR); + + while (errno = 0, (xlde = readdir(xldir)) != NULL) + { + if (strspn(xlde->d_name, "0123456789ABCDEF") == WALSUMMARY_NHEXCHARS && + strcmp(xlde->d_name + WALSUMMARY_NHEXCHARS, ".summary") == 0) + { + snprintf(path, sizeof(path), "%s/%s", WALSUMMARYDIR, xlde->d_name); + if (unlink(path) < 0) + pg_fatal("could not delete file \"%s\": %m", path); + } + } + + if (errno) + pg_fatal("could not read directory \"%s\": %m", WALSUMMARYDIR); + + if (closedir(xldir)) + pg_fatal("could not close directory \"%s\": %m", ARCHSTATDIR); +} /* * Write an empty XLOG file, containing only the checkpoint record diff --git a/src/common/Makefile b/src/common/Makefile index e4cd26762b..ef38cc2f03 100644 --- a/src/common/Makefile +++ b/src/common/Makefile @@ -48,6 +48,7 @@ LIBS += $(PTHREAD_LIBS) OBJS_COMMON = \ archive.o \ base64.o \ + blkreftable.o \ checksum_helper.o \ compression.o \ config_info.o \ diff --git a/src/common/blkreftable.c b/src/common/blkreftable.c new file mode 100644 index 0000000000..012a443584 --- /dev/null +++ b/src/common/blkreftable.c @@ -0,0 +1,1309 @@ +/*------------------------------------------------------------------------- + * + * blkreftable.c + * Block reference tables. + * + * A block reference table is used to keep track of which blocks have + * been modified by WAL records within a certain LSN range. + * + * For each relation fork, we keep track of all blocks that have appeared + * in block reference in the WAL. We also keep track of the "limit block", + * which is the smallest relation length in blocks known to have occurred + * during that range of WAL records. This should be set to 0 if the relation + * fork is created or destroyed, and to the post-truncation length if + * truncated. + * + * Whenever we set the limit block, we also forget about any modified blocks + * beyond that point. Those blocks don't exist any more. Such blocks can + * later be marked as modified again; if that happens, it means the relation + * was re-extended. + * + * Portions Copyright (c) 2010-2022, PostgreSQL Global Development Group + * + * src/common/blkreftable.c + * + *------------------------------------------------------------------------- + */ + + +#ifndef FRONTEND +#include "postgres.h" +#else +#include "postgres_fe.h" +#endif + +#ifdef FRONTEND +#include "common/logging.h" +#endif + +#include "common/blkreftable.h" +#include "common/hashfn.h" +#include "port/pg_crc32c.h" + +/* + * A block reference table keeps track of the status of each relation + * fork individually. + */ +typedef struct BlockRefTableKey +{ + RelFileLocator rlocator; + ForkNumber forknum; +} BlockRefTableKey; + +/* + * We could need to store data either for a relation in which only a + * tiny fraction of the blocks have been modified or for a relation in + * which nearly every block has been modified, and we want a + * space-efficient representation in both cases. To accomplish this, + * we divide the relation into chunks of 2^16 blocks and choose between + * an array representation and a bitmap representation for each chunk. + * + * When the number of modified blocks in a given chunk is small, we + * essentially store an array of block numbers, but we need not store the + * entire block number: instead, we store each block number as a 2-byte + * offset from the start of the chunk. + * + * When the number of modified blocks in a given chunk is large, we switch + * to a bitmap representation. + * + * These same basic representational choices are used both when a block + * reference table is stored in memory and when it is serialized to disk. + * + * In the in-memory representation, we initially allocate each chunk with + * space for a number of entries given by INITIAL_ENTRIES_PER_CHUNK and + * increase that as necessary until we reach MAX_ENTRIES_PER_CHUNK. + * Any chunk whose allocated size reaches MAX_ENTRIES_PER_CHUNK is converted + * to a bitmap, and thus never needs to grow further. + */ +#define BLOCKS_PER_CHUNK (1 << 16) +#define BLOCKS_PER_ENTRY (BITS_PER_BYTE * sizeof(uint16)) +#define MAX_ENTRIES_PER_CHUNK (BLOCKS_PER_CHUNK / BLOCKS_PER_ENTRY) +#define INITIAL_ENTRIES_PER_CHUNK 16 +typedef uint16 *BlockRefTableChunk; + +/* + * State for one relation fork. + * + * 'rlocator' and 'forknum' identify the relation fork to which this entry + * pertains. + * + * 'limit_block' is the shortest known length of the relation in blocks + * within the LSN range covered by a particular block reference table. + * It should be set to 0 if the relation fork is created or dropped. If the + * relation fork is truncated, it should be set to the number of blocks that + * remain after truncation. + * + * 'nchunks' is the allocated length of each of the three arrays that follow. + * We can only represent the status of block numbers less than nchunks * + * BLOCKS_PER_CHUNK. + * + * 'chunk_size' is an array storing the allocated size of each chunk. + * + * 'chunk_usage' is an array storing the number of elements used in each + * chunk. If that value is less than MAX_ENTRIES_PER_CHUNK, the corresonding + * chunk is used as an array; else the corresponding chunk is used as a bitmap. + * When used as a bitmap, the least significant bit of the first array element + * is the status of the lowest-numbered block covered by this chunk. + * + * 'chunk_data' is the array of chunks. + */ +struct BlockRefTableEntry +{ + BlockRefTableKey key; + BlockNumber limit_block; + char status; + uint32 nchunks; + uint16 *chunk_size; + uint16 *chunk_usage; + BlockRefTableChunk *chunk_data; +}; + +/* Declare and define a hash table over type BlockRefTableEntry. */ +#define SH_PREFIX blockreftable +#define SH_ELEMENT_TYPE BlockRefTableEntry +#define SH_KEY_TYPE BlockRefTableKey +#define SH_KEY key +#define SH_HASH_KEY(tb, key) \ + hash_bytes((const unsigned char *) &key, sizeof(BlockRefTableKey)) +#define SH_EQUAL(tb, a, b) memcmp(&a, &b, sizeof(BlockRefTableKey)) == 0 +#define SH_SCOPE static inline +#ifdef FRONTEND +#define SH_RAW_ALLOCATOR pg_malloc0 +#endif +#define SH_DEFINE +#define SH_DECLARE +#include "lib/simplehash.h" + +/* + * A block reference table is basically just the hash table, but we don't + * want to expose that to outside callers. + * + * We keep track of the memory context in use explicitly too, so that it's + * easy to place all of our allocations in the same context. + */ +struct BlockRefTable +{ + blockreftable_hash *hash; +#ifndef FRONTEND + MemoryContext mcxt; +#endif +}; + +/* + * On-disk serialization format for block reference table entries. + */ +typedef struct BlockRefTableSerializedEntry +{ + RelFileLocator rlocator; + ForkNumber forknum; + BlockNumber limit_block; + uint32 nchunks; +} BlockRefTableSerializedEntry; + +/* + * Buffer size, so that we avoid doing many small I/Os. + */ +#define BUFSIZE 65536 + +/* + * Ad-hoc buffer for file I/O. + */ +typedef struct BlockRefTableBuffer +{ + io_callback_fn io_callback; + void *io_callback_arg; + char data[BUFSIZE]; + int used; + int cursor; + pg_crc32c crc; +} BlockRefTableBuffer; + +/* + * State for keeping track of progress while incrementally reading a block + * table reference file from disk. + * + * total_chunks means the number of chunks for the RelFileLocator/ForkNumber + * combination that is curently being read, and consumed_chunks is the number + * of those that have been read. (We always read all the information for + * a single chunk at one time, so we don't need to be able to represent the + * state where a chunk has been partially read.) + * + * chunk_size is the array of chunk sizes. The length is given by total_chunks. + * + * chunk_data holds the current chunk. + * + * chunk_position helps us figure out how much progress we've made in returning + * the block numbers for the current chunk to the caller. If the chunk is a + * bitmap, it's the number of bits we've scanned; otherwise, it's the number + * of chunk entries we've scanned. + */ +struct BlockRefTableReader +{ + BlockRefTableBuffer buffer; + char *error_filename; + report_error_fn error_callback; + void *error_callback_arg; + uint32 total_chunks; + uint32 consumed_chunks; + uint16 *chunk_size; + uint16 chunk_data[MAX_ENTRIES_PER_CHUNK]; + uint32 chunk_position; +}; + +/* + * State for keeping track of progress while incrementally writing a block + * reference table file to disk. + */ +struct BlockRefTableWriter +{ + BlockRefTableBuffer buffer; +}; + +/* Function prototypes. */ +static int BlockRefTableComparator(const void *a, const void *b); +static void BlockRefTableFlush(BlockRefTableBuffer *buffer); +static void BlockRefTableRead(BlockRefTableReader *reader, void *data, + int length); +static void BlockRefTableWrite(BlockRefTableBuffer *buffer, void *data, + int length); +static void BlockRefTableFileTerminate(BlockRefTableBuffer *buffer); + +/* + * Create an empty block reference table. + */ +BlockRefTable * +CreateEmptyBlockRefTable(void) +{ + BlockRefTable *brtab = palloc(sizeof(BlockRefTable)); + + /* + * Even completely empty database has a few hundred relation forks, so it + * seems best to size the hash on the assumption that we're going to have + * at least a few thousand entries. + */ +#ifdef FRONTEND + brtab->hash = blockreftable_create(4096, NULL); +#else + brtab->mcxt = CurrentMemoryContext; + brtab->hash = blockreftable_create(brtab->mcxt, 4096, NULL); +#endif + + return brtab; +} + +/* + * Set the "limit block" for a relation fork and forget any modified blocks + * with equal or higher block numbers. + * + * The "limit block" is the shortest known length of the relation within the + * range of WAL records covered by this block reference table. + */ +void +BlockRefTableSetLimitBlock(BlockRefTable *brtab, + const RelFileLocator *rlocator, + ForkNumber forknum, + BlockNumber limit_block) +{ + BlockRefTableEntry *brtentry; + BlockRefTableKey key; + bool found; + + memcpy(&key.rlocator, rlocator, sizeof(RelFileLocator)); + key.forknum = forknum; + brtentry = blockreftable_insert(brtab->hash, key, &found); + + if (!found) + { + /* + * We have no existing data about this relation fork, so just record + * the limit_block value supplied by the caller, and make sure other + * parts of the entry are properly initialized. + */ + brtentry->limit_block = limit_block; + brtentry->nchunks = 0; + brtentry->chunk_size = NULL; + brtentry->chunk_usage = NULL; + brtentry->chunk_data = NULL; + return; + } + + BlockRefTableEntrySetLimitBlock(brtentry, limit_block); +} + +/* + * Mark a block in a given relation fork as known to have been modified. + */ +void +BlockRefTableMarkBlockModified(BlockRefTable *brtab, + const RelFileLocator *rlocator, + ForkNumber forknum, + BlockNumber blknum) +{ + BlockRefTableEntry *brtentry; + BlockRefTableKey key; + bool found; +#ifndef FRONTEND + MemoryContext oldcontext = MemoryContextSwitchTo(brtab->mcxt); +#endif + + memcpy(&key.rlocator, rlocator, sizeof(RelFileLocator)); + key.forknum = forknum; + brtentry = blockreftable_insert(brtab->hash, key, &found); + + if (!found) + { + /* + * We want to set the initial limit block value to something higher + * than any legal block number. InvalidBlockNumber fits the bill. + */ + brtentry->limit_block = InvalidBlockNumber; + brtentry->nchunks = 0; + brtentry->chunk_size = NULL; + brtentry->chunk_usage = NULL; + brtentry->chunk_data = NULL; + } + + BlockRefTableEntryMarkBlockModified(brtentry, forknum, blknum); + +#ifndef FRONTEND + MemoryContextSwitchTo(oldcontext); +#endif +} + +/* + * Get an entry from a block reference table. + * + * If the entry does not exist, this function returns NULL. Otherwise, it + * returns the entry and sets *limit_block to the value from the entry. + */ +BlockRefTableEntry * +BlockRefTableGetEntry(BlockRefTable *brtab, const RelFileLocator *rlocator, + ForkNumber forknum, BlockNumber *limit_block) +{ + BlockRefTableKey key; + BlockRefTableEntry *entry; + + Assert(limit_block != NULL); + + memcpy(&key.rlocator, rlocator, sizeof(RelFileLocator)); + key.forknum = forknum; + entry = blockreftable_lookup(brtab->hash, key); + + if (entry != NULL) + *limit_block = entry->limit_block; + + return entry; +} + +/* + * Get block numbers from a table entry. + * + * 'blocks' must point to enough space to hold at least 'nblocks' block + * numbers, and any block numbers we manage to get will be written there. + * The return value is the number of block numbers actually written. + * + * We do not return block numbers unless they are greater than or equal to + * start_blkno and strictly less than stop_blkno. + */ +int +BlockRefTableEntryGetBlocks(BlockRefTableEntry *entry, + BlockNumber start_blkno, + BlockNumber stop_blkno, + BlockNumber *blocks, + int nblocks) +{ + uint32 start_chunkno; + uint32 stop_chunkno; + uint32 chunkno; + int nresults = 0; + + Assert(entry != NULL); + + /* + * Figure out which chunks could potentially contain blocks of interest. + * + * We need to be careful about overflow here, because stop_blkno could be + * InvalidBlockNumber or something very close to it. + */ + start_chunkno = start_blkno / BLOCKS_PER_CHUNK; + stop_chunkno = stop_blkno / BLOCKS_PER_CHUNK; + if ((stop_blkno % BLOCKS_PER_CHUNK) != 0) + ++stop_chunkno; + if (stop_chunkno > entry->nchunks) + stop_chunkno = entry->nchunks; + + /* + * Loop over chunks. + */ + for (chunkno = start_chunkno; chunkno < stop_chunkno; ++chunkno) + { + uint16 chunk_usage = entry->chunk_usage[chunkno]; + BlockRefTableChunk chunk_data = entry->chunk_data[chunkno]; + unsigned start_offset = 0; + unsigned stop_offset = BLOCKS_PER_CHUNK; + + /* + * If the start and/or stop block number falls within this chunk, the + * whole chunk may not be of interest. Figure out which portion we + * care about, if it's not the whole thing. + */ + if (chunkno == start_chunkno) + start_offset = start_blkno % BLOCKS_PER_CHUNK; + if (chunkno == stop_chunkno) + stop_offset = stop_blkno % BLOCKS_PER_CHUNK; + + /* + * Handling differs depending on whether this is an array of offsets + * or a bitmap. + */ + if (chunk_usage == MAX_ENTRIES_PER_CHUNK) + { + unsigned i; + + /* It's a bitmap, so test every relevant bit. */ + for (i = start_offset; i < BLOCKS_PER_CHUNK; ++i) + { + uint16 w = chunk_data[i / BLOCKS_PER_ENTRY]; + + if ((w & (1 << (i % BLOCKS_PER_ENTRY))) != 0) + { + BlockNumber blkno = chunkno * BLOCKS_PER_CHUNK + i; + + blocks[nresults++] = blkno; + + /* Early exit if we run out of output space. */ + if (nresults == nblocks) + return nresults; + } + } + } + else + { + unsigned i; + + /* It's an array of offsets, so check each one. */ + for (i = 0; i < chunk_usage; ++i) + { + uint16 offset = chunk_data[i]; + + if (offset >= start_offset && offset < stop_offset) + { + BlockNumber blkno = chunkno * BLOCKS_PER_CHUNK + offset; + + blocks[nresults++] = blkno; + + /* Early exit if we run out of output space. */ + if (nresults == nblocks) + return nresults; + } + } + } + } + + return nresults; +} + +/* + * Serialize a block reference table to a file. + */ +void +WriteBlockRefTable(BlockRefTable *brtab, + io_callback_fn write_callback, + void *write_callback_arg) +{ + BlockRefTableSerializedEntry *sdata = NULL; + BlockRefTableBuffer buffer; + uint32 magic = BLOCKREFTABLE_MAGIC; + + /* Prepare buffer. */ + memset(&buffer, 0, sizeof(BlockRefTableBuffer)); + buffer.io_callback = write_callback; + buffer.io_callback_arg = write_callback_arg; + INIT_CRC32C(buffer.crc); + + /* Write magic number. */ + BlockRefTableWrite(&buffer, &magic, sizeof(uint32)); + + /* Write the entries, assuming there are some. */ + if (brtab->hash->members > 0) + { + unsigned i = 0; + blockreftable_iterator it; + BlockRefTableEntry *brtentry; + + /* Extract entries into serializable format and sort them. */ + sdata = + palloc(brtab->hash->members * sizeof(BlockRefTableSerializedEntry)); + blockreftable_start_iterate(brtab->hash, &it); + while ((brtentry = blockreftable_iterate(brtab->hash, &it)) != NULL) + { + BlockRefTableSerializedEntry *sentry = &sdata[i++]; + + sentry->rlocator = brtentry->key.rlocator; + sentry->forknum = brtentry->key.forknum; + sentry->limit_block = brtentry->limit_block; + sentry->nchunks = brtentry->nchunks; + + /* trim trailing zero entries */ + while (sentry->nchunks > 0 && + brtentry->chunk_usage[sentry->nchunks - 1] == 0) + sentry->nchunks--; + } + Assert(i == brtab->hash->members); + qsort(sdata, i, sizeof(BlockRefTableSerializedEntry), + BlockRefTableComparator); + + /* Loop over entries in sorted order and serialize each one. */ + for (i = 0; i < brtab->hash->members; ++i) + { + BlockRefTableSerializedEntry *sentry = &sdata[i]; + BlockRefTableEntry *brtentry; + BlockRefTableKey key; + unsigned j; + + /* Write the serialized entry itself. */ + BlockRefTableWrite(&buffer, sentry, + sizeof(BlockRefTableSerializedEntry)); + + /* Look up the original entry so we can access the chunks. */ + memcpy(&key.rlocator, &sentry->rlocator, sizeof(RelFileLocator)); + key.forknum = sentry->forknum; + brtentry = blockreftable_lookup(brtab->hash, key); + Assert(brtentry != NULL); + + /* Write the untruncated portion of the chunk length array. */ + if (sentry->nchunks != 0) + BlockRefTableWrite(&buffer, brtentry->chunk_usage, + sentry->nchunks * sizeof(uint16)); + + /* Write the contents of each chunk. */ + for (j = 0; j < brtentry->nchunks; ++j) + { + if (brtentry->chunk_usage[j] == 0) + continue; + BlockRefTableWrite(&buffer, brtentry->chunk_data[j], + brtentry->chunk_usage[j] * sizeof(uint16)); + } + } + } + + /* Write out appropriate terminator and CRC and flush buffer. */ + BlockRefTableFileTerminate(&buffer); +} + +/* + * Prepare to incrementally read a block reference table file. + * + * 'read_callback' is a function that can be called to read data from the + * underlying file (or other data source) into our internal buffer. + * + * 'read_callback_arg' is an opaque argument to be passed to read_callback. + * + * 'error_filename' is the filename that should be included in error messages + * if the file is found to be malformed. The value is not copied, so the + * caller should ensure that it remains valid until done with this + * BlockRefTableReader. + * + * 'error_callback' is a function to be called if the file is found to be + * malformed. This is not used for I/O errors, which must be handled internally + * by read_callback. + * + * 'error_callback_arg' is an opaque arguent to be passed to error_callback. + */ +BlockRefTableReader * +CreateBlockRefTableReader(io_callback_fn read_callback, + void *read_callback_arg, + char *error_filename, + report_error_fn error_callback, + void *error_callback_arg) +{ + BlockRefTableReader *reader; + uint32 magic; + + /* Initialize data structure. */ + reader = palloc0(sizeof(BlockRefTableReader)); + reader->buffer.io_callback = read_callback; + reader->buffer.io_callback_arg = read_callback_arg; + reader->error_filename = error_filename; + reader->error_callback = error_callback; + reader->error_callback_arg = error_callback_arg; + INIT_CRC32C(reader->buffer.crc); + + /* Verify magic number. */ + BlockRefTableRead(reader, &magic, sizeof(uint32)); + if (magic != BLOCKREFTABLE_MAGIC) + error_callback(error_callback_arg, + "file \"%s\" has wrong magic number: expected %u, found %u", + error_filename, + BLOCKREFTABLE_MAGIC, magic); + + return reader; +} + +/* + * Read next relation fork covered by this block reference table file. + * + * After calling this function, you must call BlockRefTableReaderGetBlocks + * until it returns 0 before calling it again. + */ +bool +BlockRefTableReaderNextRelation(BlockRefTableReader *reader, + RelFileLocator *rlocator, + ForkNumber *forknum, + BlockNumber *limit_block) +{ + BlockRefTableSerializedEntry sentry; + BlockRefTableSerializedEntry zentry = {0}; + + /* + * Sanity check: caller must read all blocks from all chunks before moving + * on to the next relation. + */ + Assert(reader->total_chunks == reader->consumed_chunks); + + /* Read serialized entry. */ + BlockRefTableRead(reader, &sentry, + sizeof(BlockRefTableSerializedEntry)); + + /* + * If we just read the sentinel entry indicating that we've reached the + * end, read and check the CRC. + */ + if (memcmp(&sentry, &zentry, sizeof(BlockRefTableSerializedEntry)) == 0) + { + pg_crc32c expected_crc; + pg_crc32c actual_crc; + + /* + * We want to know the CRC of the file excluding the 4-byte CRC + * itself, so copy the current value of the CRC accumulator before + * reading those bytes, and use the copy to finalize the calculation. + */ + expected_crc = reader->buffer.crc; + FIN_CRC32C(expected_crc); + + /* Now we can read the actual value. */ + BlockRefTableRead(reader, &actual_crc, sizeof(pg_crc32c)); + + /* Throw an error if there is a mismatch. */ + if (!EQ_CRC32C(expected_crc, actual_crc)) + reader->error_callback(reader->error_callback_arg, + "file \"%s\" has wrong checksum: expected %08X, found %08X", + reader->error_filename, expected_crc, actual_crc); + + return false; + } + + /* Read chunk size array. */ + if (reader->chunk_size != NULL) + pfree(reader->chunk_size); + reader->chunk_size = palloc(sentry.nchunks * sizeof(uint16)); + BlockRefTableRead(reader, reader->chunk_size, + sentry.nchunks * sizeof(uint16)); + + /* Set up for chunk scan. */ + reader->total_chunks = sentry.nchunks; + reader->consumed_chunks = 0; + + /* Return data to caller. */ + memcpy(rlocator, &sentry.rlocator, sizeof(RelFileLocator)); + *forknum = sentry.forknum; + *limit_block = sentry.limit_block; + return true; +} + +/* + * Get modified blocks associated with the relation fork returned by + * the most recent call to BlockRefTableReaderNextRelation. + * + * On return, block numbers will be written into the 'blocks' array, whose + * length should be passed via 'nblocks'. The return value is the number of + * entries actually written into the 'blocks' array, which may be less than + * 'nblocks' if we run out of modified blocks in the relation fork before + * we run out of room in the array. + */ +unsigned +BlockRefTableReaderGetBlocks(BlockRefTableReader *reader, + BlockNumber *blocks, + int nblocks) +{ + unsigned blocks_found = 0; + + /* Must provide space for at least one block number to be returned. */ + Assert(nblocks > 0); + + /* Loop collecting blocks to return to caller. */ + for (;;) + { + uint16 next_chunk_size; + + /* + * If we've read at least one chunk, maybe it contains some block + * numbers that could satisfy caller's request. + */ + if (reader->consumed_chunks > 0) + { + uint32 chunkno = reader->consumed_chunks - 1; + uint16 chunk_size = reader->chunk_size[chunkno]; + + if (chunk_size == MAX_ENTRIES_PER_CHUNK) + { + /* Bitmap format, so search for bits that are set. */ + while (reader->chunk_position < BLOCKS_PER_CHUNK && + blocks_found < nblocks) + { + uint16 chunkoffset = reader->chunk_position; + uint16 w; + + w = reader->chunk_data[chunkoffset / BLOCKS_PER_ENTRY]; + if ((w & (1u << (chunkoffset % BLOCKS_PER_ENTRY))) != 0) + blocks[blocks_found++] = + chunkno * BLOCKS_PER_CHUNK + chunkoffset; + ++reader->chunk_position; + } + } + else + { + /* Not in bitmap format, so each entry is a 2-byte offset. */ + while (reader->chunk_position < chunk_size && + blocks_found < nblocks) + { + blocks[blocks_found++] = chunkno * BLOCKS_PER_CHUNK + + reader->chunk_data[reader->chunk_position]; + ++reader->chunk_position; + } + } + } + + /* We found enough blocks, so we're done. */ + if (blocks_found >= nblocks) + break; + + /* + * We didn't find enough blocks, so we must need the next chunk. If + * there are none left, though, then we're done anyway. + */ + if (reader->consumed_chunks == reader->total_chunks) + break; + + /* + * Read data for next chunk and reset scan position to beginning of + * chunk. Note that the next chunk might be empty, in which case we + * consume the chunk without actually consuming any bytes from the + * underlying file. + */ + next_chunk_size = reader->chunk_size[reader->consumed_chunks]; + if (next_chunk_size > 0) + BlockRefTableRead(reader, reader->chunk_data, + next_chunk_size * sizeof(uint16)); + ++reader->consumed_chunks; + reader->chunk_position = 0; + } + + return blocks_found; +} + +/* + * Release memory used while reading a block reference table from a file. + */ +void +DestroyBlockRefTableReader(BlockRefTableReader *reader) +{ + if (reader->chunk_size != NULL) + { + pfree(reader->chunk_size); + reader->chunk_size = NULL; + } + pfree(reader); +} + +/* + * Prepare to write a block reference table file incrementally. + * + * Caller must be able to supply BlockRefTableEntry objects sorted in the + * appropriate order. + */ +BlockRefTableWriter * +CreateBlockRefTableWriter(io_callback_fn write_callback, + void *write_callback_arg) +{ + BlockRefTableWriter *writer; + uint32 magic = BLOCKREFTABLE_MAGIC; + + /* Prepare buffer and CRC check and save callbacks. */ + writer = palloc0(sizeof(BlockRefTableWriter)); + writer->buffer.io_callback = write_callback; + writer->buffer.io_callback_arg = write_callback_arg; + INIT_CRC32C(writer->buffer.crc); + + /* Write magic number. */ + BlockRefTableWrite(&writer->buffer, &magic, sizeof(uint32)); + + return writer; +} + +/* + * Append one entry to a block reference table file. + * + * Note that entries must be written in the proper order, that is, sorted by + * tablespace, then database, then relfilenumber, then fork number. Caller + * is responsible for supplying data in the correct order. If that seems hard, + * use an in-memory BlockRefTable instead. + */ +void +BlockRefTableWriteEntry(BlockRefTableWriter *writer, BlockRefTableEntry *entry) +{ + BlockRefTableSerializedEntry sentry; + unsigned j; + + /* Convert to serialized entry format. */ + sentry.rlocator = entry->key.rlocator; + sentry.forknum = entry->key.forknum; + sentry.limit_block = entry->limit_block; + sentry.nchunks = entry->nchunks; + + /* Trim trailing zero entries. */ + while (sentry.nchunks > 0 && entry->chunk_usage[sentry.nchunks - 1] == 0) + sentry.nchunks--; + + /* Write the serialized entry itself. */ + BlockRefTableWrite(&writer->buffer, &sentry, + sizeof(BlockRefTableSerializedEntry)); + + /* Write the untruncated portion of the chunk length array. */ + if (sentry.nchunks != 0) + BlockRefTableWrite(&writer->buffer, entry->chunk_usage, + sentry.nchunks * sizeof(uint16)); + + /* Write the contents of each chunk. */ + for (j = 0; j < entry->nchunks; ++j) + { + if (entry->chunk_usage[j] == 0) + continue; + BlockRefTableWrite(&writer->buffer, entry->chunk_data[j], + entry->chunk_usage[j] * sizeof(uint16)); + } +} + +/* + * Finalize an incremental write of a block reference table file. + */ +void +DestroyBlockRefTableWriter(BlockRefTableWriter *writer) +{ + BlockRefTableFileTerminate(&writer->buffer); + pfree(writer); +} + +/* + * Allocate a standalone BlockRefTableEntry. + * + * When we're manipulating a full in-memory BlockRefTable, the entries are + * part of the hash table and are allocated by simplehash. This routine is + * used by callers that want to write out a BlockRefTable to a file without + * needing to store the whole thing in memory at once. + * + * Entries allocated by this function can be manipulated using the functions + * BlockRefTableEntrySetLimitBlock and BlockRefTableEntryMarkBlockModified + * and then written using BlockRefTableWriteEntry and freed using + * BlockRefTableFreeEntry. + */ +BlockRefTableEntry * +CreateBlockRefTableEntry(RelFileLocator rlocator, ForkNumber forknum) +{ + BlockRefTableEntry *entry = palloc0(sizeof(BlockRefTableEntry)); + + memcpy(&entry->key.rlocator, &rlocator, sizeof(RelFileLocator)); + entry->key.forknum = forknum; + entry->limit_block = InvalidBlockNumber; + + return entry; +} + +/* + * Update a BlockRefTableEntry with a new value for the "limit block" and + * forget any equal-or-higher-numbered modified blocks. + * + * The "limit block" is the shortest known length of the relation within the + * range of WAL records covered by this block reference table. + */ +void +BlockRefTableEntrySetLimitBlock(BlockRefTableEntry *entry, + BlockNumber limit_block) +{ + unsigned chunkno; + unsigned limit_chunkno; + unsigned limit_chunkoffset; + BlockRefTableChunk limit_chunk; + + /* If we already have an equal or lower limit block, do nothing. */ + if (limit_block >= entry->limit_block) + return; + + /* Record the new limit block value. */ + entry->limit_block = limit_block; + + /* + * Figure out which chunk would store the state of the new limit block, + * and which offset within that chunk. + */ + limit_chunkno = limit_block / BLOCKS_PER_CHUNK; + limit_chunkoffset = limit_block % BLOCKS_PER_CHUNK; + + /* + * If the number of chunks is not large enough for any blocks with equal + * or higher block numbers to exist, then there is nothing further to do. + */ + if (limit_chunkno >= entry->nchunks) + return; + + /* Discard entire contents of any higher-numbered chunks. */ + for (chunkno = limit_chunkno + 1; chunkno < entry->nchunks; ++chunkno) + entry->chunk_usage[chunkno] = 0; + + /* + * Next, we need to discard any offsets within the chunk that would + * contain the limit_block. We must handle this differenly depending on + * whether the chunk that would contain limit_block is a bitmap or an + * array of offsets. + */ + limit_chunk = entry->chunk_data[limit_chunkno]; + if (entry->chunk_usage[limit_chunkno] == MAX_ENTRIES_PER_CHUNK) + { + unsigned chunkoffset; + + /* It's a bitmap. Unset bits. */ + for (chunkoffset = limit_chunkoffset; chunkoffset < BLOCKS_PER_CHUNK; + ++chunkoffset) + limit_chunk[chunkoffset / BLOCKS_PER_ENTRY] &= + ~(1 << (chunkoffset % BLOCKS_PER_ENTRY)); + } + else + { + unsigned i, + j = 0; + + /* It's an offset array. Filter out large offsets. */ + for (i = 0; i < entry->chunk_usage[limit_chunkno]; ++i) + { + Assert(j <= i); + if (limit_chunk[i] < limit_chunkoffset) + limit_chunk[j++] = limit_chunk[i]; + } + Assert(j <= entry->chunk_usage[limit_chunkno]); + entry->chunk_usage[limit_chunkno] = j; + } +} + +/* + * Mark a block in a given BlkRefTableEntry as known to have been modified. + */ +void +BlockRefTableEntryMarkBlockModified(BlockRefTableEntry *entry, + ForkNumber forknum, + BlockNumber blknum) +{ + unsigned chunkno; + unsigned chunkoffset; + unsigned i; + + /* + * Which chunk should store the state of this block? And what is the + * offset of this block relative to the start of that chunk? + */ + chunkno = blknum / BLOCKS_PER_CHUNK; + chunkoffset = blknum % BLOCKS_PER_CHUNK; + + /* + * If 'nchunks' isn't big enough for us to be able to represent the state + * of this block, we need to enlarge our arrays. + */ + if (chunkno >= entry->nchunks) + { + unsigned max_chunks; + unsigned extra_chunks; + + /* + * New array size is a power of 2, at least 16, big enough so that + * chunkno will be a valid array index. + */ + max_chunks = Max(16, entry->nchunks); + while (max_chunks < chunkno + 1) + chunkno *= 2; + Assert(max_chunks > chunkno); + extra_chunks = max_chunks - entry->nchunks; + + if (entry->nchunks == 0) + { + entry->chunk_size = palloc0(sizeof(uint16) * max_chunks); + entry->chunk_usage = palloc0(sizeof(uint16) * max_chunks); + entry->chunk_data = + palloc0(sizeof(BlockRefTableChunk) * max_chunks); + } + else + { + entry->chunk_size = repalloc(entry->chunk_size, + sizeof(uint16) * max_chunks); + memset(&entry->chunk_size[entry->nchunks], 0, + extra_chunks * sizeof(uint16)); + entry->chunk_usage = repalloc(entry->chunk_usage, + sizeof(uint16) * max_chunks); + memset(&entry->chunk_usage[entry->nchunks], 0, + extra_chunks * sizeof(uint16)); + entry->chunk_data = repalloc(entry->chunk_data, + sizeof(BlockRefTableChunk) * max_chunks); + memset(&entry->chunk_data[entry->nchunks], 0, + extra_chunks * sizeof(BlockRefTableChunk)); + } + entry->nchunks = max_chunks; + } + + /* + * If the chunk that covers this block number doesn't exist yet, create it + * as an array and add the appropriate offset to it. We make it pretty + * small initially, because there might only be 1 or a few block + * references in this chunk and we don't want to use up too much memory. + */ + if (entry->chunk_size[chunkno] == 0) + { + entry->chunk_data[chunkno] = + palloc(sizeof(uint16) * INITIAL_ENTRIES_PER_CHUNK); + entry->chunk_size[chunkno] = INITIAL_ENTRIES_PER_CHUNK; + entry->chunk_data[chunkno][0] = chunkoffset; + entry->chunk_usage[chunkno] = 1; + return; + } + + /* + * If the number of entries in this chunk is already maximum, it must be a + * bitmap. Just set the appropriate bit. + */ + if (entry->chunk_usage[chunkno] == MAX_ENTRIES_PER_CHUNK) + { + BlockRefTableChunk chunk = entry->chunk_data[chunkno]; + + chunk[chunkoffset / BLOCKS_PER_ENTRY] |= + 1 << (chunkoffset % BLOCKS_PER_ENTRY); + return; + } + + /* + * There is an existing chunk and it's in array format. Let's find out + * whether it already has an entry for this block. If so, we do not need + * to do anything. + */ + for (i = 0; i < entry->chunk_usage[chunkno]; ++i) + { + if (entry->chunk_data[chunkno][i] == chunkoffset) + return; + } + + /* + * If the number of entries currently used is one less than the maximum, + * it's time to convert to bitmap format. + */ + if (entry->chunk_usage[chunkno] == MAX_ENTRIES_PER_CHUNK - 1) + { + BlockRefTableChunk newchunk; + unsigned j; + + /* Allocate a new chunk. */ + newchunk = palloc0(MAX_ENTRIES_PER_CHUNK * sizeof(uint16)); + + /* Set the bit for each existing entry. */ + for (j = 0; j < entry->chunk_usage[chunkno]; ++j) + { + unsigned coff = entry->chunk_data[chunkno][j]; + + newchunk[coff / BLOCKS_PER_ENTRY] |= + 1 << (coff % BLOCKS_PER_ENTRY); + } + + /* Set the bit for the new entry. */ + newchunk[chunkoffset / BLOCKS_PER_ENTRY] |= + 1 << (chunkoffset % BLOCKS_PER_ENTRY); + + /* Swap the new chunk into place and update metadata. */ + pfree(entry->chunk_data[chunkno]); + entry->chunk_data[chunkno] = newchunk; + entry->chunk_size[chunkno] = MAX_ENTRIES_PER_CHUNK; + entry->chunk_usage[chunkno] = MAX_ENTRIES_PER_CHUNK; + return; + } + + /* + * OK, we currently have an array, and we don't need to convert to a + * bitmap, but we do need to add a new element. If there's not enough + * room, we'll have to expand the array. + */ + if (entry->chunk_usage[chunkno] == entry->chunk_size[chunkno]) + { + unsigned newsize = entry->chunk_size[chunkno] * 2; + + Assert(newsize <= MAX_ENTRIES_PER_CHUNK); + entry->chunk_data[chunkno] = repalloc(entry->chunk_data[chunkno], + newsize * sizeof(uint16)); + entry->chunk_size[chunkno] = newsize; + } + + /* Now we can add the new entry. */ + entry->chunk_data[chunkno][entry->chunk_usage[chunkno]] = + chunkoffset; + entry->chunk_usage[chunkno]++; +} + +/* + * Release memory for a BlockRefTablEntry that was created by + * CreateBlockRefTableEntry. + */ +void +BlockRefTableFreeEntry(BlockRefTableEntry *entry) +{ + if (entry->chunk_size != NULL) + { + pfree(entry->chunk_size); + entry->chunk_size = NULL; + } + + if (entry->chunk_usage != NULL) + { + pfree(entry->chunk_usage); + entry->chunk_usage = NULL; + } + + if (entry->chunk_data != NULL) + { + pfree(entry->chunk_data); + entry->chunk_data = NULL; + } + + pfree(entry); +} + +/* + * Comparator for BlockRefTableSerializedEntry objects. + * + * We make the tablespace OID the first column of the sort key to match + * the on-disk tree structure. + */ +static int +BlockRefTableComparator(const void *a, const void *b) +{ + const BlockRefTableSerializedEntry *sa = a; + const BlockRefTableSerializedEntry *sb = b; + + if (sa->rlocator.spcOid > sb->rlocator.spcOid) + return 1; + if (sa->rlocator.spcOid < sb->rlocator.spcOid) + return -1; + + if (sa->rlocator.dbOid > sb->rlocator.dbOid) + return 1; + if (sa->rlocator.dbOid < sb->rlocator.dbOid) + return -1; + + if (sa->rlocator.relNumber > sb->rlocator.relNumber) + return 1; + if (sa->rlocator.relNumber < sb->rlocator.relNumber) + return -1; + + if (sa->forknum > sb->forknum) + return 1; + if (sa->forknum < sb->forknum) + return -1; + + return 0; +} + +/* + * Flush any buffered data out of a BlockRefTableBuffer. + */ +static void +BlockRefTableFlush(BlockRefTableBuffer *buffer) +{ + buffer->io_callback(buffer->io_callback_arg, buffer->data, buffer->used); + buffer->used = 0; +} + +/* + * Read data from a BlockRefTableBuffer, and update the running CRC + * calculation for the returned data (but not any data that we may have + * buffered but not yet actually returned). + */ +static void +BlockRefTableRead(BlockRefTableReader *reader, void *data, int length) +{ + BlockRefTableBuffer *buffer = &reader->buffer; + + /* Loop until read is fully satisfied. */ + while (length > 0) + { + if (buffer->cursor < buffer->used) + { + /* + * If any buffered data is available, use that to satisfy as much + * of the request as possible. + */ + int bytes_to_copy = Min(length, buffer->used - buffer->cursor); + + memcpy(data, &buffer->data[buffer->cursor], bytes_to_copy); + COMP_CRC32C(buffer->crc, &buffer->data[buffer->cursor], + bytes_to_copy); + buffer->cursor += bytes_to_copy; + data = ((char *) data) + bytes_to_copy; + length -= bytes_to_copy; + } + else if (length >= BUFSIZE) + { + /* + * If the request length is long, read directly into caller's + * buffer. + */ + int bytes_read; + + bytes_read = buffer->io_callback(buffer->io_callback_arg, + data, length); + COMP_CRC32C(buffer->crc, data, bytes_read); + data = ((char *) data) + bytes_read; + length -= bytes_read; + + /* If we didn't get anything, that's bad. */ + if (bytes_read == 0) + reader->error_callback(reader->error_callback_arg, + "file \"%s\" ends unexpectedly", + reader->error_filename); + } + else + { + /* + * Refill our buffer. + */ + buffer->used = buffer->io_callback(buffer->io_callback_arg, + buffer->data, BUFSIZE); + buffer->cursor = 0; + + /* If we didn't get anything, that's bad. */ + if (buffer->used == 0) + reader->error_callback(reader->error_callback_arg, + "file \"%s\" ends unexpectedly", + reader->error_filename); + } + } +} + +/* + * Supply data to a BlockRefTableBuffer for write to the underlying File, + * and update the running CRC calculation for that data. + */ +static void +BlockRefTableWrite(BlockRefTableBuffer *buffer, void *data, int length) +{ + /* Update running CRC calculation. */ + COMP_CRC32C(buffer->crc, data, length); + + /* If the new data can't fit into the buffer, flush the buffer. */ + if (buffer->used + length > BUFSIZE) + { + buffer->io_callback(buffer->io_callback_arg, buffer->data, + buffer->used); + buffer->used = 0; + } + + /* If the new data would fill the buffer, or more, write it directly. */ + if (length >= BUFSIZE) + { + buffer->io_callback(buffer->io_callback_arg, data, length); + return; + } + + /* Otherwise, copy the new data into the buffer. */ + memcpy(&buffer->data[buffer->used], data, length); + buffer->used += length; + Assert(buffer->used <= BUFSIZE); +} + +/* + * Generate the sentinel and CRC required at the end of a block reference + * table file and flush them out of our internal buffer. + */ +static void +BlockRefTableFileTerminate(BlockRefTableBuffer *buffer) +{ + BlockRefTableSerializedEntry zentry = {0}; + pg_crc32c crc; + + /* Write a sentinel indicating that there are no more entries. */ + BlockRefTableWrite(buffer, &zentry, + sizeof(BlockRefTableSerializedEntry)); + + /* + * Writing the checksum will perturb the ongoing checksum calculation, so + * copy the state first and finalize the computation using the copy. + */ + crc = buffer->crc; + FIN_CRC32C(crc); + BlockRefTableWrite(buffer, &crc, sizeof(pg_crc32c)); + + /* Flush any leftover data out of our buffer. */ + BlockRefTableFlush(buffer); +} diff --git a/src/common/meson.build b/src/common/meson.build index a9ff7f9db8..dda018f6d1 100644 --- a/src/common/meson.build +++ b/src/common/meson.build @@ -3,6 +3,7 @@ common_sources = files( 'archive.c', 'base64.c', + 'blkreftable.c', 'checksum_helper.c', 'compression.c', 'controldata_utils.c', diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 48ca852381..fed5d790cc 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -206,6 +206,7 @@ extern int XLogFileOpen(XLogSegNo segno, TimeLineID tli); extern void CheckXLogRemoved(XLogSegNo segno, TimeLineID tli); extern XLogSegNo XLogGetLastRemovedSegno(void); +extern XLogSegNo XLogGetOldestSegno(TimeLineID tli); extern void XLogSetAsyncXactLSN(XLogRecPtr asyncXactLSN); extern void XLogSetReplicationSlotMinimumLSN(XLogRecPtr lsn); diff --git a/src/include/access/xlogbackup.h b/src/include/access/xlogbackup.h index 1611358137..90e04cad56 100644 --- a/src/include/access/xlogbackup.h +++ b/src/include/access/xlogbackup.h @@ -28,6 +28,8 @@ typedef struct BackupState XLogRecPtr checkpointloc; /* last checkpoint location */ pg_time_t starttime; /* backup start time */ bool started_in_recovery; /* backup started in recovery? */ + XLogRecPtr istartpoint; /* incremental based on backup at this LSN */ + TimeLineID istarttli; /* incremental based on backup on this TLI */ /* Fields saved at the end of backup */ XLogRecPtr stoppoint; /* backup stop WAL location */ diff --git a/src/include/backup/basebackup.h b/src/include/backup/basebackup.h index 1432d9c206..345bd22534 100644 --- a/src/include/backup/basebackup.h +++ b/src/include/backup/basebackup.h @@ -34,6 +34,9 @@ typedef struct int64 size; /* total size as sent; -1 if not known */ } tablespaceinfo; -extern void SendBaseBackup(BaseBackupCmd *cmd); +struct IncrementalBackupInfo; + +extern void SendBaseBackup(BaseBackupCmd *cmd, + struct IncrementalBackupInfo *ib); #endif /* _BASEBACKUP_H */ diff --git a/src/include/backup/basebackup_incremental.h b/src/include/backup/basebackup_incremental.h new file mode 100644 index 0000000000..c300235a2f --- /dev/null +++ b/src/include/backup/basebackup_incremental.h @@ -0,0 +1,56 @@ +/*------------------------------------------------------------------------- + * + * basebackup_incremental.h + * API for incremental backup support + * + * Portions Copyright (c) 2010-2022, PostgreSQL Global Development Group + * + * src/include/backup/basebackup_incremental.h + * + *------------------------------------------------------------------------- + */ +#ifndef BASEBACKUP_INCREMENTAL_H +#define BASEBACKUP_INCREMENTAL_H + +#include "access/xlogbackup.h" +#include "common/relpath.h" +#include "storage/block.h" +#include "utils/palloc.h" + +#define INCREMENTAL_MAGIC 0xd3ae1f0d + +typedef enum +{ + BACK_UP_FILE_FULLY, + BACK_UP_FILE_INCREMENTALLY, + DO_NOT_BACK_UP_FILE +} FileBackupMethod; + +struct IncrementalBackupInfo; +typedef struct IncrementalBackupInfo IncrementalBackupInfo; + +extern IncrementalBackupInfo *CreateIncrementalBackupInfo(MemoryContext); + +extern void AppendIncrementalManifestData(IncrementalBackupInfo *ib, + const char *data, + int len); +extern void FinalizeIncrementalManifest(IncrementalBackupInfo *ib); + +extern void PrepareForIncrementalBackup(IncrementalBackupInfo *ib, + BackupState *backup_state); + +extern char *GetIncrementalFilePath(Oid dboid, Oid spcoid, + RelFileNumber relfilenumber, + ForkNumber forknum, unsigned segno); +extern FileBackupMethod GetFileBackupMethod(IncrementalBackupInfo *ib, + char *path, + Oid dboid, Oid spcoid, + RelFileNumber relfilenumber, + ForkNumber forknum, + unsigned segno, size_t size, + unsigned *num_blocks_required, + BlockNumber *relative_block_numbers, + unsigned *truncation_block_length); +extern size_t GetIncrementalFileSize(unsigned num_blocks_required); + +#endif diff --git a/src/include/backup/walsummary.h b/src/include/backup/walsummary.h new file mode 100644 index 0000000000..d086e64019 --- /dev/null +++ b/src/include/backup/walsummary.h @@ -0,0 +1,49 @@ +/*------------------------------------------------------------------------- + * + * walsummary.h + * WAL summary management + * + * Portions Copyright (c) 2010-2022, PostgreSQL Global Development Group + * + * src/include/backup/walsummary.h + * + *------------------------------------------------------------------------- + */ +#ifndef WALSUMMARY_H +#define WALSUMMARY_H + +#include <time.h> + +#include "access/xlogdefs.h" +#include "nodes/pg_list.h" +#include "storage/fd.h" + +typedef struct WalSummaryIO +{ + File file; + off_t filepos; +} WalSummaryIO; + +typedef struct WalSummaryFile +{ + XLogRecPtr start_lsn; + XLogRecPtr end_lsn; + TimeLineID tli; +} WalSummaryFile; + +extern List *GetWalSummaries(TimeLineID tli, XLogRecPtr start_lsn, + XLogRecPtr end_lsn); +extern List *FilterWalSummaries(List *wslist, TimeLineID tli, + XLogRecPtr start_lsn, XLogRecPtr end_lsn); +extern bool WalSummariesAreComplete(List *wslist, + XLogRecPtr start_lsn, XLogRecPtr end_lsn, + XLogRecPtr *missing_lsn); +extern File OpenWalSummaryFile(WalSummaryFile *ws, bool missing_ok); +extern void RemoveWalSummaryIfOlderThan(WalSummaryFile *ws, + time_t cutoff_time); + +extern int ReadWalSummary(void *wal_summary_io, void *data, int length); +extern int WriteWalSummary(void *wal_summary_io, void *data, int length); +extern void ReportWalSummaryError(void *callback_arg, char *fmt,...); + +#endif /* WALSUMMARY_H */ diff --git a/src/include/catalog/pg_proc.dat b/src/include/catalog/pg_proc.dat index 9805bc6118..148f35d9b1 100644 --- a/src/include/catalog/pg_proc.dat +++ b/src/include/catalog/pg_proc.dat @@ -12062,4 +12062,23 @@ proname => 'any_value_transfn', prorettype => 'anyelement', proargtypes => 'anyelement anyelement', prosrc => 'any_value_transfn' }, +{ oid => '8436', + descr => 'list of available WAL summary files', + proname => 'pg_available_wal_summaries', prorows => '100', + proretset => 't', provolatile => 'v', proparallel => 's', + prorettype => 'record', proargtypes => '', + proallargtypes => '{int8,pg_lsn,pg_lsn}', + proargmodes => '{o,o,o}', + proargnames => '{tli,start_lsn,end_lsn}', + prosrc => 'pg_available_wal_summaries' }, +{ oid => '8437', + descr => 'contents of a WAL sumamry file', + proname => 'pg_wal_summary_contents', prorows => '100', + proretset => 't', provolatile => 'v', proparallel => 's', + prorettype => 'record', proargtypes => 'int8 pg_lsn pg_lsn', + proallargtypes => '{int8,pg_lsn,pg_lsn,oid,oid,oid,int2,int8,bool}', + proargmodes => '{i,i,i,o,o,o,o,o,o}', + proargnames => '{tli,start_lsn,end_lsn,relfilenode,reltablespace,reldatabase,relforknumber,relblocknumber,is_limit_block}', + prosrc => 'pg_wal_summary_contents' }, + ] diff --git a/src/include/common/blkreftable.h b/src/include/common/blkreftable.h new file mode 100644 index 0000000000..22d9883dc5 --- /dev/null +++ b/src/include/common/blkreftable.h @@ -0,0 +1,120 @@ +/*------------------------------------------------------------------------- + * + * blkreftable.h + * Block reference tables. + * + * A block reference table is used to keep track of which blocks have + * been modified by WAL records within a certain LSN range. + * + * For each relation fork, there is a "limit block number". All existing + * blocks greater than or equal to the limit block number must be + * considered modified; for those less than the limit block number, + * we maintain a bitmap. When a relation fork is created or dropped, + * the limit block number should be set to 0. When it's truncated, + * the limit block number should be set to the length in blocks to + * which it was truncated. + * + * Portions Copyright (c) 2010-2022, PostgreSQL Global Development Group + * + * src/include/common/blkreftable.h + * + *------------------------------------------------------------------------- + */ +#ifndef BLKREFTABLE_H +#define BLKREFTABLE_H + +#include "storage/block.h" +#include "storage/relfilelocator.h" + +/* Magic number for serialization file format. */ +#define BLOCKREFTABLE_MAGIC 0x652b137b + +struct BlockRefTable; +struct BlockRefTableEntry; +struct BlockRefTableReader; +struct BlockRefTableWriter; +typedef struct BlockRefTable BlockRefTable; +typedef struct BlockRefTableEntry BlockRefTableEntry; +typedef struct BlockRefTableReader BlockRefTableReader; +typedef struct BlockRefTableWriter BlockRefTableWriter; + +/* + * The return value of io_callback_fn should be the number of bytes read + * or written. If an error occurs, the functions should report it and + * not return. When used as a write callback, short writes should be retried + * or treated as errors, so that if the callback returns, the return value + * is always the request length. + * + * report_error_fn should not return. + */ +typedef int (*io_callback_fn) (void *callback_arg, void *data, int length); +typedef void (*report_error_fn) (void *calblack_arg, char *msg,...); + + +/* + * Functions for manipulating an entire in-memory block reference table. + */ +extern BlockRefTable *CreateEmptyBlockRefTable(void); +extern void BlockRefTableSetLimitBlock(BlockRefTable *brtab, + const RelFileLocator *rlocator, + ForkNumber forknum, + BlockNumber limit_block); +extern void BlockRefTableMarkBlockModified(BlockRefTable *brtab, + const RelFileLocator *rlocator, + ForkNumber forknum, + BlockNumber blknum); +extern void WriteBlockRefTable(BlockRefTable *brtab, + io_callback_fn write_callback, + void *write_callback_arg); + +extern BlockRefTableEntry *BlockRefTableGetEntry(BlockRefTable *brtab, + const RelFileLocator *rlocator, + ForkNumber forknum, + BlockNumber *limit_block); +extern int BlockRefTableEntryGetBlocks(BlockRefTableEntry *entry, + BlockNumber start_blkno, + BlockNumber stop_blkno, + BlockNumber *blocks, + int nblocks); + +/* + * Functions for reading a block reference table incrementally from disk. + */ +extern BlockRefTableReader *CreateBlockRefTableReader(io_callback_fn read_callback, + void *read_callback_arg, + char *error_filename, + report_error_fn error_callback, + void *error_callback_arg); +extern bool BlockRefTableReaderNextRelation(BlockRefTableReader *reader, + RelFileLocator *rlocator, + ForkNumber *forknum, + BlockNumber *limit_block); +extern unsigned BlockRefTableReaderGetBlocks(BlockRefTableReader *reader, + BlockNumber *blocks, + int nblocks); +extern void DestroyBlockRefTableReader(BlockRefTableReader *reader); + +/* + * Functions for writing a block reference table incrementally to disk. + * + * Note that entries must be written in the proper order, that is, sorted by + * database, then tablespace, then relfilenumber, then fork number. Caller + * is responsible for supplying data in the correct order. If that seems hard, + * use an in-memory BlockRefTable instead. + */ +extern BlockRefTableWriter *CreateBlockRefTableWriter(io_callback_fn write_callback, + void *write_callback_arg); +extern void BlockRefTableWriteEntry(BlockRefTableWriter *writer, + BlockRefTableEntry *entry); +extern void DestroyBlockRefTableWriter(BlockRefTableWriter *writer); + +extern BlockRefTableEntry *CreateBlockRefTableEntry(RelFileLocator rlocator, + ForkNumber forknum); +extern void BlockRefTableEntrySetLimitBlock(BlockRefTableEntry *entry, + BlockNumber limit_block); +extern void BlockRefTableEntryMarkBlockModified(BlockRefTableEntry *entry, + ForkNumber forknum, + BlockNumber blknum); +extern void BlockRefTableFreeEntry(BlockRefTableEntry *entry); + +#endif /* BLKREFTABLE_H */ diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 14bd574fc2..898adccb25 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -338,6 +338,7 @@ typedef enum BackendType B_STARTUP, B_WAL_RECEIVER, B_WAL_SENDER, + B_WAL_SUMMARIZER, B_WAL_WRITER, } BackendType; @@ -443,6 +444,7 @@ typedef enum CheckpointerProcess, WalWriterProcess, WalReceiverProcess, + WalSummarizerProcess, NUM_AUXPROCTYPES /* Must be last! */ } AuxProcType; @@ -455,6 +457,7 @@ extern PGDLLIMPORT AuxProcType MyAuxProcType; #define AmCheckpointerProcess() (MyAuxProcType == CheckpointerProcess) #define AmWalWriterProcess() (MyAuxProcType == WalWriterProcess) #define AmWalReceiverProcess() (MyAuxProcType == WalReceiverProcess) +#define AmWalSummarizerProcess() (MyAuxProcType == WalSummarizerProcess) /***************************************************************************** diff --git a/src/include/nodes/replnodes.h b/src/include/nodes/replnodes.h index 4321ba8f86..856491eecd 100644 --- a/src/include/nodes/replnodes.h +++ b/src/include/nodes/replnodes.h @@ -108,4 +108,13 @@ typedef struct TimeLineHistoryCmd TimeLineID timeline; } TimeLineHistoryCmd; +/* ---------------------- + * UPLOAD_MANIFEST command + * ---------------------- + */ +typedef struct UploadManifestCmd +{ + NodeTag type; +} UploadManifestCmd; + #endif /* REPLNODES_H */ diff --git a/src/include/postmaster/walsummarizer.h b/src/include/postmaster/walsummarizer.h new file mode 100644 index 0000000000..7584cb69a7 --- /dev/null +++ b/src/include/postmaster/walsummarizer.h @@ -0,0 +1,31 @@ +/*------------------------------------------------------------------------- + * + * walsummarizer.h + * + * Header file for background WAL summarization process. + * + * Portions Copyright (c) 1996-2022, PostgreSQL Global Development Group + * + * IDENTIFICATION + * src/include/postmaster/walsummarizer.h + * + *------------------------------------------------------------------------- + */ +#ifndef WALSUMMARIZER_H +#define WALSUMMARIZER_H + +#include "access/xlogdefs.h" + +extern int wal_summarize_mb; +extern int wal_summarize_keep_time; + +extern Size WalSummarizerShmemSize(void); +extern void WalSummarizerShmemInit(void); +extern void WalSummarizerMain(void) pg_attribute_noreturn(); + +extern XLogRecPtr GetOldestUnsummarizedLSN(TimeLineID *tli, + bool *lsn_is_exact); +extern void SetWalSummarizerLatch(void); +extern XLogRecPtr WaitForWalSummarization(XLogRecPtr lsn, long timeout); + +#endif diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index ef74f32693..ee55008082 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -417,11 +417,12 @@ extern PGDLLIMPORT PGPROC *PreparedXactProcs; * We set aside some extra PGPROC structures for auxiliary processes, * ie things that aren't full-fledged backends but need shmem access. * - * Background writer, checkpointer, WAL writer and archiver run during normal - * operation. Startup process and WAL receiver also consume 2 slots, but WAL - * writer is launched only after startup has exited, so we only need 5 slots. + * Background writer, checkpointer, WAL writer, WAL summarizer, and archiver + * run during normal operation. Startup process and WAL receiver also consume + * 2 slots, but WAL writer is launched only after startup has exited, so we + * only need 6 slots. */ -#define NUM_AUXILIARY_PROCS 5 +#define NUM_AUXILIARY_PROCS 6 /* configurable options */ extern PGDLLIMPORT int DeadlockTimeout; diff --git a/src/include/utils/guc_tables.h b/src/include/utils/guc_tables.h index d5a0880678..7d3bc0f671 100644 --- a/src/include/utils/guc_tables.h +++ b/src/include/utils/guc_tables.h @@ -72,6 +72,7 @@ enum config_group WAL_RECOVERY, WAL_ARCHIVE_RECOVERY, WAL_RECOVERY_TARGET, + WAL_SUMMARIZATION, REPLICATION_SENDING, REPLICATION_PRIMARY, REPLICATION_STANDBY, diff --git a/src/test/recovery/t/001_stream_rep.pl b/src/test/recovery/t/001_stream_rep.pl index 0c72ba0944..353db33a9f 100644 --- a/src/test/recovery/t/001_stream_rep.pl +++ b/src/test/recovery/t/001_stream_rep.pl @@ -15,6 +15,8 @@ my $node_primary = PostgreSQL::Test::Cluster->new('primary'); $node_primary->init( allows_streaming => 1, auth_extra => [ '--create-role', 'repl_role' ]); +# WAL summarization can postpone WAL recycling, leading to test failures +$node_primary->append_conf('postgresql.conf', "wal_summarize_mb = 0"); $node_primary->start; my $backup_name = 'my_backup'; diff --git a/src/test/recovery/t/019_replslot_limit.pl b/src/test/recovery/t/019_replslot_limit.pl index 33e50ad933..6ba5eca700 100644 --- a/src/test/recovery/t/019_replslot_limit.pl +++ b/src/test/recovery/t/019_replslot_limit.pl @@ -22,6 +22,7 @@ $node_primary->append_conf( min_wal_size = 2MB max_wal_size = 4MB log_checkpoints = yes +wal_summarize_mb = 0 )); $node_primary->start; $node_primary->safe_psql('postgres', @@ -256,6 +257,7 @@ $node_primary2->append_conf( min_wal_size = 32MB max_wal_size = 32MB log_checkpoints = yes +wal_summarize_mb = 0 )); $node_primary2->start; $node_primary2->safe_psql('postgres', @@ -310,6 +312,7 @@ $node_primary3->append_conf( max_wal_size = 2MB log_checkpoints = yes max_slot_wal_keep_size = 1MB + wal_summarize_mb = 0 )); $node_primary3->start; $node_primary3->safe_psql('postgres', diff --git a/src/test/recovery/t/035_standby_logical_decoding.pl b/src/test/recovery/t/035_standby_logical_decoding.pl index 480e6d6caa..a91437dfa7 100644 --- a/src/test/recovery/t/035_standby_logical_decoding.pl +++ b/src/test/recovery/t/035_standby_logical_decoding.pl @@ -250,6 +250,7 @@ $node_primary->append_conf( wal_level = 'logical' max_replication_slots = 4 max_wal_senders = 4 +wal_summarize_mb = 0 }); $node_primary->dump_info; $node_primary->start; diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 49a33c0387..56b8270dda 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -3989,3 +3989,27 @@ yyscan_t z_stream z_streamp zic_t +BlockRefTable +BlockRefTableBuffer +BlockRefTableEntry +BlockRefTableKey +BlockRefTableReader +BlockRefTableSerializedEntry +BlockRefTableWriter +FileBackupMethod +FileChunkContext +IncrementalBackupInfo +SummarizerReadLocalXLogPrivate +UploadManifestCmd +WalSummarizerData +WalSummaryFile +WalSummaryIO +backup_file_entry +backup_wal_range +cb_cleanup_dir +cb_options +cb_tablespace +cb_tablespace_mapping +manifest_data +manifest_writer +rfile -- 2.37.1 (Apple Git-137.1) ^ permalink raw reply [nested|flat] 22+ messages in thread
* Re: trying again to get incremental backup @ 2023-09-04 12:41 Dilip Kumar <[email protected]> parent: Robert Haas <[email protected]> 2 siblings, 1 reply; 22+ messages in thread From: Dilip Kumar @ 2023-09-04 12:41 UTC (permalink / raw) To: Robert Haas <[email protected]>; +Cc: Andres Freund <[email protected]>; pgsql-hackers On Wed, Aug 30, 2023 at 9:20 PM Robert Haas <[email protected]> wrote: > Unless someone has a brilliant idea that I lack, this suggests to me > that this whole line of testing is a dead end. I can, of course, write > tests that compare clusters *logically* -- do the correct relations > exist, are they accessible, do they have the right contents? Can't we think of comparing at the block level, like we can compare each block but ignore the content of the hole? -- Regards, Dilip Kumar EnterpriseDB: http://www.enterprisedb.com ^ permalink raw reply [nested|flat] 22+ messages in thread
* Re: trying again to get incremental backup @ 2023-09-05 15:05 Robert Haas <[email protected]> parent: Dilip Kumar <[email protected]> 0 siblings, 0 replies; 22+ messages in thread From: Robert Haas @ 2023-09-05 15:05 UTC (permalink / raw) To: Dilip Kumar <[email protected]>; +Cc: Andres Freund <[email protected]>; pgsql-hackers On Mon, Sep 4, 2023 at 8:42 AM Dilip Kumar <[email protected]> wrote: > Can't we think of comparing at the block level, like we can compare > each block but ignore the content of the hole? We could do that, but I don't think that's a full solution. I think I'd end up having to reimplement the equivalent of heap_mask, btree_mask, et. al. in Perl, which doesn't seem very reasonable. It's fairly complicated logic even written in C, and doing the right thing in Perl would be more complex, I think, because it wouldn't have access to all the same #defines which depend on things like word size and Endianness and stuff. If we want to allow this sort of comparison, I feel we should think of changing the C code in some way to make it work reliably rather than try to paper over the problems in Perl. -- Robert Haas EDB: http://www.enterprisedb.com ^ permalink raw reply [nested|flat] 22+ messages in thread
* Re: trying again to get incremental backup @ 2023-09-12 09:56 Dilip Kumar <[email protected]> parent: Robert Haas <[email protected]> 2 siblings, 1 reply; 22+ messages in thread From: Dilip Kumar @ 2023-09-12 09:56 UTC (permalink / raw) To: Robert Haas <[email protected]>; +Cc: Andres Freund <[email protected]>; pgsql-hackers On Wed, Aug 30, 2023 at 9:20 PM Robert Haas <[email protected]> wrote: > > Meanwhile, here's a rebased set of patches. The somewhat-primitive > attempts at writing tests are in 0009, but they don't work, for the > reasons explained above. I think I'd probably like to go ahead and > commit 0001 and 0002 soon if there are no objections, since I think > those are good refactorings independently of the rest of this. > I have started reading the patch today, I haven't yet completed one pass but here are my comments in 0007 1. + BlockNumber relative_block_numbers[RELSEG_SIZE]; This is close to 400kB of memory, so I think it is better we palloc it instead of keeping it in the stack. 2. /* * Try to parse the directory name as an unsigned integer. * - * Tablespace directories should be positive integers that can - * be represented in 32 bits, with no leading zeroes or trailing + * Tablespace directories should be positive integers that can be + * represented in 32 bits, with no leading zeroes or trailing * garbage. If we come across a name that doesn't meet those * criteria, skip it. Unrelated code refactoring hunk 3. +typedef struct +{ + const char *filename; + pg_checksum_context *checksum_ctx; + bbsink *sink; + size_t bytes_sent; +} FileChunkContext; This structure is not used anywhere. 4. + * If the file is to be set incrementally, then num_incremental_blocks + * should be the number of blocks to be sent, and incremental_blocks /If the file is to be set incrementally/If the file is to be sent incrementally 5. - while (bytes_done < statbuf->st_size) + while (1) { - size_t remaining = statbuf->st_size - bytes_done; + /* I do not really like this change, because after removing this you have put 2 independent checks for sending the full file[1] and sending it incrementally[1]. Actually for sending incrementally 'statbuf->st_size' is computed from the 'num_incremental_blocks' itself so why don't we keep this breaking condition in the while loop itself? So that we can avoid these two separate conditions. [1] + /* + * If we've read the required number of bytes, then it's time to + * stop. + */ + if (bytes_done >= statbuf->st_size) + break; [2] + /* + * If we've read all the blocks, then it's time to stop. + */ + if (ibindex >= num_incremental_blocks) + break; 6. +typedef struct +{ + TimeLineID tli; + XLogRecPtr start_lsn; + XLogRecPtr end_lsn; +} backup_wal_range; + +typedef struct +{ + uint32 status; + const char *path; + size_t size; +} backup_file_entry; Better we add some comments for these structures. -- Regards, Dilip Kumar EnterpriseDB: http://www.enterprisedb.com ^ permalink raw reply [nested|flat] 22+ messages in thread
* Re: trying again to get incremental backup @ 2023-09-28 10:22 Jakub Wartak <[email protected]> parent: Robert Haas <[email protected]> 2 siblings, 2 replies; 22+ messages in thread From: Jakub Wartak @ 2023-09-28 10:22 UTC (permalink / raw) To: Robert Haas <[email protected]>; +Cc: Andres Freund <[email protected]>; pgsql-hackers On Wed, Aug 30, 2023 at 4:50 PM Robert Haas <[email protected]> wrote: [..] I've played a little bit more this second batch of patches on e8d74ad625f7344f6b715254d3869663c1569a51 @ 31Aug (days before wait events refactor): test_across_wallevelminimal.sh test_many_incrementals_dbcreate.sh test_many_incrementals.sh test_multixact.sh test_pending_2pc.sh test_reindex_and_vacuum_full.sh test_truncaterollback.sh test_unlogged_table.sh all those basic tests had GOOD results. Please find attached. I'll try to schedule some more realistic (in terms of workload and sizes) test in a couple of days + maybe have some fun with cross-backup-and restores across standbys. As per earlier doubt: raw wal_level = minimal situation, shouldn't be a concern, sadly because it requires max_wal_senders==0, while pg_basebackup requires it above 0 (due to "FATAL: number of requested standby connections exceeds max_wal_senders (currently 0)"). I wanted to also introduce corruption onto pg_walsummaries files, but later saw in code that is already covered with CRC32, cool. In v07: > +#define MINIMUM_VERSION_FOR_WAL_SUMMARIES 160000 170000 ? > A related design question is whether we should really be sending the > whole backup manifest to the server at all. If it turns out that we > don't really need anything except for the LSN of the previous backup, > we could send that one piece of information instead of everything. On > the other hand, if we need the list of files from the previous backup, > then sending the whole manifest makes sense. If that is still an area open for discussion: wouldn't it be better to just specify LSN as it would allow resyncing standby across major lag where the WAL to replay would be enormous? Given that we had primary->standby where standby would be stuck on some LSN, right now it would be: 1) calculate backup manifest of desynced 10TB standby (how? using which tool?) - even if possible, that means reading 10TB of data instead of just putting a number, isn't it? 2) backup primary with such incremental backup >= LSN 3) copy the incremental backup to standby 4) apply it to the impaired standby 5) restart the WAL replay > - We only know how to operate on directories, not tar files. I thought > about that when working on pg_verifybackup as well, but I didn't do > anything about it. It would be nice to go back and make that tool work > on tar-format backups, and this one, too. I don't think there would be > a whole lot of point trying to operate on compressed tar files because > you need random access and that seems hard on a compressed file, but > on uncompressed files it seems at least theoretically doable. I'm not > sure whether anyone would care that much about this, though, even > though it does sound pretty cool. Also maybe it's too early to ask, but wouldn't it be nice if we could have an future option in pg_combinebackup to avoid double writes when used from restore hosts (right now we need to first to reconstruct the original datadir from full and incremental backups on host hosting backups and then TRANSFER it again and on target host?). So something like that could work well from restorehost: pg_combinebackup /tmp/backup1 /tmp/incbackup2 /tmp/incbackup3 -O tar -o - | ssh dbserver 'tar xvf -C /path/to/restored/cluster - ' . The bad thing is that such a pipe prevents parallelism from day 1 and I'm afraid I do not have a better easy idea on how to have both at the same time in the long term. -J. Attachments: [application/x-compressed] incrbackuptests-0.1.tgz (2.7K, ../../CAKZiRmzT+bX2ZYdORO32cADtfQ9DvyaOE8fsOEWZc2V5FkEWVg@mail.gmail.com/2-incrbackuptests-0.1.tgz) download ^ permalink raw reply [nested|flat] 22+ messages in thread
* Re: trying again to get incremental backup @ 2023-10-03 18:21 Robert Haas <[email protected]> parent: Dilip Kumar <[email protected]> 0 siblings, 1 reply; 22+ messages in thread From: Robert Haas @ 2023-10-03 18:21 UTC (permalink / raw) To: Dilip Kumar <[email protected]>; +Cc: Andres Freund <[email protected]>; pgsql-hackers On Tue, Sep 12, 2023 at 5:56 AM Dilip Kumar <[email protected]> wrote: > + BlockNumber relative_block_numbers[RELSEG_SIZE]; > > This is close to 400kB of memory, so I think it is better we palloc it > instead of keeping it in the stack. Fixed. > Unrelated code refactoring hunk Fixed. > This structure is not used anywhere. Removed. > /If the file is to be set incrementally/If the file is to be sent incrementally Fixed. > I do not really like this change, because after removing this you have > put 2 independent checks for sending the full file[1] and sending it > incrementally[1]. Actually for sending incrementally > 'statbuf->st_size' is computed from the 'num_incremental_blocks' > itself so why don't we keep this breaking condition in the while loop > itself? So that we can avoid these two separate conditions. I don't think that would be correct. The number of bytes that need to be read from the original file is not equal to the number of bytes that will be written to the incremental file. Admittedly, they're currently different by less than a block, but that could change if we change the format of the incremental file (e.g. suppose we compressed the blocks in the incremental file with gzip, or smushed out the holes in the pages). I wrote the loop as I did precisely so that the two cases could have different loop exit conditions. > Better we add some comments for these structures. Done. Here's a new patch set, also addressing Jakub's observation that MINIMUM_VERSION_FOR_WAL_SUMMARIES needed updating. -- Robert Haas EDB: http://www.enterprisedb.com Attachments: [application/octet-stream] v3-0002-Refactor-parse_filename_for_nontemp_relation-to-p.patch (11.9K, ../../CA+Tgmoa_Gq2Qc9dh-WiZEUyF-nk3XULuQR-zFY2vOGNWRCZjkQ@mail.gmail.com/2-v3-0002-Refactor-parse_filename_for_nontemp_relation-to-p.patch) download | inline diff: From b7f1acaead7fdf87a01ea88a7a381580c1705b38 Mon Sep 17 00:00:00 2001 From: Robert Haas <[email protected]> Date: Tue, 3 Oct 2023 13:30:44 -0400 Subject: [PATCH v3 2/7] Refactor parse_filename_for_nontemp_relation to parse more. Instead of returning the number of characters in the RelFileNumber, return the RelFileNumber itself. Continue to return the fork number, as before, and additionally return the segment number. parse_filename_for_nontemp_relation now rejects a RelFileNumber or segment number that begins with a leading zero. Before, we accepted such cases as relation filenames, but if we continued to do so after this change, the function might return the same values for two different files (e.g. 1234.5 and 001234.5 or 1234.005) which could be annoying for callers. Since we don't actually ever generate filenames with leading zeroes in the names, any such files that we find must have been created by something other than PostgreSQL, and it is therefore reasonable to treat them as non-relation files. Along the way, change unlogged_relation_entry to store a RelFileNumber rather than an OID. This update should have been made in 851f4cc75cdd8c831f1baa9a7abf8c8248b65890, but it was overlooked. It could be done separately from the rest of this commit, but that would be more involved, whereas this way it's a 1-line change. --- src/backend/backup/basebackup.c | 15 ++-- src/backend/storage/file/reinit.c | 137 ++++++++++++++++++------------ src/include/storage/reinit.h | 5 +- 3 files changed, 93 insertions(+), 64 deletions(-) diff --git a/src/backend/backup/basebackup.c b/src/backend/backup/basebackup.c index dc6892011d..b537f46219 100644 --- a/src/backend/backup/basebackup.c +++ b/src/backend/backup/basebackup.c @@ -1198,9 +1198,9 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, { int excludeIdx; bool excludeFound; - ForkNumber relForkNum; /* Type of fork if file is a relation */ - int relnumchars; /* Chars in filename that are the - * relnumber */ + RelFileNumber relNumber; + ForkNumber relForkNum; + unsigned segno; /* Skip special stuff */ if (strcmp(de->d_name, ".") == 0 || strcmp(de->d_name, "..") == 0) @@ -1250,23 +1250,20 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, /* Exclude all forks for unlogged tables except the init fork */ if (isDbDir && - parse_filename_for_nontemp_relation(de->d_name, &relnumchars, - &relForkNum)) + parse_filename_for_nontemp_relation(de->d_name, &relNumber, + &relForkNum, &segno)) { /* Never exclude init forks */ if (relForkNum != INIT_FORKNUM) { char initForkFile[MAXPGPATH]; - char relNumber[OIDCHARS + 1]; /* * If any other type of fork, check if there is an init fork * with the same RelFileNumber. If so, the file can be * excluded. */ - memcpy(relNumber, de->d_name, relnumchars); - relNumber[relnumchars] = '\0'; - snprintf(initForkFile, sizeof(initForkFile), "%s/%s_init", + snprintf(initForkFile, sizeof(initForkFile), "%s/%u_init", path, relNumber); if (lstat(initForkFile, &statbuf) == 0) diff --git a/src/backend/storage/file/reinit.c b/src/backend/storage/file/reinit.c index fb55371b1b..5df2517b46 100644 --- a/src/backend/storage/file/reinit.c +++ b/src/backend/storage/file/reinit.c @@ -31,7 +31,7 @@ static void ResetUnloggedRelationsInDbspaceDir(const char *dbspacedirname, typedef struct { - Oid reloid; /* hash key */ + RelFileNumber relnumber; /* hash key */ } unlogged_relation_entry; /* @@ -195,12 +195,13 @@ ResetUnloggedRelationsInDbspaceDir(const char *dbspacedirname, int op) while ((de = ReadDir(dbspace_dir, dbspacedirname)) != NULL) { ForkNumber forkNum; - int relnumchars; + unsigned segno; unlogged_relation_entry ent; /* Skip anything that doesn't look like a relation data file. */ - if (!parse_filename_for_nontemp_relation(de->d_name, &relnumchars, - &forkNum)) + if (!parse_filename_for_nontemp_relation(de->d_name, + &ent.relnumber, + &forkNum, &segno)) continue; /* Also skip it unless this is the init fork. */ @@ -208,10 +209,8 @@ ResetUnloggedRelationsInDbspaceDir(const char *dbspacedirname, int op) continue; /* - * Put the OID portion of the name into the hash table, if it - * isn't already. + * Put the RelFileNumber into the hash table, if it isn't already. */ - ent.reloid = atooid(de->d_name); (void) hash_search(hash, &ent, HASH_ENTER, NULL); } @@ -235,12 +234,13 @@ ResetUnloggedRelationsInDbspaceDir(const char *dbspacedirname, int op) while ((de = ReadDir(dbspace_dir, dbspacedirname)) != NULL) { ForkNumber forkNum; - int relnumchars; + unsigned segno; unlogged_relation_entry ent; /* Skip anything that doesn't look like a relation data file. */ - if (!parse_filename_for_nontemp_relation(de->d_name, &relnumchars, - &forkNum)) + if (!parse_filename_for_nontemp_relation(de->d_name, + &ent.relnumber, + &forkNum, &segno)) continue; /* We never remove the init fork. */ @@ -251,7 +251,6 @@ ResetUnloggedRelationsInDbspaceDir(const char *dbspacedirname, int op) * See whether the OID portion of the name shows up in the hash * table. If so, nuke it! */ - ent.reloid = atooid(de->d_name); if (hash_search(hash, &ent, HASH_FIND, NULL)) { snprintf(rm_path, sizeof(rm_path), "%s/%s", @@ -285,14 +284,14 @@ ResetUnloggedRelationsInDbspaceDir(const char *dbspacedirname, int op) while ((de = ReadDir(dbspace_dir, dbspacedirname)) != NULL) { ForkNumber forkNum; - int relnumchars; - char relnumbuf[OIDCHARS + 1]; + RelFileNumber relNumber; + unsigned segno; char srcpath[MAXPGPATH * 2]; char dstpath[MAXPGPATH]; /* Skip anything that doesn't look like a relation data file. */ - if (!parse_filename_for_nontemp_relation(de->d_name, &relnumchars, - &forkNum)) + if (!parse_filename_for_nontemp_relation(de->d_name, &relNumber, + &forkNum, &segno)) continue; /* Also skip it unless this is the init fork. */ @@ -304,11 +303,12 @@ ResetUnloggedRelationsInDbspaceDir(const char *dbspacedirname, int op) dbspacedirname, de->d_name); /* Construct destination pathname. */ - memcpy(relnumbuf, de->d_name, relnumchars); - relnumbuf[relnumchars] = '\0'; - snprintf(dstpath, sizeof(dstpath), "%s/%s%s", - dbspacedirname, relnumbuf, de->d_name + relnumchars + 1 + - strlen(forkNames[INIT_FORKNUM])); + if (segno == 0) + snprintf(dstpath, sizeof(dstpath), "%s/%u", + dbspacedirname, relNumber); + else + snprintf(dstpath, sizeof(dstpath), "%s/%u.%u", + dbspacedirname, relNumber, segno); /* OK, we're ready to perform the actual copy. */ elog(DEBUG2, "copying %s to %s", srcpath, dstpath); @@ -327,14 +327,14 @@ ResetUnloggedRelationsInDbspaceDir(const char *dbspacedirname, int op) dbspace_dir = AllocateDir(dbspacedirname); while ((de = ReadDir(dbspace_dir, dbspacedirname)) != NULL) { + RelFileNumber relNumber; ForkNumber forkNum; - int relnumchars; - char relnumbuf[OIDCHARS + 1]; + unsigned segno; char mainpath[MAXPGPATH]; /* Skip anything that doesn't look like a relation data file. */ - if (!parse_filename_for_nontemp_relation(de->d_name, &relnumchars, - &forkNum)) + if (!parse_filename_for_nontemp_relation(de->d_name, &relNumber, + &forkNum, &segno)) continue; /* Also skip it unless this is the init fork. */ @@ -342,11 +342,12 @@ ResetUnloggedRelationsInDbspaceDir(const char *dbspacedirname, int op) continue; /* Construct main fork pathname. */ - memcpy(relnumbuf, de->d_name, relnumchars); - relnumbuf[relnumchars] = '\0'; - snprintf(mainpath, sizeof(mainpath), "%s/%s%s", - dbspacedirname, relnumbuf, de->d_name + relnumchars + 1 + - strlen(forkNames[INIT_FORKNUM])); + if (segno == 0) + snprintf(mainpath, sizeof(mainpath), "%s/%u", + dbspacedirname, relNumber); + else + snprintf(mainpath, sizeof(mainpath), "%s/%u.%u", + dbspacedirname, relNumber, segno); fsync_fname(mainpath, false); } @@ -371,52 +372,82 @@ ResetUnloggedRelationsInDbspaceDir(const char *dbspacedirname, int op) * This function returns true if the file appears to be in the correct format * for a non-temporary relation and false otherwise. * - * NB: If this function returns true, the caller is entitled to assume that - * *relnumchars has been set to a value no more than OIDCHARS, and thus - * that a buffer of OIDCHARS+1 characters is sufficient to hold the - * RelFileNumber portion of the filename. This is critical to protect against - * a possible buffer overrun. + * If it returns true, it sets *relnumber, *fork, and *segno to the values + * extracted from the filename. If it returns false, these values are set to + * InvalidRelFileNumber, InvalidForkNumber, and 0, respectively. */ bool -parse_filename_for_nontemp_relation(const char *name, int *relnumchars, - ForkNumber *fork) +parse_filename_for_nontemp_relation(const char *name, RelFileNumber *relnumber, + ForkNumber *fork, unsigned *segno) { - int pos; + unsigned long n, + s; + ForkNumber f; + char *endp; - /* Look for a non-empty string of digits (that isn't too long). */ - for (pos = 0; isdigit((unsigned char) name[pos]); ++pos) - ; - if (pos == 0 || pos > OIDCHARS) + *relnumber = InvalidRelFileNumber; + *fork = InvalidForkNumber; + *segno = 0; + + /* + * Relation filenames should begin with a digit that is not a zero. By + * rejecting cases involving leading zeroes, the caller can assume that + * there's only one possible string of characters that could have produced + * any given value for *relnumber. + * + * (To be clear, we don't expect files with names like 0017.3 to exist at + * all -- but if 0017.3 does exist, it's a non-relation file, not part of + * the main fork for relfilenode 17.) + */ + if (name[0] < '1' || name[0] > '9') + return false; + + /* + * Parse the leading digit string. If the value is out of range, we + * conclude that this isn't a relation file at all. + */ + errno = 0; + n = strtoul(name, &endp, 10); + if (errno || name == endp || n <= 0 || n > PG_UINT32_MAX) return false; - *relnumchars = pos; + name = endp; /* Check for a fork name. */ - if (name[pos] != '_') - *fork = MAIN_FORKNUM; + if (*name != '_') + f = MAIN_FORKNUM; else { int forkchar; - forkchar = forkname_chars(&name[pos + 1], fork); + forkchar = forkname_chars(name + 1, &f); if (forkchar <= 0) return false; - pos += forkchar + 1; + name += forkchar + 1; } /* Check for a segment number. */ - if (name[pos] == '.') + if (*name != '.') + s = 0; + else { - int segchar; + /* Reject leading zeroes, just like we do for RelFileNumber. */ + if (name[0] < '1' || name[0] > '9') + return false; - for (segchar = 1; isdigit((unsigned char) name[pos + segchar]); ++segchar) - ; - if (segchar <= 1) + errno = 0; + s = strtoul(name + 1, &endp, 10); + if (errno || name + 1 == endp || s <= 0 || s > PG_UINT32_MAX) return false; - pos += segchar; + name = endp; } /* Now we should be at the end. */ - if (name[pos] != '\0') + if (*name != '\0') return false; + + /* Set out parameters and return. */ + *relnumber = (RelFileNumber) n; + *fork = f; + *segno = (unsigned) s; return true; } diff --git a/src/include/storage/reinit.h b/src/include/storage/reinit.h index e2bbb5abe9..f8eb7ce234 100644 --- a/src/include/storage/reinit.h +++ b/src/include/storage/reinit.h @@ -20,8 +20,9 @@ extern void ResetUnloggedRelations(int op); extern bool parse_filename_for_nontemp_relation(const char *name, - int *relnumchars, - ForkNumber *fork); + RelFileNumber *relnumber, + ForkNumber *fork, + unsigned *segno); #define UNLOGGED_RELATION_CLEANUP 0x0001 #define UNLOGGED_RELATION_INIT 0x0002 -- 2.37.1 (Apple Git-137.1) [application/octet-stream] v3-0004-Move-src-bin-pg_verifybackup-parse_manifest.c-int.patch (4.3K, ../../CA+Tgmoa_Gq2Qc9dh-WiZEUyF-nk3XULuQR-zFY2vOGNWRCZjkQ@mail.gmail.com/3-v3-0004-Move-src-bin-pg_verifybackup-parse_manifest.c-int.patch) download | inline diff: From 333933d0d14588a43826e0089d97579c65cef94e Mon Sep 17 00:00:00 2001 From: Robert Haas <[email protected]> Date: Tue, 3 Oct 2023 13:32:45 -0400 Subject: [PATCH v3 4/7] Move src/bin/pg_verifybackup/parse_manifest.c into src/common. This makes it possible for the code to be easily reused by other client-side tools, and/or by the server. --- src/bin/pg_verifybackup/Makefile | 1 - src/bin/pg_verifybackup/meson.build | 1 - src/bin/pg_verifybackup/pg_verifybackup.c | 2 +- src/common/Makefile | 1 + src/common/meson.build | 1 + src/{bin/pg_verifybackup => common}/parse_manifest.c | 4 ++-- src/{bin/pg_verifybackup => include/common}/parse_manifest.h | 2 +- 7 files changed, 6 insertions(+), 6 deletions(-) rename src/{bin/pg_verifybackup => common}/parse_manifest.c (99%) rename src/{bin/pg_verifybackup => include/common}/parse_manifest.h (97%) diff --git a/src/bin/pg_verifybackup/Makefile b/src/bin/pg_verifybackup/Makefile index 596df15118..8f04fa662c 100644 --- a/src/bin/pg_verifybackup/Makefile +++ b/src/bin/pg_verifybackup/Makefile @@ -21,7 +21,6 @@ LDFLAGS_INTERNAL += -L$(top_builddir)/src/fe_utils -lpgfeutils $(libpq_pgport) OBJS = \ $(WIN32RES) \ - parse_manifest.o \ pg_verifybackup.o all: pg_verifybackup diff --git a/src/bin/pg_verifybackup/meson.build b/src/bin/pg_verifybackup/meson.build index 9369da1bc6..58f780d1a6 100644 --- a/src/bin/pg_verifybackup/meson.build +++ b/src/bin/pg_verifybackup/meson.build @@ -1,7 +1,6 @@ # Copyright (c) 2022-2023, PostgreSQL Global Development Group pg_verifybackup_sources = files( - 'parse_manifest.c', 'pg_verifybackup.c' ) diff --git a/src/bin/pg_verifybackup/pg_verifybackup.c b/src/bin/pg_verifybackup/pg_verifybackup.c index 059836f0e6..ce423a03d4 100644 --- a/src/bin/pg_verifybackup/pg_verifybackup.c +++ b/src/bin/pg_verifybackup/pg_verifybackup.c @@ -20,9 +20,9 @@ #include "common/hashfn.h" #include "common/logging.h" +#include "common/parse_manifest.h" #include "fe_utils/simple_list.h" #include "getopt_long.h" -#include "parse_manifest.h" #include "pgtime.h" /* diff --git a/src/common/Makefile b/src/common/Makefile index cc5c54dcee..ff60666f5c 100644 --- a/src/common/Makefile +++ b/src/common/Makefile @@ -66,6 +66,7 @@ OBJS_COMMON = \ kwlookup.o \ link-canary.o \ md5_common.o \ + parse_manifest.o \ percentrepl.o \ pg_get_line.o \ pg_lzcompress.o \ diff --git a/src/common/meson.build b/src/common/meson.build index 3b97497d1a..fcc0c4fe8d 100644 --- a/src/common/meson.build +++ b/src/common/meson.build @@ -18,6 +18,7 @@ common_sources = files( 'kwlookup.c', 'link-canary.c', 'md5_common.c', + 'parse_manifest.c', 'percentrepl.c', 'pg_get_line.c', 'pg_lzcompress.c', diff --git a/src/bin/pg_verifybackup/parse_manifest.c b/src/common/parse_manifest.c similarity index 99% rename from src/bin/pg_verifybackup/parse_manifest.c rename to src/common/parse_manifest.c index 2379f7be7b..672e8bcf25 100644 --- a/src/bin/pg_verifybackup/parse_manifest.c +++ b/src/common/parse_manifest.c @@ -6,15 +6,15 @@ * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group * Portions Copyright (c) 1994, Regents of the University of California * - * src/bin/pg_verifybackup/parse_manifest.c + * src/common/parse_manifest.c * *------------------------------------------------------------------------- */ #include "postgres_fe.h" -#include "parse_manifest.h" #include "common/jsonapi.h" +#include "common/parse_manifest.h" /* * Semantic states for JSON manifest parsing. diff --git a/src/bin/pg_verifybackup/parse_manifest.h b/src/include/common/parse_manifest.h similarity index 97% rename from src/bin/pg_verifybackup/parse_manifest.h rename to src/include/common/parse_manifest.h index 7387a917a2..7b24c5d785 100644 --- a/src/bin/pg_verifybackup/parse_manifest.h +++ b/src/include/common/parse_manifest.h @@ -6,7 +6,7 @@ * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group * Portions Copyright (c) 1994, Regents of the University of California * - * src/bin/pg_verifybackup/parse_manifest.h + * src/include/common/parse_manifest.h * *------------------------------------------------------------------------- */ -- 2.37.1 (Apple Git-137.1) [application/octet-stream] v3-0001-Change-struct-tablespaceinfo-s-oid-member-from-ch.patch (11.7K, ../../CA+Tgmoa_Gq2Qc9dh-WiZEUyF-nk3XULuQR-zFY2vOGNWRCZjkQ@mail.gmail.com/4-v3-0001-Change-struct-tablespaceinfo-s-oid-member-from-ch.patch) download | inline diff: From fcc50ccfc814763e7b12dfa3b265ee737802b74f Mon Sep 17 00:00:00 2001 From: Robert Haas <[email protected]> Date: Tue, 3 Oct 2023 13:30:16 -0400 Subject: [PATCH v3 1/7] Change struct tablespaceinfo's oid member from 'char *' to 'Oid' This shouldn't change behavior except in the unusual case where there are file in the tablespace directory that have entirely numeric names but are nevertheless not possible names for a tablespace directory, either because their names has leading zeroes that shouldn't be there, or the value is actually zero, or because the value is too large to represent as an OID. In those cases, the directory would previously have made it into the list of tablespaceinfo objects and no longer will. Thus, base backups will now ignore such directories, instead of treating them as legitimate tablespace directories. Similarly, if entries for such tablespaces occur in a tablespace_map file, they will now be rejected as erroneous, instead of being honored. This is infrastructure for future work that wants to be able to know the tablespace of each relation that is part of a backup *as an OID*. By strengthening the up-front validation, we don't have to worry about weird cases later, and can more easily avoid repeated string->integer conversions. --- src/backend/access/transam/xlog.c | 19 ++++++++++-- src/backend/access/transam/xlogrecovery.c | 12 ++++++-- src/backend/backup/backup_manifest.c | 6 ++-- src/backend/backup/basebackup.c | 35 ++++++++++++----------- src/backend/backup/basebackup_copy.c | 2 +- src/include/backup/backup_manifest.h | 2 +- src/include/backup/basebackup.h | 2 +- 7 files changed, 49 insertions(+), 29 deletions(-) diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index fcbde10529..677a5bf51b 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -8483,9 +8483,22 @@ do_pg_backup_start(const char *backupidstr, bool fast, List **tablespaces, char *relpath = NULL; char *s; PGFileType de_type; + char *badp; + Oid tsoid; - /* Skip anything that doesn't look like a tablespace */ - if (strspn(de->d_name, "0123456789") != strlen(de->d_name)) + /* + * Try to parse the directory name as an unsigned integer. + * + * Tablespace directories should be positive integers that can be + * represented in 32 bits, with no leading zeroes or trailing + * garbage. If we come across a name that doesn't meet those + * criteria, skip it. + */ + if (de->d_name[0] < '1' || de->d_name[1] > '9') + continue; + errno = 0; + tsoid = strtoul(de->d_name, &badp, 10); + if (*badp != '\0' || errno == EINVAL || errno == ERANGE) continue; snprintf(fullpath, sizeof(fullpath), "pg_tblspc/%s", de->d_name); @@ -8560,7 +8573,7 @@ do_pg_backup_start(const char *backupidstr, bool fast, List **tablespaces, } ti = palloc(sizeof(tablespaceinfo)); - ti->oid = pstrdup(de->d_name); + ti->oid = tsoid; ti->path = pstrdup(linkpath); ti->rpath = relpath; ti->size = -1; diff --git a/src/backend/access/transam/xlogrecovery.c b/src/backend/access/transam/xlogrecovery.c index becc2bda62..5549e1afc5 100644 --- a/src/backend/access/transam/xlogrecovery.c +++ b/src/backend/access/transam/xlogrecovery.c @@ -678,7 +678,7 @@ InitWalRecovery(ControlFileData *ControlFile, bool *wasShutdown_ptr, tablespaceinfo *ti = lfirst(lc); char *linkloc; - linkloc = psprintf("pg_tblspc/%s", ti->oid); + linkloc = psprintf("pg_tblspc/%u", ti->oid); /* * Remove the existing symlink if any and Create the symlink @@ -692,7 +692,6 @@ InitWalRecovery(ControlFileData *ControlFile, bool *wasShutdown_ptr, errmsg("could not create symbolic link \"%s\": %m", linkloc))); - pfree(ti->oid); pfree(ti->path); pfree(ti); } @@ -1341,6 +1340,8 @@ read_tablespace_map(List **tablespaces) { if (!was_backslash && (ch == '\n' || ch == '\r')) { + char *endp; + if (i == 0) continue; /* \r immediately followed by \n */ @@ -1360,7 +1361,12 @@ read_tablespace_map(List **tablespaces) str[n++] = '\0'; ti = palloc0(sizeof(tablespaceinfo)); - ti->oid = pstrdup(str); + errno = 0; + ti->oid = strtoul(str, &endp, 10); + if (*endp != '\0' || errno == EINVAL || errno == ERANGE) + ereport(FATAL, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("invalid data in file \"%s\"", TABLESPACE_MAP))); ti->path = pstrdup(str + n); *tablespaces = lappend(*tablespaces, ti); diff --git a/src/backend/backup/backup_manifest.c b/src/backend/backup/backup_manifest.c index cee6216524..aeed362a9a 100644 --- a/src/backend/backup/backup_manifest.c +++ b/src/backend/backup/backup_manifest.c @@ -97,7 +97,7 @@ FreeBackupManifest(backup_manifest_info *manifest) * Add an entry to the backup manifest for a file. */ void -AddFileToBackupManifest(backup_manifest_info *manifest, const char *spcoid, +AddFileToBackupManifest(backup_manifest_info *manifest, Oid spcoid, const char *pathname, size_t size, pg_time_t mtime, pg_checksum_context *checksum_ctx) { @@ -114,9 +114,9 @@ AddFileToBackupManifest(backup_manifest_info *manifest, const char *spcoid, * pathname relative to the data directory (ignoring the intermediate * symlink traversal). */ - if (spcoid != NULL) + if (OidIsValid(spcoid)) { - snprintf(pathbuf, sizeof(pathbuf), "pg_tblspc/%s/%s", spcoid, + snprintf(pathbuf, sizeof(pathbuf), "pg_tblspc/%u/%s", spcoid, pathname); pathname = pathbuf; } diff --git a/src/backend/backup/basebackup.c b/src/backend/backup/basebackup.c index 7d025bcf38..dc6892011d 100644 --- a/src/backend/backup/basebackup.c +++ b/src/backend/backup/basebackup.c @@ -75,14 +75,15 @@ typedef struct pg_checksum_type manifest_checksum_type; } basebackup_options; -static int64 sendTablespace(bbsink *sink, char *path, char *spcoid, bool sizeonly, +static int64 sendTablespace(bbsink *sink, char *path, Oid spcoid, bool sizeonly, struct backup_manifest_info *manifest); static int64 sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, List *tablespaces, bool sendtblspclinks, - backup_manifest_info *manifest, const char *spcoid); + backup_manifest_info *manifest, Oid spcoid); static bool sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, - struct stat *statbuf, bool missing_ok, Oid dboid, - backup_manifest_info *manifest, const char *spcoid); + struct stat *statbuf, bool missing_ok, + Oid dboid, Oid spcoid, + backup_manifest_info *manifest); static off_t read_file_data_into_buffer(bbsink *sink, const char *readfilename, int fd, off_t offset, size_t length, @@ -305,7 +306,7 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) if (tmp->path == NULL) tmp->size = sendDir(sink, ".", 1, true, state.tablespaces, - true, NULL, NULL); + true, NULL, InvalidOid); else tmp->size = sendTablespace(sink, tmp->path, tmp->oid, true, NULL); @@ -346,7 +347,7 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) /* Then the bulk of the files... */ sendDir(sink, ".", 1, false, state.tablespaces, - sendtblspclinks, &manifest, NULL); + sendtblspclinks, &manifest, InvalidOid); /* ... and pg_control after everything else. */ if (lstat(XLOG_CONTROL_FILE, &statbuf) != 0) @@ -355,11 +356,11 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) errmsg("could not stat file \"%s\": %m", XLOG_CONTROL_FILE))); sendFile(sink, XLOG_CONTROL_FILE, XLOG_CONTROL_FILE, &statbuf, - false, InvalidOid, &manifest, NULL); + false, InvalidOid, InvalidOid, &manifest); } else { - char *archive_name = psprintf("%s.tar", ti->oid); + char *archive_name = psprintf("%u.tar", ti->oid); bbsink_begin_archive(sink, archive_name); @@ -623,8 +624,8 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) (errcode_for_file_access(), errmsg("could not stat file \"%s\": %m", pathbuf))); - sendFile(sink, pathbuf, pathbuf, &statbuf, false, InvalidOid, - &manifest, NULL); + sendFile(sink, pathbuf, pathbuf, &statbuf, false, + InvalidOid, InvalidOid, &manifest); /* unconditionally mark file as archived */ StatusFilePath(pathbuf, fname, ".done"); @@ -1087,7 +1088,7 @@ sendFileWithContent(bbsink *sink, const char *filename, const char *content, _tarWritePadding(sink, len); - AddFileToBackupManifest(manifest, NULL, filename, len, + AddFileToBackupManifest(manifest, InvalidOid, filename, len, (pg_time_t) statbuf.st_mtime, &checksum_ctx); } @@ -1099,7 +1100,7 @@ sendFileWithContent(bbsink *sink, const char *filename, const char *content, * Only used to send auxiliary tablespaces, not PGDATA. */ static int64 -sendTablespace(bbsink *sink, char *path, char *spcoid, bool sizeonly, +sendTablespace(bbsink *sink, char *path, Oid spcoid, bool sizeonly, backup_manifest_info *manifest) { int64 size; @@ -1154,7 +1155,7 @@ sendTablespace(bbsink *sink, char *path, char *spcoid, bool sizeonly, static int64 sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, List *tablespaces, bool sendtblspclinks, backup_manifest_info *manifest, - const char *spcoid) + Oid spcoid) { DIR *dir; struct dirent *de; @@ -1419,8 +1420,8 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, if (!sizeonly) sent = sendFile(sink, pathbuf, pathbuf + basepathlen + 1, &statbuf, - true, isDbDir ? atooid(lastDir + 1) : InvalidOid, - manifest, spcoid); + true, isDbDir ? atooid(lastDir + 1) : InvalidOid, spcoid, + manifest); if (sent || sizeonly) { @@ -1489,8 +1490,8 @@ is_checksummed_file(const char *fullpath, const char *filename) */ static bool sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, - struct stat *statbuf, bool missing_ok, Oid dboid, - backup_manifest_info *manifest, const char *spcoid) + struct stat *statbuf, bool missing_ok, Oid dboid, Oid spcoid, + backup_manifest_info *manifest) { int fd; BlockNumber blkno = 0; diff --git a/src/backend/backup/basebackup_copy.c b/src/backend/backup/basebackup_copy.c index fee30c21e1..3bdbe1f989 100644 --- a/src/backend/backup/basebackup_copy.c +++ b/src/backend/backup/basebackup_copy.c @@ -407,7 +407,7 @@ SendTablespaceList(List *tablespaces) } else { - values[0] = ObjectIdGetDatum(strtoul(ti->oid, NULL, 10)); + values[0] = ObjectIdGetDatum(ti->oid); values[1] = CStringGetTextDatum(ti->path); } if (ti->size >= 0) diff --git a/src/include/backup/backup_manifest.h b/src/include/backup/backup_manifest.h index d41b439980..5a481dbcf5 100644 --- a/src/include/backup/backup_manifest.h +++ b/src/include/backup/backup_manifest.h @@ -39,7 +39,7 @@ extern void InitializeBackupManifest(backup_manifest_info *manifest, backup_manifest_option want_manifest, pg_checksum_type manifest_checksum_type); extern void AddFileToBackupManifest(backup_manifest_info *manifest, - const char *spcoid, + Oid spcoid, const char *pathname, size_t size, pg_time_t mtime, pg_checksum_context *checksum_ctx); diff --git a/src/include/backup/basebackup.h b/src/include/backup/basebackup.h index 3e68abc2bb..1432d9c206 100644 --- a/src/include/backup/basebackup.h +++ b/src/include/backup/basebackup.h @@ -27,7 +27,7 @@ */ typedef struct { - char *oid; /* tablespace's OID, as a decimal string */ + Oid oid; /* tablespace's OID */ char *path; /* full path to tablespace's directory */ char *rpath; /* relative path if it's within PGDATA, else * NULL */ -- 2.37.1 (Apple Git-137.1) [application/octet-stream] v3-0003-Change-how-a-base-backup-decides-which-files-have.patch (11.4K, ../../CA+Tgmoa_Gq2Qc9dh-WiZEUyF-nk3XULuQR-zFY2vOGNWRCZjkQ@mail.gmail.com/5-v3-0003-Change-how-a-base-backup-decides-which-files-have.patch) download | inline diff: From 65f7f683ecbcebc6575f73716ce86e2a2f70368b Mon Sep 17 00:00:00 2001 From: Robert Haas <[email protected]> Date: Tue, 3 Oct 2023 13:32:28 -0400 Subject: [PATCH v3 3/7] Change how a base backup decides which files have checksums. Previously, it thought that any plain file located under global, base, or a tablespace directory had checksums unless it was in a short list of excluded files. Now, it thinks that files in those directories have checksums if parse_filename_for_nontemp_relation says that they are relation files. (Temporary relation files don't matter because they're excluded from the backup anyway.) This changes the behavior if you have stray files not managed by PostgreSQL in the relevant directories. Previously, you'd get some kind of checksum-related complaint if such files existed, assuming that the cluster had checksums enabled and that the base backup wasn't run with NOVERIFY_CHECKSUMS. Now, you won't get those complaints any more. That seems like an improvement to me, because those files were presumably not created by PostgreSQL and so there is no reason to think that they would be checksummed like a PostgreSQL relation file. (If we want to complain about such files, we should complain about them existing at all, not just about their checksums.) The point of this change is to make the code more consistent. sendDir() was already calling parse_filename_for_nontemp_relation() as part of an effort to determine which files to include in the backup. So, it already had the information about whether a certain file was a relation file. sendFile() then used a separate method, embodied in is_checksummed_file(), to make what is essentially the same determination. It's better not to make the same decision using two different methods, especially in closely-related code. --- src/backend/backup/basebackup.c | 172 ++++++++++---------------------- 1 file changed, 55 insertions(+), 117 deletions(-) diff --git a/src/backend/backup/basebackup.c b/src/backend/backup/basebackup.c index b537f46219..4ba63ad8a6 100644 --- a/src/backend/backup/basebackup.c +++ b/src/backend/backup/basebackup.c @@ -82,7 +82,8 @@ static int64 sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeo backup_manifest_info *manifest, Oid spcoid); static bool sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, struct stat *statbuf, bool missing_ok, - Oid dboid, Oid spcoid, + Oid dboid, Oid spcoid, RelFileNumber relfilenumber, + unsigned segno, backup_manifest_info *manifest); static off_t read_file_data_into_buffer(bbsink *sink, const char *readfilename, int fd, @@ -104,7 +105,6 @@ static void convert_link_to_directory(const char *pathbuf, struct stat *statbuf) static void perform_base_backup(basebackup_options *opt, bbsink *sink); static void parse_basebackup_options(List *options, basebackup_options *opt); static int compareWalFileNames(const ListCell *a, const ListCell *b); -static bool is_checksummed_file(const char *fullpath, const char *filename); static int basebackup_read_file(int fd, char *buf, size_t nbytes, off_t offset, const char *filename, bool partial_read_ok); @@ -213,23 +213,6 @@ static const struct exclude_list_item excludeFiles[] = {NULL, false} }; -/* - * List of files excluded from checksum validation. - * - * Note: this list should be kept in sync with what pg_checksums.c - * includes. - */ -static const struct exclude_list_item noChecksumFiles[] = { - {"pg_control", false}, - {"pg_filenode.map", false}, - {"pg_internal.init", true}, - {"PG_VERSION", false}, -#ifdef EXEC_BACKEND - {"config_exec_params", true}, -#endif - {NULL, false} -}; - /* * Actually do a base backup for the specified tablespaces. * @@ -356,7 +339,8 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) errmsg("could not stat file \"%s\": %m", XLOG_CONTROL_FILE))); sendFile(sink, XLOG_CONTROL_FILE, XLOG_CONTROL_FILE, &statbuf, - false, InvalidOid, InvalidOid, &manifest); + false, InvalidOid, InvalidOid, + InvalidRelFileNumber, 0, &manifest); } else { @@ -625,7 +609,8 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) errmsg("could not stat file \"%s\": %m", pathbuf))); sendFile(sink, pathbuf, pathbuf, &statbuf, false, - InvalidOid, InvalidOid, &manifest); + InvalidOid, InvalidOid, InvalidRelFileNumber, 0, + &manifest); /* unconditionally mark file as archived */ StatusFilePath(pathbuf, fname, ".done"); @@ -1163,7 +1148,8 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, struct stat statbuf; int64 size = 0; const char *lastDir; /* Split last dir from parent path. */ - bool isDbDir = false; /* Does this directory contain relations? */ + bool isRelationDir = false; /* Does directory contain relations? */ + Oid dboid = InvalidOid; /* * Determine if the current path is a database directory that can contain @@ -1190,17 +1176,23 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, strncmp(lastDir - (sizeof(TABLESPACE_VERSION_DIRECTORY) - 1), TABLESPACE_VERSION_DIRECTORY, sizeof(TABLESPACE_VERSION_DIRECTORY) - 1) == 0)) - isDbDir = true; + { + isRelationDir = true; + dboid = atooid(lastDir + 1); + } } + else if (strcmp(path, "./global") == 0) + isRelationDir = true; dir = AllocateDir(path); while ((de = ReadDir(dir, path)) != NULL) { int excludeIdx; bool excludeFound; - RelFileNumber relNumber; - ForkNumber relForkNum; - unsigned segno; + RelFileNumber relfilenumber = InvalidRelFileNumber; + ForkNumber relForkNum = InvalidForkNumber; + unsigned segno = 0; + bool isRelationFile = false; /* Skip special stuff */ if (strcmp(de->d_name, ".") == 0 || strcmp(de->d_name, "..") == 0) @@ -1248,37 +1240,40 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, if (excludeFound) continue; + /* + * If there could be non-temporary relation files in this directory, + * try to parse the filename. + */ + if (isRelationDir) + isRelationFile = + parse_filename_for_nontemp_relation(de->d_name, + &relfilenumber, + &relForkNum, &segno); + /* Exclude all forks for unlogged tables except the init fork */ - if (isDbDir && - parse_filename_for_nontemp_relation(de->d_name, &relNumber, - &relForkNum, &segno)) + if (isRelationFile && relForkNum != INIT_FORKNUM) { - /* Never exclude init forks */ - if (relForkNum != INIT_FORKNUM) - { - char initForkFile[MAXPGPATH]; + char initForkFile[MAXPGPATH]; - /* - * If any other type of fork, check if there is an init fork - * with the same RelFileNumber. If so, the file can be - * excluded. - */ - snprintf(initForkFile, sizeof(initForkFile), "%s/%u_init", - path, relNumber); + /* + * If any other type of fork, check if there is an init fork with + * the same RelFileNumber. If so, the file can be excluded. + */ + snprintf(initForkFile, sizeof(initForkFile), "%s/%u_init", + path, relfilenumber); - if (lstat(initForkFile, &statbuf) == 0) - { - elog(DEBUG2, - "unlogged relation file \"%s\" excluded from backup", - de->d_name); + if (lstat(initForkFile, &statbuf) == 0) + { + elog(DEBUG2, + "unlogged relation file \"%s\" excluded from backup", + de->d_name); - continue; - } + continue; } } /* Exclude temporary relations */ - if (isDbDir && looks_like_temp_rel_name(de->d_name)) + if (OidIsValid(dboid) && looks_like_temp_rel_name(de->d_name)) { elog(DEBUG2, "temporary relation file \"%s\" excluded from backup", @@ -1417,8 +1412,8 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, if (!sizeonly) sent = sendFile(sink, pathbuf, pathbuf + basepathlen + 1, &statbuf, - true, isDbDir ? atooid(lastDir + 1) : InvalidOid, spcoid, - manifest); + true, dboid, spcoid, + relfilenumber, segno, manifest); if (sent || sizeonly) { @@ -1440,40 +1435,6 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, return size; } -/* - * Check if a file should have its checksum validated. - * We validate checksums on files in regular tablespaces - * (including global and default) only, and in those there - * are some files that are explicitly excluded. - */ -static bool -is_checksummed_file(const char *fullpath, const char *filename) -{ - /* Check that the file is in a tablespace */ - if (strncmp(fullpath, "./global/", 9) == 0 || - strncmp(fullpath, "./base/", 7) == 0 || - strncmp(fullpath, "/", 1) == 0) - { - int excludeIdx; - - /* Compare file against noChecksumFiles skip list */ - for (excludeIdx = 0; noChecksumFiles[excludeIdx].name != NULL; excludeIdx++) - { - int cmplen = strlen(noChecksumFiles[excludeIdx].name); - - if (!noChecksumFiles[excludeIdx].match_prefix) - cmplen++; - if (strncmp(filename, noChecksumFiles[excludeIdx].name, - cmplen) == 0) - return false; - } - - return true; - } - else - return false; -} - /* * Given the member, write the TAR header & send the file. * @@ -1488,6 +1449,7 @@ is_checksummed_file(const char *fullpath, const char *filename) static bool sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, struct stat *statbuf, bool missing_ok, Oid dboid, Oid spcoid, + RelFileNumber relfilenumber, unsigned segno, backup_manifest_info *manifest) { int fd; @@ -1495,8 +1457,6 @@ sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, int checksum_failures = 0; off_t cnt; pgoff_t bytes_done = 0; - int segmentno = 0; - char *segmentpath; bool verify_checksum = false; pg_checksum_context checksum_ctx; @@ -1522,36 +1482,14 @@ sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, */ Assert((sink->bbs_buffer_length % BLCKSZ) == 0); - if (!noverify_checksums && DataChecksumsEnabled()) - { - char *filename; - - /* - * Get the filename (excluding path). As last_dir_separator() - * includes the last directory separator, we chop that off by - * incrementing the pointer. - */ - filename = last_dir_separator(readfilename) + 1; - - if (is_checksummed_file(readfilename, filename)) - { - verify_checksum = true; - - /* - * Cut off at the segment boundary (".") to get the segment number - * in order to mix it into the checksum. - */ - segmentpath = strstr(filename, "."); - if (segmentpath != NULL) - { - segmentno = atoi(segmentpath + 1); - if (segmentno == 0) - ereport(ERROR, - (errmsg("invalid segment number %d in file \"%s\"", - segmentno, filename))); - } - } - } + /* + * If we weren't told not to verify checksums, and if checksums are + * enabled for this cluster, and if this is a relation file, then verify + * the checksum. + */ + if (!noverify_checksums && DataChecksumsEnabled() && + RelFileNumberIsValid(relfilenumber)) + verify_checksum = true; /* * Loop until we read the amount of data the caller told us to expect. The @@ -1566,7 +1504,7 @@ sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, /* Try to read some more data. */ cnt = read_file_data_into_buffer(sink, readfilename, fd, bytes_done, remaining, - blkno + segmentno * RELSEG_SIZE, + blkno + segno * RELSEG_SIZE, verify_checksum, &checksum_failures); -- 2.37.1 (Apple Git-137.1) [application/octet-stream] v3-0005-Prototype-patch-for-incremental-and-differential-.patch (307.7K, ../../CA+Tgmoa_Gq2Qc9dh-WiZEUyF-nk3XULuQR-zFY2vOGNWRCZjkQ@mail.gmail.com/6-v3-0005-Prototype-patch-for-incremental-and-differential-.patch) download | inline diff: From 1a93d452eee57de92d3fc281b1a0bd65f7bb9acb Mon Sep 17 00:00:00 2001 From: Robert Haas <[email protected]> Date: Wed, 14 Jun 2023 11:31:29 -0400 Subject: [PATCH v3 5/7] Prototype patch for incremental and differential backup. We don't differentiate between incremental and differential backups; the term "incremental" as used herein means "either incremental or differential". This adds a new background process, the WAL summarizer, whose behavor is governed by new GUCs wal_summarize_mb and wal_summarize_keep_time. This writes out WAL summary files to $PGDATA/pg_wal/summaries. Each summary file contains information for a certain range of LSNs on a certain TLI. For each relation, it stores a "limit block" which is 0 if a relation is created or destroyed within a certain range of WAL records, or otherwise the shortest length to which the relation was truncated during that range of WAL records, or otherwise InvalidBlockNumber. In addition, it stores any blocks which have been modified during that range of WAL records, but excluding blocks which were removed by truncation after they were modified and which were never modified thereafter. In other words, it tells us which blocks need to copied in case of an incremental backup covering that range of WAL records. To take an incremental backup, you use the new replication command UPLOAD_MANIFEST to upload the manifest for the prior backup. This prior backup could either be a full backup or another incremental backup. You then use BASE_BACKUP with the INCREMENTAL option to take the backup. pg_basebackup now has an --incremental=PATH_TO_MANIFEST option to trigger this behavior. An incremental backup is like a regular full backup except that some relation files are replaced with files with names like INCREMENTAL.${ORIGINAL_NAME}, and the backup_label file contains additional lines identifying it as an incremental backup. The new pg_combinebackup tool can be used to reconstruct a data directory from a full backup and a series of incremental backups. XXX. It would be nice if we could do something about incremental JSON parsing. XXX. This needs a lot of work on documentation and tests. Patch by me. Thanks to Dilip Kumar and Andres Freund for some helpful design discussions. Reviewed by Dilip Kumar and Jakub Wartak. --- src/backend/access/transam/xlog.c | 93 +- src/backend/access/transam/xlogbackup.c | 10 + src/backend/access/transam/xlogrecovery.c | 6 + src/backend/backup/Makefile | 5 +- src/backend/backup/basebackup.c | 334 +++- src/backend/backup/basebackup_incremental.c | 873 ++++++++++ src/backend/backup/meson.build | 3 + src/backend/backup/walsummary.c | 356 +++++ src/backend/backup/walsummaryfuncs.c | 169 ++ src/backend/postmaster/Makefile | 1 + src/backend/postmaster/auxprocess.c | 8 + src/backend/postmaster/meson.build | 1 + src/backend/postmaster/postmaster.c | 53 + src/backend/postmaster/walsummarizer.c | 1414 +++++++++++++++++ src/backend/replication/repl_gram.y | 14 +- src/backend/replication/repl_scanner.l | 2 + src/backend/replication/walsender.c | 162 +- src/backend/storage/ipc/ipci.c | 3 + src/backend/storage/lmgr/lwlocknames.txt | 1 + src/backend/utils/activity/pgstat_io.c | 4 +- .../utils/activity/wait_event_names.txt | 5 + src/backend/utils/init/miscinit.c | 3 + src/backend/utils/misc/guc_tables.c | 29 + src/backend/utils/misc/postgresql.conf.sample | 5 + src/bin/Makefile | 1 + src/bin/initdb/initdb.c | 1 + src/bin/meson.build | 1 + src/bin/pg_basebackup/bbstreamer_file.c | 1 + src/bin/pg_basebackup/pg_basebackup.c | 108 +- src/bin/pg_basebackup/t/010_pg_basebackup.pl | 4 +- src/bin/pg_combinebackup/.gitignore | 1 + src/bin/pg_combinebackup/Makefile | 46 + src/bin/pg_combinebackup/backup_label.c | 281 ++++ src/bin/pg_combinebackup/backup_label.h | 29 + src/bin/pg_combinebackup/copy_file.c | 169 ++ src/bin/pg_combinebackup/copy_file.h | 19 + src/bin/pg_combinebackup/load_manifest.c | 245 +++ src/bin/pg_combinebackup/load_manifest.h | 67 + src/bin/pg_combinebackup/meson.build | 29 + src/bin/pg_combinebackup/pg_combinebackup.c | 1276 +++++++++++++++ src/bin/pg_combinebackup/reconstruct.c | 618 +++++++ src/bin/pg_combinebackup/reconstruct.h | 32 + src/bin/pg_combinebackup/write_manifest.c | 293 ++++ src/bin/pg_combinebackup/write_manifest.h | 33 + src/bin/pg_resetwal/pg_resetwal.c | 36 + src/common/Makefile | 1 + src/common/blkreftable.c | 1309 +++++++++++++++ src/common/meson.build | 1 + src/include/access/xlog.h | 1 + src/include/access/xlogbackup.h | 2 + src/include/backup/basebackup.h | 5 +- src/include/backup/basebackup_incremental.h | 56 + src/include/backup/walsummary.h | 49 + src/include/catalog/pg_proc.dat | 19 + src/include/common/blkreftable.h | 120 ++ src/include/miscadmin.h | 3 + src/include/nodes/replnodes.h | 9 + src/include/postmaster/walsummarizer.h | 31 + src/include/storage/proc.h | 9 +- src/include/utils/guc_tables.h | 1 + src/test/recovery/t/001_stream_rep.pl | 2 + src/test/recovery/t/019_replslot_limit.pl | 3 + .../t/035_standby_logical_decoding.pl | 1 + src/tools/pgindent/typedefs.list | 23 + 64 files changed, 8429 insertions(+), 60 deletions(-) create mode 100644 src/backend/backup/basebackup_incremental.c create mode 100644 src/backend/backup/walsummary.c create mode 100644 src/backend/backup/walsummaryfuncs.c create mode 100644 src/backend/postmaster/walsummarizer.c create mode 100644 src/bin/pg_combinebackup/.gitignore create mode 100644 src/bin/pg_combinebackup/Makefile create mode 100644 src/bin/pg_combinebackup/backup_label.c create mode 100644 src/bin/pg_combinebackup/backup_label.h create mode 100644 src/bin/pg_combinebackup/copy_file.c create mode 100644 src/bin/pg_combinebackup/copy_file.h create mode 100644 src/bin/pg_combinebackup/load_manifest.c create mode 100644 src/bin/pg_combinebackup/load_manifest.h create mode 100644 src/bin/pg_combinebackup/meson.build create mode 100644 src/bin/pg_combinebackup/pg_combinebackup.c create mode 100644 src/bin/pg_combinebackup/reconstruct.c create mode 100644 src/bin/pg_combinebackup/reconstruct.h create mode 100644 src/bin/pg_combinebackup/write_manifest.c create mode 100644 src/bin/pg_combinebackup/write_manifest.h create mode 100644 src/common/blkreftable.c create mode 100644 src/include/backup/basebackup_incremental.h create mode 100644 src/include/backup/walsummary.h create mode 100644 src/include/common/blkreftable.h create mode 100644 src/include/postmaster/walsummarizer.h diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 677a5bf51b..6cfeee63e8 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -77,6 +77,7 @@ #include "port/pg_iovec.h" #include "postmaster/bgwriter.h" #include "postmaster/startup.h" +#include "postmaster/walsummarizer.h" #include "postmaster/walwriter.h" #include "replication/logical.h" #include "replication/origin.h" @@ -3499,6 +3500,43 @@ XLogGetLastRemovedSegno(void) return lastRemovedSegNo; } +/* + * Return the oldest WAL segment on the given TLI that still exists in + * XLOGDIR, or 0 if none. + */ +XLogSegNo +XLogGetOldestSegno(TimeLineID tli) +{ + DIR *xldir; + struct dirent *xlde; + XLogSegNo oldest_segno = 0; + + xldir = AllocateDir(XLOGDIR); + while ((xlde = ReadDir(xldir, XLOGDIR)) != NULL) + { + TimeLineID file_tli; + XLogSegNo file_segno; + + /* Ignore files that are not XLOG segments */ + if (!IsXLogFileName(xlde->d_name)) + continue; + + /* Parse filename to get TLI and segno. */ + XLogFromFileName(xlde->d_name, &file_tli, &file_segno, + wal_segment_size); + + /* Ignore anything that's not from the TLI of interest. */ + if (tli != file_tli) + continue; + + /* If it's the oldest so far, update oldest_segno. */ + if (oldest_segno == 0 || file_segno < oldest_segno) + oldest_segno = file_segno; + } + + FreeDir(xldir); + return oldest_segno; +} /* * Update the last removed segno pointer in shared memory, to reflect that the @@ -3778,8 +3816,8 @@ RemoveXlogFile(const struct dirent *segment_de, } /* - * Verify whether pg_wal and pg_wal/archive_status exist. - * If the latter does not exist, recreate it. + * Verify whether pg_wal, pg_wal/archive_status, and pg_wal/summaries exist. + * If the latter do not exist, recreate them. * * It is not the goal of this function to verify the contents of these * directories, but to help in cases where someone has performed a cluster @@ -3822,6 +3860,26 @@ ValidateXLOGDirectoryStructure(void) (errmsg("could not create missing directory \"%s\": %m", path))); } + + /* Check for summaries */ + snprintf(path, MAXPGPATH, XLOGDIR "/summaries"); + if (stat(path, &stat_buf) == 0) + { + /* Check for weird cases where it exists but isn't a directory */ + if (!S_ISDIR(stat_buf.st_mode)) + ereport(FATAL, + (errmsg("required WAL directory \"%s\" does not exist", + path))); + } + else + { + ereport(LOG, + (errmsg("creating missing WAL directory \"%s\"", path))); + if (MakePGDirectory(path) < 0) + ereport(FATAL, + (errmsg("could not create missing directory \"%s\": %m", + path))); + } } /* @@ -5146,9 +5204,9 @@ StartupXLOG(void) #endif /* - * Verify that pg_wal and pg_wal/archive_status exist. In cases where - * someone has performed a copy for PITR, these directories may have been - * excluded and need to be re-created. + * Verify that pg_wal, pg_wal/archive_status, and pg_wal/summaries exist. + * In cases where someone has performed a copy for PITR, these directories + * may have been excluded and need to be re-created. */ ValidateXLOGDirectoryStructure(); @@ -6829,6 +6887,17 @@ CreateCheckPoint(int flags) */ END_CRIT_SECTION(); + /* + * If there hasn't been much system activity in a while, the WAL + * summarizer may be sleeping for relatively long periods, which could + * delay an incremental backup that has started concurrently. In the hopes + * of avoiding that, poke the WAL summarizer here. + * + * Possibly this should instead be done at some earlier point in this + * function, but it's not clear that it matters much. + */ + SetWalSummarizerLatch(); + /* * Let smgr do post-checkpoint cleanup (eg, deleting old files). */ @@ -7503,6 +7572,20 @@ KeepLogSeg(XLogRecPtr recptr, XLogSegNo *logSegNo) } } + /* + * If WAL summarization is in use, don't remove WAL that has yet to be + * summarized. + */ + keep = GetOldestUnsummarizedLSN(NULL, NULL); + if (keep != InvalidXLogRecPtr) + { + XLogSegNo unsummarized_segno; + + XLByteToSeg(keep, unsummarized_segno, wal_segment_size); + if (unsummarized_segno < segno) + segno = unsummarized_segno; + } + /* but, keep at least wal_keep_size if that's set */ if (wal_keep_size_mb > 0) { diff --git a/src/backend/access/transam/xlogbackup.c b/src/backend/access/transam/xlogbackup.c index 21d68133ae..f51d4282bb 100644 --- a/src/backend/access/transam/xlogbackup.c +++ b/src/backend/access/transam/xlogbackup.c @@ -77,6 +77,16 @@ build_backup_content(BackupState *state, bool ishistoryfile) appendStringInfo(result, "STOP TIMELINE: %u\n", state->stoptli); } + /* either both istartpoint and istarttli should be set, or neither */ + Assert(XLogRecPtrIsInvalid(state->istartpoint) == (state->istarttli == 0)); + if (!XLogRecPtrIsInvalid(state->istartpoint)) + { + appendStringInfo(result, "INCREMENTAL FROM LSN: %X/%X\n", + LSN_FORMAT_ARGS(state->istartpoint)); + appendStringInfo(result, "INCREMENTAL FROM TLI: %u\n", + state->istarttli); + } + data = result->data; pfree(result); diff --git a/src/backend/access/transam/xlogrecovery.c b/src/backend/access/transam/xlogrecovery.c index 5549e1afc5..89ddec5bf9 100644 --- a/src/backend/access/transam/xlogrecovery.c +++ b/src/backend/access/transam/xlogrecovery.c @@ -1284,6 +1284,12 @@ read_backup_label(XLogRecPtr *checkPointLoc, TimeLineID *backupLabelTLI, tli_from_file, BACKUP_LABEL_FILE))); } + if (fscanf(lfp, "INCREMENTAL FROM LSN: %X/%X\n", &hi, &lo) > 0) + ereport(FATAL, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("this is an incremental backup, not a data directory"), + errhint("Use pg_combinebackup to reconstruct a valid data directory."))); + if (ferror(lfp) || FreeFile(lfp)) ereport(FATAL, (errcode_for_file_access(), diff --git a/src/backend/backup/Makefile b/src/backend/backup/Makefile index b21bd8ff43..751e6d3d5e 100644 --- a/src/backend/backup/Makefile +++ b/src/backend/backup/Makefile @@ -19,12 +19,15 @@ OBJS = \ basebackup.o \ basebackup_copy.o \ basebackup_gzip.o \ + basebackup_incremental.o \ basebackup_lz4.o \ basebackup_zstd.o \ basebackup_progress.o \ basebackup_server.o \ basebackup_sink.o \ basebackup_target.o \ - basebackup_throttle.o + basebackup_throttle.o \ + walsummary.o \ + walsummaryfuncs.o include $(top_srcdir)/src/backend/common.mk diff --git a/src/backend/backup/basebackup.c b/src/backend/backup/basebackup.c index 4ba63ad8a6..8a70a9ae41 100644 --- a/src/backend/backup/basebackup.c +++ b/src/backend/backup/basebackup.c @@ -20,8 +20,10 @@ #include "access/xlogbackup.h" #include "backup/backup_manifest.h" #include "backup/basebackup.h" +#include "backup/basebackup_incremental.h" #include "backup/basebackup_sink.h" #include "backup/basebackup_target.h" +#include "catalog/pg_tablespace_d.h" #include "commands/defrem.h" #include "common/compression.h" #include "common/file_perm.h" @@ -64,6 +66,7 @@ typedef struct bool fastcheckpoint; bool nowait; bool includewal; + bool incremental; uint32 maxrate; bool sendtblspcmapfile; bool send_to_client; @@ -76,21 +79,28 @@ typedef struct } basebackup_options; static int64 sendTablespace(bbsink *sink, char *path, Oid spcoid, bool sizeonly, - struct backup_manifest_info *manifest); + struct backup_manifest_info *manifest, + IncrementalBackupInfo *ib); static int64 sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, List *tablespaces, bool sendtblspclinks, - backup_manifest_info *manifest, Oid spcoid); + backup_manifest_info *manifest, Oid spcoid, + IncrementalBackupInfo *ib); static bool sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, struct stat *statbuf, bool missing_ok, Oid dboid, Oid spcoid, RelFileNumber relfilenumber, unsigned segno, - backup_manifest_info *manifest); + backup_manifest_info *manifest, + unsigned num_incremental_blocks, + BlockNumber *incremental_blocks, + unsigned truncation_block_length); static off_t read_file_data_into_buffer(bbsink *sink, const char *readfilename, int fd, off_t offset, size_t length, BlockNumber blkno, bool verify_checksum, int *checksum_failures); +static void push_to_sink(bbsink *sink, pg_checksum_context *checksum_ctx, + size_t *bytes_done, void *data, size_t length); static bool verify_page_checksum(Page page, XLogRecPtr start_lsn, BlockNumber blkno, uint16 *expected_checksum); @@ -102,7 +112,8 @@ static int64 _tarWriteHeader(bbsink *sink, const char *filename, bool sizeonly); static void _tarWritePadding(bbsink *sink, int len); static void convert_link_to_directory(const char *pathbuf, struct stat *statbuf); -static void perform_base_backup(basebackup_options *opt, bbsink *sink); +static void perform_base_backup(basebackup_options *opt, bbsink *sink, + IncrementalBackupInfo *ib); static void parse_basebackup_options(List *options, basebackup_options *opt); static int compareWalFileNames(const ListCell *a, const ListCell *b); static int basebackup_read_file(int fd, char *buf, size_t nbytes, off_t offset, @@ -220,7 +231,8 @@ static const struct exclude_list_item excludeFiles[] = * clobbered by longjmp" from stupider versions of gcc. */ static void -perform_base_backup(basebackup_options *opt, bbsink *sink) +perform_base_backup(basebackup_options *opt, bbsink *sink, + IncrementalBackupInfo *ib) { bbsink_state state; XLogRecPtr endptr; @@ -270,6 +282,10 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) ListCell *lc; tablespaceinfo *newti; + /* If this is an incremental backup, execute preparatory steps. */ + if (ib != NULL) + PrepareForIncrementalBackup(ib, backup_state); + /* Add a node for the base directory at the end */ newti = palloc0(sizeof(tablespaceinfo)); newti->size = -1; @@ -289,10 +305,10 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) if (tmp->path == NULL) tmp->size = sendDir(sink, ".", 1, true, state.tablespaces, - true, NULL, InvalidOid); + true, NULL, InvalidOid, NULL); else tmp->size = sendTablespace(sink, tmp->path, tmp->oid, true, - NULL); + NULL, NULL); state.bytes_total += tmp->size; } state.bytes_total_is_valid = true; @@ -330,7 +346,7 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) /* Then the bulk of the files... */ sendDir(sink, ".", 1, false, state.tablespaces, - sendtblspclinks, &manifest, InvalidOid); + sendtblspclinks, &manifest, InvalidOid, ib); /* ... and pg_control after everything else. */ if (lstat(XLOG_CONTROL_FILE, &statbuf) != 0) @@ -340,7 +356,7 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) XLOG_CONTROL_FILE))); sendFile(sink, XLOG_CONTROL_FILE, XLOG_CONTROL_FILE, &statbuf, false, InvalidOid, InvalidOid, - InvalidRelFileNumber, 0, &manifest); + InvalidRelFileNumber, 0, &manifest, 0, NULL, 0); } else { @@ -348,7 +364,7 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) bbsink_begin_archive(sink, archive_name); - sendTablespace(sink, ti->path, ti->oid, false, &manifest); + sendTablespace(sink, ti->path, ti->oid, false, &manifest, ib); } /* @@ -610,7 +626,7 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) sendFile(sink, pathbuf, pathbuf, &statbuf, false, InvalidOid, InvalidOid, InvalidRelFileNumber, 0, - &manifest); + &manifest, 0, NULL, 0); /* unconditionally mark file as archived */ StatusFilePath(pathbuf, fname, ".done"); @@ -686,6 +702,7 @@ parse_basebackup_options(List *options, basebackup_options *opt) bool o_checkpoint = false; bool o_nowait = false; bool o_wal = false; + bool o_incremental = false; bool o_maxrate = false; bool o_tablespace_map = false; bool o_noverify_checksums = false; @@ -764,6 +781,15 @@ parse_basebackup_options(List *options, basebackup_options *opt) opt->includewal = defGetBoolean(defel); o_wal = true; } + else if (strcmp(defel->defname, "incremental") == 0) + { + if (o_incremental) + ereport(ERROR, + (errcode(ERRCODE_SYNTAX_ERROR), + errmsg("duplicate option \"%s\"", defel->defname))); + opt->incremental = defGetBoolean(defel); + o_incremental = true; + } else if (strcmp(defel->defname, "max_rate") == 0) { int64 maxrate; @@ -956,7 +982,7 @@ parse_basebackup_options(List *options, basebackup_options *opt) * the filesystem, bypassing the buffer cache. */ void -SendBaseBackup(BaseBackupCmd *cmd) +SendBaseBackup(BaseBackupCmd *cmd, IncrementalBackupInfo *ib) { basebackup_options opt; bbsink *sink; @@ -980,6 +1006,20 @@ SendBaseBackup(BaseBackupCmd *cmd) set_ps_display(activitymsg); } + /* + * If we're asked to perform an incremental backup and the user has not + * supplied a manifest, that's an ERROR. + * + * If we're asked to perform a full backup and the user did supply a + * manifest, just ignore it. + */ + if (!opt.incremental) + ib = NULL; + else if (ib == NULL) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("must UPLOAD_MANIFEST before performing an incremental BASE_BACKUP"))); + /* * If the target is specifically 'client' then set up to stream the backup * to the client; otherwise, it's being sent someplace else and should not @@ -1011,7 +1051,7 @@ SendBaseBackup(BaseBackupCmd *cmd) */ PG_TRY(); { - perform_base_backup(&opt, sink); + perform_base_backup(&opt, sink, ib); } PG_FINALLY(); { @@ -1086,7 +1126,7 @@ sendFileWithContent(bbsink *sink, const char *filename, const char *content, */ static int64 sendTablespace(bbsink *sink, char *path, Oid spcoid, bool sizeonly, - backup_manifest_info *manifest) + backup_manifest_info *manifest, IncrementalBackupInfo *ib) { int64 size; char pathbuf[MAXPGPATH]; @@ -1120,7 +1160,7 @@ sendTablespace(bbsink *sink, char *path, Oid spcoid, bool sizeonly, /* Send all the files in the tablespace version directory */ size += sendDir(sink, pathbuf, strlen(path), sizeonly, NIL, true, manifest, - spcoid); + spcoid, ib); return size; } @@ -1140,7 +1180,7 @@ sendTablespace(bbsink *sink, char *path, Oid spcoid, bool sizeonly, static int64 sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, List *tablespaces, bool sendtblspclinks, backup_manifest_info *manifest, - Oid spcoid) + Oid spcoid, IncrementalBackupInfo *ib) { DIR *dir; struct dirent *de; @@ -1149,7 +1189,16 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, int64 size = 0; const char *lastDir; /* Split last dir from parent path. */ bool isRelationDir = false; /* Does directory contain relations? */ + bool isGlobalDir = false; Oid dboid = InvalidOid; + BlockNumber *relative_block_numbers = NULL; + + /* + * Since this array is relatively large, avoid putting it on the stack. + * But we don't need it at all if this is not an incremental backup. + */ + if (ib != NULL) + relative_block_numbers = palloc(sizeof(BlockNumber) * RELSEG_SIZE); /* * Determine if the current path is a database directory that can contain @@ -1182,7 +1231,10 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, } } else if (strcmp(path, "./global") == 0) + { isRelationDir = true; + isGlobalDir = true; + } dir = AllocateDir(path); while ((de = ReadDir(dir, path)) != NULL) @@ -1331,11 +1383,13 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, &statbuf, sizeonly); /* - * Also send archive_status directory (by hackishly reusing - * statbuf from above ...). + * Also send archive_status and summaries directories (by + * hackishly reusing statbuf from above ...). */ size += _tarWriteHeader(sink, "./pg_wal/archive_status", NULL, &statbuf, sizeonly); + size += _tarWriteHeader(sink, "./pg_wal/summaries", NULL, + &statbuf, sizeonly); continue; /* don't recurse into pg_wal */ } @@ -1404,33 +1458,88 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, if (!skip_this_dir) size += sendDir(sink, pathbuf, basepathlen, sizeonly, tablespaces, - sendtblspclinks, manifest, spcoid); + sendtblspclinks, manifest, spcoid, ib); } else if (S_ISREG(statbuf.st_mode)) { bool sent = false; + unsigned num_blocks_required = 0; + unsigned truncation_block_length = 0; + char tarfilenamebuf[MAXPGPATH * 2]; + char *tarfilename = pathbuf + basepathlen + 1; + FileBackupMethod method = BACK_UP_FILE_FULLY; - if (!sizeonly) - sent = sendFile(sink, pathbuf, pathbuf + basepathlen + 1, &statbuf, - true, dboid, spcoid, - relfilenumber, segno, manifest); + if (ib != NULL && isRelationFile) + { + Oid relspcoid; + char *lookup_path; - if (sent || sizeonly) + if (OidIsValid(spcoid)) + { + relspcoid = spcoid; + lookup_path = psprintf("pg_tblspc/%u/%s", spcoid, + pathbuf + basepathlen + 1); + } + else + { + if (isGlobalDir) + relspcoid = GLOBALTABLESPACE_OID; + else + relspcoid = DEFAULTTABLESPACE_OID; + lookup_path = pstrdup(pathbuf + basepathlen + 1); + } + + method = GetFileBackupMethod(ib, lookup_path, dboid, relspcoid, + relfilenumber, relForkNum, + segno, statbuf.st_size, + &num_blocks_required, + relative_block_numbers, + &truncation_block_length); + if (method == BACK_UP_FILE_INCREMENTALLY) + { + statbuf.st_size = + GetIncrementalFileSize(num_blocks_required); + snprintf(tarfilenamebuf, sizeof(tarfilenamebuf), + "%s/INCREMENTAL.%s", + path + basepathlen + 1, + de->d_name); + tarfilename = tarfilenamebuf; + } + + pfree(lookup_path); + } + + if (method != DO_NOT_BACK_UP_FILE) { - /* Add size. */ - size += statbuf.st_size; + if (!sizeonly) + sent = sendFile(sink, pathbuf, tarfilename, &statbuf, + true, dboid, spcoid, + relfilenumber, segno, manifest, + num_blocks_required, + method == BACK_UP_FILE_INCREMENTALLY ? relative_block_numbers : NULL, + truncation_block_length); + + if (sent || sizeonly) + { + /* Add size. */ + size += statbuf.st_size; - /* Pad to a multiple of the tar block size. */ - size += tarPaddingBytesRequired(statbuf.st_size); + /* Pad to a multiple of the tar block size. */ + size += tarPaddingBytesRequired(statbuf.st_size); - /* Size of the header for the file. */ - size += TAR_BLOCK_SIZE; + /* Size of the header for the file. */ + size += TAR_BLOCK_SIZE; + } } } else ereport(WARNING, (errmsg("skipping special file \"%s\"", pathbuf))); } + + if (relative_block_numbers != NULL) + pfree(relative_block_numbers); + FreeDir(dir); return size; } @@ -1443,6 +1552,12 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, * If dboid is anything other than InvalidOid then any checksum failures * detected will get reported to the cumulative stats system. * + * If the file is to be sent incrementally, then num_incremental_blocks + * should be the number of blocks to be sent, and incremental_blocks + * an array of block numbers relative to the start of the current segment. + * If the whole file is to be sent, then incremental_blocks should be NULL, + * and num_incremental_blocks can have any value, as it will be ignored. + * * Returns true if the file was successfully sent, false if 'missing_ok', * and the file did not exist. */ @@ -1450,7 +1565,8 @@ static bool sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, struct stat *statbuf, bool missing_ok, Oid dboid, Oid spcoid, RelFileNumber relfilenumber, unsigned segno, - backup_manifest_info *manifest) + backup_manifest_info *manifest, unsigned num_incremental_blocks, + BlockNumber *incremental_blocks, unsigned truncation_block_length) { int fd; BlockNumber blkno = 0; @@ -1459,6 +1575,7 @@ sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, pgoff_t bytes_done = 0; bool verify_checksum = false; pg_checksum_context checksum_ctx; + int ibindex = 0; if (pg_checksum_init(&checksum_ctx, manifest->checksum_type) < 0) elog(ERROR, "could not initialize checksum of file \"%s\"", @@ -1491,22 +1608,111 @@ sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, RelFileNumberIsValid(relfilenumber)) verify_checksum = true; + /* + * If we're sending an incremental file, write the file header. + */ + if (incremental_blocks != NULL) + { + unsigned magic = INCREMENTAL_MAGIC; + size_t header_bytes_done = 0; + + /* Emit header data. */ + push_to_sink(sink, &checksum_ctx, &header_bytes_done, + &magic, sizeof(magic)); + push_to_sink(sink, &checksum_ctx, &header_bytes_done, + &num_incremental_blocks, sizeof(num_incremental_blocks)); + push_to_sink(sink, &checksum_ctx, &header_bytes_done, + &truncation_block_length, sizeof(truncation_block_length)); + push_to_sink(sink, &checksum_ctx, &header_bytes_done, + incremental_blocks, + sizeof(BlockNumber) * num_incremental_blocks); + + /* Flush out any data still in the buffer so it's again empty. */ + if (header_bytes_done > 0) + { + bbsink_archive_contents(sink, header_bytes_done); + if (pg_checksum_update(&checksum_ctx, + (uint8 *) sink->bbs_buffer, + header_bytes_done) < 0) + elog(ERROR, "could not update checksum of base backup"); + } + + /* Update our notion of file position. */ + bytes_done += sizeof(magic); + bytes_done += sizeof(num_incremental_blocks); + bytes_done += sizeof(truncation_block_length); + bytes_done += sizeof(BlockNumber) * num_incremental_blocks; + } + /* * Loop until we read the amount of data the caller told us to expect. The * file could be longer, if it was extended while we were sending it, but * for a base backup we can ignore such extended data. It will be restored * from WAL. */ - while (bytes_done < statbuf->st_size) + while (1) { - size_t remaining = statbuf->st_size - bytes_done; + /* + * Determine whether we've read all the data that we need, and if not, + * read some more. + */ + if (incremental_blocks == NULL) + { + size_t remaining = statbuf->st_size - bytes_done; + + /* + * If we've read the required number of bytes, then it's time to + * stop. + */ + if (bytes_done >= statbuf->st_size) + break; + + /* + * Read as many bytes as will fit in the buffer, or however many + * are left to read, whichever is less. + */ + cnt = read_file_data_into_buffer(sink, readfilename, fd, + bytes_done, remaining, + blkno + segno * RELSEG_SIZE, + verify_checksum, + &checksum_failures); + } + else + { + BlockNumber relative_blkno; - /* Try to read some more data. */ - cnt = read_file_data_into_buffer(sink, readfilename, fd, bytes_done, - remaining, - blkno + segno * RELSEG_SIZE, - verify_checksum, - &checksum_failures); + /* + * If we've read all the blocks, then it's time to stop. + */ + if (ibindex >= num_incremental_blocks) + break; + + /* + * Read just one block, whichever one is the next that we're + * supposed to include. + */ + relative_blkno = incremental_blocks[ibindex++]; + cnt = read_file_data_into_buffer(sink, readfilename, fd, + relative_blkno * BLCKSZ, + BLCKSZ, + relative_blkno + segno * RELSEG_SIZE, + verify_checksum, + &checksum_failures); + + /* + * If we get a partial read, that must mean that the relation is + * being truncated. Ultimately, it should be truncated to a + * multiple of BLCKSZ, since this path should only be reached for + * relation files, but we might transiently observe an + * intermediate value. + * + * It should be fine to treat this just as if the entire block had + * been truncated away - i.e. fill this and all later blocks with + * zeroes. WAL replay will fix things up. + */ + if (cnt < BLCKSZ) + break; + } /* * If the amount of data we were able to read was not a multiple of @@ -1689,6 +1895,56 @@ read_file_data_into_buffer(bbsink *sink, const char *readfilename, int fd, return cnt; } +/* + * Push data into a bbsink. + * + * It's better, when possible, to read data directly into the bbsink's buffer, + * rather than using this function to copy it into the buffer; this function is + * for cases where that approach is not practical. + * + * bytes_done should point to a count of the number of bytes that are + * currently used in the bbsink's buffer. Upon return, the bytes identified by + * data and length will have been copied into the bbsink's buffer, flushing + * as required, and *bytes_done will have been updated accordingly. If the + * buffer was flushed, the previous contents will also have been fed to + * checksum_ctx. + * + * Note that after one or more calls to this function it is the caller's + * responsibility to perform any required final flush. + */ +static void +push_to_sink(bbsink *sink, pg_checksum_context *checksum_ctx, + size_t *bytes_done, void *data, size_t length) +{ + while (length > 0) + { + size_t bytes_to_copy; + + /* + * We use < here rather than <= so that if the data exactly fills the + * remaining buffer space, we trigger a flush now. + */ + if (length < sink->bbs_buffer_length - *bytes_done) + { + /* Append remaining data to buffer. */ + memcpy(sink->bbs_buffer + *bytes_done, data, length); + *bytes_done += length; + return; + } + + /* Copy until buffer is full and flush it. */ + bytes_to_copy = sink->bbs_buffer_length - *bytes_done; + memcpy(sink->bbs_buffer + *bytes_done, data, bytes_to_copy); + data = ((char *) data) + bytes_to_copy; + length -= bytes_to_copy; + bbsink_archive_contents(sink, sink->bbs_buffer_length); + if (pg_checksum_update(checksum_ctx, (uint8 *) sink->bbs_buffer, + sink->bbs_buffer_length) < 0) + elog(ERROR, "could not update checksum"); + *bytes_done = 0; + } +} + /* * Try to verify the checksum for the provided page, if it seems appropriate * to do so. diff --git a/src/backend/backup/basebackup_incremental.c b/src/backend/backup/basebackup_incremental.c new file mode 100644 index 0000000000..20cc00bded --- /dev/null +++ b/src/backend/backup/basebackup_incremental.c @@ -0,0 +1,873 @@ +/*------------------------------------------------------------------------- + * + * basebackup_incremental.c + * code for incremental backup support + * + * This code isn't actually in charge of taking an incremental backup; + * the actual construction of the incremental backup happens in + * basebackup.c. Here, we're concerned with providing the necessary + * supports for that operation. In particular, we need to parse the + * backup manifest supplied by the user taking the incremental backup + * and extract the required information from it. + * + * Portions Copyright (c) 2010-2022, PostgreSQL Global Development Group + * + * IDENTIFICATION + * src/backend/backup/basebackup_incremental.c + * + *------------------------------------------------------------------------- + */ +#include "postgres.h" + +#include "access/timeline.h" +#include "access/xlog.h" +#include "access/xlogrecovery.h" +#include "backup/basebackup_incremental.h" +#include "backup/walsummary.h" +#include "common/blkreftable.h" +#include "common/parse_manifest.h" +#include "common/hashfn.h" +#include "postmaster/walsummarizer.h" + +#define BLOCKS_PER_READ 512 + +/* + * Details extracted from the WAL ranges present in the supplied backup manifest. + */ +typedef struct +{ + TimeLineID tli; + XLogRecPtr start_lsn; + XLogRecPtr end_lsn; +} backup_wal_range; + +/* + * Details extracted from the file list present in the supplied backup manifest. + */ +typedef struct +{ + uint32 status; + const char *path; + size_t size; +} backup_file_entry; + +static uint32 hash_string_pointer(const char *s); +#define SH_PREFIX backup_file +#define SH_ELEMENT_TYPE backup_file_entry +#define SH_KEY_TYPE const char * +#define SH_KEY path +#define SH_HASH_KEY(tb, key) hash_string_pointer(key) +#define SH_EQUAL(tb, a, b) (strcmp(a, b) == 0) +#define SH_SCOPE static inline +#define SH_DECLARE +#define SH_DEFINE +#include "lib/simplehash.h" + +struct IncrementalBackupInfo +{ + /* Memory context for this object and its subsidiary objects. */ + MemoryContext mcxt; + + /* Temporary buffer for storing the manifest while parsing it. */ + StringInfoData buf; + + /* WAL ranges extracted from the backup manifest. */ + List *manifest_wal_ranges; + + /* + * Files extracted from the backup manifest. + * + * We don't really need this information, because we use WAL summaries to + * figure what's changed. It would be unsafe to just rely on the list of + * files that existed before, because it's possible for a file to be + * removed and a new one created with the same name and different + * contents. In such cases, the whole file must still be sent. We can tell + * from the WAL summaries whether that happened, but not from the file + * list. + * + * Nonetheless, this data is useful for sanity checking. If a file that we + * think we shouldn't need to send is not present in the manifest for the + * prior backup, something has gone terribly wrong. We retain the file + * names and sizes, but not the checksums or last modified times, for + * which we have no use. + * + * One significant downside of storing this data is that it consumes + * memory. If that turns out to be a problem, we might have to decide not + * to retain this information, or to make it optional. + */ + backup_file_hash *manifest_files; + + /* + * Block-reference table for the incremental backup. + * + * It's possible that storing the entire block-reference table in memory + * will be a problem for some users. The in-memory format that we're using + * here is pretty efficient, converging to little more than 1 bit per + * block for relation forks with large numbers of modified blocks. It's + * possible, however, that if you try to perform an incremental backup of + * a database with a sufficiently large number of relations on a + * sufficiently small machine, you could run out of memory here. If that + * turns out to be a problem in practice, we'll need to be more clever. + */ + BlockRefTable *brtab; +}; + +static void manifest_process_file(JsonManifestParseContext *, + char *pathname, + size_t size, + pg_checksum_type checksum_type, + int checksum_length, + uint8 *checksum_payload); +static void manifest_process_wal_range(JsonManifestParseContext *, + TimeLineID tli, + XLogRecPtr start_lsn, + XLogRecPtr end_lsn); +static void manifest_report_error(JsonManifestParseContext *ib, + const char *fmt,...) + pg_attribute_printf(2, 3) pg_attribute_noreturn(); + +/* + * Create a new object for storing information extracted from the manifest + * supplied when creating an incremental backup. + */ +IncrementalBackupInfo * +CreateIncrementalBackupInfo(MemoryContext mcxt) +{ + IncrementalBackupInfo *ib; + MemoryContext oldcontext; + + oldcontext = MemoryContextSwitchTo(mcxt); + + ib = palloc0(sizeof(IncrementalBackupInfo)); + ib->mcxt = mcxt; + initStringInfo(&ib->buf); + + /* + * It's hard to guess how many files a "typical" installation will have in + * the data directory, but a fresh initdb creates almost 1000 files as of + * this writing, so it seems to make sense for our estimate to + * substantially higher. + */ + ib->manifest_files = backup_file_create(mcxt, 10000, NULL); + + MemoryContextSwitchTo(oldcontext); + + return ib; +} + +/* + * Before taking an incremental backup, the caller must supply the backup + * manifest from a prior backup. Each chunk of manifest data recieved + * from the client should be passed to this function. + */ +void +AppendIncrementalManifestData(IncrementalBackupInfo *ib, const char *data, + int len) +{ + MemoryContext oldcontext; + + /* Switch to our memory context. */ + oldcontext = MemoryContextSwitchTo(ib->mcxt); + + /* + * XXX. Our json parser is at present incapable of parsing json blobs + * incrementally, so we have to accumulate the entire backup manifest + * before we can do anything with it. This should really be fixed, since + * some users might have very large numbers of files in the data + * directory. + */ + appendBinaryStringInfo(&ib->buf, data, len); + + /* Switch back to previous memory context. */ + MemoryContextSwitchTo(oldcontext); +} + +/* + * Finalize an IncrementalBackupInfo object after all manifest data has + * been supplied via calls to AppendIncrementalManifestData. + */ +void +FinalizeIncrementalManifest(IncrementalBackupInfo *ib) +{ + JsonManifestParseContext context; + MemoryContext oldcontext; + + /* Switch to our memory context. */ + oldcontext = MemoryContextSwitchTo(ib->mcxt); + + /* Parse the manifest. */ + context.private_data = ib; + context.perfile_cb = manifest_process_file; + context.perwalrange_cb = manifest_process_wal_range; + context.error_cb = manifest_report_error; + json_parse_manifest(&context, ib->buf.data, ib->buf.len); + + /* Done with the buffer, so release memory. */ + pfree(ib->buf.data); + ib->buf.data = NULL; + + /* Switch back to previous memory context. */ + MemoryContextSwitchTo(oldcontext); +} + +/* + * Prepare to take an incremental backup. + * + * Before this function is called, AppendIncrementalManifestData and + * FinalizeIncrementalManifest should have already been called to pass all + * the manifest data to this object. + * + * This function performs sanity checks on the data extracted from the + * manifest and figures out for which WAL ranges we need summaries, and + * whether those summaries are available. Then, it reads and combines the + * data from those summary files. It also updates the backup_state with the + * reference TLI and LSN for the prior backup. + */ +void +PrepareForIncrementalBackup(IncrementalBackupInfo *ib, + BackupState *backup_state) +{ + MemoryContext oldcontext; + List *expectedTLEs; + List *all_wslist, + *required_wslist = NIL; + ListCell *lc; + TimeLineHistoryEntry **tlep; + int num_wal_ranges; + int i; + bool found_backup_start_tli = false; + TimeLineID earliest_wal_range_tli = 0; + XLogRecPtr earliest_wal_range_start_lsn; + TimeLineID latest_wal_range_tli = 0; + XLogRecPtr summarized_lsn; + + Assert(ib->buf.data == NULL); + + /* Switch to our memory context. */ + oldcontext = MemoryContextSwitchTo(ib->mcxt); + + /* + * Match up the TLIs that appear in the WAL ranges of the backup manifest + * with those that appear in this server's timeline history. We expect + * every backup_wal_range to match to a TimeLineHistoryEntry; if it does + * not, that's an error. + * + * This loop also decides which of the WAL ranges is the manifest is most + * ancient and which one is the newest, according to the timeline history + * of this server, and stores TLIs of those WAL ranges into + * earliest_wal_range_tli and latest_wal_range_tli. It also updates + * earliest_wal_range_start_lsn to the start LSN of the WAL range for + * earliest_wal_range_tli. + * + * Note that the return value of readTimeLineHistory puts the latest + * timeline at the beginning of the list, not the end. Hence, the earliest + * TLI is the one that occurs nearest the end of the list returned by + * readTimeLineHistory, and the latest TLI is the one that occurs closest + * to the beginning. + */ + expectedTLEs = readTimeLineHistory(backup_state->starttli); + num_wal_ranges = list_length(ib->manifest_wal_ranges); + tlep = palloc0(num_wal_ranges * sizeof(TimeLineHistoryEntry *)); + for (i = 0; i < num_wal_ranges; ++i) + { + backup_wal_range *range = list_nth(ib->manifest_wal_ranges, i); + bool saw_earliest_wal_range_tli = false; + bool saw_latest_wal_range_tli = false; + + /* Search this server's history for this WAL range's TLI. */ + foreach(lc, expectedTLEs) + { + TimeLineHistoryEntry *tle = lfirst(lc); + + if (tle->tli == range->tli) + { + tlep[i] = tle; + break; + } + + if (tle->tli == earliest_wal_range_tli) + saw_earliest_wal_range_tli = true; + if (tle->tli == latest_wal_range_tli) + saw_latest_wal_range_tli = true; + } + + /* + * An incremental backup can only be taken relative to a backup that + * represents a previous state of this server. If the backup requires + * WAL from a timeline that's not in our history, that definitely + * isn't the case. + */ + if (tlep[i] == NULL) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("timeline %u found in manifest, but not in this server's history", + range->tli))); + + /* + * If we found this TLI in the server's history before encountering + * the latest TLI seen so far in the server's history, then this TLI + * is the latest one seen so far. + * + * If on the other hand we saw the earliest TLI seen so far before + * finding this TLI, this TLI is earlier than the earliest one seen so + * far. And if this is the first TLI for which we've searched, it's + * also the earliest one seen so far. + * + * On the first loop iteration, both things should necessarily be + * true. + */ + if (!saw_latest_wal_range_tli) + latest_wal_range_tli = range->tli; + if (earliest_wal_range_tli == 0 || saw_earliest_wal_range_tli) + { + earliest_wal_range_tli = range->tli; + earliest_wal_range_start_lsn = range->start_lsn; + } + } + + /* + * Propagate information about the prior backup into the backup_label that + * will be generated for this backup. + */ + backup_state->istartpoint = earliest_wal_range_start_lsn; + backup_state->istarttli = earliest_wal_range_tli; + + /* + * Sanity check start and end LSNs for the WAL ranges in the manifest. + * + * Commonly, there won't be any timeline switches during the prior backup + * at all, but if there are, they should happen at the same LSNs that this + * server switched timelines. + * + * Whether there are any timeline switches during the prior backup or not, + * the prior backup shouldn't require any WAL from a timeline prior to the + * start of that timeline. It also shouldn't require any WAL from later + * than the start of this backup. + * + * If any of these sanity checks fail, one possible explanation is that + * the user has generated WAL on the same timeline with the same LSNs more + * than once. For instance, if two standbys running on timeline 1 were + * both promoted and (due to a broken archiving setup) both selected new + * timeline ID 2, then it's possible that one of these checks might trip. + * + * Note that there are lots of ways for the user to do something very bad + * without tripping any of these checks, and they are not intended to be + * comprehensive. It's pretty hard to see how we could be certain of + * anything here. However, if there's a problem staring us right in the + * face, it's best to report it, so we do. + */ + for (i = 0; i < num_wal_ranges; ++i) + { + backup_wal_range *range = list_nth(ib->manifest_wal_ranges, i); + + if (range->tli == earliest_wal_range_tli) + { + if (range->start_lsn < tlep[i]->begin) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("manifest requires WAL from initial timeline %u starting at %X/%X, but that timeline begins at %X/%X", + range->tli, + LSN_FORMAT_ARGS(range->start_lsn), + LSN_FORMAT_ARGS(tlep[i]->begin)))); + } + else + { + if (range->start_lsn != tlep[i]->begin) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("manifest requires WAL from continuation timeline %u starting at %X/%X, but that timeline begins at %X/%X", + range->tli, + LSN_FORMAT_ARGS(range->start_lsn), + LSN_FORMAT_ARGS(tlep[i]->begin)))); + } + + if (range->tli == latest_wal_range_tli) + { + if (range->end_lsn > backup_state->startpoint) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("manifest requires WAL from final timeline %u ending at %X/%X, but this backup starts at %X/%X", + range->tli, + LSN_FORMAT_ARGS(range->end_lsn), + LSN_FORMAT_ARGS(backup_state->startpoint)))); + } + else + { + if (range->end_lsn != tlep[i]->end) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("manifest requires WAL from non-final timeline %u ending at %X/%X, but this server switched timelines at %X/%X", + range->tli, + LSN_FORMAT_ARGS(range->end_lsn), + LSN_FORMAT_ARGS(tlep[i]->end)))); + } + + } + + /* + * Wait for WAL summarization to catch up to the backup start LSN (but + * time out if it doesn't do so quickly enough). + */ + /* XXX make timeout configurable */ + summarized_lsn = WaitForWalSummarization(backup_state->startpoint, 60000); + if (summarized_lsn < backup_state->startpoint) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("timeout waiting for WAL summarization"), + errdetail("This backup requires WAL to be summarized up to %X/%X, but summarizer has only reached %X/%X.", + LSN_FORMAT_ARGS(backup_state->startpoint), + LSN_FORMAT_ARGS(summarized_lsn)))); + + /* + * Retrieve a list of all WAL summaries on any timeline that overlap with + * the LSN range of interest. We could instead call GetWalSummaries() once + * per timeline in the loop that follows, but that would involve reading + * the directory multiple times. It should be mildly faster - and perhaps + * a bit safer - to do it just once. + */ + all_wslist = GetWalSummaries(0, earliest_wal_range_start_lsn, + backup_state->startpoint); + + /* + * We need WAL summaries for everything that happened during the prior + * backup and everything that happened afterward up until the point where + * the current backup started. + */ + foreach(lc, expectedTLEs) + { + TimeLineHistoryEntry *tle = lfirst(lc); + XLogRecPtr tli_start_lsn = tle->begin; + XLogRecPtr tli_end_lsn = tle->end; + XLogRecPtr tli_missing_lsn = InvalidXLogRecPtr; + List *tli_wslist; + + /* + * Working through the history of this server from the current + * timeline backwards, we skip everything until we find the timeline + * where this backup started. Most of the time, this means we won't + * skip anything at all, as it's unlikely that the timeline has + * changed since the beginning of the backup moments ago. + */ + if (tle->tli == backup_state->starttli) + { + found_backup_start_tli = true; + tli_end_lsn = backup_state->startpoint; + } + else if (!found_backup_start_tli) + continue; + + /* + * Find the summaries that overlap the LSN range of interest for this + * timeline. If this is the earliest timeline involved, the range of + * interest begins with the start LSN of the prior backup; otherwise, + * it begins at the LSN at which this timeline came into existence. If + * this is the latest TLI involved, the range of interest ends at the + * start LSN of the current backup; otherwise, it ends at the point + * where we switched from this timeline to the next one. + */ + if (tle->tli == earliest_wal_range_tli) + tli_start_lsn = earliest_wal_range_start_lsn; + tli_wslist = FilterWalSummaries(all_wslist, tle->tli, + tli_start_lsn, tli_end_lsn); + + /* + * There is no guarantee that the WAL summaries we found cover the + * entire range of LSNs for which summaries are required, or indeed + * that we found any WAL summaries at all. Check whether we have a + * problem of that sort. + */ + if (!WalSummariesAreComplete(tli_wslist, tli_start_lsn, tli_end_lsn, + &tli_missing_lsn)) + { + if (XLogRecPtrIsInvalid(tli_missing_lsn)) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("WAL summaries are required on timeline %u from %X/%X to %X/%X, but no summaries for that timeline and LSN range exist", + tle->tli, + LSN_FORMAT_ARGS(tli_start_lsn), + LSN_FORMAT_ARGS(tli_end_lsn)))); + else + ereport(ERROR, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("WAL summaries are required on timeline %u from %X/%X to %X/%X, but the summaries for that timeline and LSN range are incomplete", + tle->tli, + LSN_FORMAT_ARGS(tli_start_lsn), + LSN_FORMAT_ARGS(tli_end_lsn)), + errdetail("The first unsummarized LSN is this range is %X/%X.", + LSN_FORMAT_ARGS(tli_missing_lsn)))); + } + + /* + * Remember that we need to read these summaries. + * + * Technically, it's possible that this could read more files than + * required, since tli_wslist in theory could contain redundant + * summaries. For instance, if we have a summary from 0/10000000 to + * 0/20000000 and also one from 0/00000000 to 0/30000000, then the + * latter subsumes the former and the former could be ignored. + * + * We ignore this possibility because the WAL summarizer only tries to + * generate summaries that do not overlap. If somehow they exist, + * we'll do a bit of extra work but the results should still be + * correct. + */ + required_wslist = list_concat(required_wslist, tli_wslist); + + /* + * Timelines earlier than the one in which the prior backup began are + * not relevant. + */ + if (tle->tli == earliest_wal_range_tli) + break; + } + + /* + * Read all of the required block reference table files and merge all of + * the data into a single in-memory block reference table. + * + * See the comments for struct IncrementalBackupInfo for some thoughts on + * memory usage. + */ + ib->brtab = CreateEmptyBlockRefTable(); + foreach(lc, required_wslist) + { + WalSummaryFile *ws = lfirst(lc); + WalSummaryIO wsio; + BlockRefTableReader *reader; + RelFileLocator rlocator; + ForkNumber forknum; + BlockNumber limit_block; + BlockNumber blocks[BLOCKS_PER_READ]; + + wsio.file = OpenWalSummaryFile(ws, false); + wsio.filepos = 0; + ereport(DEBUG1, + (errmsg_internal("reading WAL summary file \"%s\"", + FilePathName(wsio.file)))); + reader = CreateBlockRefTableReader(ReadWalSummary, &wsio, + FilePathName(wsio.file), + ReportWalSummaryError, NULL); + while (BlockRefTableReaderNextRelation(reader, &rlocator, &forknum, + &limit_block)) + { + BlockRefTableSetLimitBlock(ib->brtab, &rlocator, + forknum, limit_block); + + while (1) + { + unsigned nblocks; + unsigned i; + + nblocks = BlockRefTableReaderGetBlocks(reader, blocks, + BLOCKS_PER_READ); + if (nblocks == 0) + break; + + for (i = 0; i < nblocks; ++i) + BlockRefTableMarkBlockModified(ib->brtab, &rlocator, + forknum, blocks[i]); + } + } + DestroyBlockRefTableReader(reader); + FileClose(wsio.file); + } + + /* Switch back to previous memory context. */ + MemoryContextSwitchTo(oldcontext); +} + +/* + * Get the pathname that should be used when a file is sent incrementally. + * + * The result is a palloc'd string. + */ +char * +GetIncrementalFilePath(Oid dboid, Oid spcoid, RelFileNumber relfilenumber, + ForkNumber forknum, unsigned segno) +{ + char *path; + char *lastslash; + char *ipath; + + path = GetRelationPath(dboid, spcoid, relfilenumber, InvalidBackendId, + forknum); + + lastslash = strrchr(path, '/'); + Assert(lastslash != NULL); + *lastslash = '\0'; + + if (segno > 0) + ipath = psprintf("%s/INCREMENTAL.%s.%u", path, lastslash + 1, segno); + else + ipath = psprintf("%s/INCREMENTAL.%s", path, lastslash + 1); + + pfree(path); + + return ipath; +} + +/* + * How should we back up a particular file as part of an incremental backup? + * + * If the return value is BACK_UP_FILE_FULLY, caller should back up the whole + * file just as if this were not an incremental backup. + * + * If the return value is BACK_UP_FILE_INCREMENTALLY, caller should include + * an incremental file in the backup instead of the entire file. On return, + * *num_blocks_required will be set to the number of blocks that need to be + * sent, and the actual block numbers will have been stored in + * relative_block_numbers, which should be an array of at least RELSEG_SIZE. + * In addition, *truncation_block_length will be set to the value that should + * be included in the incremental file. + * + * If the return value is DO_NOT_BACK_UP_FILE, the caller should not include + * the file in the backup at all. + */ +FileBackupMethod +GetFileBackupMethod(IncrementalBackupInfo *ib, char *path, + Oid dboid, Oid spcoid, + RelFileNumber relfilenumber, ForkNumber forknum, + unsigned segno, size_t size, + unsigned *num_blocks_required, + BlockNumber *relative_block_numbers, + unsigned *truncation_block_length) +{ + BlockNumber absolute_block_numbers[RELSEG_SIZE]; + BlockNumber limit_block; + BlockNumber start_blkno; + BlockNumber stop_blkno; + RelFileLocator rlocator; + BlockRefTableEntry *brtentry; + unsigned i; + unsigned nblocks; + + /* Should only be called after PrepareForIncrementalBackup. */ + Assert(ib->buf.data == NULL); + + /* + * dboid could be InvalidOid if shared rel, but spcoid and relfilenumber + * should have legal values. + */ + Assert(OidIsValid(spcoid)); + Assert(RelFileNumberIsValid(relfilenumber)); + + /* + * If the file size is too large or not a multiple of BLCKSZ, then + * something weird is happening, so give up and send the whole file. + */ + if ((size % BLCKSZ) != 0 || size / BLCKSZ > RELSEG_SIZE) + return BACK_UP_FILE_FULLY; + + /* + * The free-space map fork is not properly WAL-logged, so we need to + * backup the entire file every time. + */ + if (forknum == FSM_FORKNUM) + return BACK_UP_FILE_FULLY; + + /* + * Check whether this file is part of the prior backup. If it isn't, back + * up the whole file. + */ + if (backup_file_lookup(ib->manifest_files, path) == NULL) + { + char *ipath; + + ipath = GetIncrementalFilePath(dboid, spcoid, relfilenumber, + forknum, segno); + if (backup_file_lookup(ib->manifest_files, ipath) == NULL) + return BACK_UP_FILE_FULLY; + } + + /* Look up the block reference table entry. */ + rlocator.spcOid = spcoid; + rlocator.dbOid = dboid; + rlocator.relNumber = relfilenumber; + brtentry = BlockRefTableGetEntry(ib->brtab, &rlocator, forknum, + &limit_block); + + /* + * If there is no entry, then there have been no WAL-logged changes to the + * relation since the predecessor backup was taken, so we can back it up + * incrementally and need not include any modified blocks. + * + * However, if the file is zero-length, we should do a full backup, + * because an incremental file is always more than zero length, and it's + * silly to take an incremental backup when a full backup would be + * smaller. + */ + if (brtentry == NULL) + { + *num_blocks_required = 0; + *truncation_block_length = size / BLCKSZ; + if (size == 0) + return BACK_UP_FILE_FULLY; + return BACK_UP_FILE_INCREMENTALLY; + } + + /* + * If the limit_block is less than or equal to the point where this + * segment starts, send the whole file. + */ + if (limit_block <= segno * RELSEG_SIZE) + return BACK_UP_FILE_FULLY; + + /* + * Get relevant entries from the block reference table entry. + * + * We shouldn't overflow computing the start or stop block numbers, but if + * it manages to happen somehow, detect it and throw an error. + */ + start_blkno = segno * RELSEG_SIZE; + stop_blkno = start_blkno + (size / BLCKSZ); + if (start_blkno / RELSEG_SIZE != segno || stop_blkno < start_blkno) + ereport(ERROR, + errcode(ERRCODE_INTERNAL_ERROR), + errmsg_internal("overflow computing block number bounds for segment %u with size %lu", + segno, size)); + nblocks = BlockRefTableEntryGetBlocks(brtentry, start_blkno, stop_blkno, + absolute_block_numbers, RELSEG_SIZE); + Assert(nblocks <= RELSEG_SIZE); + + /* + * If we're going to have to send nearly all of the blocks, then just send + * the whole file, because that won't require much extra storage or + * transfer and will speed up and simplify backup restoration. It's not + * clear what threshold is most appropriate here and perhaps it ought to + * be configurable, but for now we're just going to say that if we'd need + * to send 90% of the blocks anyway, give up and send the whole file. + * + * NB: If you change the threshold here, at least make sure to back up the + * file fully when every single block must be sent, because there's + * nothing good about sending an incremental file in that case. + */ + if (nblocks * BLCKSZ > size * 0.9) + return BACK_UP_FILE_FULLY; + + /* + * Looks like we can send an incremental file. + * + * Return the relevant details to the caller, transposing absolute block + * numbers to relative block numbers. + * + * The truncation block length is the minimum length of the reconstructed + * file. Any block numbers below this threshold that are not present in + * the backup need to be fetched from the prior backup. At or above this + * threshold, blocks should only be included in the result if they are + * present in the backup. (This may require inserting zero blocks if the + * blocks included in the backup are non-consecutive.) + */ + for (i = 0; i < nblocks; ++i) + relative_block_numbers[i] = absolute_block_numbers[i] - start_blkno; + *num_blocks_required = nblocks; + *truncation_block_length = + Min(size / BLCKSZ, limit_block - segno * RELSEG_SIZE); + return BACK_UP_FILE_INCREMENTALLY; +} + +/* + * Compute the size for an incremental file containing a given number of blocks. + */ +extern size_t +GetIncrementalFileSize(unsigned num_blocks_required) +{ + size_t result; + + /* Make sure we're not going to overflow. */ + Assert(num_blocks_required <= RELSEG_SIZE); + + /* + * Three four byte quantities (magic number, truncation block length, + * block count) followed by block numbers followed by block contents. + */ + result = 3 * sizeof(uint32); + result += (BLCKSZ + sizeof(BlockNumber)) * num_blocks_required; + + return result; +} + +/* + * Helper function for filemap hash table. + */ +static uint32 +hash_string_pointer(const char *s) +{ + unsigned char *ss = (unsigned char *) s; + + return hash_bytes(ss, strlen(s)); +} + +/* + * This callback is invoked for each file mentioned in the backup manifest. + * + * We store the path to each file and the size of each file for sanity-checking + * purposes. For further details, see comments for IncrementalBackupInfo. + */ +static void +manifest_process_file(JsonManifestParseContext *context, + char *pathname, size_t size, + pg_checksum_type checksum_type, + int checksum_length, + uint8 *checksum_payload) +{ + IncrementalBackupInfo *ib = context->private_data; + backup_file_entry *entry; + bool found; + + entry = backup_file_insert(ib->manifest_files, pathname, &found); + if (!found) + { + entry->path = MemoryContextStrdup(ib->manifest_files->ctx, + pathname); + entry->size = size; + } +} + +/* + * This callback is invoked for each WAL range mentioned in the backup + * manifest. + * + * We're just interested in learning the oldest LSN and the corresponding TLI + * that appear in any WAL range. + */ +static void +manifest_process_wal_range(JsonManifestParseContext *context, + TimeLineID tli, XLogRecPtr start_lsn, + XLogRecPtr end_lsn) +{ + IncrementalBackupInfo *ib = context->private_data; + backup_wal_range *range = palloc(sizeof(backup_wal_range)); + + range->tli = tli; + range->start_lsn = start_lsn; + range->end_lsn = end_lsn; + ib->manifest_wal_ranges = lappend(ib->manifest_wal_ranges, range); +} + +/* + * This callback is invoked if an error occurs while parsing the backup + * manifest. + */ +static void +manifest_report_error(JsonManifestParseContext *context, const char *fmt,...) +{ + StringInfoData errbuf; + + initStringInfo(&errbuf); + + for (;;) + { + va_list ap; + int needed; + + va_start(ap, fmt); + needed = appendStringInfoVA(&errbuf, fmt, ap); + va_end(ap); + if (needed == 0) + break; + enlargeStringInfo(&errbuf, needed); + } + + ereport(ERROR, + errmsg_internal("%s", errbuf.data)); +} diff --git a/src/backend/backup/meson.build b/src/backend/backup/meson.build index 11a79bbf80..19c355ceca 100644 --- a/src/backend/backup/meson.build +++ b/src/backend/backup/meson.build @@ -5,6 +5,7 @@ backend_sources += files( 'basebackup.c', 'basebackup_copy.c', 'basebackup_gzip.c', + 'basebackup_incremental.c', 'basebackup_lz4.c', 'basebackup_progress.c', 'basebackup_server.c', @@ -12,4 +13,6 @@ backend_sources += files( 'basebackup_target.c', 'basebackup_throttle.c', 'basebackup_zstd.c', + 'walsummary.c', + 'walsummaryfuncs.c' ) diff --git a/src/backend/backup/walsummary.c b/src/backend/backup/walsummary.c new file mode 100644 index 0000000000..ebf4ea038d --- /dev/null +++ b/src/backend/backup/walsummary.c @@ -0,0 +1,356 @@ +/*------------------------------------------------------------------------- + * + * walsummary.c + * Functions for accessing and managing WAL summary data. + * + * Portions Copyright (c) 2010-2022, PostgreSQL Global Development Group + * + * src/backend/backup/walsummary.c + * + *------------------------------------------------------------------------- + */ + +#include "postgres.h" + +#include <sys/stat.h> +#include <unistd.h> + +#include "access/xlog_internal.h" +#include "backup/walsummary.h" +#include "utils/wait_event.h" + +static bool IsWalSummaryFilename(char *filename); +static int ListComparatorForWalSummaryFiles(const ListCell *a, + const ListCell *b); + +/* + * Get a list of WAL summaries. + * + * If tli != 0, only WAL summaries with the indicated TLI will be included. + * + * If start_lsn != InvalidXLogRecPtr, only summaries that end before the + * indicated LSN will be included. + * + * If end_lsn != InvalidXLogRecPtr, only summaries that start before the + * indicated LSN will be included. + * + * The intent is that you can call GetWalSummaries(tli, start_lsn, end_lsn) + * to get all WAL summaries on the indicated timeline that overlap the + * specified LSN range. + */ +List * +GetWalSummaries(TimeLineID tli, XLogRecPtr start_lsn, XLogRecPtr end_lsn) +{ + DIR *sdir; + struct dirent *dent; + List *result = NIL; + + sdir = AllocateDir(XLOGDIR "/summaries"); + while ((dent = ReadDir(sdir, XLOGDIR "/summaries")) != NULL) + { + WalSummaryFile *ws; + uint32 tmp[5]; + TimeLineID file_tli; + XLogRecPtr file_start_lsn; + XLogRecPtr file_end_lsn; + + /* Decode filename, or skip if it's not in the expected format. */ + if (!IsWalSummaryFilename(dent->d_name)) + continue; + sscanf(dent->d_name, "%08X%08X%08X%08X%08X", + &tmp[0], &tmp[1], &tmp[2], &tmp[3], &tmp[4]); + file_tli = tmp[0]; + file_start_lsn = ((uint64) tmp[1]) << 32 | tmp[2]; + file_end_lsn = ((uint64) tmp[3]) << 32 | tmp[4]; + + /* Skip if it doesn't match the filter criteria. */ + if (tli != 0 && tli != file_tli) + continue; + if (!XLogRecPtrIsInvalid(start_lsn) && start_lsn > file_end_lsn) + continue; + if (!XLogRecPtrIsInvalid(end_lsn) && end_lsn < file_start_lsn) + continue; + + /* Add it to the list. */ + ws = palloc(sizeof(WalSummaryFile)); + ws->tli = file_tli; + ws->start_lsn = file_start_lsn; + ws->end_lsn = file_end_lsn; + result = lappend(result, ws); + } + FreeDir(sdir); + + return result; +} + +/* + * Build a new list of WAL summaries based on an existing list, but filtering + * out summaries that don't match the search parameters. + * + * If tli != 0, only WAL summaries with the indicated TLI will be included. + * + * If start_lsn != InvalidXLogRecPtr, only summaries that end before the + * indicated LSN will be included. + * + * If end_lsn != InvalidXLogRecPtr, only summaries that start before the + * indicated LSN will be included. + */ +List * +FilterWalSummaries(List *wslist, TimeLineID tli, + XLogRecPtr start_lsn, XLogRecPtr end_lsn) +{ + List *result = NIL; + ListCell *lc; + + /* Loop over input. */ + foreach(lc, wslist) + { + WalSummaryFile *ws = lfirst(lc); + + /* Skip if it doesn't match the filter criteria. */ + if (tli != 0 && tli != ws->tli) + continue; + if (!XLogRecPtrIsInvalid(start_lsn) && start_lsn > ws->end_lsn) + continue; + if (!XLogRecPtrIsInvalid(end_lsn) && end_lsn < ws->start_lsn) + continue; + + /* Add it to the result list. */ + result = lappend(result, ws); + } + + return result; +} + +/* + * Check whether the supplied list of WalSummaryFile objects covers the + * whole range of LSNs from start_lsn to end_lsn. This function ignores + * timelines, so the caller should probably filter using the appropriate + * timeline before calling this. + * + * If the whole range of LSNs is covered, returns true, otherwise false. + * If false is returned, *missing_lsn is set either to InvalidXLogRecPtr + * if there are no WAL summary files in the input list, or to the first LSN + * in the range that is not covered by a WAL summary file in the input list. + */ +bool +WalSummariesAreComplete(List *wslist, XLogRecPtr start_lsn, + XLogRecPtr end_lsn, XLogRecPtr *missing_lsn) +{ + XLogRecPtr current_lsn = start_lsn; + ListCell *lc; + + /* Special case for empty list. */ + if (wslist == NIL) + { + *missing_lsn = InvalidXLogRecPtr; + return false; + } + + /* Make a private copy of the list and sort it by start LSN. */ + wslist = list_copy(wslist); + list_sort(wslist, ListComparatorForWalSummaryFiles); + + /* + * Consider summary files in order of increasing start_lsn, advancing the + * known-summarized range from start_lsn toward end_lsn. + * + * Normally, the summary files should cover non-overlapping WAL ranges, + * but this algorithm is intended to be correct even in case of overlap. + */ + foreach(lc, wslist) + { + WalSummaryFile *ws = lfirst(lc); + + if (ws->start_lsn > current_lsn) + { + /* We found a gap. */ + break; + } + if (ws->end_lsn > current_lsn) + { + /* + * Next summary extends beyond end of previous summary, so extend + * the end of the range known to be summarized. + */ + current_lsn = ws->end_lsn; + + /* + * If the range we know to be summarized has reached the required + * end LSN, we have proved completeness. + */ + if (current_lsn >= end_lsn) + return true; + } + } + + /* + * We either ran out of summary files without reaching the end LSN, or we + * hit a gap in the sequence that resulted in us bailing out of the loop + * above. + */ + *missing_lsn = current_lsn; + return false; +} + +/* + * Open a WAL summary file. + * + * This will throw an error in case of trouble. As an exception, if + * missing_ok = true and the trouble is specifically that the file does + * not exist, it will not throw an error and will return a value less than 0. + */ +File +OpenWalSummaryFile(WalSummaryFile *ws, bool missing_ok) +{ + char path[MAXPGPATH]; + File file; + + snprintf(path, MAXPGPATH, + XLOGDIR "/summaries/%08X%08X%08X%08X%08X.summary", + ws->tli, + LSN_FORMAT_ARGS(ws->start_lsn), + LSN_FORMAT_ARGS(ws->end_lsn)); + + file = PathNameOpenFile(path, O_RDONLY); + if (file < 0 && (errno != EEXIST || !missing_ok)) + ereport(ERROR, + (errcode_for_file_access(), + errmsg("could not open file \"%s\": %m", path))); + + return file; +} + +/* + * Remove a WAL summary file if the last modification time precedes the + * cutoff time. + */ +void +RemoveWalSummaryIfOlderThan(WalSummaryFile *ws, time_t cutoff_time) +{ + char path[MAXPGPATH]; + struct stat statbuf; + + snprintf(path, MAXPGPATH, + XLOGDIR "/summaries/%08X%08X%08X%08X%08X.summary", + ws->tli, + LSN_FORMAT_ARGS(ws->start_lsn), + LSN_FORMAT_ARGS(ws->end_lsn)); + + if (lstat(path, &statbuf) != 0) + { + if (errno == ENOENT) + return; + ereport(ERROR, + (errcode_for_file_access(), + errmsg("could not stat file \"%s\": %m", path))); + } + if (statbuf.st_mtime >= cutoff_time) + return; + if (unlink(path) != 0) + ereport(ERROR, + (errcode_for_file_access(), + errmsg("could not stat file \"%s\": %m", path))); + ereport(DEBUG2, + (errmsg_internal("removing file \"%s\"", path))); +} + +/* + * Test whether a filename looks like a WAL summary file. + */ +static bool +IsWalSummaryFilename(char *filename) +{ + return strspn(filename, "0123456789ABCDEF") == 40 && + strcmp(filename + 40, ".summary") == 0; +} + +/* + * Data read callback for use with CreateBlockRefTableReader. + */ +int +ReadWalSummary(void *wal_summary_io, void *data, int length) +{ + WalSummaryIO *io = wal_summary_io; + int nbytes; + + nbytes = FileRead(io->file, data, length, io->filepos, + WAIT_EVENT_WAL_SUMMARY_READ); + if (nbytes < 0) + ereport(ERROR, + (errcode_for_file_access(), + errmsg("could not write file \"%s\": %m", + FilePathName(io->file)))); + + io->filepos += nbytes; + return nbytes; +} + +/* + * Data write callback for use with WriteBlockRefTable. + */ +int +WriteWalSummary(void *wal_summary_io, void *data, int length) +{ + WalSummaryIO *io = wal_summary_io; + int nbytes; + + nbytes = FileWrite(io->file, data, length, io->filepos, + WAIT_EVENT_WAL_SUMMARY_WRITE); + if (nbytes < 0) + ereport(ERROR, + (errcode_for_file_access(), + errmsg("could not write file \"%s\": %m", + FilePathName(io->file)))); + if (nbytes != length) + ereport(ERROR, + (errcode_for_file_access(), + errmsg("could not write file \"%s\": wrote only %d of %d bytes at offset %u", + FilePathName(io->file), nbytes, + length, (unsigned) io->filepos), + errhint("Check free disk space."))); + + io->filepos += nbytes; + return nbytes; +} + +/* + * Error-reporting callback for use with CreateBlockRefTableReader. + */ +void +ReportWalSummaryError(void *callback_arg, char *fmt,...) +{ + StringInfoData buf; + va_list ap; + int needed; + + initStringInfo(&buf); + for (;;) + { + va_start(ap, fmt); + needed = appendStringInfoVA(&buf, fmt, ap); + va_end(ap); + if (needed == 0) + break; + enlargeStringInfo(&buf, needed); + } + ereport(ERROR, + errcode(ERRCODE_DATA_CORRUPTED), + errmsg_internal("%s", buf.data)); +} + +/* + * Comparator to sort a List of WalSummaryFile objects by start_lsn. + */ +static int +ListComparatorForWalSummaryFiles(const ListCell *a, const ListCell *b) +{ + WalSummaryFile *ws1 = lfirst(a); + WalSummaryFile *ws2 = lfirst(b); + + if (ws1->start_lsn < ws2->start_lsn) + return -1; + if (ws1->start_lsn > ws2->start_lsn) + return 1; + return 0; +} diff --git a/src/backend/backup/walsummaryfuncs.c b/src/backend/backup/walsummaryfuncs.c new file mode 100644 index 0000000000..2e77d38b4a --- /dev/null +++ b/src/backend/backup/walsummaryfuncs.c @@ -0,0 +1,169 @@ +/*------------------------------------------------------------------------- + * + * walsummaryfuncs.c + * SQL-callable functions for accessing WAL summary data. + * + * Portions Copyright (c) 2010-2022, PostgreSQL Global Development Group + * + * src/backend/backup/walsummaryfuncs.c + * + *------------------------------------------------------------------------- + */ + +#include "postgres.h" + +#include "backup/walsummary.h" +#include "common/blkreftable.h" +#include "funcapi.h" +#include "miscadmin.h" +#include "utils/fmgrprotos.h" +#include "utils/pg_lsn.h" + +#define NUM_WS_ATTS 3 +#define NUM_SUMMARY_ATTS 6 +#define MAX_BLOCKS_PER_CALL 256 + +/* + * List the WAL summary files available in pg_wal/summaries. + */ +Datum +pg_available_wal_summaries(PG_FUNCTION_ARGS) +{ + ReturnSetInfo *rsi; + List *wslist; + ListCell *lc; + Datum values[NUM_WS_ATTS]; + bool nulls[NUM_WS_ATTS]; + + InitMaterializedSRF(fcinfo, 0); + rsi = (ReturnSetInfo *) fcinfo->resultinfo; + + memset(nulls, 0, sizeof(nulls)); + + wslist = GetWalSummaries(0, InvalidXLogRecPtr, InvalidXLogRecPtr); + foreach(lc, wslist) + { + WalSummaryFile *ws = (WalSummaryFile *) lfirst(lc); + HeapTuple tuple; + + CHECK_FOR_INTERRUPTS(); + + values[0] = Int64GetDatum((int64) ws->tli); + values[1] = LSNGetDatum(ws->start_lsn); + values[2] = LSNGetDatum(ws->end_lsn); + + tuple = heap_form_tuple(rsi->setDesc, values, nulls); + tuplestore_puttuple(rsi->setResult, tuple); + } + + return (Datum) 0; +} + +/* + * List the contents of a WAL summary file identified by TLI, start LSN, + * and end LSN. + */ +Datum +pg_wal_summary_contents(PG_FUNCTION_ARGS) +{ + ReturnSetInfo *rsi; + Datum values[NUM_SUMMARY_ATTS]; + bool nulls[NUM_SUMMARY_ATTS]; + WalSummaryFile ws; + WalSummaryIO io; + BlockRefTableReader *reader; + int64 raw_tli; + RelFileLocator rlocator; + ForkNumber forknum; + BlockNumber limit_block; + + InitMaterializedSRF(fcinfo, 0); + rsi = (ReturnSetInfo *) fcinfo->resultinfo; + memset(nulls, 0, sizeof(nulls)); + + /* + * Since the timeline could at least in theory be more than 2^31, and + * since we don't have unsigned types at the SQL level, it is passed as a + * 64-bit integer. Test whether it's out of range. + */ + raw_tli = PG_GETARG_INT64(0); + if (raw_tli < 1 || raw_tli > PG_INT32_MAX) + ereport(ERROR, + errcode(ERRCODE_INVALID_PARAMETER_VALUE), + errmsg("invalid timeline %lld", (long long) raw_tli)); + + /* Prepare to read the specified WAL summry file. */ + ws.tli = (TimeLineID) raw_tli; + ws.start_lsn = PG_GETARG_LSN(1); + ws.end_lsn = PG_GETARG_LSN(2); + io.filepos = 0; + io.file = OpenWalSummaryFile(&ws, false); + reader = CreateBlockRefTableReader(ReadWalSummary, &io, + FilePathName(io.file), + ReportWalSummaryError, NULL); + + /* Loop over relation forks. */ + while (BlockRefTableReaderNextRelation(reader, &rlocator, &forknum, + &limit_block)) + { + BlockNumber blocks[MAX_BLOCKS_PER_CALL]; + HeapTuple tuple; + + CHECK_FOR_INTERRUPTS(); + + values[0] = ObjectIdGetDatum(rlocator.relNumber); + values[1] = ObjectIdGetDatum(rlocator.spcOid); + values[2] = ObjectIdGetDatum(rlocator.dbOid); + values[3] = Int16GetDatum((int16) forknum); + + /* Loop over blocks within the current relation fork. */ + while (true) + { + unsigned nblocks; + unsigned i; + + CHECK_FOR_INTERRUPTS(); + + nblocks = BlockRefTableReaderGetBlocks(reader, blocks, + MAX_BLOCKS_PER_CALL); + if (nblocks == 0) + break; + + /* + * For each block that we specifically know to have been modified, + * emit a row with that block number and limit_block = false. + */ + values[5] = BoolGetDatum(false); + for (i = 0; i < nblocks; ++i) + { + values[4] = Int64GetDatum((int64) blocks[i]); + + tuple = heap_form_tuple(rsi->setDesc, values, nulls); + tuplestore_puttuple(rsi->setResult, tuple); + } + + /* + * If the limit block is not InvalidBlockNumber, emit an exta row + * with that block number and limit_block = true. + * + * There is no point in doing this when the limit_block is + * InvalidBlockNumber, because no block with that number or any + * higher number can ever exist. + */ + if (BlockNumberIsValid(limit_block)) + { + values[4] = Int64GetDatum((int64) limit_block); + values[5] = BoolGetDatum(true); + + tuple = heap_form_tuple(rsi->setDesc, values, nulls); + tuplestore_puttuple(rsi->setResult, tuple); + } + } + } + + /* Cleanup */ + DestroyBlockRefTableReader(reader); + FileClose(io.file); + + return (Datum) 0; +} diff --git a/src/backend/postmaster/Makefile b/src/backend/postmaster/Makefile index 047448b34e..367a46c617 100644 --- a/src/backend/postmaster/Makefile +++ b/src/backend/postmaster/Makefile @@ -24,6 +24,7 @@ OBJS = \ postmaster.o \ startup.o \ syslogger.o \ + walsummarizer.o \ walwriter.o include $(top_srcdir)/src/backend/common.mk diff --git a/src/backend/postmaster/auxprocess.c b/src/backend/postmaster/auxprocess.c index cae6feb356..0c15c1777d 100644 --- a/src/backend/postmaster/auxprocess.c +++ b/src/backend/postmaster/auxprocess.c @@ -21,6 +21,7 @@ #include "postmaster/auxprocess.h" #include "postmaster/bgwriter.h" #include "postmaster/startup.h" +#include "postmaster/walsummarizer.h" #include "postmaster/walwriter.h" #include "replication/walreceiver.h" #include "storage/bufmgr.h" @@ -80,6 +81,9 @@ AuxiliaryProcessMain(AuxProcType auxtype) case WalReceiverProcess: MyBackendType = B_WAL_RECEIVER; break; + case WalSummarizerProcess: + MyBackendType = B_WAL_SUMMARIZER; + break; default: elog(PANIC, "unrecognized process type: %d", (int) MyAuxProcType); MyBackendType = B_INVALID; @@ -161,6 +165,10 @@ AuxiliaryProcessMain(AuxProcType auxtype) WalReceiverMain(); proc_exit(1); + case WalSummarizerProcess: + WalSummarizerMain(); + proc_exit(1); + default: elog(PANIC, "unrecognized process type: %d", (int) MyAuxProcType); proc_exit(1); diff --git a/src/backend/postmaster/meson.build b/src/backend/postmaster/meson.build index cda921fd10..a30eb6692f 100644 --- a/src/backend/postmaster/meson.build +++ b/src/backend/postmaster/meson.build @@ -12,5 +12,6 @@ backend_sources += files( 'postmaster.c', 'startup.c', 'syslogger.c', + 'walsummarizer.c', 'walwriter.c', ) diff --git a/src/backend/postmaster/postmaster.c b/src/backend/postmaster/postmaster.c index 54e9bfb8c4..0538b84ef8 100644 --- a/src/backend/postmaster/postmaster.c +++ b/src/backend/postmaster/postmaster.c @@ -115,6 +115,7 @@ #include "postmaster/pgarch.h" #include "postmaster/postmaster.h" #include "postmaster/syslogger.h" +#include "postmaster/walsummarizer.h" #include "replication/logicallauncher.h" #include "replication/walsender.h" #include "storage/fd.h" @@ -251,6 +252,7 @@ static pid_t StartupPID = 0, CheckpointerPID = 0, WalWriterPID = 0, WalReceiverPID = 0, + WalSummarizerPID = 0, AutoVacPID = 0, PgArchPID = 0, SysLoggerPID = 0; @@ -442,6 +444,7 @@ static bool CreateOptsFile(int argc, char *argv[], char *fullprogname); static pid_t StartChildProcess(AuxProcType type); static void StartAutovacuumWorker(void); static void MaybeStartWalReceiver(void); +static void MaybeStartWalSummarizer(void); static void InitPostmasterDeathWatchHandle(void); /* @@ -561,6 +564,7 @@ static void ShmemBackendArrayRemove(Backend *bn); #define StartCheckpointer() StartChildProcess(CheckpointerProcess) #define StartWalWriter() StartChildProcess(WalWriterProcess) #define StartWalReceiver() StartChildProcess(WalReceiverProcess) +#define StartWalSummarizer() StartChildProcess(WalSummarizerProcess) /* Macros to check exit status of a child process */ #define EXIT_STATUS_0(st) ((st) == 0) @@ -1847,6 +1851,9 @@ ServerLoop(void) if (WalReceiverRequested) MaybeStartWalReceiver(); + /* If we need to start a WAL summarizer, try to do that now */ + MaybeStartWalSummarizer(); + /* Get other worker processes running, if needed */ if (StartWorkerNeeded || HaveCrashedWorker) maybe_start_bgworkers(); @@ -2714,6 +2721,8 @@ process_pm_reload_request(void) signal_child(WalWriterPID, SIGHUP); if (WalReceiverPID != 0) signal_child(WalReceiverPID, SIGHUP); + if (WalSummarizerPID != 0) + signal_child(WalSummarizerPID, SIGHUP); if (AutoVacPID != 0) signal_child(AutoVacPID, SIGHUP); if (PgArchPID != 0) @@ -3067,6 +3076,7 @@ process_pm_child_exit(void) BgWriterPID = StartBackgroundWriter(); if (WalWriterPID == 0) WalWriterPID = StartWalWriter(); + MaybeStartWalSummarizer(); /* * Likewise, start other special children as needed. In a restart @@ -3185,6 +3195,20 @@ process_pm_child_exit(void) continue; } + /* + * Was it the wal summarizer? Normal exit can be ignored; we'll start + * a new one at the next iteration of the postmaster's main loop, if + * necessary. Any other exit condition is treated as a crash. + */ + if (pid == WalSummarizerPID) + { + WalSummarizerPID = 0; + if (!EXIT_STATUS_0(exitstatus)) + HandleChildCrash(pid, exitstatus, + _("WAL summarizer process")); + continue; + } + /* * Was it the autovacuum launcher? Normal exit can be ignored; we'll * start a new one at the next iteration of the postmaster's main @@ -3580,6 +3604,12 @@ HandleChildCrash(int pid, int exitstatus, const char *procname) else if (WalReceiverPID != 0 && take_action) sigquit_child(WalReceiverPID); + /* Take care of the walsummarizer too */ + if (pid == WalSummarizerPID) + WalSummarizerPID = 0; + else if (WalSummarizerPID != 0 && take_action) + sigquit_child(WalSummarizerPID); + /* Take care of the autovacuum launcher too */ if (pid == AutoVacPID) AutoVacPID = 0; @@ -3730,6 +3760,8 @@ PostmasterStateMachine(void) signal_child(StartupPID, SIGTERM); if (WalReceiverPID != 0) signal_child(WalReceiverPID, SIGTERM); + if (WalSummarizerPID != 0) + signal_child(WalSummarizerPID, SIGTERM); /* checkpointer, archiver, stats, and syslogger may continue for now */ /* Now transition to PM_WAIT_BACKENDS state to wait for them to die */ @@ -3756,6 +3788,7 @@ PostmasterStateMachine(void) if (CountChildren(BACKEND_TYPE_ALL - BACKEND_TYPE_WALSND) == 0 && StartupPID == 0 && WalReceiverPID == 0 && + WalSummarizerPID == 0 && BgWriterPID == 0 && (CheckpointerPID == 0 || (!FatalError && Shutdown < ImmediateShutdown)) && @@ -3853,6 +3886,7 @@ PostmasterStateMachine(void) /* These other guys should be dead already */ Assert(StartupPID == 0); Assert(WalReceiverPID == 0); + Assert(WalSummarizerPID == 0); Assert(BgWriterPID == 0); Assert(CheckpointerPID == 0); Assert(WalWriterPID == 0); @@ -4074,6 +4108,8 @@ TerminateChildren(int signal) signal_child(WalWriterPID, signal); if (WalReceiverPID != 0) signal_child(WalReceiverPID, signal); + if (WalSummarizerPID != 0) + signal_child(WalSummarizerPID, signal); if (AutoVacPID != 0) signal_child(AutoVacPID, signal); if (PgArchPID != 0) @@ -5380,6 +5416,10 @@ StartChildProcess(AuxProcType type) ereport(LOG, (errmsg("could not fork WAL receiver process: %m"))); break; + case WalSummarizerProcess: + ereport(LOG, + (errmsg("could not fork WAL summarizer process: %m"))); + break; default: ereport(LOG, (errmsg("could not fork process: %m"))); @@ -5516,6 +5556,19 @@ MaybeStartWalReceiver(void) } } +/* + * MaybeStartWalSummarizer + * Start the WAL summarizer process, if not running and our state allows. + */ +static void +MaybeStartWalSummarizer(void) +{ + if (wal_summarize_mb != 0 && WalSummarizerPID == 0 && + (pmState == PM_RUN || pmState == PM_HOT_STANDBY) && + Shutdown <= SmartShutdown) + WalSummarizerPID = StartWalSummarizer(); +} + /* * Create the opts file diff --git a/src/backend/postmaster/walsummarizer.c b/src/backend/postmaster/walsummarizer.c new file mode 100644 index 0000000000..34bd254183 --- /dev/null +++ b/src/backend/postmaster/walsummarizer.c @@ -0,0 +1,1414 @@ +/*------------------------------------------------------------------------- + * + * walsummarizer.c + * + * Background process to perform WAL summarization, if it is enabled. + * It continuously scans the write-ahead log and periodically emits a + * summary file which indicates which blocks in which relation forks + * were modified by WAL records in the LSN range covered by the summary + * file. See walsummary.c and blkreftable.c for more details on the + * naming and contents of WAL summary files. + * + * If configured to do, this background process will also remove WAL + * summary files when the file timestamp is older than a configurable + * threshold (but only if the WAL has been removed first). + * + * Portions Copyright (c) 1996-2022, PostgreSQL Global Development Group + * + * IDENTIFICATION + * src/backend/postmaster/walsummarizer.c + * + *------------------------------------------------------------------------- + */ +#include "postgres.h" + +#include "access/timeline.h" +#include "access/xlog.h" +#include "access/xlog_internal.h" +#include "access/xlogrecovery.h" +#include "access/xlogutils.h" +#include "backup/walsummary.h" +#include "catalog/storage_xlog.h" +#include "common/blkreftable.h" +#include "libpq/pqsignal.h" +#include "miscadmin.h" +#include "postmaster/bgwriter.h" +#include "postmaster/interrupt.h" +#include "postmaster/walsummarizer.h" +#include "replication/walreceiver.h" +#include "storage/fd.h" +#include "storage/ipc.h" +#include "storage/lwlock.h" +#include "storage/latch.h" +#include "storage/proc.h" +#include "storage/procsignal.h" +#include "storage/shmem.h" +#include "storage/spin.h" +#include "utils/guc.h" +#include "utils/memutils.h" +#include "utils/wait_event.h" + +/* + * Data in shared memory related to WAL summarization. + */ +typedef struct +{ + /* + * These fields are protected by WALSummarizerLock. + * + * Until we've discovered what summary files already exist on disk and + * stored that information in shared memory, initialized is false and the + * other fields here contain no meaningful information. After that has + * been done, initialized is true. + * + * summarized_tli and summarized_lsn indicate the last LSN and TLI at + * which the next summary file will start. Normally, these are the LSN and + * TLI at which the last file ended; in such case, lsn_is_exact is true. + * If, however, the LSN is just an approximation, then lsn_is_exact is + * false. This can happen if, for example, there are no existing WAL + * summary files at startup. In that case, we have to derive the position + * at which to start summarizing from the WAL files that exist on disk, + * and so the LSN might point to the start of the next file even though + * that might happen to be in the middle of a WAL record. + * + * summarizer_pgprocno is the pgprocno value for the summarizer process, + * if one is running, or else INVALID_PGPROCNO. + * + * pending_lsn is used by the summarizer to advertise the ending LSN of a + * record it has recently read. It shouldn't ever be less than + * summarized_lsn, but might be greater, because the summarizer buffers + * data for a range of LSNs in memory before writing out a new file. + * + * switch_requested can be set to true to notify the summarizer that a new + * WAL summary file should be written as soon as possible, without trying + * to read more WAL first. + */ + bool initialized; + TimeLineID summarized_tli; + XLogRecPtr summarized_lsn; + bool lsn_is_exact; + int summarizer_pgprocno; + XLogRecPtr pending_lsn; + bool switch_requested; + + /* + * This field handles its own synchronizaton. + */ + ConditionVariable summary_file_cv; +} WalSummarizerData; + +/* + * Private data for our xlogreader's page read callback. + */ +typedef struct +{ + TimeLineID tli; + bool historic; + XLogRecPtr read_upto; + bool end_of_wal; + bool waited; + XLogRecPtr redo_pointer; + bool redo_pointer_reached; + XLogRecPtr redo_pointer_refresh_lsn; +} SummarizerReadLocalXLogPrivate; + +/* Pointer to shared memory state. */ +static WalSummarizerData *WalSummarizerCtl; + +/* + * When we reach end of WAL and need to read more, we sleep for a number of + * milliseconds that is a integer multiple of MS_PER_SLEEP_QUANTUM. This is + * the multiplier. It should vary between 1 and MAX_SLEEP_QUANTA, depending + * on system activity. See summarizer_wait_for_wal() for how we adjust this. + */ +static long sleep_quanta = 1; + +/* + * The sleep time will always be a multiple of 200ms and will not exceed + * one minute (300 * 200 = 60 * 1000). + */ +#define MAX_SLEEP_QUANTA 300 +#define MS_PER_SLEEP_QUANTUM 200 + +/* + * This is a count of the number of pages of WAL that we've read since the + * last time we waited for more WAL to appear. + */ +static long pages_read_since_last_sleep = 0; + +/* + * Most recent RedoRecPtr value observed by MaybeRemoveOldWalSummaries. + */ +static XLogRecPtr redo_pointer_at_last_summary_removal = InvalidXLogRecPtr; + +/* + * GUC parameters + */ +int wal_summarize_mb = 256; +int wal_summarize_keep_time = 7 * 24 * 60; + +static XLogRecPtr GetLatestLSN(TimeLineID *tli); +static void HandleWalSummarizerInterrupts(void); +static XLogRecPtr SummarizeWAL(TimeLineID tli, bool historic, + XLogRecPtr start_lsn, bool exact, + XLogRecPtr cutoff_lsn, XLogRecPtr maximum_lsn); +static void SummarizeSmgrRecord(XLogReaderState *xlogreader, + BlockRefTable *brtab); +static void SummarizeXactRecord(XLogReaderState *xlogreader, + BlockRefTable *brtab); +static int summarizer_read_local_xlog_page(XLogReaderState *state, + XLogRecPtr targetPagePtr, + int reqLen, + XLogRecPtr targetRecPtr, + char *cur_page); +static void summarizer_wait_for_wal(void); +static void MaybeRemoveOldWalSummaries(void); + +/* + * Amount of shared memory required for this module. + */ +Size +WalSummarizerShmemSize(void) +{ + return sizeof(WalSummarizerData); +} + +/* + * Create or attach to shared memory segment for this module. + */ +void +WalSummarizerShmemInit(void) +{ + bool found; + + WalSummarizerCtl = (WalSummarizerData *) + ShmemInitStruct("Wal Summarizer Ctl", WalSummarizerShmemSize(), + &found); + + if (!found) + { + /* + * First time through, so initialize. + * + * We're just filling in dummy values here -- the real initialization + * will happen when GetOldestUnsummarizedLSN() is called for the first + * time. + */ + WalSummarizerCtl->initialized = false; + WalSummarizerCtl->summarized_tli = 0; + WalSummarizerCtl->summarized_lsn = InvalidXLogRecPtr; + WalSummarizerCtl->lsn_is_exact = false; + WalSummarizerCtl->summarizer_pgprocno = INVALID_PGPROCNO; + WalSummarizerCtl->pending_lsn = InvalidXLogRecPtr; + WalSummarizerCtl->switch_requested = false; + ConditionVariableInit(&WalSummarizerCtl->summary_file_cv); + } +} + +/* + * Entry point for walsummarizer process. + */ +void +WalSummarizerMain(void) +{ + sigjmp_buf local_sigjmp_buf; + MemoryContext context; + + /* + * Within this function, 'current_lsn' and 'current_tli' refer to the + * point from which the next WAL summary file should start. 'exact' is + * true if 'current_lsn' is known to be the start of a WAL recod or WAL + * segment, and false if it might be in the middle of a record someplace. + * + * 'switch_lsn' and 'switch_tli', if set, are the LSN at which we need to + * switch to a new timeline and the timeline to which we need to switch. + * If not set, we either haven't figured out the answers yet or we're + * already on the latest timeline. + */ + XLogRecPtr current_lsn; + TimeLineID current_tli; + bool exact; + XLogRecPtr switch_lsn = InvalidXLogRecPtr; + TimeLineID switch_tli = 0; + + ereport(DEBUG1, + (errmsg_internal("WAL summarizer started"))); + + /* + * Properly accept or ignore signals the postmaster might send us + * + * We have no particular use for SIGINT at the moment, but seems + * reasonable to treat like SIGTERM. + */ + pqsignal(SIGHUP, SignalHandlerForConfigReload); + pqsignal(SIGINT, SignalHandlerForShutdownRequest); + pqsignal(SIGTERM, SignalHandlerForShutdownRequest); + /* SIGQUIT handler was already set up by InitPostmasterChild */ + pqsignal(SIGALRM, SIG_IGN); + pqsignal(SIGPIPE, SIG_IGN); + pqsignal(SIGUSR1, procsignal_sigusr1_handler); + pqsignal(SIGUSR2, SIG_IGN); /* not used */ + + /* Advertise ourselves. */ + LWLockAcquire(WALSummarizerLock, LW_EXCLUSIVE); + WalSummarizerCtl->summarizer_pgprocno = MyProc->pgprocno; + LWLockRelease(WALSummarizerLock); + + /* Create and switch to a memory context that we can reset on error. */ + context = AllocSetContextCreate(TopMemoryContext, + "Wal Summarizer", + ALLOCSET_DEFAULT_SIZES); + MemoryContextSwitchTo(context); + + /* + * Reset some signals that are accepted by postmaster but not here + */ + pqsignal(SIGCHLD, SIG_DFL); + + /* + * If an exception is encountered, processing resumes here. + */ + if (sigsetjmp(local_sigjmp_buf, 1) != 0) + { + /* Since not using PG_TRY, must reset error stack by hand */ + error_context_stack = NULL; + + /* Prevent interrupts while cleaning up */ + HOLD_INTERRUPTS(); + + /* Report the error to the server log */ + EmitErrorReport(); + + /* Release resources we might have acquired. */ + LWLockReleaseAll(); + ConditionVariableCancelSleep(); + pgstat_report_wait_end(); + ReleaseAuxProcessResources(false); + AtEOXact_Files(false); + AtEOXact_HashTables(false); + + /* + * Now return to normal top-level context and clear ErrorContext for + * next time. + */ + MemoryContextSwitchTo(context); + FlushErrorState(); + + /* Flush any leaked data in the top-level context */ + MemoryContextResetAndDeleteChildren(context); + + /* Now we can allow interrupts again */ + RESUME_INTERRUPTS(); + + /* + * Sleep for 10 seconds before attempting to resume operations in + * order to avoid excessing logging. + * + * Many of the likely error conditions are things that will repeat + * every time. For example, if the WAL can't be read or the summary + * can't be written, only administrator action will cure the problem. + * So a really fast retry time doesn't seem to be especially + * beneficial, and it will clutter the logs. + */ + (void) WaitLatch(MyLatch, + WL_TIMEOUT | WL_EXIT_ON_PM_DEATH, + 10000, + WAIT_EVENT_WAL_SUMMARIZER_ERROR); + } + + /* We can now handle ereport(ERROR) */ + PG_exception_stack = &local_sigjmp_buf; + + /* + * Unblock signals (they were blocked when the postmaster forked us) + */ + sigprocmask(SIG_SETMASK, &UnBlockSig, NULL); + + /* + * Fetch information about previous progress from shared memory. + * + * If we discover that WAL summarization is not enabled, just exit. + */ + current_lsn = GetOldestUnsummarizedLSN(¤t_tli, &exact); + if (XLogRecPtrIsInvalid(current_lsn)) + proc_exit(0); + + /* + * Loop forever + */ + for (;;) + { + XLogRecPtr latest_lsn; + TimeLineID latest_tli; + XLogRecPtr cutoff_lsn; + XLogRecPtr end_of_summary_lsn; + + /* Flush any leaked data in the top-level context */ + MemoryContextResetAndDeleteChildren(context); + + /* Process any signals received recently. */ + HandleWalSummarizerInterrupts(); + + /* If it's time to remove any old WAL summaries, do that now. */ + MaybeRemoveOldWalSummaries(); + + /* Find the LSN and TLI up to which we can safely summarize. */ + latest_lsn = GetLatestLSN(&latest_tli); + + /* + * If we're summarizing a historic timeline and we haven't yet + * computed the point at which to switch to the next timeline, do that + * now. + * + * Note that if this is a standby, what was previously the current + * timeline could become historic at any time. + * + * We could try to make this more efficient by caching the results of + * readTimeLineHistory when latest_tli has not changed, but since we + * only have to do this once per timeline switch, we probably wouldn't + * save any significant amount of work in practice. + */ + if (current_tli != latest_tli && XLogRecPtrIsInvalid(switch_lsn)) + { + List *tles = readTimeLineHistory(latest_tli); + + switch_lsn = tliSwitchPoint(current_tli, tles, &switch_tli); + elog(DEBUG2, + "switch point from TLI %u to TLI %u is at %X/%X", + current_tli, switch_tli, LSN_FORMAT_ARGS(switch_lsn)); + } + + /* + * wal_summarize_mb sets a soft limit on the amont of WAL covered by a + * single summary file. If we read a WAL record that ends after the + * cutoff LSN computed here, we'll stop the summary. In most cases, it + * will actually stop earlier than that, but this is here as a + * backstop. + */ + cutoff_lsn = current_lsn + wal_summarize_mb * 1024 * 1024; + if (!XLogRecPtrIsInvalid(switch_lsn) && cutoff_lsn > switch_lsn) + cutoff_lsn = switch_lsn; + elog(DEBUG2, + "WAL summarization cutoff is TLI %d @ %X/%X, flush position is %X/%X", + current_tli, LSN_FORMAT_ARGS(cutoff_lsn), LSN_FORMAT_ARGS(latest_lsn)); + + /* Summarize WAL. */ + end_of_summary_lsn = SummarizeWAL(current_tli, + current_tli != latest_tli, + current_lsn, exact, + cutoff_lsn, latest_lsn); + Assert(!XLogRecPtrIsInvalid(end_of_summary_lsn)); + Assert(end_of_summary_lsn >= current_lsn); + + /* + * Update state for next loop iteration. + * + * Next summary file should start from exactly where this one ended. + * Timeline remains unchanged unless a switch LSN was computed and we + * have reached it. + */ + current_lsn = end_of_summary_lsn; + exact = true; + if (!XLogRecPtrIsInvalid(switch_lsn) && cutoff_lsn >= switch_lsn) + { + current_tli = switch_tli; + switch_lsn = InvalidXLogRecPtr; + switch_tli = 0; + } + + /* Update state in shared memory. */ + LWLockAcquire(WALSummarizerLock, LW_EXCLUSIVE); + Assert(WalSummarizerCtl->pending_lsn <= end_of_summary_lsn); + WalSummarizerCtl->summarized_lsn = end_of_summary_lsn; + WalSummarizerCtl->summarized_tli = current_tli; + WalSummarizerCtl->lsn_is_exact = true; + WalSummarizerCtl->pending_lsn = end_of_summary_lsn; + WalSummarizerCtl->switch_requested = false; + LWLockRelease(WALSummarizerLock); + + /* Wake up anyone waiting for more summary files to be written. */ + ConditionVariableBroadcast(&WalSummarizerCtl->summary_file_cv); + } +} + +/* + * Get the oldest LSN in this server's timeline history that has not yet been + * summarized. + * + * If *tli != NULL, it will be set to the TLI for the LSN that is returned. + * + * If *lsn_is_exact != NULL, it will be set to true if the returned LSN is + * necessarily the start of a WAL record and false if it's just the beginning + * of a WAL segment. + */ +XLogRecPtr +GetOldestUnsummarizedLSN(TimeLineID *tli, bool *lsn_is_exact) +{ + TimeLineID latest_tli; + LWLockMode mode = LW_SHARED; + int n; + List *tles; + XLogRecPtr unsummarized_lsn; + TimeLineID unsummarized_tli = 0; + bool should_make_exact = false; + List *existing_summaries; + ListCell *lc; + + /* If not summarizing WAL, do nothing. */ + if (wal_summarize_mb == 0) + return InvalidXLogRecPtr; + + /* + * Initially, we acquire the lock in shared mode and try to fetch the + * required information. If the data structure hasn't been initialized, we + * reacquire the lock in shared mode so that we can initialize it. + * However, if someone else does that first before we get the lock, then + * we can just return the requested information after all. + */ + while (true) + { + LWLockAcquire(WALSummarizerLock, mode); + + if (WalSummarizerCtl->initialized) + { + unsummarized_lsn = WalSummarizerCtl->summarized_lsn; + if (tli != NULL) + *tli = WalSummarizerCtl->summarized_tli; + if (lsn_is_exact != NULL) + *lsn_is_exact = WalSummarizerCtl->lsn_is_exact; + LWLockRelease(WALSummarizerLock); + return unsummarized_lsn; + } + + if (mode == LW_EXCLUSIVE) + break; + + LWLockRelease(WALSummarizerLock); + mode = LW_EXCLUSIVE; + } + + /* + * The data structure needs to be initialized, and we are the first to + * obtain the lock in exclusive mode, so it's our job to do that + * initialization. + * + * So, find the oldest timeline on which WAL still exists, and the + * earliest segment for which it exists. + */ + (void) GetLatestLSN(&latest_tli); + tles = readTimeLineHistory(latest_tli); + for (n = list_length(tles) - 1; n >= 0; --n) + { + TimeLineHistoryEntry *tle = list_nth(tles, n); + XLogSegNo oldest_segno; + + oldest_segno = XLogGetOldestSegno(tle->tli); + if (oldest_segno != 0) + { + /* Compute oldest LSN that still exists on disk. */ + XLogSegNoOffsetToRecPtr(oldest_segno, 0, wal_segment_size, + unsummarized_lsn); + + unsummarized_tli = tle->tli; + break; + } + } + + /* It really should not be possible for us to find no WAL. */ + if (unsummarized_tli == 0) + ereport(ERROR, + errcode(ERRCODE_INTERNAL_ERROR), + errmsg_internal("no WAL found on timeline %d", latest_tli)); + + /* + * Don't try to summarize anything older than the end LSN of the newest + * summary file that exists for this timeline. + */ + existing_summaries = + GetWalSummaries(unsummarized_tli, + InvalidXLogRecPtr, InvalidXLogRecPtr); + foreach(lc, existing_summaries) + { + WalSummaryFile *ws = lfirst(lc); + + if (ws->end_lsn > unsummarized_lsn) + { + unsummarized_lsn = ws->end_lsn; + should_make_exact = true; + } + } + + /* Update shared memory with the discovered values. */ + WalSummarizerCtl->initialized = true; + WalSummarizerCtl->summarized_lsn = unsummarized_lsn; + WalSummarizerCtl->summarized_tli = unsummarized_tli; + WalSummarizerCtl->lsn_is_exact = should_make_exact; + WalSummarizerCtl->pending_lsn = unsummarized_lsn; + + /* Also return the to the caller as required. */ + if (tli != NULL) + *tli = WalSummarizerCtl->summarized_tli; + if (lsn_is_exact != NULL) + *lsn_is_exact = WalSummarizerCtl->lsn_is_exact; + LWLockRelease(WALSummarizerLock); + + return unsummarized_lsn; +} + +/* + * Attempt to set the WAL summarizer's latch. + * + * This might not work, because there's no guarantee that the WAL summarizer + * process was successfully started, and it also might have started but + * subsequently terminated. So, under normal circumstances, this will get the + * latch set, but there's no guarantee. + */ +void +SetWalSummarizerLatch(void) +{ + int pgprocno; + + LWLockAcquire(WALSummarizerLock, LW_EXCLUSIVE); + pgprocno = WalSummarizerCtl->summarizer_pgprocno; + LWLockRelease(WALSummarizerLock); + + if (pgprocno != INVALID_PGPROCNO) + SetLatch(&ProcGlobal->allProcs[pgprocno].procLatch); +} + +/* + * Wait until WAL summarization reaches the given LSN, but not longer than + * the given timeout. + * + * The return value is the first still-unsummarized LSN. If it's greater than + * or equal to the passed LSN, then that LSN was reached. If not, we timed out. + */ +XLogRecPtr +WaitForWalSummarization(XLogRecPtr lsn, long timeout) +{ + TimestampTz start_time = GetCurrentTimestamp(); + TimestampTz deadline = TimestampTzPlusMilliseconds(start_time, timeout); + XLogRecPtr summarized_lsn; + + Assert(!XLogRecPtrIsInvalid(lsn)); + Assert(timeout > 0); + + while (1) + { + TimestampTz now; + long remaining_timeout; + + /* + * If the LSN summarized on disk has reached the target value, stop. + * If it hasn't, but the in-memory value has reached the target value, + * request that a file be written as soon as possible. + */ + LWLockAcquire(WALSummarizerLock, LW_EXCLUSIVE); + summarized_lsn = WalSummarizerCtl->summarized_lsn; + if (summarized_lsn < lsn && + WalSummarizerCtl->pending_lsn >= lsn) + WalSummarizerCtl->switch_requested = true; + LWLockRelease(WALSummarizerLock); + if (summarized_lsn >= lsn) + break; + + /* Timeout reached? If yes, stop. */ + now = GetCurrentTimestamp(); + remaining_timeout = TimestampDifferenceMilliseconds(now, deadline); + if (remaining_timeout <= 0) + break; + + /* + * Limit the sleep to 1 second, because we may need to request a + * switch. + */ + if (remaining_timeout > 1000) + remaining_timeout = 1000; + + /* Wait and see. */ + ConditionVariableTimedSleep(&WalSummarizerCtl->summary_file_cv, + remaining_timeout, + WAIT_EVENT_WAL_SUMMARY_READY); + } + + return summarized_lsn; +} + +/* + * Get the latest LSN that is eligible to be summarized, and set *tli to the + * corresponding timeline. + */ +static XLogRecPtr +GetLatestLSN(TimeLineID *tli) +{ + if (!RecoveryInProgress()) + { + /* Don't summarize WAL before it's flushed. */ + return GetFlushRecPtr(tli); + } + else + { + XLogRecPtr flush_lsn; + TimeLineID flush_tli; + XLogRecPtr replay_lsn; + TimeLineID replay_tli; + + /* + * What we really want to know is how much WAL has been flushed to + * disk, but the only flush position available is the one provided by + * the walreceiver, which may not be running, because this could be + * crash recovery or recovery via restore_command. So use either the + * WAL receiver's flush position or the replay position, whichever is + * further ahead, on the theory that if the WAL has been replayed then + * it must also have been flushed to disk. + */ + flush_lsn = GetWalRcvFlushRecPtr(NULL, &flush_tli); + replay_lsn = GetXLogReplayRecPtr(&replay_tli); + if (flush_lsn > replay_lsn) + { + *tli = flush_tli; + return flush_lsn; + } + else + { + *tli = replay_tli; + return replay_lsn; + } + } +} + +/* + * Interrupt handler for main loop of WAL summarizer process. + */ +static void +HandleWalSummarizerInterrupts(void) +{ + if (ProcSignalBarrierPending) + ProcessProcSignalBarrier(); + + if (ConfigReloadPending) + { + ConfigReloadPending = false; + ProcessConfigFile(PGC_SIGHUP); + } + + if (ShutdownRequestPending || wal_summarize_mb == 0) + { + ereport(DEBUG1, + errmsg_internal("WAL summarizer shutting down")); + proc_exit(0); + } + + /* Perform logging of memory contexts of this process */ + if (LogMemoryContextPending) + ProcessLogMemoryContextInterrupt(); +} + +/* + * Summarize a range of WAL records on a single timeline. + * + * 'tli' is the timeline to be summarized. 'historic' should be false if the + * timeline in question is the latest one and true otherwise. + * + * 'start_lsn' is the point at which we should start summarizing. If this + * value comes from the end LSN of the previous record as returned by the + * xlograder machinery, 'exact' should be true; otherwise, 'exact' should + * be false, and this function will search forward for the start of a valid + * WAL record. + * + * 'cutoff_lsn' is the point at which we should stop summarizing. The first + * record that ends at or after cutoff_lsn will be the last one included + * in the summary. + * + * 'maximum_lsn' identifies the point beyond which we can't count on being + * able to read any more WAL. It should be the switch point when reading a + * historic timeline, or the most-recently-measured end of WAL when reading + * the current timeline. + * + * The return value is the LSN at which the WAL summary actually ends. Most + * often, a summary file ends because we notice that a checkpoint has + * occurred and reach the redo pointer of that checkpoint, but sometimes + * we stop for other reasons, such as a timeline switch, or reading a record + * that ends after the cutoff_lsn. + */ +static XLogRecPtr +SummarizeWAL(TimeLineID tli, bool historic, + XLogRecPtr start_lsn, bool exact, + XLogRecPtr cutoff_lsn, XLogRecPtr maximum_lsn) +{ + SummarizerReadLocalXLogPrivate *private_data; + XLogReaderState *xlogreader; + XLogRecPtr summary_start_lsn; + XLogRecPtr summary_end_lsn = cutoff_lsn; + char temp_path[MAXPGPATH]; + char final_path[MAXPGPATH]; + WalSummaryIO io; + BlockRefTable *brtab = CreateEmptyBlockRefTable(); + + /* Initialize private data for xlogreader. */ + private_data = (SummarizerReadLocalXLogPrivate *) + palloc0(sizeof(SummarizerReadLocalXLogPrivate)); + private_data->tli = tli; + private_data->historic = historic; + private_data->read_upto = maximum_lsn; + private_data->redo_pointer = GetRedoRecPtr(); + private_data->redo_pointer_refresh_lsn = start_lsn; + private_data->redo_pointer_reached = + (start_lsn >= private_data->redo_pointer); + + /* Create xlogreader. */ + xlogreader = XLogReaderAllocate(wal_segment_size, NULL, + XL_ROUTINE(.page_read = &summarizer_read_local_xlog_page, + .segment_open = &wal_segment_open, + .segment_close = &wal_segment_close), + private_data); + if (xlogreader == NULL) + ereport(ERROR, + (errcode(ERRCODE_OUT_OF_MEMORY), + errmsg("out of memory"), + errdetail("Failed while allocating a WAL reading processor."))); + + /* + * When exact = false, we're starting from an arbitrary point in the WAL + * and must search forward for the start of the next record. + * + * When exact = true, start_lsn should be either the LSN where a record + * begins, or the LSN of a page where the page header is immediately + * followed by the start of a new record. XLogBeginRead should tolerate + * either case. + * + * We need to allow for both cases because the behavior of xlogreader + * varies. When a record spans two or more xlog pages, the ending LSN + * reported by xlogreader will be the starting LSN of the following + * record, but when an xlog page boundary falls between two records, the + * end LSN for the first will be reported as the first byte of the + * following page. We can't know until we read that page how large the + * header will be, but we'll have to skip over it to find the next record. + */ + if (exact) + { + /* + * Even if start_lsn is the beginning of a page rather than the + * beginning of the first record on that page, we should still use it + * as the start LSN for the summary file. That's because we detect + * missing summary files by looking for cases where the end LSN of one + * file is less than the start LSN of the next file. When only a page + * header is skipped, nothing has been missed. + */ + XLogBeginRead(xlogreader, start_lsn); + summary_start_lsn = start_lsn; + } + else + { + summary_start_lsn = XLogFindNextRecord(xlogreader, start_lsn); + if (XLogRecPtrIsInvalid(summary_start_lsn)) + { + /* + * If we hit end-of-WAL while trying to find the next valid + * record, we must be on a historic timeline that has no valid + * records that begin after start_lsn and before end of WAL. + */ + if (private_data->end_of_wal) + { + ereport(DEBUG1, + errmsg_internal("could not read WAL from timeline %d at %X/%X: end of WAL at %X/%X", + tli, + LSN_FORMAT_ARGS(start_lsn), + LSN_FORMAT_ARGS(private_data->read_upto))); + + /* + * The timeline ends at or after start_lsn, without containing + * any records. Thus, we must make sure the main loop does not + * iterate. If start_lsn is the end of the timeline, then we + * won't actually emit an empty summary file, but otherwise, + * we must, to capture the fact that the LSN range in question + * contains no interesting WAL records. + */ + summary_start_lsn = start_lsn; + summary_end_lsn = private_data->read_upto; + cutoff_lsn = xlogreader->EndRecPtr; + } + else + ereport(ERROR, + (errmsg("could not find a valid record after %X/%X", + LSN_FORMAT_ARGS(start_lsn)))); + } + + /* We shouldn't go backward. */ + Assert(summary_start_lsn >= start_lsn); + } + + /* + * Main loop: read xlog records one by one. + */ + while (xlogreader->EndRecPtr < cutoff_lsn) + { + int block_id; + char *errormsg; + XLogRecord *record; + bool switch_requested; + + /* We shouldn't go backward. */ + Assert(summary_start_lsn <= xlogreader->EndRecPtr); + + /* + * This flag tracks whether the read of a particular record had to + * wait for more WAL to arrive, so reset it before reading the next + * record. + */ + private_data->waited = false; + + /* Now read the next record. */ + record = XLogReadRecord(xlogreader, &errormsg); + if (record == NULL) + { + SummarizerReadLocalXLogPrivate *private_data; + + private_data = (SummarizerReadLocalXLogPrivate *) + xlogreader->private_data; + if (private_data->end_of_wal) + { + /* + * This timeline must be historic and must end before we were + * able to read a complete record. + */ + ereport(DEBUG1, + errmsg_internal("could not read WAL from timeline %d at %X/%X: end of WAL at %X/%X", + tli, + LSN_FORMAT_ARGS(xlogreader->EndRecPtr), + LSN_FORMAT_ARGS(private_data->read_upto))); + /* Summary ends at end of WAL. */ + summary_end_lsn = private_data->read_upto; + break; + } + if (errormsg) + ereport(ERROR, + (errcode_for_file_access(), + errmsg("could not read WAL at %X/%X: %s", + LSN_FORMAT_ARGS(xlogreader->EndRecPtr), errormsg))); + else + ereport(ERROR, + (errcode_for_file_access(), + errmsg("could not read WAL at %X/%X", + LSN_FORMAT_ARGS(xlogreader->EndRecPtr)))); + } + + /* We shouldn't go backward. */ + Assert(summary_start_lsn <= xlogreader->EndRecPtr); + + if (xlogreader->ReadRecPtr >= cutoff_lsn) + { + /* + * Woops! We've read a record that *starts* after the cutoff LSN, + * contrary to our goal of reading only until we hit the first + * record that ends at or after the cutoff LSN. Pretend we didn't + * read it after all by bailing out of this loop right here, + * before we do anything with this record. + * + * This can happen because the last record before the cutoff LSN + * might be continued across multiple pages, and then we might + * come to a page with XLP_FIRST_IS_OVERWRITE_CONTRECORD set. In + * that case, the record that was continued across multiple pages + * is incomplete and will be disregarded, and the read will + * restart from the beginning of the page that is flagged + * XLP_FIRST_IS_OVERWRITE_CONTRECORD. + * + * If this case occurs, we can fairly say that the current summary + * file ends at the cutoff LSN exactly. The first record on the + * page marked XLP_FIRST_IS_OVERWRITE_CONTRECORD will be + * discovered when generating the next summary file. + */ + summary_end_lsn = cutoff_lsn; + break; + } + + /* + * We attempt, on a best effort basis only, to make WAL summary file + * boundaries line up with checkpoint cycles. So, if the last redo + * pointer we've seen was in the future, and this record starts at + * that redo pointer, stop before processing and let it be included in + * the next summary file. + * + * Note that in the case of a checkpoint triggered by a backup, the + * redo pointer is likely to be pointing to the first record on a + * page. Before reading the record, xlogreader->EndRecPtr will have + * pointed to the start of the page, which precedes the redo LSN. But + * after reading the next record, we'll advance over the page header + * and realize that the next record starts at the redo LSN exactly, + * making this the first point at which we can realize that it's time + * to stop. + */ + if (!private_data->redo_pointer_reached && + xlogreader->ReadRecPtr >= private_data->redo_pointer) + { + summary_end_lsn = xlogreader->ReadRecPtr; + break; + } + + /* Special handling for particular types of WAL records. */ + switch (XLogRecGetRmid(xlogreader)) + { + case RM_SMGR_ID: + SummarizeSmgrRecord(xlogreader, brtab); + break; + case RM_XACT_ID: + SummarizeXactRecord(xlogreader, brtab); + break; + default: + break; + } + + /* Feed block references from xlog record to block reference table. */ + for (block_id = 0; block_id <= XLogRecMaxBlockId(xlogreader); + block_id++) + { + RelFileLocator rlocator; + ForkNumber forknum; + BlockNumber blocknum; + + if (!XLogRecGetBlockTagExtended(xlogreader, block_id, &rlocator, + &forknum, &blocknum, NULL)) + continue; + + BlockRefTableMarkBlockModified(brtab, &rlocator, forknum, + blocknum); + } + + /* Update our notion of where this summary file ends. */ + summary_end_lsn = xlogreader->EndRecPtr; + + /* + * Also update shared memory, and handle any request for a WAL summary + * file switch. + */ + LWLockAcquire(WALSummarizerLock, LW_EXCLUSIVE); + Assert(summary_end_lsn >= WalSummarizerCtl->pending_lsn); + Assert(summary_end_lsn >= WalSummarizerCtl->summarized_lsn); + WalSummarizerCtl->pending_lsn = summary_end_lsn; + switch_requested = WalSummarizerCtl->switch_requested; + LWLockRelease(WALSummarizerLock); + if (switch_requested) + break; + + /* + * Periodically update our notion of the redo pointer, because it + * might be changing concurrently. There's no interlocking here: we + * might race past the new redo pointer before we learn about it. + * That's OK; we only use the redo pointer as a heuristic for where to + * stop summarizing. + * + * It would be nice if we could just fetch the updated redo pointer on + * every pass through this loop, but that seems a bit too expensive: + * GetRedoRecPtr acquires a heavily-contended spinlock. So, instead, + * just fetch the updated value if we've just had to sleep, or if + * we've read more than a segment's worth of WAL without sleeping. + */ + if (private_data->waited || xlogreader->EndRecPtr > + private_data->redo_pointer_refresh_lsn + wal_segment_size) + { + private_data->redo_pointer = GetRedoRecPtr(); + private_data->redo_pointer_refresh_lsn = xlogreader->EndRecPtr; + private_data->redo_pointer_reached = + (xlogreader->EndRecPtr >= private_data->redo_pointer); + } + + /* + * Recheck whether we've just caught up with the redo pointer, and if + * so, stop. This has the same purpose as the earlier check for the + * same condition above, but there we've just read a record and might + * decide against including it in the current summary file, whereas + * here we've already included it and might decide against reading the + * next one. Note that we may have just refreshed our notion of the + * redo pointer, so it's smart to check here before we do any more + * work. + */ + if (!private_data->redo_pointer_reached && + xlogreader->EndRecPtr >= private_data->redo_pointer) + break; + } + + /* Destroy xlogreader. */ + pfree(xlogreader->private_data); + XLogReaderFree(xlogreader); + + /* + * If a timeline switch occurs, we may fail to make any progress at all + * before exiting the loop above. If that happens, we don't write a WAL + * summary file at all. + */ + if (summary_end_lsn > summary_start_lsn) + { + /* Generate temporary and final path name. */ + snprintf(temp_path, MAXPGPATH, + XLOGDIR "/summaries/temp.summary"); + snprintf(final_path, MAXPGPATH, + XLOGDIR "/summaries/%08X%08X%08X%08X%08X.summary", + tli, + LSN_FORMAT_ARGS(summary_start_lsn), + LSN_FORMAT_ARGS(summary_end_lsn)); + + /* Open the temporary file for writing. */ + io.filepos = 0; + io.file = PathNameOpenFile(temp_path, O_WRONLY | O_CREAT | O_TRUNC); + if (io.file < 0) + ereport(ERROR, + (errcode_for_file_access(), + errmsg("could not create file \"%s\": %m", temp_path))); + + /* Write the data. */ + WriteBlockRefTable(brtab, WriteWalSummary, &io); + + /* Close temporary file and shut down xlogreader. */ + FileClose(io.file); + + /* Tell the user what we did. */ + ereport(LOG, + errmsg("summarized WAL on TLI %d from %X/%X to %X/%X", + tli, + LSN_FORMAT_ARGS(summary_start_lsn), + LSN_FORMAT_ARGS(summary_end_lsn))); + + /* Durably rename the new summary into place. */ + durable_rename(temp_path, final_path, ERROR); + } + + return summary_end_lsn; +} + +/* + * Special handling for WAL records with RM_SMGR_ID. + */ +static void +SummarizeSmgrRecord(XLogReaderState *xlogreader, BlockRefTable *brtab) +{ + uint8 info = XLogRecGetInfo(xlogreader) & ~XLR_INFO_MASK; + + if (info == XLOG_SMGR_CREATE) + { + xl_smgr_create *xlrec; + + /* + * If a new relation fork is created on disk, there is no point + * tracking anything about which blocks have been modified, because + * the whole thing will be new. Hence, set the limit block for this + * fork to 0. + * + * Ignore the FSM fork, which is not fully WAL-logged. + */ + xlrec = (xl_smgr_create *) XLogRecGetData(xlogreader); + + if (xlrec->forkNum != FSM_FORKNUM) + BlockRefTableSetLimitBlock(brtab, &xlrec->rlocator, + xlrec->forkNum, 0); + } + else if (info == XLOG_SMGR_TRUNCATE) + { + xl_smgr_truncate *xlrec; + + xlrec = (xl_smgr_truncate *) XLogRecGetData(xlogreader); + + /* + * If a relation fork is truncated on disk, there is in point in + * tracking anything about block modifications beyond the truncation + * point. + * + * We ignore SMGR_TRUNCATE_FSM here because the FSM isn't fully + * WAL-logged and thus we can't track modified blocks for it anyway. + */ + if ((xlrec->flags & SMGR_TRUNCATE_HEAP) != 0) + BlockRefTableSetLimitBlock(brtab, &xlrec->rlocator, + MAIN_FORKNUM, xlrec->blkno); + if ((xlrec->flags & SMGR_TRUNCATE_VM) != 0) + BlockRefTableSetLimitBlock(brtab, &xlrec->rlocator, + VISIBILITYMAP_FORKNUM, xlrec->blkno); + } +} + +/* + * Special handling for WAL recods with RM_XACT_ID. + */ +static void +SummarizeXactRecord(XLogReaderState *xlogreader, BlockRefTable *brtab) +{ + uint8 info = XLogRecGetInfo(xlogreader) & ~XLR_INFO_MASK; + uint8 xact_info = info & XLOG_XACT_OPMASK; + + if (xact_info == XLOG_XACT_COMMIT || + xact_info == XLOG_XACT_COMMIT_PREPARED) + { + xl_xact_commit *xlrec = (xl_xact_commit *) XLogRecGetData(xlogreader); + xl_xact_parsed_commit parsed; + int i; + + ParseCommitRecord(XLogRecGetInfo(xlogreader), xlrec, &parsed); + for (i = 0; i < parsed.nrels; ++i) + { + ForkNumber forknum; + + for (forknum = 0; forknum <= MAX_FORKNUM; ++forknum) + if (forknum != FSM_FORKNUM) + BlockRefTableSetLimitBlock(brtab, &parsed.xlocators[i], + forknum, 0); + } + } + else if (xact_info == XLOG_XACT_ABORT || + xact_info == XLOG_XACT_ABORT_PREPARED) + { + xl_xact_abort *xlrec = (xl_xact_abort *) XLogRecGetData(xlogreader); + xl_xact_parsed_abort parsed; + int i; + + ParseAbortRecord(XLogRecGetInfo(xlogreader), xlrec, &parsed); + for (i = 0; i < parsed.nrels; ++i) + { + ForkNumber forknum; + + for (forknum = 0; forknum <= MAX_FORKNUM; ++forknum) + if (forknum != FSM_FORKNUM) + BlockRefTableSetLimitBlock(brtab, &parsed.xlocators[i], + forknum, 0); + } + } +} + +/* + * Similar to read_local_xlog_page, but limited to read from one particular + * timeline. If the end of WAL is reached, it will wait for more if reading + * from the current timeline, or give up if reading from a historic timeline. + * In the latter case, it will also set private_data->end_of_wal = true. + * + * Caller must set private_data->tli to the TLI of interest, + * private_data->read_upto to the lowest LSN that is not known to be safe + * to read on that timeline, and private_data->historic to true if and only + * if the timeline is not the current timeline. This function will update + * private_data->read_upto and private_data->historic if more WAL appears + * on the current timeline or if the current timeline becomes historic. + */ +static int +summarizer_read_local_xlog_page(XLogReaderState *state, + XLogRecPtr targetPagePtr, int reqLen, + XLogRecPtr targetRecPtr, char *cur_page) +{ + int count; + WALReadError errinfo; + SummarizerReadLocalXLogPrivate *private_data; + + private_data = (SummarizerReadLocalXLogPrivate *) + state->private_data; + + while (true) + { + if (targetPagePtr + XLOG_BLCKSZ <= private_data->read_upto) + { + /* + * more than one block available; read only that block, have + * caller come back if they need more. + */ + count = XLOG_BLCKSZ; + break; + } + else if (targetPagePtr + reqLen > private_data->read_upto) + { + /* We don't seem to have enough data. */ + if (private_data->historic) + { + /* + * This is a historic timeline, so there will never be any + * more data than we have currently. + */ + private_data->end_of_wal = true; + return -1; + } + else + { + XLogRecPtr latest_lsn; + TimeLineID latest_tli; + + /* + * This is - or at least was up until very recently - the + * current timeline, so more data might show up. Delay here + * so we don't tight-loop. + */ + HandleWalSummarizerInterrupts(); + summarizer_wait_for_wal(); + private_data->waited = true; + + /* Recheck end-of-WAL. */ + latest_lsn = GetLatestLSN(&latest_tli); + if (private_data->tli == latest_tli) + { + /* Still the current timeline, update max LSN. */ + Assert(latest_lsn >= private_data->read_upto); + private_data->read_upto = latest_lsn; + } + else + { + List *tles = readTimeLineHistory(latest_tli); + XLogRecPtr switchpoint; + + /* + * The timeline we're scanning is no longer the latest + * one. Figure out when it ended and allow reads up to + * exactly that point. + */ + private_data->historic = true; + switchpoint = tliSwitchPoint(private_data->tli, tles, + NULL); + Assert(switchpoint >= private_data->read_upto); + private_data->read_upto = switchpoint; + } + + /* Go around and try again. */ + } + } + else + { + /* enough bytes available to satisfy the request */ + count = private_data->read_upto - targetPagePtr; + break; + } + } + + /* + * Even though we just determined how much of the page can be validly read + * as 'count', read the whole page anyway. It's guaranteed to be + * zero-padded up to the page boundary if it's incomplete. + */ + if (!WALRead(state, cur_page, targetPagePtr, XLOG_BLCKSZ, + private_data->tli, &errinfo)) + WALReadRaiseError(&errinfo); + + /* Track that we read a page, for sleep time calculation. */ + ++pages_read_since_last_sleep; + + /* number of valid bytes in the buffer */ + return count; +} + +/* + * Sleep for long enough that we believe it's likely that more WAL will + * be available afterwards. + */ +static void +summarizer_wait_for_wal(void) +{ + if (pages_read_since_last_sleep == 0) + { + /* + * No pages were read since the last sleep, so double the sleep time, + * but not beyond the maximum allowable value. + */ + sleep_quanta = Min(sleep_quanta * 2, MAX_SLEEP_QUANTA); + } + else if (pages_read_since_last_sleep > 1) + { + /* + * Multiple pages were read since the last sleep, so reduce the sleep + * time. + * + * A large burst of activity should be able to quickly reduce the + * sleep time to the minimum, but we don't want a handful of extra WAL + * records to provoke a strong reaction. We choose to reduce the sleep + * time by 1 quantum for each page read beyond the first, which is a + * fairly arbitrary way of trying to be reactive without + * overrreacting. + */ + if (pages_read_since_last_sleep > sleep_quanta - 1) + sleep_quanta = 1; + else + sleep_quanta -= pages_read_since_last_sleep; + } + + /* OK, now sleep. */ + (void) WaitLatch(MyLatch, + WL_LATCH_SET | WL_TIMEOUT | WL_EXIT_ON_PM_DEATH, + sleep_quanta * MS_PER_SLEEP_QUANTUM, + WAIT_EVENT_WAL_SUMMARIZER_WAL); + ResetLatch(MyLatch); + + /* Reset count of pages read. */ + pages_read_since_last_sleep = 0; +} + +/* + * Most recent RedoRecPtr value observed by RemoveOldWalSummaries. + */ +static void +MaybeRemoveOldWalSummaries(void) +{ + XLogRecPtr redo_pointer = GetRedoRecPtr(); + List *wslist; + time_t cutoff_time; + + /* If WAL summary removal is disabled, don't do anything. */ + if (wal_summarize_keep_time == 0) + return; + + /* + * If the redo pointer has not advanced, don't do anything. + * + * This has the effect that we only try to remove old WAL summary files + * once per checkpoint cycle. + */ + if (redo_pointer == redo_pointer_at_last_summary_removal) + return; + redo_pointer_at_last_summary_removal = redo_pointer; + + /* + * Files should only be removed if the last modification time precedes the + * cutoff time we compute here. + */ + cutoff_time = time(NULL) - 60 * wal_summarize_keep_time; + + /* Get all the summaries that currently exist. */ + wslist = GetWalSummaries(0, InvalidXLogRecPtr, InvalidXLogRecPtr); + + /* Loop until all summaries have been considered for removal. */ + while (wslist != NIL) + { + ListCell *lc; + XLogSegNo oldest_segno; + XLogRecPtr oldest_lsn = InvalidXLogRecPtr; + TimeLineID selected_tli; + + CHECK_FOR_INTERRUPTS(); + + /* + * Pick a timeline for which some summary files still exist on disk, + * and find the oldest LSN that still exists on disk for that + * timeline. + */ + selected_tli = ((WalSummaryFile *) linitial(wslist))->tli; + oldest_segno = XLogGetOldestSegno(selected_tli); + if (oldest_segno != 0) + XLogSegNoOffsetToRecPtr(oldest_segno, 0, wal_segment_size, + oldest_lsn); + + + /* Consider each WAL file on the selected timeline in turn. */ + foreach(lc, wslist) + { + WalSummaryFile *ws = lfirst(lc); + + CHECK_FOR_INTERRUPTS(); + + /* If it's not on this timeline, it's not time to consider it. */ + if (selected_tli != ws->tli) + continue; + + /* + * If the WAL doesn't exist any more, we can remove it if the file + * modification time is old enough. + */ + if (XLogRecPtrIsInvalid(oldest_lsn) || ws->end_lsn <= oldest_lsn) + RemoveWalSummaryIfOlderThan(ws, cutoff_time); + + /* + * Whether we we removed the file or not, we need not consider it + * again. + */ + wslist = foreach_delete_current(wslist, lc); + pfree(ws); + } + } +} diff --git a/src/backend/replication/repl_gram.y b/src/backend/replication/repl_gram.y index 0c874e33cf..a5d118ed68 100644 --- a/src/backend/replication/repl_gram.y +++ b/src/backend/replication/repl_gram.y @@ -76,11 +76,12 @@ Node *replication_parse_result; %token K_EXPORT_SNAPSHOT %token K_NOEXPORT_SNAPSHOT %token K_USE_SNAPSHOT +%token K_UPLOAD_MANIFEST %type <node> command %type <node> base_backup start_replication start_logical_replication create_replication_slot drop_replication_slot identify_system - read_replication_slot timeline_history show + read_replication_slot timeline_history show upload_manifest %type <list> generic_option_list %type <defelt> generic_option %type <uintval> opt_timeline @@ -114,6 +115,7 @@ command: | read_replication_slot | timeline_history | show + | upload_manifest ; /* @@ -307,6 +309,15 @@ timeline_history: } ; +/* UPLOAD_MANIFEST doesn't currently accept any arguments */ +upload_manifest: + K_UPLOAD_MANIFEST + { + UploadManifestCmd *cmd = makeNode(UploadManifestCmd); + + $$ = (Node *) cmd; + } + opt_physical: K_PHYSICAL | /* EMPTY */ @@ -411,6 +422,7 @@ ident_or_keyword: | K_EXPORT_SNAPSHOT { $$ = "export_snapshot"; } | K_NOEXPORT_SNAPSHOT { $$ = "noexport_snapshot"; } | K_USE_SNAPSHOT { $$ = "use_snapshot"; } + | K_UPLOAD_MANIFEST { $$ = "upload_manifest"; } ; %% diff --git a/src/backend/replication/repl_scanner.l b/src/backend/replication/repl_scanner.l index 1cc7fb858c..4805da08ee 100644 --- a/src/backend/replication/repl_scanner.l +++ b/src/backend/replication/repl_scanner.l @@ -136,6 +136,7 @@ EXPORT_SNAPSHOT { return K_EXPORT_SNAPSHOT; } NOEXPORT_SNAPSHOT { return K_NOEXPORT_SNAPSHOT; } USE_SNAPSHOT { return K_USE_SNAPSHOT; } WAIT { return K_WAIT; } +UPLOAD_MANIFEST { return K_UPLOAD_MANIFEST; } {space}+ { /* do nothing */ } @@ -303,6 +304,7 @@ replication_scanner_is_replication_command(void) case K_DROP_REPLICATION_SLOT: case K_READ_REPLICATION_SLOT: case K_TIMELINE_HISTORY: + case K_UPLOAD_MANIFEST: case K_SHOW: /* Yes; push back the first token so we can parse later. */ repl_pushed_back_token = first_token; diff --git a/src/backend/replication/walsender.c b/src/backend/replication/walsender.c index e250b0567e..b33b86671b 100644 --- a/src/backend/replication/walsender.c +++ b/src/backend/replication/walsender.c @@ -58,6 +58,7 @@ #include "access/xlogrecovery.h" #include "access/xlogutils.h" #include "backup/basebackup.h" +#include "backup/basebackup_incremental.h" #include "catalog/pg_authid.h" #include "catalog/pg_type.h" #include "commands/dbcommands.h" @@ -137,6 +138,17 @@ bool wake_wal_senders = false; */ static XLogReaderState *xlogreader = NULL; +/* + * If the UPLOAD_MANIFEST command is used to provide a backup manifest in + * preparation for an incremental backup, uploaded_manifest will be point + * to an object containing information about its contexts, and + * uploaded_manifest_mcxt will point to the memory context that contains + * that object and all of its subordinate data. Otherwise, both values will + * be NULL. + */ +static IncrementalBackupInfo *uploaded_manifest = NULL; +static MemoryContext uploaded_manifest_mcxt = NULL; + /* * These variables keep track of the state of the timeline we're currently * sending. sendTimeLine identifies the timeline. If sendTimeLineIsHistoric, @@ -233,6 +245,9 @@ static void XLogSendLogical(void); static void WalSndDone(WalSndSendDataCallback send_data); static XLogRecPtr GetStandbyFlushRecPtr(TimeLineID *tli); static void IdentifySystem(void); +static void UploadManifest(void); +static bool HandleUploadManifestPacket(StringInfo buf, off_t *offset, + IncrementalBackupInfo *ib); static void ReadReplicationSlot(ReadReplicationSlotCmd *cmd); static void CreateReplicationSlot(CreateReplicationSlotCmd *cmd); static void DropReplicationSlot(DropReplicationSlotCmd *cmd); @@ -660,6 +675,143 @@ SendTimeLineHistory(TimeLineHistoryCmd *cmd) pq_endmessage(&buf); } +/* + * Handle UPLOAD_MANIFEST command. + */ +static void +UploadManifest(void) +{ + MemoryContext mcxt; + IncrementalBackupInfo *ib; + off_t offset = 0; + StringInfoData buf; + + /* + * parsing the manifest will use the cryptohash stuff, which requires a + * resource owner + */ + Assert(CurrentResourceOwner == NULL); + CurrentResourceOwner = ResourceOwnerCreate(NULL, "base backup"); + + /* Prepare to read manifest data into a temporary context. */ + mcxt = AllocSetContextCreate(CurrentMemoryContext, + "incremental backup information", + ALLOCSET_DEFAULT_SIZES); + ib = CreateIncrementalBackupInfo(mcxt); + + /* Send a CopyInResponse message */ + pq_beginmessage(&buf, 'G'); + pq_sendbyte(&buf, 0); + pq_sendint16(&buf, 0); + pq_endmessage_reuse(&buf); + pq_flush(); + + /* Recieve packets from client until done. */ + while (HandleUploadManifestPacket(&buf, &offset, ib)) + ; + + /* Finish up manifest processing. */ + FinalizeIncrementalManifest(ib); + + /* + * Discard any old manifest information and arrange to preserve the new + * information we just got. + * + * We assume that MemoryContextDelete and MemoryContextSetParent won't + * fail, and thus we shouldn't end up bailing out of here in such a way as + * to leave dangling pointrs. + */ + if (uploaded_manifest_mcxt != NULL) + MemoryContextDelete(uploaded_manifest_mcxt); + MemoryContextSetParent(mcxt, CacheMemoryContext); + uploaded_manifest = ib; + uploaded_manifest_mcxt = mcxt; + + /* clean up the resource owner we created */ + WalSndResourceCleanup(true); +} + +/* + * Process one packet received during the handling of an UPLOAD_MANIFEST + * operation. + * + * 'buf' is scratch space. This function expects it to be initialized, doesn't + * care what the current contents are, and may override them with completely + * new contents. + * + * The return value is true if the caller should continue processing + * additional packets and false if the UPLOAD_MANIFEST operation is complete. + */ +static bool +HandleUploadManifestPacket(StringInfo buf, off_t *offset, + IncrementalBackupInfo *ib) +{ + int mtype; + int maxmsglen; + + HOLD_CANCEL_INTERRUPTS(); + + pq_startmsgread(); + mtype = pq_getbyte(); + if (mtype == EOF) + ereport(ERROR, + (errcode(ERRCODE_CONNECTION_FAILURE), + errmsg("unexpected EOF on client connection with an open transaction"))); + + switch (mtype) + { + case 'd': /* CopyData */ + maxmsglen = PQ_LARGE_MESSAGE_LIMIT; + break; + case 'c': /* CopyDone */ + case 'f': /* CopyFail */ + case 'H': /* Flush */ + case 'S': /* Sync */ + maxmsglen = PQ_SMALL_MESSAGE_LIMIT; + break; + default: + ereport(ERROR, + (errcode(ERRCODE_PROTOCOL_VIOLATION), + errmsg("unexpected message type 0x%02X during COPY from stdin", + mtype))); + maxmsglen = 0; /* keep compiler quiet */ + break; + } + + /* Now collect the message body */ + if (pq_getmessage(buf, maxmsglen)) + ereport(ERROR, + (errcode(ERRCODE_CONNECTION_FAILURE), + errmsg("unexpected EOF on client connection with an open transaction"))); + RESUME_CANCEL_INTERRUPTS(); + + /* Process the message */ + switch (mtype) + { + case 'd': /* CopyData */ + AppendIncrementalManifestData(ib, buf->data, buf->len); + return true; + + case 'c': /* CopyDone */ + return false; + + case 'H': /* Sync */ + case 'S': /* Flush */ + /* Ignore these while in CopyOut mode as we do elsewhere. */ + return true; + + case 'f': + ereport(ERROR, + (errcode(ERRCODE_QUERY_CANCELED), + errmsg("COPY from stdin failed: %s", + pq_getmsgstring(buf)))); + } + + /* Not reached. */ + Assert(false); + return false; +} + /* * Handle START_REPLICATION command. * @@ -1801,7 +1953,7 @@ exec_replication_command(const char *cmd_string) cmdtag = "BASE_BACKUP"; set_ps_display(cmdtag); PreventInTransactionBlock(true, cmdtag); - SendBaseBackup((BaseBackupCmd *) cmd_node); + SendBaseBackup((BaseBackupCmd *) cmd_node, uploaded_manifest); EndReplicationCommand(cmdtag); break; @@ -1863,6 +2015,14 @@ exec_replication_command(const char *cmd_string) } break; + case T_UploadManifestCmd: + cmdtag = "UPLOAD_MANIFEST"; + set_ps_display(cmdtag); + PreventInTransactionBlock(true, cmdtag); + UploadManifest(); + EndReplicationCommand(cmdtag); + break; + default: elog(ERROR, "unrecognized replication command node tag: %u", cmd_node->type); diff --git a/src/backend/storage/ipc/ipci.c b/src/backend/storage/ipc/ipci.c index a3d8eacb8d..3a6729003a 100644 --- a/src/backend/storage/ipc/ipci.c +++ b/src/backend/storage/ipc/ipci.c @@ -31,6 +31,7 @@ #include "postmaster/bgworker_internals.h" #include "postmaster/bgwriter.h" #include "postmaster/postmaster.h" +#include "postmaster/walsummarizer.h" #include "replication/logicallauncher.h" #include "replication/origin.h" #include "replication/slot.h" @@ -136,6 +137,7 @@ CalculateShmemSize(int *num_semaphores) size = add_size(size, ReplicationOriginShmemSize()); size = add_size(size, WalSndShmemSize()); size = add_size(size, WalRcvShmemSize()); + size = add_size(size, WalSummarizerShmemSize()); size = add_size(size, PgArchShmemSize()); size = add_size(size, ApplyLauncherShmemSize()); size = add_size(size, BTreeShmemSize()); @@ -291,6 +293,7 @@ CreateSharedMemoryAndSemaphores(void) ReplicationOriginShmemInit(); WalSndShmemInit(); WalRcvShmemInit(); + WalSummarizerShmemInit(); PgArchShmemInit(); ApplyLauncherShmemInit(); diff --git a/src/backend/storage/lmgr/lwlocknames.txt b/src/backend/storage/lmgr/lwlocknames.txt index f72f2906ce..d621f5507f 100644 --- a/src/backend/storage/lmgr/lwlocknames.txt +++ b/src/backend/storage/lmgr/lwlocknames.txt @@ -54,3 +54,4 @@ XactTruncationLock 44 WrapLimitsVacuumLock 46 NotifyQueueTailLock 47 WaitEventExtensionLock 48 +WALSummarizerLock 49 diff --git a/src/backend/utils/activity/pgstat_io.c b/src/backend/utils/activity/pgstat_io.c index eb7d35d422..bd0a921a3e 100644 --- a/src/backend/utils/activity/pgstat_io.c +++ b/src/backend/utils/activity/pgstat_io.c @@ -292,7 +292,8 @@ pgstat_io_snapshot_cb(void) * - Syslogger because it is not connected to shared memory * - Archiver because most relevant archiving IO is delegated to a * specialized command or module -* - WAL Receiver and WAL Writer IO is not tracked in pg_stat_io for now +* - WAL Receiver, WAL Writer, and WAL Summarizer IO are not tracked in +* pg_stat_io for now * * Function returns true if BackendType participates in the cumulative stats * subsystem for IO and false if it does not. @@ -314,6 +315,7 @@ pgstat_tracks_io_bktype(BackendType bktype) case B_LOGGER: case B_WAL_RECEIVER: case B_WAL_WRITER: + case B_WAL_SUMMARIZER: return false; case B_AUTOVAC_LAUNCHER: diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 9c5fdeb3ca..17ad986c98 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -56,6 +56,7 @@ RECOVERY_WAL_STREAM "Waiting in main loop of startup process for WAL to arrive, SYSLOGGER_MAIN "Waiting in main loop of syslogger process." WAL_RECEIVER_MAIN "Waiting in main loop of WAL receiver process." WAL_SENDER_MAIN "Waiting in main loop of WAL sender process." +WAL_SUMMARIZER_WAL "Waiting in WAL summarizer for more WAL to be generated." WAL_WRITER_MAIN "Waiting in main loop of WAL writer process." @@ -140,6 +141,7 @@ SAFE_SNAPSHOT "Waiting to obtain a valid snapshot for a <literal>READ ONLY DEFER SYNC_REP "Waiting for confirmation from a remote server during synchronous replication." WAL_RECEIVER_EXIT "Waiting for the WAL receiver to exit." WAL_RECEIVER_WAIT_START "Waiting for startup process to send initial data for streaming replication." +WAL_SUMMARY_READY "Waiting for a new WAL summary to be generated." XACT_GROUP_UPDATE "Waiting for the group leader to update transaction status at end of a parallel operation." @@ -160,6 +162,7 @@ REGISTER_SYNC_REQUEST "Waiting while sending synchronization requests to the che SPIN_DELAY "Waiting while acquiring a contended spinlock." VACUUM_DELAY "Waiting in a cost-based vacuum delay point." VACUUM_TRUNCATE "Waiting to acquire an exclusive lock to truncate off any empty pages at the end of a table vacuumed." +WAL_SUMMARIZER_ERROR "Waiting after a WAL summarizer error." # @@ -241,6 +244,8 @@ WAL_COPY_WRITE "Waiting for a write when creating a new WAL segment by copying a WAL_INIT_SYNC "Waiting for a newly initialized WAL file to reach durable storage." WAL_INIT_WRITE "Waiting for a write while initializing a new WAL file." WAL_READ "Waiting for a read from a WAL file." +WAL_SUMMARY_READ "Waiting for a read from a WAL summary file." +WAL_SUMMARY_WRITE "Waiting for a write to a WAL summary file." WAL_SYNC "Waiting for a WAL file to reach durable storage." WAL_SYNC_METHOD_ASSIGN "Waiting for data to reach durable storage while assigning a new WAL sync method." WAL_WRITE "Waiting for a write to a WAL file." diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c index 1e671c560c..037111b89f 100644 --- a/src/backend/utils/init/miscinit.c +++ b/src/backend/utils/init/miscinit.c @@ -306,6 +306,9 @@ GetBackendTypeDesc(BackendType backendType) case B_WAL_SENDER: backendDesc = "walsender"; break; + case B_WAL_SUMMARIZER: + backendDesc = "walsummarizer"; + break; case B_WAL_WRITER: backendDesc = "walwriter"; break; diff --git a/src/backend/utils/misc/guc_tables.c b/src/backend/utils/misc/guc_tables.c index 16ec6c5ef0..a532f57af1 100644 --- a/src/backend/utils/misc/guc_tables.c +++ b/src/backend/utils/misc/guc_tables.c @@ -63,6 +63,7 @@ #include "postmaster/postmaster.h" #include "postmaster/startup.h" #include "postmaster/syslogger.h" +#include "postmaster/walsummarizer.h" #include "postmaster/walwriter.h" #include "replication/logicallauncher.h" #include "replication/slot.h" @@ -704,6 +705,8 @@ const char *const config_group_names[] = gettext_noop("Write-Ahead Log / Archive Recovery"), /* WAL_RECOVERY_TARGET */ gettext_noop("Write-Ahead Log / Recovery Target"), + /* WAL_SUMMARIZATION */ + gettext_noop("Write-Ahead Log / Summarization"), /* REPLICATION_SENDING */ gettext_noop("Replication / Sending Servers"), /* REPLICATION_PRIMARY */ @@ -3181,6 +3184,32 @@ struct config_int ConfigureNamesInt[] = check_wal_segment_size, NULL, NULL }, + { + {"wal_summarize_mb", PGC_SIGHUP, WAL_SUMMARIZATION, + gettext_noop("Number of bytes of WAL per summary file."), + gettext_noop("Smaller values minimize extra work performed by incremental backup, but increase the number of files on disk."), + GUC_UNIT_MB, + }, + &wal_summarize_mb, + 256, + 0, + INT_MAX, + NULL, NULL, NULL + }, + + { + {"wal_summarize_keep_time", PGC_SIGHUP, WAL_SUMMARIZATION, + gettext_noop("Time for which WAL summary files should be kept."), + NULL, + GUC_UNIT_MIN, + }, + &wal_summarize_keep_time, + 7 * 24 * 60, /* 1 week */ + 0, + INT_MAX, + NULL, NULL, NULL + }, + { {"autovacuum_naptime", PGC_SIGHUP, AUTOVACUUM, gettext_noop("Time to sleep between autovacuum runs."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index d08d55c3fe..4736606ac1 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -299,6 +299,11 @@ #recovery_target_action = 'pause' # 'pause', 'promote', 'shutdown' # (change requires restart) +# - WAL Summarization - + +#wal_summarize_mb = 256 # MB of WAL per summary file, 0 disables +#wal_summarize_keep_time = '7d' # when to remove old summary files, 0 = never + #------------------------------------------------------------------------------ # REPLICATION diff --git a/src/bin/Makefile b/src/bin/Makefile index 373077bf52..aa2210925e 100644 --- a/src/bin/Makefile +++ b/src/bin/Makefile @@ -19,6 +19,7 @@ SUBDIRS = \ pg_archivecleanup \ pg_basebackup \ pg_checksums \ + pg_combinebackup \ pg_config \ pg_controldata \ pg_ctl \ diff --git a/src/bin/initdb/initdb.c b/src/bin/initdb/initdb.c index 0c6f5ceb0a..e68b40d2b5 100644 --- a/src/bin/initdb/initdb.c +++ b/src/bin/initdb/initdb.c @@ -227,6 +227,7 @@ static char *extra_options = ""; static const char *const subdirs[] = { "global", "pg_wal/archive_status", + "pg_wal/summaries", "pg_commit_ts", "pg_dynshmem", "pg_notify", diff --git a/src/bin/meson.build b/src/bin/meson.build index 67cb50630c..4cb6fd59bb 100644 --- a/src/bin/meson.build +++ b/src/bin/meson.build @@ -5,6 +5,7 @@ subdir('pg_amcheck') subdir('pg_archivecleanup') subdir('pg_basebackup') subdir('pg_checksums') +subdir('pg_combinebackup') subdir('pg_config') subdir('pg_controldata') subdir('pg_ctl') diff --git a/src/bin/pg_basebackup/bbstreamer_file.c b/src/bin/pg_basebackup/bbstreamer_file.c index 45f32974ff..6b78ee283d 100644 --- a/src/bin/pg_basebackup/bbstreamer_file.c +++ b/src/bin/pg_basebackup/bbstreamer_file.c @@ -296,6 +296,7 @@ should_allow_existing_directory(const char *pathname) if (strcmp(filename, "pg_wal") == 0 || strcmp(filename, "pg_xlog") == 0 || strcmp(filename, "archive_status") == 0 || + strcmp(filename, "summaries") == 0 || strcmp(filename, "pg_tblspc") == 0) return true; diff --git a/src/bin/pg_basebackup/pg_basebackup.c b/src/bin/pg_basebackup/pg_basebackup.c index 1a8cef345d..2e27fb58c6 100644 --- a/src/bin/pg_basebackup/pg_basebackup.c +++ b/src/bin/pg_basebackup/pg_basebackup.c @@ -101,6 +101,11 @@ typedef void (*WriteDataCallback) (size_t nbytes, char *buf, */ #define MINIMUM_VERSION_FOR_TERMINATED_TARFILE 150000 +/* + * pg_wal/summaries exists beginning with version 17. + */ +#define MINIMUM_VERSION_FOR_WAL_SUMMARIES 170000 + /* * Different ways to include WAL */ @@ -217,7 +222,8 @@ static void ReceiveBackupManifestInMemoryChunk(size_t r, char *copybuf, void *callback_data); static void BaseBackup(char *compression_algorithm, char *compression_detail, CompressionLocation compressloc, - pg_compress_specification *client_compress); + pg_compress_specification *client_compress, + char *incremental_manifest); static bool reached_end_position(XLogRecPtr segendpos, uint32 timeline, bool segment_finished); @@ -688,6 +694,23 @@ StartLogStreamer(char *startpos, uint32 timeline, char *sysidentifier, if (pg_mkdir_p(statusdir, pg_dir_create_mode) != 0 && errno != EEXIST) pg_fatal("could not create directory \"%s\": %m", statusdir); + + /* + * For newer server versions, likewise create pg_wal/summaries + */ + if (PQserverVersion(conn) < MINIMUM_VERSION_FOR_WAL_SUMMARIES) + { + char summarydir[MAXPGPATH]; + + snprintf(summarydir, sizeof(summarydir), "%s/%s/summaries", + basedir, + PQserverVersion(conn) < MINIMUM_VERSION_FOR_PG_WAL ? + "pg_xlog" : "pg_wal"); + + if (pg_mkdir_p(statusdir, pg_dir_create_mode) != 0 && + errno != EEXIST) + pg_fatal("could not create directory \"%s\": %m", summarydir); + } } /* @@ -1728,7 +1751,9 @@ ReceiveBackupManifestInMemoryChunk(size_t r, char *copybuf, static void BaseBackup(char *compression_algorithm, char *compression_detail, - CompressionLocation compressloc, pg_compress_specification *client_compress) + CompressionLocation compressloc, + pg_compress_specification *client_compress, + char *incremental_manifest) { PGresult *res; char *sysidentifier; @@ -1794,7 +1819,74 @@ BaseBackup(char *compression_algorithm, char *compression_detail, exit(1); /* - * Start the actual backup + * If the user wants an incremental backup, we must upload the manifest + * for the previous backup upon which it is to be based. + */ + if (incremental_manifest != NULL) + { + int fd; + char mbuf[65536]; + int nbytes; + + /* XXX add a server version check here */ + + /* Open the file. */ + fd = open(incremental_manifest, O_RDONLY | PG_BINARY, 0); + if (fd < 0) + pg_fatal("could not open file \"%s\": %m", incremental_manifest); + + /* Tell the server what we want to do. */ + if (PQsendQuery(conn, "UPLOAD_MANIFEST") == 0) + pg_fatal("could not send replication command \"%s\": %s", + "UPLOAD_MANIFEST", PQerrorMessage(conn)); + res = PQgetResult(conn); + if (PQresultStatus(res) != PGRES_COPY_IN) + { + if (PQresultStatus(res) == PGRES_FATAL_ERROR) + pg_fatal("could not upload manifest: %s", + PQerrorMessage(conn)); + else + pg_fatal("could not upload manifest: unexpected status %s", + PQresStatus(PQresultStatus(res))); + } + + /* Loop, reading from the file and sending the data to the server. */ + while ((nbytes = read(fd, mbuf, sizeof mbuf)) > 0) + { + if (PQputCopyData(conn, mbuf, nbytes) < 0) + pg_fatal("could not send COPY data: %s", + PQerrorMessage(conn)); + } + + /* Bail out if we exited the loop due to an error. */ + if (nbytes < 0) + pg_fatal("could not read file \"%s\": %m", incremental_manifest); + + /* End the COPY operation. */ + if (PQputCopyEnd(conn, NULL) < 0) + pg_fatal("could not send end-of-COPY: %s", + PQerrorMessage(conn)); + + /* See whether the server is happy with what we sent. */ + res = PQgetResult(conn); + if (PQresultStatus(res) == PGRES_FATAL_ERROR) + pg_fatal("could not upload manifest: %s", + PQerrorMessage(conn)); + else if (PQresultStatus(res) != PGRES_COMMAND_OK) + pg_fatal("could not upload manifest: unexpected status %s", + PQresStatus(PQresultStatus(res))); + + /* Consume ReadyForQuery message from server. */ + res = PQgetResult(conn); + if (res != NULL) + pg_fatal("unexpected extra result while sending manifest"); + + /* Add INCREMENTAL option to BASE_BACKUP command. */ + AppendPlainCommandOption(&buf, use_new_option_syntax, "INCREMENTAL"); + } + + /* + * Continue building up the options list for the BASE_BACKUP command. */ AppendStringCommandOption(&buf, use_new_option_syntax, "LABEL", label); if (estimatesize) @@ -1901,6 +1993,7 @@ BaseBackup(char *compression_algorithm, char *compression_detail, else basebkp = psprintf("BASE_BACKUP %s", buf.data); + /* OK, try to start the backup. */ if (PQsendQuery(conn, basebkp) == 0) pg_fatal("could not send replication command \"%s\": %s", "BASE_BACKUP", PQerrorMessage(conn)); @@ -2256,6 +2349,7 @@ main(int argc, char **argv) {"version", no_argument, NULL, 'V'}, {"pgdata", required_argument, NULL, 'D'}, {"format", required_argument, NULL, 'F'}, + {"incremental", required_argument, NULL, 'i'}, {"checkpoint", required_argument, NULL, 'c'}, {"create-slot", no_argument, NULL, 'C'}, {"max-rate", required_argument, NULL, 'r'}, @@ -2293,6 +2387,7 @@ main(int argc, char **argv) int option_index; char *compression_algorithm = "none"; char *compression_detail = NULL; + char *incremental_manifest = NULL; CompressionLocation compressloc = COMPRESS_LOCATION_UNSPECIFIED; pg_compress_specification client_compress; @@ -2317,7 +2412,7 @@ main(int argc, char **argv) atexit(cleanup_directories_atexit); - while ((c = getopt_long(argc, argv, "c:Cd:D:F:h:l:nNp:Pr:Rs:S:t:T:U:vwWX:zZ:", + while ((c = getopt_long(argc, argv, "c:Cd:D:F:h:i:l:nNp:Pr:Rs:S:t:T:U:vwWX:zZ:", long_options, &option_index)) != -1) { switch (c) @@ -2352,6 +2447,9 @@ main(int argc, char **argv) case 'h': dbhost = pg_strdup(optarg); break; + case 'i': + incremental_manifest = pg_strdup(optarg); + break; case 'l': label = pg_strdup(optarg); break; @@ -2765,7 +2863,7 @@ main(int argc, char **argv) } BaseBackup(compression_algorithm, compression_detail, compressloc, - &client_compress); + &client_compress, incremental_manifest); success = true; return 0; diff --git a/src/bin/pg_basebackup/t/010_pg_basebackup.pl b/src/bin/pg_basebackup/t/010_pg_basebackup.pl index b9f5e1266b..bf765291e7 100644 --- a/src/bin/pg_basebackup/t/010_pg_basebackup.pl +++ b/src/bin/pg_basebackup/t/010_pg_basebackup.pl @@ -223,10 +223,10 @@ SKIP: "check backup dir permissions"); } -# Only archive_status directory should be copied in pg_wal/. +# Only archive_status and summaries directories should be copied in pg_wal/. is_deeply( [ sort(slurp_dir("$tempdir/backup/pg_wal/")) ], - [ sort qw(. .. archive_status) ], + [ sort qw(. .. archive_status summaries) ], 'no WAL files copied'); # Contents of these directories should not be copied. diff --git a/src/bin/pg_combinebackup/.gitignore b/src/bin/pg_combinebackup/.gitignore new file mode 100644 index 0000000000..d7e617438c --- /dev/null +++ b/src/bin/pg_combinebackup/.gitignore @@ -0,0 +1 @@ +pg_combinebackup diff --git a/src/bin/pg_combinebackup/Makefile b/src/bin/pg_combinebackup/Makefile new file mode 100644 index 0000000000..cb20480aae --- /dev/null +++ b/src/bin/pg_combinebackup/Makefile @@ -0,0 +1,46 @@ +#------------------------------------------------------------------------- +# +# Makefile for src/bin/pg_combinebackup +# +# Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group +# Portions Copyright (c) 1994, Regents of the University of California +# +# src/bin/pg_combinebackup/Makefile +# +#------------------------------------------------------------------------- + +PGFILEDESC = "pg_combinebackup - combine incremental backups" +PGAPPICON=win32 + +subdir = src/bin/pg_combinebackup +top_builddir = ../../.. +include $(top_builddir)/src/Makefile.global + +override CPPFLAGS := -I$(libpq_srcdir) $(CPPFLAGS) +LDFLAGS_INTERNAL += -L$(top_builddir)/src/fe_utils -lpgfeutils + +OBJS = \ + $(WIN32RES) \ + pg_combinebackup.o \ + backup_label.o \ + copy_file.o \ + load_manifest.o \ + reconstruct.o \ + write_manifest.o + +all: pg_combinebackup + +pg_combinebackup: $(OBJS) | submake-libpgport submake-libpgfeutils + $(CC) $(CFLAGS) $^ $(LDFLAGS) $(LDFLAGS_EX) $(LIBS) -o $@$(X) + +install: all installdirs + $(INSTALL_PROGRAM) pg_combinebackup$(X) '$(DESTDIR)$(bindir)/pg_combinebackup$(X)' + +installdirs: + $(MKDIR_P) '$(DESTDIR)$(bindir)' + +uninstall: + rm -f '$(DESTDIR)$(bindir)/pg_combinebackup$(X)' + +clean distclean maintainer-clean: + rm -f pg_combinebackup$(X) $(OBJS) diff --git a/src/bin/pg_combinebackup/backup_label.c b/src/bin/pg_combinebackup/backup_label.c new file mode 100644 index 0000000000..2a62aa6fad --- /dev/null +++ b/src/bin/pg_combinebackup/backup_label.c @@ -0,0 +1,281 @@ +/*------------------------------------------------------------------------- + * + * Read and manipulate backup label files + * + * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group + * Portions Copyright (c) 1994, Regents of the University of California + * + * src/bin/pg_combinebackup/backup_label.c + * + *------------------------------------------------------------------------- + */ +#include "postgres_fe.h" + +#include <unistd.h> + +#include "access/xlogdefs.h" +#include "backup_label.h" +#include "common/logging.h" +#include "common/file_perm.h" +#include "write_manifest.h" + +static int get_eol_offset(StringInfo buf); +static bool line_starts_with(char *s, char *e, char *match, char **sout); +static bool parse_lsn(char *s, char *e, XLogRecPtr *lsn, char **c); +static bool parse_tli(char *s, char *e, TimeLineID *tli); + +/* + * Parse a backup label file, starting at buf->cursor. + * + * We expect to find a START WAL LOCATION line, followed by a LSN, followed + * by a space; the resulting LSN is stored into *start_lsn. + * + * We expect to find a START TIMELINE line, followed by a TLI, followed by + * a newline; the resulting TLI is stored into *start_tli. + * + * We expect to find either both INCREMENTAL FROM LSN and INCREMENTAL FROM TLI + * or neither. If these are found, they should be followed by an LSN or TLI + * respectively and then by a newline, and the values will be stored into + * *previous_lsn and *previous_tli, respectively. + * + * Other lines in the provided backup_label data are ignored. filename is used + * for error reporting; errors are fatal. + */ +void +parse_backup_label(char *filename, StringInfo buf, + TimeLineID *start_tli, XLogRecPtr *start_lsn, + TimeLineID *previous_tli, XLogRecPtr *previous_lsn) +{ + int found = 0; + + *start_tli = 0; + *start_lsn = InvalidXLogRecPtr; + *previous_tli = 0; + *previous_lsn = InvalidXLogRecPtr; + + while (buf->cursor < buf->len) + { + char *s = &buf->data[buf->cursor]; + int eo = get_eol_offset(buf); + char *e = &buf->data[eo]; + char *c; + + if (line_starts_with(s, e, "START WAL LOCATION: ", &s)) + { + if (!parse_lsn(s, e, start_lsn, &c)) + pg_fatal("%s: could not parse START WAL LOCATION", + filename); + if (c >= e || *c != ' ') + pg_fatal("%s: improper terminator for START WAL LOCATION", + filename); + found |= 1; + } + else if (line_starts_with(s, e, "START TIMELINE: ", &s)) + { + if (!parse_tli(s, e, start_tli)) + pg_fatal("%s: could not parse TLI for START TIMELINE", + filename); + if (*start_tli == 0) + pg_fatal("%s: invalid TLI", filename); + found |= 2; + } + else if (line_starts_with(s, e, "INCREMENTAL FROM LSN: ", &s)) + { + if (!parse_lsn(s, e, previous_lsn, &c)) + pg_fatal("%s: could not parse INCREMENTAL FROM LSN", + filename); + if (c >= e || *c != '\n') + pg_fatal("%s: improper terminator for INCREMENTAL FROM LSN", + filename); + found |= 4; + } + else if (line_starts_with(s, e, "INCREMENTAL FROM TLI: ", &s)) + { + if (!parse_tli(s, e, previous_tli)) + pg_fatal("%s: could not parse INCREMENTAL FROM TLI", + filename); + if (*previous_tli == 0) + pg_fatal("%s: invalid TLI", filename); + found |= 8; + } + + buf->cursor = eo; + } + + if ((found & 1) == 0) + pg_fatal("%s: could not find START WAL LOCATION", filename); + if ((found & 2) == 0) + pg_fatal("%s: could not find START TIMELINE", filename); + if ((found & 4) != 0 && (found & 8) == 0) + pg_fatal("%s: INCREMENTAL FROM LSN requires INCREMENTAL FROM TLI", filename); + if ((found & 8) != 0 && (found & 4) == 0) + pg_fatal("%s: INCREMENTAL FROM TLI requires INCREMENTAL FROM LSN", filename); +} + +/* + * Write a backup label file to the output directory. + * + * This will be identical to the provided backup_label file, except that the + * INCREMENTAL FROM LSN and INCREMENTAL FROM TLI lines will be omitted. + * + * The new file will be checksummed using the specified algorithm. If + * mwriter != NULL, it will be added to the manifest. + */ +void +write_backup_label(char *output_directory, StringInfo buf, + pg_checksum_type checksum_type, manifest_writer *mwriter) +{ + char output_filename[MAXPGPATH]; + int output_fd; + pg_checksum_context checksum_ctx; + uint8 checksum_payload[PG_CHECKSUM_MAX_LENGTH]; + int checksum_length; + + pg_checksum_init(&checksum_ctx, checksum_type); + + snprintf(output_filename, MAXPGPATH, "%s/backup_label", output_directory); + + if ((output_fd = open(output_filename, + O_WRONLY | O_CREAT | O_EXCL | PG_BINARY, + pg_file_create_mode)) < 0) + pg_fatal("could not open file \"%s\": %m", output_filename); + + while (buf->cursor < buf->len) + { + char *s = &buf->data[buf->cursor]; + int eo = get_eol_offset(buf); + char *e = &buf->data[eo]; + + if (!line_starts_with(s, e, "INCREMENTAL FROM LSN: ", NULL) && + !line_starts_with(s, e, "INCREMENTAL FROM TLI: ", NULL)) + { + ssize_t wb; + + wb = write(output_fd, s, e - s); + if (wb != e - s) + { + if (wb < 0) + pg_fatal("could not write file \"%s\": %m", output_filename); + else + pg_fatal("could not write file \"%s\": wrote only %d of %d bytes", + output_filename, (int) wb, (int) (e - s)); + } + if (pg_checksum_update(&checksum_ctx, (uint8 *) s, e - s) < 0) + pg_fatal("could not update checksum of file \"%s\"", + output_filename); + } + + buf->cursor = eo; + } + + if (close(output_fd) != 0) + pg_fatal("could not close \"%s\": %m", output_filename); + + checksum_length = pg_checksum_final(&checksum_ctx, checksum_payload); + + if (mwriter != NULL) + { + struct stat sb; + + /* + * We could track the length ourselves, but must stat() to get the + * mtime. + */ + if (stat(output_filename, &sb) < 0) + pg_fatal("could not stat file \"%s\": %m", output_filename); + add_file_to_manifest(mwriter, "backup_label", sb.st_size, + sb.st_mtime, checksum_type, + checksum_length, checksum_payload); + } +} + +/* + * Return the offset at which the next line in the buffer starts, or there + * is none, the offset at which the buffer ends. + * + * The search begins at buf->cursor. + */ +static int +get_eol_offset(StringInfo buf) +{ + int eo = buf->cursor; + + while (eo < buf->len) + { + if (buf->data[eo] == '\n') + return eo + 1; + ++eo; + } + + return eo; +} + +/* + * Test whether the line that runs from s to e (inclusive of *s, but not + * inclusive of *e) starts with the match string provided, and return true + * or false according to whether or not this is the case. + * + * If the function returns true and if *sout != NULL, stores a pointer to the + * byte following the match into *sout. + */ +static bool +line_starts_with(char *s, char *e, char *match, char **sout) +{ + while (s < e && *match != '\0' && *s == *match) + ++s, ++match; + + if (*match == '\0' && sout != NULL) + *sout = s; + + return (*match == '\0'); +} + +/* + * Parse an LSN starting at s and not stopping at or before e. The return value + * is true on success and otherwise false. On success, stores the result into + * *lsn and sets *c to the first character that is not part of the LSN. + */ +static bool +parse_lsn(char *s, char *e, XLogRecPtr *lsn, char **c) +{ + char save = *e; + int nchars; + bool success; + unsigned hi; + unsigned lo; + + *e = '\0'; + success = (sscanf(s, "%X/%X%n", &hi, &lo, &nchars) == 2); + *e = save; + + if (success) + { + *lsn = ((XLogRecPtr) hi) << 32 | (XLogRecPtr) lo; + *c = s + nchars; + } + + return success; +} + +/* + * Parse a TLI starting at s and stopping at or before e. The return value is + * true on success and otherwise false. On success, stores the result into + * *tli. If the first character that is not part of the TLI is anything other + * than a newline, that is deemed a failure. + */ +static bool +parse_tli(char *s, char *e, TimeLineID *tli) +{ + char save = *e; + int nchars; + bool success; + + *e = '\0'; + success = (sscanf(s, "%u%n", tli, &nchars) == 1); + *e = save; + + if (success && s[nchars] != '\n') + success = false; + + return success; +} diff --git a/src/bin/pg_combinebackup/backup_label.h b/src/bin/pg_combinebackup/backup_label.h new file mode 100644 index 0000000000..08d6ed67a9 --- /dev/null +++ b/src/bin/pg_combinebackup/backup_label.h @@ -0,0 +1,29 @@ +/*------------------------------------------------------------------------- + * + * Read and manipulate backup label files + * + * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group + * Portions Copyright (c) 1994, Regents of the University of California + * + * src/bin/pg_combinebackup/backup_label.h + * + *------------------------------------------------------------------------- + */ +#ifndef BACKUP_LABEL_H +#define BACKUP_LABEL_H + +#include "common/checksum_helper.h" +#include "lib/stringinfo.h" + +struct manifest_writer; + +extern void parse_backup_label(char *filename, StringInfo buf, + TimeLineID *start_tli, + XLogRecPtr *start_lsn, + TimeLineID *previous_tli, + XLogRecPtr *previous_lsn); +extern void write_backup_label(char *output_directory, StringInfo buf, + pg_checksum_type checksum_type, + struct manifest_writer *mwriter); + +#endif /* BACKUP_LABEL_H */ diff --git a/src/bin/pg_combinebackup/copy_file.c b/src/bin/pg_combinebackup/copy_file.c new file mode 100644 index 0000000000..8ba6cc09e4 --- /dev/null +++ b/src/bin/pg_combinebackup/copy_file.c @@ -0,0 +1,169 @@ +/* + * Copy entire files. + * + * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group + * Portions Copyright (c) 1994, Regents of the University of California + * + * src/bin/pg_combinebackup/copy_file.h + * + *------------------------------------------------------------------------- + */ +#include "postgres_fe.h" + +#ifdef HAVE_COPYFILE_H +#include <copyfile.h> +#endif +#include <fcntl.h> +#include <sys/stat.h> +#include <unistd.h> + +#include "common/file_perm.h" +#include "common/logging.h" +#include "copy_file.h" + +static void copy_file_blocks(const char *src, const char *dst, + pg_checksum_context *checksum_ctx); + +#ifdef WIN32 +static void copy_file_copyfile(const char *src, const char *dst); +#endif + +/* + * Copy a regular file, optionally computing a checksum, and emitting + * appropriate debug messages. But if we're in dry-run mode, then just emit + * the messages and don't copy anything. + */ +void +copy_file(const char *src, const char *dst, + pg_checksum_context *checksum_ctx, bool dry_run) +{ + /* + * In dry-run mode, we don't actually copy anything, nor do we read any + * data from the source file, but we do verify that we can open it. + */ + if (dry_run) + { + int fd; + + if ((fd = open(src, O_RDONLY | PG_BINARY)) < 0) + pg_fatal("could not open \"%s\": %m", src); + if (close(fd) < 0) + pg_fatal("could not close \"%s\": %m", src); + } + + /* + * If we don't need to compute a checksum, then we can use any special + * operating system primitives that we know about to copy the file; this + * may be quicker than a naive block copy. + */ + if (checksum_ctx->type != CHECKSUM_TYPE_NONE) + { + char *strategy_name = NULL; + void (*strategy_implementation) (const char *, const char *) = NULL; + +#ifdef WIN32 + strategy_name = "CopyFile"; + strategy_implementation = copy_file_copyfile; +#endif + + if (strategy_name != NULL) + { + if (dry_run) + pg_log_debug("would copy \"%s\" to \"%s\" using strategy %s", + src, dst, strategy_name); + else + { + pg_log_debug("copying \"%s\" to \"%s\" using strategy %s", + src, dst, strategy_name); + (*strategy_implementation) (src, dst); + } + return; + } + } + + /* + * Fall back to the simple approach of reading and writing all the blocks, + * feeding them into the checksum context as we go. + */ + if (dry_run) + { + if (checksum_ctx->type == CHECKSUM_TYPE_NONE) + pg_log_debug("would copy \"%s\" to \"%s\"", + src, dst); + else + pg_log_debug("would copy \"%s\" to \"%s\" and checksum with %s", + src, dst, pg_checksum_type_name(checksum_ctx->type)); + } + else + { + if (checksum_ctx->type == CHECKSUM_TYPE_NONE) + pg_log_debug("copying \"%s\" to \"%s\"", + src, dst); + else + pg_log_debug("copying \"%s\" to \"%s\" and checksumming with %s", + src, dst, pg_checksum_type_name(checksum_ctx->type)); + copy_file_blocks(src, dst, checksum_ctx); + } +} + +/* + * Copy a file block by block, and optionally compute a checksum as we go. + */ +static void +copy_file_blocks(const char *src, const char *dst, + pg_checksum_context *checksum_ctx) +{ + int src_fd; + int dest_fd; + uint8 *buffer; + const int buffer_size = 50 * BLCKSZ; + ssize_t rb; + unsigned offset = 0; + + if ((src_fd = open(src, O_RDONLY | PG_BINARY, 0)) < 0) + pg_fatal("could not open file \"%s\": %m", src); + + if ((dest_fd = open(dst, O_WRONLY | O_CREAT | O_EXCL | PG_BINARY, + pg_file_create_mode)) < 0) + pg_fatal("could not open file \"%s\": %m", dst); + + buffer = pg_malloc(buffer_size); + + while ((rb = read(src_fd, buffer, buffer_size)) > 0) + { + ssize_t wb; + + if ((wb = write(dest_fd, buffer, rb)) != rb) + { + if (wb < 0) + pg_fatal("could not write file \"%s\": %m", dst); + else + pg_fatal("could not write file \"%s\": wrote only %d of %d bytes at offset %u", + dst, (int) wb, (int) rb, offset); + } + + if (pg_checksum_update(checksum_ctx, buffer, rb) < 0) + pg_fatal("could not update checksum of file \"%s\"", dst); + + offset += rb; + } + + if (rb < 0) + pg_fatal("could not read file \"%s\": %m", dst); + + pg_free(buffer); + close(src_fd); + close(dest_fd); +} + +#ifdef WIN32 +static void +copy_file_copyfile(const char *src, const char *dst) +{ + if (CopyFile(src, dst, true) == 0) + { + _dosmaperr(GetLastError()); + pg_fatal("could not copy \"%s\" to \"%s\": %m", src, dst); + } +} +#endif /* WIN32 */ diff --git a/src/bin/pg_combinebackup/copy_file.h b/src/bin/pg_combinebackup/copy_file.h new file mode 100644 index 0000000000..031030bacb --- /dev/null +++ b/src/bin/pg_combinebackup/copy_file.h @@ -0,0 +1,19 @@ +/* + * Copy entire files. + * + * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group + * Portions Copyright (c) 1994, Regents of the University of California + * + * src/bin/pg_combinebackup/copy_file.h + * + *------------------------------------------------------------------------- + */ +#ifndef COPY_FILE_H +#define COPY_FILE_H + +#include "common/checksum_helper.h" + +extern void copy_file(const char *src, const char *dst, + pg_checksum_context *checksum_ctx, bool dry_run); + +#endif /* COPY_FILE_H */ diff --git a/src/bin/pg_combinebackup/load_manifest.c b/src/bin/pg_combinebackup/load_manifest.c new file mode 100644 index 0000000000..d0b8de7912 --- /dev/null +++ b/src/bin/pg_combinebackup/load_manifest.c @@ -0,0 +1,245 @@ +/*------------------------------------------------------------------------- + * + * Load data from a backup manifest into memory. + * + * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group + * Portions Copyright (c) 1994, Regents of the University of California + * + * src/bin/pg_combinebackup/load_manifest.c + * + *------------------------------------------------------------------------- + */ + +#include "postgres_fe.h" + +#include <sys/stat.h> +#include <unistd.h> + +#include "common/hashfn.h" +#include "common/logging.h" +#include "common/parse_manifest.h" +#include "load_manifest.h" + +/* + * For efficiency, we'd like our hash table containing information about the + * manifest to start out with approximately the correct number of entries. + * There's no way to know the exact number of entries without reading the whole + * file, but we can get an estimate by dividing the file size by the estimated + * number of bytes per line. + * + * This could be off by about a factor of two in either direction, because the + * checksum algorithm has a big impact on the line lengths; e.g. a SHA512 + * checksum is 128 hex bytes, whereas a CRC-32C value is only 8, and there + * might be no checksum at all. + */ +#define ESTIMATED_BYTES_PER_MANIFEST_LINE 100 + +/* + * Define a hash table which we can use to store information about the files + * mentioned in the backup manifest. + */ +static uint32 hash_string_pointer(char *s); +#define SH_PREFIX manifest_files +#define SH_ELEMENT_TYPE manifest_file +#define SH_KEY_TYPE char * +#define SH_KEY pathname +#define SH_HASH_KEY(tb, key) hash_string_pointer(key) +#define SH_EQUAL(tb, a, b) (strcmp(a, b) == 0) +#define SH_SCOPE extern +#define SH_RAW_ALLOCATOR pg_malloc0 +#define SH_DEFINE +#include "lib/simplehash.h" + +static void record_manifest_details_for_file(JsonManifestParseContext *context, + char *pathname, size_t size, + pg_checksum_type checksum_type, + int checksum_length, + uint8 *checksum_payload); +static void record_manifest_details_for_wal_range(JsonManifestParseContext *context, + TimeLineID tli, + XLogRecPtr start_lsn, + XLogRecPtr end_lsn); +static void report_manifest_error(JsonManifestParseContext *context, + const char *fmt,...) + pg_attribute_printf(2, 3) pg_attribute_noreturn(); + +/* + * Load backup_manifest files from an array of backups and produces an array + * of manifest_data objects. + * + * NB: Since load_backup_manifest() can return NULL, the resulting array could + * contain NULL entries. + */ +manifest_data ** +load_backup_manifests(int n_backups, char **backup_directories) +{ + manifest_data **result; + int i; + + result = pg_malloc(sizeof(manifest_data *) * n_backups); + for (i = 0; i < n_backups; ++i) + result[i] = load_backup_manifest(backup_directories[i]); + + return result; +} + +/* + * Parse the backup_manifest file in the named backup directory. Construct a + * hash table with information about all the files it mentions, and a linked + * list of all the WAL ranges it mentions. + * + * If the backup_manifest file simply doesn't exist, logs a warning and returns + * NULL. Any other error, or any error parsing the contents of the file, is + * fatal. + */ +manifest_data * +load_backup_manifest(char *backup_directory) +{ + char pathname[MAXPGPATH]; + int fd; + struct stat statbuf; + off_t estimate; + uint32 initial_size; + manifest_files_hash *ht; + char *buffer; + int rc; + JsonManifestParseContext context; + manifest_data *result; + + /* Open the manifest file. */ + snprintf(pathname, MAXPGPATH, "%s/backup_manifest", backup_directory); + if ((fd = open(pathname, O_RDONLY | PG_BINARY, 0)) < 0) + { + if (errno == EEXIST) + { + pg_log_warning("\"%s\" does not exist", pathname); + return NULL; + } + pg_fatal("could not open file \"%s\": %m", pathname); + } + + /* Figure out how big the manifest is. */ + if (fstat(fd, &statbuf) != 0) + pg_fatal("could not stat file \"%s\": %m", pathname); + + /* Guess how large to make the hash table based on the manifest size. */ + estimate = statbuf.st_size / ESTIMATED_BYTES_PER_MANIFEST_LINE; + initial_size = Min(PG_UINT32_MAX, Max(estimate, 256)); + + /* Create the hash table. */ + ht = manifest_files_create(initial_size, NULL); + + /* + * Slurp in the whole file. + * + * This is not ideal, but there's currently no way to get pg_parse_json() + * to perform incremental parsing. + */ + buffer = pg_malloc(statbuf.st_size); + rc = read(fd, buffer, statbuf.st_size); + if (rc != statbuf.st_size) + { + if (rc < 0) + pg_fatal("could not read file \"%s\": %m", pathname); + else + pg_fatal("could not read file \"%s\": read %d of %lld", + pathname, rc, (long long int) statbuf.st_size); + } + + /* Close the manifest file. */ + close(fd); + + /* Parse the manifest. */ + result = pg_malloc0(sizeof(manifest_data)); + result->files = ht; + context.private_data = result; + context.perfile_cb = record_manifest_details_for_file; + context.perwalrange_cb = record_manifest_details_for_wal_range; + context.error_cb = report_manifest_error; + json_parse_manifest(&context, buffer, statbuf.st_size); + + /* All done. */ + pfree(buffer); + return result; +} + +/* + * Report an error while parsing the manifest. + * + * We consider all such errors to be fatal errors. The manifest parser + * expects this function not to return. + */ +static void +report_manifest_error(JsonManifestParseContext *context, const char *fmt,...) +{ + va_list ap; + + va_start(ap, fmt); + pg_log_generic_v(PG_LOG_ERROR, PG_LOG_PRIMARY, gettext(fmt), ap); + va_end(ap); + + exit(1); +} + +/* + * Record details extracted from the backup manifest for one file. + */ +static void +record_manifest_details_for_file(JsonManifestParseContext *context, + char *pathname, size_t size, + pg_checksum_type checksum_type, + int checksum_length, uint8 *checksum_payload) +{ + manifest_data *manifest = context->private_data; + manifest_file *m; + bool found; + + /* Make a new entry in the hash table for this file. */ + m = manifest_files_insert(manifest->files, pathname, &found); + if (found) + pg_fatal("duplicate path name in backup manifest: \"%s\"", pathname); + + /* Initialize the entry. */ + m->size = size; + m->checksum_type = checksum_type; + m->checksum_length = checksum_length; + m->checksum_payload = checksum_payload; +} + +/* + * Record details extracted from the backup manifest for one WAL range. + */ +static void +record_manifest_details_for_wal_range(JsonManifestParseContext *context, + TimeLineID tli, + XLogRecPtr start_lsn, XLogRecPtr end_lsn) +{ + manifest_data *manifest = context->private_data; + manifest_wal_range *range; + + /* Allocate and initialize a struct describing this WAL range. */ + range = palloc(sizeof(manifest_wal_range)); + range->tli = tli; + range->start_lsn = start_lsn; + range->end_lsn = end_lsn; + range->prev = manifest->last_wal_range; + range->next = NULL; + + /* Add it to the end of the list. */ + if (manifest->first_wal_range == NULL) + manifest->first_wal_range = range; + else + manifest->last_wal_range->next = range; + manifest->last_wal_range = range; +} + +/* + * Helper function for manifest_files hash table. + */ +static uint32 +hash_string_pointer(char *s) +{ + unsigned char *ss = (unsigned char *) s; + + return hash_bytes(ss, strlen(s)); +} diff --git a/src/bin/pg_combinebackup/load_manifest.h b/src/bin/pg_combinebackup/load_manifest.h new file mode 100644 index 0000000000..2bfeeff156 --- /dev/null +++ b/src/bin/pg_combinebackup/load_manifest.h @@ -0,0 +1,67 @@ +/*------------------------------------------------------------------------- + * + * Load data from a backup manifest into memory. + * + * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group + * Portions Copyright (c) 1994, Regents of the University of California + * + * src/bin/pg_combinebackup/load_manifest.h + * + *------------------------------------------------------------------------- + */ +#ifndef LOAD_MANIFEST_H +#define LOAD_MANIFEST_H + +#include "access/xlogdefs.h" +#include "common/checksum_helper.h" + +/* + * Each file described by the manifest file is parsed to produce an object + * like this. + */ +typedef struct manifest_file +{ + uint32 status; /* hash status */ + char *pathname; + size_t size; + pg_checksum_type checksum_type; + int checksum_length; + uint8 *checksum_payload; +} manifest_file; + +#define SH_PREFIX manifest_files +#define SH_ELEMENT_TYPE manifest_file +#define SH_KEY_TYPE char * +#define SH_SCOPE extern +#define SH_RAW_ALLOCATOR pg_malloc0 +#define SH_DECLARE +#include "lib/simplehash.h" + +/* + * Each WAL range described by the manifest file is parsed to produce an + * object like this. + */ +typedef struct manifest_wal_range +{ + TimeLineID tli; + XLogRecPtr start_lsn; + XLogRecPtr end_lsn; + struct manifest_wal_range *next; + struct manifest_wal_range *prev; +} manifest_wal_range; + +/* + * All the data parsed from a backup_manifest file. + */ +typedef struct manifest_data +{ + manifest_files_hash *files; + manifest_wal_range *first_wal_range; + manifest_wal_range *last_wal_range; +} manifest_data; + +extern manifest_data *load_backup_manifest(char *backup_directory); +extern manifest_data **load_backup_manifests(int n_backups, + char **backup_directories); + +#endif /* LOAD_MANIFEST_H */ diff --git a/src/bin/pg_combinebackup/meson.build b/src/bin/pg_combinebackup/meson.build new file mode 100644 index 0000000000..bea0db405e --- /dev/null +++ b/src/bin/pg_combinebackup/meson.build @@ -0,0 +1,29 @@ +# Copyright (c) 2022-2023, PostgreSQL Global Development Group + +pg_combinebackup_sources = files( + 'pg_combinebackup.c', + 'backup_label.c', + 'copy_file.c', + 'load_manifest.c', + 'reconstruct.c', + 'write_manifest.c', +) + +if host_system == 'windows' + pg_combinebackup_sources += rc_bin_gen.process(win32ver_rc, extra_args: [ + '--NAME', 'pg_combinebackup', + '--FILEDESC', 'pg_combinebackup - combine incremental backups',]) +endif + +pg_combinebackup = executable('pg_combinebackup', + pg_combinebackup_sources, + dependencies: [frontend_code], + kwargs: default_bin_args, +) +bin_targets += pg_combinebackup + +tests += { + 'name': 'pg_combinebackup', + 'sd': meson.current_source_dir(), + 'bd': meson.current_build_dir() +} diff --git a/src/bin/pg_combinebackup/pg_combinebackup.c b/src/bin/pg_combinebackup/pg_combinebackup.c new file mode 100644 index 0000000000..369447204a --- /dev/null +++ b/src/bin/pg_combinebackup/pg_combinebackup.c @@ -0,0 +1,1276 @@ +/*------------------------------------------------------------------------- + * + * pg_combinebackup.c + * Combine incremental backups with prior backups. + * + * Copyright (c) 2017-2023, PostgreSQL Global Development Group + * + * IDENTIFICATION + * src/bin/pg_combinebackup/pg_combinebackup.c + * + *------------------------------------------------------------------------- + */ +#include "postgres_fe.h" + +#include <dirent.h> +#include <fcntl.h> +#include <limits.h> + +#include "access/xlogdefs.h" +#include "backup_label.h" +#include "common/blkreftable.h" +#include "common/checksum_helper.h" +#include "common/controldata_utils.h" +#include "common/file_perm.h" +#include "common/file_utils.h" +#include "common/logging.h" +#include "copy_file.h" +#include "fe_utils/option_utils.h" +#include "lib/stringinfo.h" +#include "load_manifest.h" +#include "getopt_long.h" +#include "reconstruct.h" +#include "write_manifest.h" + +/* Incremental file naming convention. */ +#define INCREMENTAL_PREFIX "INCREMENTAL." +#define INCREMENTAL_PREFIX_LENGTH 12 + +/* + * Tracking for directories that need to be removed, or have their contents + * removed, if the operation fails. + */ +typedef struct cb_cleanup_dir +{ + char *target_path; + bool rmtopdir; + struct cb_cleanup_dir *next; +} cb_cleanup_dir; + +/* + * Stores a tablespace mapping provided using -T, --tablespace-mapping. + */ +typedef struct cb_tablespace_mapping +{ + char old_dir[MAXPGPATH]; + char new_dir[MAXPGPATH]; + struct cb_tablespace_mapping *next; +} cb_tablespace_mapping; + +/* + * Stores data parsed from all command-line options. + */ +typedef struct cb_options +{ + bool debug; + char *output; + bool dry_run; + bool no_sync; + bool progress; + cb_tablespace_mapping *tsmappings; + pg_checksum_type manifest_checksums; + bool no_manifest; + DataDirSyncMethod sync_method; +} cb_options; + +/* + * Data about a tablespace. + * + * Every normal tablespace needs a tablespace mapping, but in-place tablespaces + * don't, so the list of tablespaces can contain more entries than the list of + * tablespace mappings. + */ +typedef struct cb_tablespace +{ + Oid oid; + bool in_place; + char old_dir[MAXPGPATH]; + char new_dir[MAXPGPATH]; + struct cb_tablespace *next; +} cb_tablespace; + +/* Directories to be removed if we exit uncleanly. */ +cb_cleanup_dir *cleanup_dir_list = NULL; + +static void add_tablespace_mapping(cb_options *opt, char *arg); +static StringInfo check_backup_label_files(int n_backups, char **backup_dirs); +static void check_control_files(int n_backups, char **backup_dirs); +static void check_input_dir_permissions(char *dir); +static void cleanup_directories_atexit(void); +static void create_output_directory(char *dirname, cb_options *opt); +static void help(const char *progname); +static bool parse_oid(char *s, Oid *result); +static void process_directory_recursively(Oid tsoid, + char *input_directory, + char *output_directory, + char *relative_path, + int n_prior_backups, + char **prior_backup_dirs, + manifest_data **manifests, + manifest_writer *mwriter, + cb_options *opt); +static int read_pg_version_file(char *directory); +static void remember_to_cleanup_directory(char *target_path, bool rmtopdir); +static void reset_directory_cleanup_list(void); +static cb_tablespace *scan_for_existing_tablespaces(char *pathname, + cb_options *opt); +static void slurp_file(int fd, char *filename, StringInfo buf, int maxlen); + +/* + * Main program. + */ +int +main(int argc, char *argv[]) +{ + static struct option long_options[] = { + {"debug", no_argument, NULL, 'd'}, + {"output", required_argument, NULL, 'o'}, + {"dry-run", no_argument, NULL, 'n'}, + {"no-sync", no_argument, NULL, 'N'}, + {"progress", no_argument, NULL, 'P'}, + {"tablespace-mapping", no_argument, NULL, 'T'}, + {"manifest-checksums", required_argument, NULL, 1}, + {"no-manifest", no_argument, NULL, 2}, + {"sync-method", required_argument, NULL, 3}, + {NULL, 0, NULL, 0} + }; + + const char *progname; + char *last_input_dir; + int optindex; + int c; + int n_backups; + int n_prior_backups; + int version; + char **prior_backup_dirs; + cb_options opt; + cb_tablespace *tablespaces; + cb_tablespace *ts; + StringInfo last_backup_label; + manifest_data **manifests; + manifest_writer *mwriter; + + pg_logging_init(argv[0]); + progname = get_progname(argv[0]); + handle_help_version_opts(argc, argv, progname, help); + + memset(&opt, 0, sizeof(opt)); + opt.manifest_checksums = CHECKSUM_TYPE_CRC32C; + opt.sync_method = DATA_DIR_SYNC_METHOD_FSYNC; + + /* process command-line options */ + while ((c = getopt_long(argc, argv, "do:nNPT:", + long_options, &optindex)) != -1) + { + switch (c) + { + case 'd': + opt.debug = true; + pg_logging_increase_verbosity(); + break; + case 'o': + opt.output = optarg; + break; + case 'n': + opt.dry_run = true; + break; + case 'N': + opt.no_sync = true; + break; + case 'P': + opt.progress = true; + break; + case 'T': + add_tablespace_mapping(&opt, optarg); + break; + case 1: + if (!pg_checksum_parse_type(optarg, + &opt.manifest_checksums)) + pg_fatal("unrecognized checksum algorithm: \"%s\"", + optarg); + break; + case 2: + opt.no_manifest = true; + break; + case 3: + if (!parse_sync_method(optarg, &opt.sync_method)) + exit(1); + break; + default: + /* getopt_long already emitted a complaint */ + pg_log_error_hint("Try \"%s --help\" for more information.", progname); + exit(1); + } + } + + if (optind >= argc) + { + pg_log_error("%s: no input directories specified", progname); + pg_log_error_hint("Try \"%s --help\" for more information.", progname); + exit(1); + } + + if (opt.output == NULL) + pg_fatal("no output directory specified"); + + /* If no manifest is needed, no checksums are needed, either. */ + if (opt.no_manifest) + opt.manifest_checksums = CHECKSUM_TYPE_NONE; + + /* Read the server version from the final backup. */ + version = read_pg_version_file(argv[argc - 1]); + + /* Sanity-check control files. */ + n_backups = argc - optind; + check_control_files(n_backups, argv + optind); + + /* Sanity-check backup_label files, and get the contents of the last one. */ + last_backup_label = check_backup_label_files(n_backups, argv + optind); + + /* Load backup manifests. */ + manifests = load_backup_manifests(n_backups, argv + optind); + + /* Figure out which tablespaces are going to be included in the output. */ + last_input_dir = argv[argc - 1]; + check_input_dir_permissions(last_input_dir); + tablespaces = scan_for_existing_tablespaces(last_input_dir, &opt); + + /* + * Create output directories. + * + * We create one output directory for the main data directory plus one for + * each non-in-place tablespace. create_output_directory() will arrange + * for those directories to be cleaned up on failure. In-place tablespaces + * aren't handled at this stage because they're located beneath the main + * output directory, and thus the cleanup of that directory will get rid + * of them. Plus, the pg_tblspc directory that needs to contain them + * doesn't exist yet. + */ + atexit(cleanup_directories_atexit); + create_output_directory(opt.output, &opt); + for (ts = tablespaces; ts != NULL; ts = ts->next) + if (!ts->in_place) + create_output_directory(ts->new_dir, &opt); + + /* If we need to write a backup_manifest, prepare to do so. */ + if (!opt.dry_run && !opt.no_manifest) + mwriter = create_manifest_writer(opt.output); + else + mwriter = NULL; + + /* Write backup label into output directory. */ + if (opt.dry_run) + pg_log_debug("would generate \"%s/backup_label\"", opt.output); + else + { + pg_log_debug("generating \"%s/backup_label\"", opt.output); + last_backup_label->cursor = 0; + write_backup_label(opt.output, last_backup_label, + opt.manifest_checksums, mwriter); + } + + /* + * We'll need the pathnames to the prior backups. By "prior" we mean all + * but the last one listed on the command line. + */ + n_prior_backups = argc - optind - 1; + prior_backup_dirs = argv + optind; + + /* Process everything that's not part of a user-defined tablespace. */ + pg_log_debug("processing backup directory \"%s\"", last_input_dir); + process_directory_recursively(InvalidOid, last_input_dir, opt.output, + NULL, n_prior_backups, prior_backup_dirs, + manifests, mwriter, &opt); + + /* Process user-defined tablespaces. */ + for (ts = tablespaces; ts != NULL; ts = ts->next) + { + pg_log_debug("processing tablespace directory \"%s\"", ts->old_dir); + + /* + * If it's a normal tablespace, we need to set up a symbolic link from + * pg_tblspc/${OID} to the target directory; if it's an in-place + * tablespace, we need to create a directory at pg_tblspc/${OID}. + */ + if (!ts->in_place) + { + char linkpath[MAXPGPATH]; + + snprintf(linkpath, MAXPGPATH, "%s/pg_tblspc/%u", opt.output, + ts->oid); + + if (opt.dry_run) + pg_log_debug("would create symbolic link from \"%s\" to \"%s\"", + linkpath, ts->new_dir); + else + { + pg_log_debug("creating symbolic link from \"%s\" to \"%s\"", + linkpath, ts->new_dir); + if (symlink(ts->new_dir, linkpath) != 0) + pg_fatal("could not create symbolic link from \"%s\" to \"%s\": %m", + linkpath, ts->new_dir); + } + } + else + { + if (opt.dry_run) + pg_log_debug("would create directory \"%s\"", ts->new_dir); + else + { + pg_log_debug("creating directory \"%s\"", ts->new_dir); + if (pg_mkdir_p(ts->new_dir, pg_dir_create_mode) == -1) + pg_fatal("could not create directory \"%s\": %m", + ts->new_dir); + } + } + + /* OK, now handle the directory contents. */ + process_directory_recursively(ts->oid, ts->old_dir, ts->new_dir, + NULL, n_prior_backups, prior_backup_dirs, + manifests, mwriter, &opt); + } + + /* Finalize the backup_manifest, if we're generating one. */ + if (mwriter != NULL) + finalize_manifest(mwriter, + manifests[n_prior_backups]->first_wal_range); + + /* fsync that output directory unless we've been told not to do so */ + if (!opt.no_sync) + { + if (opt.dry_run) + pg_log_debug("would recursively fsync \"%s\"", opt.output); + else + { + pg_log_debug("recursively fsyncing \"%s\"", opt.output); + sync_pgdata(opt.output, version * 10000, opt.sync_method); + } + } + + /* It's a success, so don't remove the output directories. */ + reset_directory_cleanup_list(); + exit(0); +} + +/* + * Process the option argument for the -T, --tablespace-mapping switch. + */ +static void +add_tablespace_mapping(cb_options *opt, char *arg) +{ + cb_tablespace_mapping *tsmap = pg_malloc0(sizeof(cb_tablespace_mapping)); + char *dst; + char *dst_ptr; + char *arg_ptr; + + /* + * Basically, we just want to copy everything before the equals sign to + * tsmap->old_dir and everything afterwards to tsmap->new_dir, but if + * there's more or less than one equals sign, that's an error, and if + * there's an equals sign preceded by a backslash, don't treat it as a + * field separator but instead copy a literal equals sign. + */ + dst_ptr = dst = tsmap->old_dir; + for (arg_ptr = arg; *arg_ptr != '\0'; arg_ptr++) + { + if (dst_ptr - dst >= MAXPGPATH) + pg_fatal("directory name too long"); + + if (*arg_ptr == '\\' && *(arg_ptr + 1) == '=') + ; /* skip backslash escaping = */ + else if (*arg_ptr == '=' && (arg_ptr == arg || *(arg_ptr - 1) != '\\')) + { + if (tsmap->new_dir[0] != '\0') + pg_fatal("multiple \"=\" signs in tablespace mapping"); + else + dst = dst_ptr = tsmap->new_dir; + } + else + *dst_ptr++ = *arg_ptr; + } + if (!tsmap->old_dir[0] || !tsmap->new_dir[0]) + pg_fatal("invalid tablespace mapping format \"%s\", must be \"OLDDIR=NEWDIR\"", arg); + + /* + * All tablespaces are created with absolute directories, so specifying a + * non-absolute path here would never match, possibly confusing users. + * + * In contrast to pg_basebackup, both the old and new directories are on + * the local machine, so the local machine's definition of an absolute + * path is the only relevant one. + */ + if (!is_absolute_path(tsmap->old_dir)) + pg_fatal("old directory is not an absolute path in tablespace mapping: %s", + tsmap->old_dir); + + if (!is_absolute_path(tsmap->new_dir)) + pg_fatal("old directory is not an absolute path in tablespace mapping: %s", + tsmap->new_dir); + + /* Canonicalize paths to avoid spurious failures when comparing. */ + canonicalize_path(tsmap->old_dir); + canonicalize_path(tsmap->new_dir); + + /* Add it to the list. */ + tsmap->next = opt->tsmappings; + opt->tsmappings = tsmap; +} + +/* + * Check that the backup_label files form a coherent backup chain, and return + * the contents of the backup_label file from the latest backup. + */ +static StringInfo +check_backup_label_files(int n_backups, char **backup_dirs) +{ + StringInfo buf = makeStringInfo(); + StringInfo lastbuf = buf; + int i; + TimeLineID check_tli = 0; + XLogRecPtr check_lsn = InvalidXLogRecPtr; + + /* Try to read each backup_label file in turn, last to first. */ + for (i = n_backups - 1; i >= 0; --i) + { + char pathbuf[MAXPGPATH]; + int fd; + TimeLineID start_tli; + TimeLineID previous_tli; + XLogRecPtr start_lsn; + XLogRecPtr previous_lsn; + + /* Open the backup_label file. */ + snprintf(pathbuf, MAXPGPATH, "%s/backup_label", backup_dirs[i]); + pg_log_debug("reading \"%s\"", pathbuf); + if ((fd = open(pathbuf, O_RDONLY, 0)) < 0) + pg_fatal("could not open file \"%s\": %m", pathbuf); + + /* + * Slurp the whole file into memory. + * + * The exact size limit that we impose here doesn't really matter -- + * most of what's supposed to be in the file is fixed size and quite + * short. However, the length of the backup_label is limited (at least + * by some parts of the code) to MAXGPATH, so include that value in + * the maximum length that we tolerate. + */ + slurp_file(fd, pathbuf, buf, 10000 + MAXPGPATH); + + /* Close the file. */ + if (close(fd) != 0) + pg_fatal("could not close \"%s\": %m", pathbuf); + + /* Parse the file contents. */ + parse_backup_label(pathbuf, buf, &start_tli, &start_lsn, + &previous_tli, &previous_lsn); + + /* + * Sanity checks. + * + * XXX. It's actually not required that start_lsn == check_lsn. It + * would be OK if start_lsn > check_lsn provided that start_lsn is + * less than or equal to the relevant switchpoint. But at the moment + * we don't have that information. + */ + if (i > 0 && previous_tli == 0) + pg_fatal("backup at \"%s\" is a full backup, but only the first backup should be a full backup", + backup_dirs[i]); + if (i == 0 && previous_tli != 0) + pg_fatal("backup at \"%s\" is an incremental backup, but the first backup should be a full backup", + backup_dirs[i]); + if (i < n_backups - 1 && start_tli != check_tli) + pg_fatal("backup at \"%s\" starts on timeline %u, but expected %u", + backup_dirs[i], start_tli, check_tli); + if (i < n_backups - 1 && start_lsn != check_lsn) + pg_fatal("backup at \"%s\" starts at LSN %X/%X, but expected %X/%X", + backup_dirs[i], + LSN_FORMAT_ARGS(start_lsn), + LSN_FORMAT_ARGS(check_lsn)); + check_tli = previous_tli; + check_lsn = previous_lsn; + + /* + * The last backup label in the chain needs to be saved for later use, + * while the others are only needed within this loop. + */ + if (lastbuf == buf) + buf = makeStringInfo(); + else + resetStringInfo(buf); + } + + /* Free memory that we don't need any more. */ + if (lastbuf != buf) + { + pfree(buf->data); + pfree(buf); + } + + /* + * Return the data from the first backup_info that we read (which is the + * backup_label from the last directory specified on the command line). + */ + return lastbuf; +} + +/* + * Sanity check control files. + */ +static void +check_control_files(int n_backups, char **backup_dirs) +{ + int i; + uint64 system_identifier; + + /* Try to read each control file in turn, last to first. */ + for (i = n_backups - 1; i >= 0; --i) + { + ControlFileData *control_file; + bool crc_ok; + + pg_log_debug("reading \"%s/global/pg_control\"", backup_dirs[i]); + control_file = get_controlfile(backup_dirs[i], &crc_ok); + + /* Control file contents not meaningful if CRC is bad. */ + if (!crc_ok) + pg_fatal("%s/global/pg_control: crc is incorrect", backup_dirs[i]); + + /* Can't interpret control file if not current version. */ + if (control_file->pg_control_version != PG_CONTROL_VERSION) + pg_fatal("%s/global/pg_control: unexpected control file version", + backup_dirs[i]); + + /* System identifiers should all match. */ + if (i == n_backups - 1) + system_identifier = control_file->system_identifier; + else if (system_identifier != control_file->system_identifier) + pg_fatal("%s/global/pg_control: expected system identifier %llu, but found %llu", + backup_dirs[i], (unsigned long long) system_identifier, + (unsigned long long) control_file->system_identifier); + + /* Release memory. */ + pfree(control_file); + } + + /* + * If debug output is enabled, make a note of the system identifier that + * we found in all of the relevant control files. + */ + pg_log_debug("system identifier is %llu", + (unsigned long long) system_identifier); +} + +/* + * Set default permissions for new files and directories based on the + * permissions of the given directory. The intent here is that the output + * directory should use the same permissions scheme as the final input + * directory. + */ +static void +check_input_dir_permissions(char *dir) +{ + struct stat st; + + if (stat(dir, &st) != 0) + pg_fatal("could not stat \"%s\": %m", dir); + + SetDataDirectoryCreatePerm(st.st_mode); +} + +/* + * Clean up output directories before exiting. + */ +static void +cleanup_directories_atexit(void) +{ + while (cleanup_dir_list != NULL) + { + cb_cleanup_dir *dir = cleanup_dir_list; + + if (dir->rmtopdir) + { + pg_log_info("removing output directory \"%s\"", dir->target_path); + if (!rmtree(dir->target_path, dir->rmtopdir)) + pg_log_error("failed to remove output directory"); + } + else + { + pg_log_info("removing contents of output directory \"%s\"", + dir->target_path); + if (!rmtree(dir->target_path, dir->rmtopdir)) + pg_log_error("failed to remove contents of output directory"); + } + + cleanup_dir_list = cleanup_dir_list->next; + pfree(dir); + } +} + +/* + * Create the named output directory, unless it already exists or we're in + * dry-run mode. If it already exists but is not empty, that's a fatal error. + * + * Adds the created directory to the list of directories to be cleaned up + * at process exit. + */ +static void +create_output_directory(char *dirname, cb_options *opt) +{ + switch (pg_check_dir(dirname)) + { + case 0: + if (opt->dry_run) + { + pg_log_debug("would create directory \"%s\"", dirname); + return; + } + pg_log_debug("creating directory \"%s\"", dirname); + if (pg_mkdir_p(dirname, pg_dir_create_mode) == -1) + pg_fatal("could not create directory \"%s\": %m", dirname); + remember_to_cleanup_directory(dirname, true); + break; + + case 1: + pg_log_debug("using existing directory \"%s\"", dirname); + remember_to_cleanup_directory(dirname, false); + break; + + case 2: + case 3: + case 4: + pg_fatal("directory \"%s\" exists but is not empty", dirname); + + case -1: + pg_fatal("could not access directory \"%s\": %m", dirname); + } +} + +/* + * help + * + * Prints help page for the program + * + * progname: the name of the executed program, such as "pg_combinebackup" + */ +static void +help(const char *progname) +{ + printf(_("%s combines incremental backups.\n\n"), progname); + printf(_("Usage:\n")); + printf(_(" %s [OPTION]... DIRECTORY...\n"), progname); + printf(_("\nOptions:\n")); + printf(_(" -d, --debug generate lots of debugging output\n")); + printf(_(" -o, --output output directory\n")); + printf(_(" -n, --dry-run don't actually do anything\n")); + printf(_(" -N, --no-sync do not wait for changes to be written safely to disk\n")); + printf(_(" -P, --progress show progress information\n")); + printf(_(" -T, --tablespace-mapping=OLDDIR=NEWDIR\n")); + printf(_(" relocate tablespace in OLDDIR to NEWDIR\n")); + printf(_(" --manifest-checksums=SHA{224,256,384,512}|CRC32C|NONE\n" + " use algorithm for manifest checksums\n")); + printf(_(" --no-manifest suppress generation of backup manifest\n")); + printf(_(" --sync-method=METHOD set method for syncing files to disk\n")); + printf(_(" -?, --help show this help, then exit\n")); + + printf(_("\nReport bugs to <%s>.\n"), PACKAGE_BUGREPORT); + printf(_("%s home page: <%s>\n"), PACKAGE_NAME, PACKAGE_URL); +} + +/* + * Try to parse a string as a non-zero OID without leading zeroes. + * + * If it works, return true and set *result to the answer, else return false. + */ +static bool +parse_oid(char *s, Oid *result) +{ + Oid oid; + char *ep; + + errno = 0; + oid = strtoul(s, &ep, 10); + if (errno != 0 || *ep != '\0' || oid < 1 || oid > PG_UINT32_MAX) + return false; + + *result = oid; + return true; +} + +/* + * Copy files from the input directory to the output directory, reconstructing + * full files from incremental files as required. + * + * If processing is a user-defined tablespace, the tsoid should be the OID + * of that tablespace and input_directory and output_directory should be the + * toplevel input and output directories for that tablespace. Otherwise, + * tsoid should be InvalidOid and input_directory and output_directory should + * be the main input and output directories. + * + * relative_path is the path beneath the given input and output directories + * that we are currently processing. If NULL, it indicates that we're + * processing the input and output directories themselves. + * + * n_prior_backups is the number of prior backups that we have available. + * This doesn't count the very last backup, which is referenced by + * output_directory, just the older ones. prior_backup_dirs is an array of + * the locations of those previous backups. + */ +static void +process_directory_recursively(Oid tsoid, + char *input_directory, + char *output_directory, + char *relative_path, + int n_prior_backups, + char **prior_backup_dirs, + manifest_data **manifests, + manifest_writer *mwriter, + cb_options *opt) +{ + char ifulldir[MAXPGPATH]; + char ofulldir[MAXPGPATH]; + char manifest_prefix[MAXPGPATH]; + DIR *dir; + struct dirent *de; + bool is_pg_tblspc; + bool is_pg_wal; + manifest_data *latest_manifest = manifests[n_prior_backups]; + pg_checksum_type checksum_type; + + StaticAssertStmt(strlen(INCREMENTAL_PREFIX) == INCREMENTAL_PREFIX_LENGTH, + "INCREMENTAL_PREFIX_LENGTH is incorrect"); + + /* + * pg_tblspc and pg_wal are special cases, so detect those here. + * + * pg_tblspc is only special at the top level, but subdirectories of + * pg_wal are just as special as the top level directory. + * + * Since incremental backup does not exist in pre-v10 versions, we don't + * have to worry about the old pg_xlog naming. + */ + is_pg_tblspc = !OidIsValid(tsoid) && relative_path != NULL && + strcmp(relative_path, "pg_tblspc") == 0; + is_pg_wal = !OidIsValid(tsoid) && relative_path != NULL && + (strcmp(relative_path, "pg_wal") == 0 || + strncmp(relative_path, "pg_wal/", 7) == 0); + + /* + * If we're under pg_wal, then we don't need checksums, because these + * files aren't included in the backup manifest. Otherwise use whatever + * type of checksum is configured. + */ + if (!is_pg_wal) + checksum_type = opt->manifest_checksums; + else + checksum_type = CHECKSUM_TYPE_NONE; + + /* + * Append the relative path to the input and output directories, and + * figure out the appropriate prefix to add to files in this directory + * when looking them up in a backup manifest. + */ + if (relative_path == NULL) + { + strncpy(ifulldir, input_directory, MAXPGPATH); + strncpy(ofulldir, output_directory, MAXPGPATH); + if (OidIsValid(tsoid)) + snprintf(manifest_prefix, MAXPGPATH, "pg_tblspc/%u/", tsoid); + else + manifest_prefix[0] = '\0'; + } + else + { + snprintf(ifulldir, MAXPGPATH, "%s/%s", input_directory, + relative_path); + snprintf(ofulldir, MAXPGPATH, "%s/%s", output_directory, + relative_path); + if (OidIsValid(tsoid)) + snprintf(manifest_prefix, MAXPGPATH, "pg_tblspc/%u/%s/", + tsoid, relative_path); + else + snprintf(manifest_prefix, MAXPGPATH, "%s/", relative_path); + } + + /* + * Toplevel output directories have already been created by the time this + * function is called, but any subdirectories are our responsibility. + */ + if (relative_path != NULL) + { + if (opt->dry_run) + pg_log_debug("would create directory \"%s\"", ofulldir); + else + { + pg_log_debug("creating directory \"%s\"", ofulldir); + if (mkdir(ofulldir, pg_dir_create_mode) == -1) + pg_fatal("could not create directory \"%s\": %m", ofulldir); + } + } + + /* It's time to scan the directory. */ + if ((dir = opendir(ifulldir)) == NULL) + pg_fatal("could not open directory \"%s\": %m", ifulldir); + while (errno = 0, (de = readdir(dir)) != NULL) + { + PGFileType type; + char ifullpath[MAXPGPATH]; + char ofullpath[MAXPGPATH]; + char manifest_path[MAXPGPATH]; + Oid oid = InvalidOid; + int checksum_length = 0; + uint8 *checksum_payload = NULL; + pg_checksum_context checksum_ctx; + + /* Ignore "." and ".." entries. */ + if (strcmp(de->d_name, ".") == 0 || + strcmp(de->d_name, "..") == 0) + continue; + + /* Construct input path. */ + snprintf(ifullpath, MAXPGPATH, "%s/%s", ifulldir, de->d_name); + + /* Figure out what kind of directory entry this is. */ + type = get_dirent_type(ifullpath, de, false, PG_LOG_ERROR); + if (type == PGFILETYPE_ERROR) + exit(1); + + /* + * If we're processing pg_tblspc, then check whether the filename + * looks like it could be a tablespace OID. If so, and if the + * directory entry is a symbolic link or a directory, skip it. + * + * Our goal here is to ignore anything that would have been considered + * by scan_for_existing_tablespaces to be a tablespace. + */ + if (is_pg_tblspc && parse_oid(de->d_name, &oid) && + (type == PGFILETYPE_LNK || type == PGFILETYPE_DIR)) + continue; + + /* If it's a directory, recurse. */ + if (type == PGFILETYPE_DIR) + { + char new_relative_path[MAXPGPATH]; + + /* Append new pathname component to relative path. */ + if (relative_path == NULL) + strncpy(new_relative_path, de->d_name, MAXPGPATH); + else + snprintf(new_relative_path, MAXPGPATH, "%s/%s", relative_path, + de->d_name); + + /* And recurse. */ + process_directory_recursively(tsoid, + input_directory, output_directory, + new_relative_path, + n_prior_backups, prior_backup_dirs, + manifests, mwriter, opt); + continue; + } + + /* Skip anything that's not a regular file. */ + if (type != PGFILETYPE_REG) + { + if (type == PGFILETYPE_LNK) + pg_log_warning("skipping symbolic link \"%s\"", ifullpath); + else + pg_log_warning("skipping special file \"%s\"", ifullpath); + continue; + } + + /* + * Skip the backup_label and backup_manifest files; they require + * special handling and are handled elsewhere. + */ + if (relative_path == NULL && + (strcmp(de->d_name, "backup_label") == 0 || + strcmp(de->d_name, "backup_manifest") == 0)) + continue; + + /* + * If it's an incremental file, hand it off to the reconstruction + * code, which will figure out what to do. + */ + if (strncmp(de->d_name, INCREMENTAL_PREFIX, + INCREMENTAL_PREFIX_LENGTH) == 0) + { + /* Output path should not include "INCREMENTAL." prefix. */ + snprintf(ofullpath, MAXPGPATH, "%s/%s", ofulldir, + de->d_name + INCREMENTAL_PREFIX_LENGTH); + + + /* Manifest path likewise omits incremental prefix. */ + snprintf(manifest_path, MAXPGPATH, "%s%s", manifest_prefix, + de->d_name + INCREMENTAL_PREFIX_LENGTH); + + /* Reconstruction logic will do the rest. */ + reconstruct_from_incremental_file(ifullpath, ofullpath, + relative_path, + de->d_name + INCREMENTAL_PREFIX_LENGTH, + n_prior_backups, + prior_backup_dirs, + manifests, + manifest_path, + checksum_type, + &checksum_length, + &checksum_payload, + opt->dry_run); + } + else + { + /* Construct the path that the backup_manifest will use. */ + snprintf(manifest_path, MAXPGPATH, "%s%s", manifest_prefix, + de->d_name); + + /* + * It's not an incremental file, so we need to copy the entire + * file to the output directory. + * + * If a checksum of the required type already exists in the + * backup_manifest for the final input directory, we can save some + * work by reusing that checksum instead of computing a new one. + */ + if (checksum_type != CHECKSUM_TYPE_NONE && + latest_manifest != NULL) + { + manifest_file *mfile; + + mfile = manifest_files_lookup(latest_manifest->files, + manifest_path); + if (mfile == NULL) + { + /* + * The directory is out of sync with the backup_manifest, + * so emit a warning. + */ + pg_log_warning("\"%s/backup_manifest\" contains no entry for \"%s\"", + input_directory, manifest_path); + } + else if (mfile->checksum_type == checksum_type) + { + checksum_length = mfile->checksum_length; + checksum_payload = mfile->checksum_payload; + } + } + + /* + * If we're reusing a checksum, then we don't need copy_file() to + * compute one for us, but otherwise, it needs to compute whatever + * type of checksum we need. + */ + if (checksum_length != 0) + pg_checksum_init(&checksum_ctx, CHECKSUM_TYPE_NONE); + else + pg_checksum_init(&checksum_ctx, checksum_type); + + /* Actually copy the file. */ + snprintf(ofullpath, MAXPGPATH, "%s/%s", ofulldir, de->d_name); + copy_file(ifullpath, ofullpath, &checksum_ctx, opt->dry_run); + + /* + * If copy_file() performed a checksum calculation for us, then + * save the results (except in dry-run mode, when there's no + * point). + */ + if (checksum_ctx.type != CHECKSUM_TYPE_NONE && !opt->dry_run) + { + checksum_payload = pg_malloc(PG_CHECKSUM_MAX_LENGTH); + checksum_length = pg_checksum_final(&checksum_ctx, + checksum_payload); + } + } + + /* Generate manifest entry, if needed. */ + if (mwriter != NULL) + { + struct stat sb; + + /* + * In order to generate a manifest entry, we need the file size + * and mtime. We have no way to know the correct mtime except to + * stat() the file, so just do that and get the size as well. + * + * If we didn't need the mtime here, we could try to obtain the + * file size from the reconstruction or file copy process above, + * although that is actually not convenient in all cases. If we + * write the file ourselves then clearly we can keep a count of + * bytes, but if we use something like CopyFile() then it's + * trickier. Since we have to stat() anyway to get the mtime, + * there's no point in worrying about it. + */ + if (stat(ofullpath, &sb) < 0) + pg_fatal("could not stat file \"%s\": %m", ofullpath); + + /* OK, now do the work. */ + add_file_to_manifest(mwriter, manifest_path, + sb.st_size, sb.st_mtime, + checksum_type, checksum_length, + checksum_payload); + } + + /* Avoid leaking memory. */ + if (checksum_payload != NULL) + pfree(checksum_payload); + } + + closedir(dir); +} + +/* + * Read the version number from PG_VERSION and convert it to the usual server + * version number format. (e.g. If PG_VERSION contains "14\n" this function + * will return 140000) + */ +static int +read_pg_version_file(char *directory) +{ + char filename[MAXPGPATH]; + StringInfoData buf; + int fd; + int version; + char *ep; + + /* Construct pathname. */ + snprintf(filename, MAXPGPATH, "%s/PG_VERSION", directory); + + /* Open file. */ + if ((fd = open(filename, O_RDONLY, 0)) < 0) + pg_fatal("could not open file \"%s\": %m", filename); + + /* Read into memory. Length limit of 128 should be more than generous. */ + initStringInfo(&buf); + slurp_file(fd, filename, &buf, 128); + + /* Close the file. */ + if (close(fd) != 0) + pg_fatal("could not close \"%s\": %m", filename); + + /* Convert to integer. */ + errno = 0; + version = strtoul(buf.data, &ep, 10); + if (errno != 0 || *ep != '\n') + { + /* + * Incremental backup is not relevant to very old server versions that + * used multi-part version number (e.g. 9.6, or 8.4). So if we see + * what looks like the beginning of such a version number, just bail + * out. + */ + if (version < 10 && *ep == '.') + pg_fatal("%s: server version too old\n", filename); + pg_fatal("%s: could not parse version number\n", filename); + } + + /* Debugging output. */ + pg_log_debug("read server version %d from \"%s\"", version, filename); + + /* Release memory and return result. */ + pfree(buf.data); + return version * 10000; +} + +/* + * Add a directory to the list of output directories to clean up. + */ +static void +remember_to_cleanup_directory(char *target_path, bool rmtopdir) +{ + cb_cleanup_dir *dir = pg_malloc(sizeof(cb_cleanup_dir)); + + dir->target_path = target_path; + dir->rmtopdir = rmtopdir; + dir->next = cleanup_dir_list; + cleanup_dir_list = dir; +} + +/* + * Empty out the list of directories scheduled for cleanup a exit. + * + * We want to remove the output directories only on a failure, so call this + * function when we know that the operation has succeeded. + * + * Since we only expect this to be called when we're about to exit, we could + * just set cleanup_dir_list to NULL and be done with it, but we free the + * memory to be tidy. + */ +static void +reset_directory_cleanup_list(void) +{ + while (cleanup_dir_list != NULL) + { + cb_cleanup_dir *dir = cleanup_dir_list; + + cleanup_dir_list = cleanup_dir_list->next; + pfree(dir); + } +} + +/* + * Scan the pg_tblspc directory of the final input backup to get a canonical + * list of what tablespaces are part of the backup. + * + * 'pathname' should be the path to the toplevel backup directory for the + * final backup in the backup chain. + */ +static cb_tablespace * +scan_for_existing_tablespaces(char *pathname, cb_options *opt) +{ + char pg_tblspc[MAXPGPATH]; + DIR *dir; + struct dirent *de; + cb_tablespace *tslist = NULL; + + snprintf(pg_tblspc, MAXPGPATH, "%s/pg_tblspc", pathname); + pg_log_debug("scanning \"%s\"", pg_tblspc); + + if ((dir = opendir(pg_tblspc)) == NULL) + pg_fatal("could not open directory \"%s\": %m", pathname); + + while (errno = 0, (de = readdir(dir)) != NULL) + { + Oid oid; + char tblspcdir[MAXPGPATH]; + char link_target[MAXPGPATH]; + int link_length; + cb_tablespace *ts; + cb_tablespace *otherts; + PGFileType type; + + /* Silently ignore "." and ".." entries. */ + if (strcmp(de->d_name, ".") == 0 || strcmp(de->d_name, "..") == 0) + continue; + + /* Construct full pathname. */ + snprintf(tblspcdir, MAXPGPATH, "%s/%s", pg_tblspc, de->d_name); + + /* Ignore any file name that doesn't look like a proper OID. */ + if (!parse_oid(de->d_name, &oid)) + { + pg_log_debug("skipping \"%s\" because the filename is not a legal tablespace OID", + tblspcdir); + continue; + } + + /* Only symbolic links and directories are tablespaces. */ + type = get_dirent_type(tblspcdir, de, false, PG_LOG_ERROR); + if (type == PGFILETYPE_ERROR) + exit(1); + if (type != PGFILETYPE_LNK && type != PGFILETYPE_DIR) + { + pg_log_debug("skipping \"%s\" because it is neither a symbolic link nor a directory", + tblspcdir); + continue; + } + + /* Create a new tablespace object. */ + ts = pg_malloc0(sizeof(cb_tablespace)); + ts->oid = oid; + + /* + * If it's a link, it's not an in-place tablespace. Otherwise, it must + * be a directory, and thus an in-place tablespace. + */ + if (type == PGFILETYPE_LNK) + { + cb_tablespace_mapping *tsmap; + + /* Read the link target. */ + link_length = readlink(tblspcdir, link_target, sizeof(link_target)); + if (link_length < 0) + pg_fatal("could not read symbolic link \"%s\": %m", + tblspcdir); + if (link_length >= sizeof(link_target)) + pg_fatal("symbolic link \"%s\" is too long", tblspcdir); + link_target[link_length] = '\0'; + if (!is_absolute_path(link_target)) + pg_fatal("symbolic link \"%s\" is relative", tblspcdir); + + /* Caonicalize the link target. */ + canonicalize_path(link_target); + + /* + * Find the corresponding tablespace mapping and copy the relevant + * details into the new tablespace entry. + */ + for (tsmap = opt->tsmappings; tsmap != NULL; tsmap = tsmap->next) + { + if (strcmp(tsmap->old_dir, link_target) == 0) + { + strncpy(ts->old_dir, tsmap->old_dir, MAXPGPATH); + strncpy(ts->new_dir, tsmap->new_dir, MAXPGPATH); + ts->in_place = false; + break; + } + } + + /* Every non-in-place tablespace must be mapped. */ + if (tsmap == NULL) + pg_fatal("tablespace at \"%s\" has no tablespace mapping", + link_target); + } + else + { + /* + * For an in-place tablespace, there's no separate directory, so + * we just record the paths within the data directories. + */ + snprintf(ts->old_dir, MAXPGPATH, "%s/%s", pg_tblspc, de->d_name); + snprintf(ts->new_dir, MAXPGPATH, "%s/pg_tblpc/%s", opt->output, + de->d_name); + ts->in_place = true; + } + + /* Tablespaces should not share a directory. */ + for (otherts = tslist; otherts != NULL; otherts = otherts->next) + if (strcmp(ts->new_dir, otherts->new_dir) == 0) + pg_fatal("tablespaces with OIDs %u and %u both point at \"%s\"", + otherts->oid, oid, ts->new_dir); + + /* Add this tablespace to the list. */ + ts->next = tslist; + tslist = ts; + } + + return tslist; +} + +/* + * Read a file into a StringInfo. + * + * fd is used for the actual file I/O, filename for error reporting purposes. + * A file longer than maxlen is a fatal error. + */ +static void +slurp_file(int fd, char *filename, StringInfo buf, int maxlen) +{ + struct stat st; + ssize_t rb; + + /* Check file size, and complain if it's too large. */ + if (fstat(fd, &st) != 0) + pg_fatal("could not stat \"%s\": %m", filename); + if (st.st_size > maxlen) + pg_fatal("file \"%s\" is too large", filename); + + /* Make sure we have enough space. */ + enlargeStringInfo(buf, st.st_size); + + /* Read the data. */ + rb = read(fd, &buf->data[buf->len], st.st_size); + + /* + * We don't expect any concurrent changes, so we should read exactly the + * expected number of bytes. + */ + if (rb != st.st_size) + { + if (rb < 0) + pg_fatal("could not read file \"%s\": %m", filename); + else + pg_fatal("could not read file \"%s\": read only %d of %d bytes", + filename, (int) rb, (int) st.st_size); + } + + /* Adjust buffer length for new data and restore trailing-\0 invariant */ + buf->len += rb; + buf->data[buf->len] = '\0'; +} diff --git a/src/bin/pg_combinebackup/reconstruct.c b/src/bin/pg_combinebackup/reconstruct.c new file mode 100644 index 0000000000..c774bf1842 --- /dev/null +++ b/src/bin/pg_combinebackup/reconstruct.c @@ -0,0 +1,618 @@ +/*------------------------------------------------------------------------- + * + * reconstruct.c + * Reconstruct full file from incremental file and backup chain. + * + * Copyright (c) 2017-2023, PostgreSQL Global Development Group + * + * IDENTIFICATION + * src/bin/pg_combinebackup/reconstruct.c + * + *------------------------------------------------------------------------- + */ +#include "postgres_fe.h" + +#include <unistd.h> + +#include "backup/basebackup_incremental.h" +#include "common/logging.h" +#include "common/file_perm.h" +#include "copy_file.h" +#include "reconstruct.h" +#include "storage/block.h" + +/* + * An rfile stores the data that we need in order to be able to use some file + * on disk for reconstruction. For any given output file, we create one rfile + * per backup that we need to consult when we constructing that output file. + * + * If we find a full version of the file in the backup chain, then only + * filename and fd are initialized; the remaining fields are 0 or NULL. + * For an incremental file, header_length, num_blocks, relative_block_numbers, + * and truncation_block_length are also set. + * + * num_blocks_read and highest_offset_read always start out as 0. + */ +typedef struct rfile +{ + char *filename; + int fd; + size_t header_length; + unsigned num_blocks; + BlockNumber *relative_block_numbers; + unsigned truncation_block_length; + unsigned num_blocks_read; + off_t highest_offset_read; +} rfile; + +static void debug_reconstruction(int n_source, + rfile **sources, + bool dry_run); +static unsigned find_reconstructed_block_length(rfile *s); +static rfile *make_incremental_rfile(char *filename); +static rfile *make_rfile(char *filename, bool missing_ok); +static void write_reconstructed_file(char *input_filename, + char *output_filename, + unsigned block_length, + rfile **sourcemap, + off_t *offsetmap, + pg_checksum_context *checksum_ctx, + bool dry_run); +static void read_bytes(rfile *rf, void *buffer, unsigned length); + +/* + * Reconstruct a full file from an incremental file and a chain of prior + * backups. + * + * input_filename should be the path to the incremental file, and + * output_filename should be the path where the reconstructed file is to be + * written. + * + * relative_path should be the relative path to the directory containing this + * file. bare_file_name should be the name of the file within that directory, + * without "INCREMENTAL.". + * + * n_prior_backups is the number of prior backups, and prior_backup_dirs is + * an array of pathnames where those backups can be found. + */ +void +reconstruct_from_incremental_file(char *input_filename, + char *output_filename, + char *relative_path, + char *bare_file_name, + int n_prior_backups, + char **prior_backup_dirs, + manifest_data **manifests, + char *manifest_path, + pg_checksum_type checksum_type, + int *checksum_length, + uint8 **checksum_payload, + bool dry_run) +{ + rfile **source; + rfile *latest_source = NULL; + rfile **sourcemap; + off_t *offsetmap; + unsigned block_length; + unsigned num_missing_blocks; + unsigned i; + unsigned sidx = n_prior_backups; + bool full_copy_possible = true; + int copy_source_index = -1; + rfile *copy_source = NULL; + pg_checksum_context checksum_ctx; + + /* + * Every block must come either from the latest version of the file or + * from one of the prior backups. + */ + source = pg_malloc0(sizeof(rfile *) * (1 + n_prior_backups)); + + /* + * Use the information from the latest incremental file to figure out how + * long the reconstructed file should be. + */ + latest_source = make_incremental_rfile(input_filename); + source[n_prior_backups] = latest_source; + block_length = find_reconstructed_block_length(latest_source); + + /* + * For each block in the output file, we need to know from which file we + * need to obtain it and at what offset in that file it's stored. + * sourcemap gives us the first of these things, and offsetmap the latter. + */ + sourcemap = pg_malloc0(sizeof(rfile *) * block_length); + offsetmap = pg_malloc0(sizeof(off_t) * block_length); + + /* + * Blocks prior to the truncation_block_length threshold must be obtained + * from some prior backup, while those after that threshold are left as + * zeroes if not present in the newest incremental file. + * num_missing_blocks counts the number of blocks that we must be found + * somewhere in the backup chain, and is thus initially equal to + * truncation_block_length. + */ + num_missing_blocks = latest_source->truncation_block_length; + + /* + * Every block that is present in the newest incremental file should be + * sourced from that file. If it precedes the truncation_block_length, + * it's a block that we would otherwise have had to find in an older + * backup and thus reduces the number of blocks remaining to be found by + * one; otherwise, it's an extra block that needs to be included in the + * output but would not have needed to be found in an older backup if it + * had not been present. + */ + for (i = 0; i < latest_source->num_blocks; ++i) + { + BlockNumber b = latest_source->relative_block_numbers[i]; + + Assert(b < block_length); + sourcemap[b] = latest_source; + offsetmap[b] = latest_source->header_length + (i * BLCKSZ); + if (b < latest_source->truncation_block_length) + num_missing_blocks--; + + /* + * A full copy of a file from an earlier backup is only possible if no + * blocks are needed from any later incremental file. + */ + full_copy_possible = false; + } + + while (num_missing_blocks > 0) + { + char source_filename[MAXPGPATH]; + rfile *s; + + /* + * Move to the next backup in the chain. If there are no more, then + * something has gone wrong and reconstruction has failed. + */ + if (sidx == 0) + pg_fatal("reconstruction for file \"%s\" failed to find %u required blocks", + output_filename, num_missing_blocks); + --sidx; + + /* + * Look for the full file in the previous backup. If not found, then + * look for an incremental file instead. + */ + snprintf(source_filename, MAXPGPATH, "%s/%s/%s", + prior_backup_dirs[sidx], relative_path, bare_file_name); + if ((s = make_rfile(source_filename, true)) == NULL) + { + snprintf(source_filename, MAXPGPATH, "%s/%s/INCREMENTAL.%s", + prior_backup_dirs[sidx], relative_path, bare_file_name); + s = make_incremental_rfile(source_filename); + } + source[sidx] = s; + + /* + * If s->header_length == 0, then this is a full file; otherwise, it's + * an incremental file. + */ + if (s->header_length != 0) + { + /* + * Since we found another incremental file, source all blocks from + * it that we need but don't yet have. + */ + for (i = 0; i < s->num_blocks; ++i) + { + BlockNumber b = s->relative_block_numbers[i]; + + if (b < latest_source->truncation_block_length && + sourcemap[b] == NULL) + { + sourcemap[b] = s; + offsetmap[b] = s->header_length + (i * BLCKSZ); + + Assert(num_missing_blocks > 0); + --num_missing_blocks; + + /* + * A full copy of a file from an earlier backup is only + * possible if no blocks are needed from any later + * incremental file. + */ + full_copy_possible = false; + } + } + } + else + { + BlockNumber b; + + /* + * Since we found a full file, source all remaining required + * blocks from it. + */ + for (b = 0; b < latest_source->truncation_block_length; ++b) + { + if (sourcemap[b] == NULL) + { + sourcemap[b] = s; + offsetmap[b] = b * BLCKSZ; + + Assert(num_missing_blocks > 0); + --num_missing_blocks; + } + } + Assert(num_missing_blocks == 0); + + /* + * If a full copy looks possible, check whether the resulting file + * should be exactly as long as the source file is. If so, a full + * copy is acceptable, otherwise not. + */ + if (full_copy_possible) + { + struct stat sb; + uint64 expected_length; + + if (fstat(s->fd, &sb) < 0) + pg_fatal("could not stat \"%s\": %m", s->filename); + expected_length = + (uint64) latest_source->truncation_block_length; + expected_length *= BLCKSZ; + if (expected_length == sb.st_size) + { + copy_source = s; + copy_source_index = sidx; + } + } + } + } + + /* + * If a checksum of the required type already exists in the + * backup_manifest for the relevant input directory, we can save some work + * by reusing that checksum instead of computing a new one. + */ + if (copy_source_index >= 0 && manifests[copy_source_index] != NULL && + checksum_type != CHECKSUM_TYPE_NONE) + { + manifest_file *mfile; + + mfile = manifest_files_lookup(manifests[copy_source_index]->files, + manifest_path); + if (mfile == NULL) + { + /* + * The directory is out of sync with the backup_manifest, so emit + * a warning. + */ + pg_log_warning("\"%s/backup_manifest\" contains no entry for \"%s\"", + prior_backup_dirs[copy_source_index], + manifest_path); + } + else if (mfile->checksum_type == checksum_type) + { + *checksum_length = mfile->checksum_length; + *checksum_payload = pg_malloc(*checksum_length); + memcpy(*checksum_payload, mfile->checksum_payload, + *checksum_length); + checksum_type = CHECKSUM_TYPE_NONE; + } + } + + /* Prepare for checksum calculation, if required. */ + pg_checksum_init(&checksum_ctx, checksum_type); + + /* + * If the full file can be created by copying a file from an older backup + * in the chain without needing to overwrite any blocks or truncate the + * result, then forget about performing reconstruction and just copy that + * file in its entirety. + * + * Otherwise, reconstruct. + */ + if (copy_source != NULL) + copy_file(copy_source->filename, output_filename, + &checksum_ctx, dry_run); + else + { + write_reconstructed_file(input_filename, output_filename, + block_length, sourcemap, offsetmap, + &checksum_ctx, dry_run); + debug_reconstruction(n_prior_backups + 1, source, dry_run); + } + + /* Save results of checksum calculation. */ + if (checksum_type != CHECKSUM_TYPE_NONE) + { + *checksum_payload = pg_malloc(PG_CHECKSUM_MAX_LENGTH); + *checksum_length = pg_checksum_final(&checksum_ctx, + *checksum_payload); + } + + /* + * Close files and release memory. + */ + for (i = 0; i <= n_prior_backups; ++i) + { + rfile *s = source[i]; + + if (s == NULL) + continue; + if (close(s->fd) != 0) + pg_fatal("could not close \"%s\": %m", s->filename); + if (s->relative_block_numbers != NULL) + pfree(s->relative_block_numbers); + pg_free(s->filename); + } + pfree(sourcemap); + pfree(offsetmap); + pfree(source); +} + +/* + * Perform post-reconstruction logging and sanity checks. + */ +static void +debug_reconstruction(int n_source, rfile **sources, bool dry_run) +{ + unsigned i; + + for (i = 0; i < n_source; ++i) + { + rfile *s = sources[i]; + + /* Ignore source if not used. */ + if (s == NULL) + continue; + + /* If no data is needed from this file, we can ignore it. */ + if (s->num_blocks_read == 0) + continue; + + /* Debug logging. */ + if (dry_run) + pg_log_debug("would have read %u blocks from \"%s\"", + s->num_blocks_read, s->filename); + else + pg_log_debug("read %u blocks from \"%s\"", + s->num_blocks_read, s->filename); + + /* + * In dry-run mode, we don't actually try to read data from the file, + * but we do try to verify that the file is long enough that we could + * have read the data if we'd tried. + * + * If this fails, then it means that a non-dry-run attempt would fail, + * complaining of not being able to read the required bytes from the + * file. + */ + if (dry_run) + { + struct stat sb; + + if (fstat(s->fd, &sb) < 0) + pg_fatal("could not stat \"%s\": %m", s->filename); + if (sb.st_size < s->highest_offset_read) + pg_fatal("file \"%s\" is too short: expected %llu, found %llu", + s->filename, + (unsigned long long) s->highest_offset_read, + (unsigned long long) sb.st_size); + } + } +} + +/* + * When we perform reconstruction using an incremental file, the output file + * should be at least as long as the truncation_block_length. Any blocks + * present in the incremental file increase the output length as far as is + * necessary to include those blocks. + */ +static unsigned +find_reconstructed_block_length(rfile *s) +{ + unsigned block_length = s->truncation_block_length; + unsigned i; + + for (i = 0; i < s->num_blocks; ++i) + if (s->relative_block_numbers[i] >= block_length) + block_length = s->relative_block_numbers[i] + 1; + + return block_length; +} + +/* + * Initialize an incremental rfile, reading the header so that we know which + * blocks it contains. + */ +static rfile * +make_incremental_rfile(char *filename) +{ + rfile *rf; + unsigned magic; + + rf = make_rfile(filename, false); + + /* Read and validate magic number. */ + read_bytes(rf, &magic, sizeof(magic)); + if (magic != INCREMENTAL_MAGIC) + pg_fatal("file \"%s\" has bad incremental magic number (0x%x not 0x%x)", + filename, magic, INCREMENTAL_MAGIC); + + /* Read block count. */ + read_bytes(rf, &rf->num_blocks, sizeof(rf->num_blocks)); + if (rf->num_blocks > RELSEG_SIZE) + pg_fatal("file \"%s\" has block count %u in excess of segment size %u", + filename, rf->num_blocks, RELSEG_SIZE); + + /* Read truncation block length. */ + read_bytes(rf, &rf->truncation_block_length, + sizeof(rf->truncation_block_length)); + if (rf->truncation_block_length > RELSEG_SIZE) + pg_fatal("file \"%s\" has truncation block length %u in excess of segment size %u", + filename, rf->truncation_block_length, RELSEG_SIZE); + + /* Read block numbers if there are any. */ + if (rf->num_blocks > 0) + { + rf->relative_block_numbers = + pg_malloc0(sizeof(BlockNumber) * rf->num_blocks); + read_bytes(rf, rf->relative_block_numbers, + sizeof(BlockNumber) * rf->num_blocks); + } + + /* Remember length of header. */ + rf->header_length = sizeof(magic) + sizeof(rf->num_blocks) + + sizeof(rf->truncation_block_length) + + sizeof(BlockNumber) * rf->num_blocks; + + return rf; +} + +/* + * Allocate and perform basic initialization of an rfile. + */ +static rfile * +make_rfile(char *filename, bool missing_ok) +{ + rfile *rf; + + rf = pg_malloc0(sizeof(rfile)); + rf->filename = pstrdup(filename); + if ((rf->fd = open(filename, O_RDONLY | PG_BINARY, 0)) < 0) + { + if (missing_ok && errno == ENOENT) + { + pg_free(rf); + return NULL; + } + pg_fatal("could not open file \"%s\": %m", filename); + } + + return rf; +} + +/* + * Read the indicated number of bytes from an rfile into the buffer. + */ +static void +read_bytes(rfile *rf, void *buffer, unsigned length) +{ + unsigned rb = read(rf->fd, buffer, length); + + if (rb != length) + { + if (rb < 0) + pg_fatal("could not read file \"%s\": %m", rf->filename); + else + pg_fatal("could not read file \"%s\": read only %d of %d bytes", + rf->filename, (int) rb, length); + } +} + +/* + * Write out a reconstructed file. + */ +static void +write_reconstructed_file(char *input_filename, + char *output_filename, + unsigned block_length, + rfile **sourcemap, + off_t *offsetmap, + pg_checksum_context *checksum_ctx, + bool dry_run) +{ + int wfd = -1; + unsigned i; + unsigned zero_blocks = 0; + + /* Debugging output. */ + if (dry_run) + pg_log_debug("would reconstruct \"%s\" (%u blocks, checksum %s)", + output_filename, block_length, + pg_checksum_type_name(checksum_ctx->type)); + else + pg_log_debug("reconstructing \"%s\" (%u blocks, checksum %s)", + output_filename, block_length, + pg_checksum_type_name(checksum_ctx->type)); + + /* Open the output file, except in dry_run mode. */ + if (!dry_run && + (wfd = open(output_filename, + O_RDWR | PG_BINARY | O_CREAT | O_EXCL, + pg_file_create_mode)) < 0) + pg_fatal("could not open file \"%s\": %m", output_filename); + + /* Read and write the blocks as required. */ + for (i = 0; i < block_length; ++i) + { + uint8 buffer[BLCKSZ]; + rfile *s = sourcemap[i]; + unsigned wb; + + /* Update accounting information. */ + if (s == NULL) + ++zero_blocks; + else + { + s->num_blocks_read++; + s->highest_offset_read = Max(s->highest_offset_read, + offsetmap[i] + BLCKSZ); + } + + /* Skip the rest of this in dry-run mode. */ + if (dry_run) + continue; + + /* Read or zero-fill the block as appropriate. */ + if (s == NULL) + { + /* + * New block not mentioned in the WAL summary. Should have been an + * uninitialized block, so just zero-fill it. + */ + memset(buffer, 0, BLCKSZ); + } + else + { + unsigned rb; + + /* Read the block from the correct source, except if dry-run. */ + rb = pg_pread(s->fd, buffer, BLCKSZ, offsetmap[i]); + if (rb != BLCKSZ) + { + if (rb < 0) + pg_fatal("could not read file \"%s\": %m", s->filename); + else + pg_fatal("could not read file \"%s\": read only %d of %d bytes at offset %u", + s->filename, (int) rb, BLCKSZ, + (unsigned) offsetmap[i]); + } + } + + /* Write out the block. */ + if ((wb = write(wfd, buffer, BLCKSZ)) != BLCKSZ) + { + if (wb < 0) + pg_fatal("could not write file \"%s\": %m", output_filename); + else + pg_fatal("could not write file \"%s\": wrote only %d of %d bytes", + output_filename, (int) wb, BLCKSZ); + } + + /* Update the checksum computation. */ + if (pg_checksum_update(checksum_ctx, buffer, BLCKSZ) < 0) + pg_fatal("could not update checksum of file \"%s\"", + output_filename); + } + + /* Debugging output. */ + if (zero_blocks > 0) + { + if (dry_run) + pg_log_debug("would have zero-filled %u blocks", zero_blocks); + else + pg_log_debug("zero-filled %u blocks", zero_blocks); + } + + /* Close the output file. */ + if (wfd >= 0 && close(wfd) != 0) + pg_fatal("could not close \"%s\": %m", output_filename); +} diff --git a/src/bin/pg_combinebackup/reconstruct.h b/src/bin/pg_combinebackup/reconstruct.h new file mode 100644 index 0000000000..c599a70d42 --- /dev/null +++ b/src/bin/pg_combinebackup/reconstruct.h @@ -0,0 +1,32 @@ +/*------------------------------------------------------------------------- + * + * reconstruct.h + * Reconstruct full file from incremental file and backup chain. + * + * Copyright (c) 2017-2023, PostgreSQL Global Development Group + * + * IDENTIFICATION + * src/bin/pg_combinebackup/reconstruct.h + * + *------------------------------------------------------------------------- + */ +#ifndef RECONSTRUCT_H +#define RECONSTRUCT_H + +#include "common/checksum_helper.h" +#include "load_manifest.h" + +extern void reconstruct_from_incremental_file(char *input_filename, + char *output_filename, + char *relative_path, + char *bare_file_name, + int n_prior_backups, + char **prior_backup_dirs, + manifest_data **manifests, + char *manifest_path, + pg_checksum_type checksum_type, + int *checksum_length, + uint8 **checksum_payload, + bool dry_run); + +#endif diff --git a/src/bin/pg_combinebackup/write_manifest.c b/src/bin/pg_combinebackup/write_manifest.c new file mode 100644 index 0000000000..82160134d8 --- /dev/null +++ b/src/bin/pg_combinebackup/write_manifest.c @@ -0,0 +1,293 @@ +/*------------------------------------------------------------------------- + * + * Write a new backup manifest. + * + * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group + * Portions Copyright (c) 1994, Regents of the University of California + * + * src/bin/pg_combinebackup/write_manifest.c + * + *------------------------------------------------------------------------- + */ + +#include "postgres_fe.h" + +#include <fcntl.h> +#include <time.h> +#include <unistd.h> + +#include "common/checksum_helper.h" +#include "common/file_perm.h" +#include "common/logging.h" +#include "lib/stringinfo.h" +#include "load_manifest.h" +#include "mb/pg_wchar.h" +#include "write_manifest.h" + +struct manifest_writer +{ + char pathname[MAXPGPATH]; + int fd; + StringInfoData buf; + bool first_file; + bool still_checksumming; + pg_checksum_context manifest_ctx; +}; + +static void escape_json(StringInfo buf, const char *str); +static void flush_manifest(manifest_writer *mwriter); +static size_t hex_encode(const uint8 *src, size_t len, char *dst); + +/* + * Create a new backup manifest writer. + * + * The backup manifest will be written into a file named backup_manifest + * in the specified directory. + */ +manifest_writer * +create_manifest_writer(char *directory) +{ + manifest_writer *mwriter = pg_malloc(sizeof(manifest_writer)); + + snprintf(mwriter->pathname, MAXPGPATH, "%s/backup_manifest", directory); + mwriter->fd = -1; + initStringInfo(&mwriter->buf); + mwriter->first_file = true; + mwriter->still_checksumming = true; + pg_checksum_init(&mwriter->manifest_ctx, CHECKSUM_TYPE_SHA256); + + appendStringInfo(&mwriter->buf, + "{ \"PostgreSQL-Backup-Manifest-Version\": 1,\n" + "\"Files\": ["); + + return mwriter; +} + +/* + * Add an entry for a file to a backup manifest. + * + * This is very similar to the backend's AddFileToBackupManifest, but + * various adjustments are required due to frontend/backend differences + * and other details. + */ +void +add_file_to_manifest(manifest_writer *mwriter, const char *manifest_path, + size_t size, pg_time_t mtime, + pg_checksum_type checksum_type, + int checksum_length, + uint8 *checksum_payload) +{ + int pathlen = strlen(manifest_path); + + if (mwriter->first_file) + { + appendStringInfoChar(&mwriter->buf, '\n'); + mwriter->first_file = false; + } + else + appendStringInfoString(&mwriter->buf, ",\n"); + + if (pg_encoding_verifymbstr(PG_UTF8, manifest_path, pathlen) == pathlen) + { + appendStringInfoString(&mwriter->buf, "{ \"Path\": "); + escape_json(&mwriter->buf, manifest_path); + appendStringInfoString(&mwriter->buf, ", "); + } + else + { + appendStringInfoString(&mwriter->buf, "{ \"Encoded-Path\": \""); + enlargeStringInfo(&mwriter->buf, 2 * pathlen); + mwriter->buf.len += hex_encode((const uint8 *) manifest_path, pathlen, + &mwriter->buf.data[mwriter->buf.len]); + appendStringInfoString(&mwriter->buf, "\", "); + } + + appendStringInfo(&mwriter->buf, "\"Size\": %zu, ", size); + + appendStringInfoString(&mwriter->buf, "\"Last-Modified\": \""); + enlargeStringInfo(&mwriter->buf, 128); + mwriter->buf.len += strftime(&mwriter->buf.data[mwriter->buf.len], 128, + "%Y-%m-%d %H:%M:%S %Z", + gmtime(&mtime)); + appendStringInfoChar(&mwriter->buf, '"'); + + if (mwriter->buf.len > 128 * 1024) + flush_manifest(mwriter); + + if (checksum_length > 0) + { + appendStringInfo(&mwriter->buf, + ", \"Checksum-Algorithm\": \"%s\", \"Checksum\": \"", + pg_checksum_type_name(checksum_type)); + + enlargeStringInfo(&mwriter->buf, 2 * checksum_length); + mwriter->buf.len += hex_encode(checksum_payload, checksum_length, + &mwriter->buf.data[mwriter->buf.len]); + + appendStringInfoChar(&mwriter->buf, '"'); + } + + appendStringInfoString(&mwriter->buf, " }"); + + if (mwriter->buf.len > 128 * 1024) + flush_manifest(mwriter); +} + +/* + * Finalize the backup_manifest. + */ +void +finalize_manifest(manifest_writer *mwriter, + manifest_wal_range *first_wal_range) +{ + uint8 checksumbuf[PG_SHA256_DIGEST_LENGTH]; + int len; + manifest_wal_range *wal_range; + + /* Terminate the list of files. */ + appendStringInfoString(&mwriter->buf, "\n],\n"); + + /* Start a list of LSN ranges. */ + appendStringInfoString(&mwriter->buf, "\"WAL-Ranges\": [\n"); + + for (wal_range = first_wal_range; wal_range != NULL; + wal_range = wal_range->next) + appendStringInfo(&mwriter->buf, + "%s{ \"Timeline\": %u, \"Start-LSN\": \"%X/%X\", \"End-LSN\": \"%X/%X\" }", + wal_range == first_wal_range ? "" : ",\n", + wal_range->tli, + LSN_FORMAT_ARGS(wal_range->start_lsn), + LSN_FORMAT_ARGS(wal_range->end_lsn)); + + /* Terminate the list of WAL ranges. */ + appendStringInfoString(&mwriter->buf, "\n],\n"); + + /* Flush accumulated data and update checksum calculation. */ + flush_manifest(mwriter); + + /* Checksum only includes data up to this point. */ + mwriter->still_checksumming = false; + + /* Compute and insert manifest checksum. */ + appendStringInfoString(&mwriter->buf, "\"Manifest-Checksum\": \""); + enlargeStringInfo(&mwriter->buf, 2 * PG_SHA256_DIGEST_STRING_LENGTH); + len = pg_checksum_final(&mwriter->manifest_ctx, checksumbuf); + Assert(len == PG_SHA256_DIGEST_LENGTH); + mwriter->buf.len += + hex_encode(checksumbuf, len, &mwriter->buf.data[mwriter->buf.len]); + appendStringInfoString(&mwriter->buf, "\"}\n"); + + /* Flush the last manifest checksum itself. */ + flush_manifest(mwriter); + + /* Close the file. */ + if (close(mwriter->fd) != 0) + pg_fatal("could not close \"%s\": %m", mwriter->pathname); + mwriter->fd = -1; +} + +/* + * Produce a JSON string literal, properly escaping characters in the text. + */ +static void +escape_json(StringInfo buf, const char *str) +{ + const char *p; + + appendStringInfoCharMacro(buf, '"'); + for (p = str; *p; p++) + { + switch (*p) + { + case '\b': + appendStringInfoString(buf, "\\b"); + break; + case '\f': + appendStringInfoString(buf, "\\f"); + break; + case '\n': + appendStringInfoString(buf, "\\n"); + break; + case '\r': + appendStringInfoString(buf, "\\r"); + break; + case '\t': + appendStringInfoString(buf, "\\t"); + break; + case '"': + appendStringInfoString(buf, "\\\""); + break; + case '\\': + appendStringInfoString(buf, "\\\\"); + break; + default: + if ((unsigned char) *p < ' ') + appendStringInfo(buf, "\\u%04x", (int) *p); + else + appendStringInfoCharMacro(buf, *p); + break; + } + } + appendStringInfoCharMacro(buf, '"'); +} + +/* + * Flush whatever portion of the backup manifest we have generated and + * buffered in memory out to a file on disk. + * + * The first call to this function will create the file. After that, we + * keep it open and just append more data. + */ +static void +flush_manifest(manifest_writer *mwriter) +{ + char pathname[MAXPGPATH]; + + if (mwriter->fd == -1 && + (mwriter->fd = open(mwriter->pathname, + O_WRONLY | O_CREAT | O_EXCL | PG_BINARY, + pg_file_create_mode)) < 0) + pg_fatal("could not open file \"%s\": %m", mwriter->pathname); + + if (mwriter->buf.len > 0) + { + ssize_t wb; + + wb = write(mwriter->fd, mwriter->buf.data, mwriter->buf.len); + if (wb != mwriter->buf.len) + { + if (wb < 0) + pg_fatal("could not write \"%s\": %m", mwriter->pathname); + else + pg_fatal("could not write file \"%s\": wrote only %d of %d bytes", + pathname, (int) wb, mwriter->buf.len); + } + + if (mwriter->still_checksumming) + pg_checksum_update(&mwriter->manifest_ctx, + (uint8 *) mwriter->buf.data, + mwriter->buf.len); + resetStringInfo(&mwriter->buf); + } +} + +/* + * Encode bytes using two hexademical digits for each one. + */ +static size_t +hex_encode(const uint8 *src, size_t len, char *dst) +{ + const uint8 *end = src + len; + + while (src < end) + { + unsigned n1 = (*src >> 4) & 0xF; + unsigned n2 = *src & 0xF; + + *dst++ = n1 < 10 ? '0' + n1 : 'a' + n1 - 10; + *dst++ = n2 < 10 ? '0' + n2 : 'a' + n2 - 10; + ++src; + } + + return len * 2; +} diff --git a/src/bin/pg_combinebackup/write_manifest.h b/src/bin/pg_combinebackup/write_manifest.h new file mode 100644 index 0000000000..8fd7fe02c8 --- /dev/null +++ b/src/bin/pg_combinebackup/write_manifest.h @@ -0,0 +1,33 @@ +/*------------------------------------------------------------------------- + * + * Write a new backup manifest. + * + * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group + * Portions Copyright (c) 1994, Regents of the University of California + * + * src/bin/pg_combinebackup/write_manifest.h + * + *------------------------------------------------------------------------- + */ +#ifndef WRITE_MANIFEST_H +#define WRITE_MANIFEST_H + +#include "common/checksum_helper.h" +#include "pgtime.h" + +struct manifest_wal_range; + +struct manifest_writer; +typedef struct manifest_writer manifest_writer; + +extern manifest_writer *create_manifest_writer(char *directory); +extern void add_file_to_manifest(manifest_writer *mwriter, + const char *manifest_path, + size_t size, pg_time_t mtime, + pg_checksum_type checksum_type, + int checksum_length, + uint8 *checksum_payload); +extern void finalize_manifest(manifest_writer *mwriter, + struct manifest_wal_range *first_wal_range); + +#endif /* WRITE_MANIFEST_H */ diff --git a/src/bin/pg_resetwal/pg_resetwal.c b/src/bin/pg_resetwal/pg_resetwal.c index 04567f349d..c3b9e07841 100644 --- a/src/bin/pg_resetwal/pg_resetwal.c +++ b/src/bin/pg_resetwal/pg_resetwal.c @@ -85,6 +85,7 @@ static void RewriteControlFile(void); static void FindEndOfXLOG(void); static void KillExistingXLOG(void); static void KillExistingArchiveStatus(void); +static void KillExistingWALSummaries(void); static void WriteEmptyXLOG(void); static void usage(void); @@ -493,6 +494,7 @@ main(int argc, char *argv[]) RewriteControlFile(); KillExistingXLOG(); KillExistingArchiveStatus(); + KillExistingWALSummaries(); WriteEmptyXLOG(); printf(_("Write-ahead log reset\n")); @@ -1034,6 +1036,40 @@ KillExistingArchiveStatus(void) pg_fatal("could not close directory \"%s\": %m", ARCHSTATDIR); } +/* + * Remove existing WAL summary files + */ +static void +KillExistingWALSummaries(void) +{ +#define WALSUMMARYDIR XLOGDIR "/summaries" +#define WALSUMMARY_NHEXCHARS 40 + + DIR *xldir; + struct dirent *xlde; + char path[MAXPGPATH + sizeof(WALSUMMARYDIR)]; + + xldir = opendir(WALSUMMARYDIR); + if (xldir == NULL) + pg_fatal("could not open directory \"%s\": %m", WALSUMMARYDIR); + + while (errno = 0, (xlde = readdir(xldir)) != NULL) + { + if (strspn(xlde->d_name, "0123456789ABCDEF") == WALSUMMARY_NHEXCHARS && + strcmp(xlde->d_name + WALSUMMARY_NHEXCHARS, ".summary") == 0) + { + snprintf(path, sizeof(path), "%s/%s", WALSUMMARYDIR, xlde->d_name); + if (unlink(path) < 0) + pg_fatal("could not delete file \"%s\": %m", path); + } + } + + if (errno) + pg_fatal("could not read directory \"%s\": %m", WALSUMMARYDIR); + + if (closedir(xldir)) + pg_fatal("could not close directory \"%s\": %m", ARCHSTATDIR); +} /* * Write an empty XLOG file, containing only the checkpoint record diff --git a/src/common/Makefile b/src/common/Makefile index ff60666f5c..ebff20b1d3 100644 --- a/src/common/Makefile +++ b/src/common/Makefile @@ -49,6 +49,7 @@ OBJS_COMMON = \ archive.o \ base64.o \ binaryheap.o \ + blkreftable.o \ checksum_helper.o \ compression.o \ config_info.o \ diff --git a/src/common/blkreftable.c b/src/common/blkreftable.c new file mode 100644 index 0000000000..012a443584 --- /dev/null +++ b/src/common/blkreftable.c @@ -0,0 +1,1309 @@ +/*------------------------------------------------------------------------- + * + * blkreftable.c + * Block reference tables. + * + * A block reference table is used to keep track of which blocks have + * been modified by WAL records within a certain LSN range. + * + * For each relation fork, we keep track of all blocks that have appeared + * in block reference in the WAL. We also keep track of the "limit block", + * which is the smallest relation length in blocks known to have occurred + * during that range of WAL records. This should be set to 0 if the relation + * fork is created or destroyed, and to the post-truncation length if + * truncated. + * + * Whenever we set the limit block, we also forget about any modified blocks + * beyond that point. Those blocks don't exist any more. Such blocks can + * later be marked as modified again; if that happens, it means the relation + * was re-extended. + * + * Portions Copyright (c) 2010-2022, PostgreSQL Global Development Group + * + * src/common/blkreftable.c + * + *------------------------------------------------------------------------- + */ + + +#ifndef FRONTEND +#include "postgres.h" +#else +#include "postgres_fe.h" +#endif + +#ifdef FRONTEND +#include "common/logging.h" +#endif + +#include "common/blkreftable.h" +#include "common/hashfn.h" +#include "port/pg_crc32c.h" + +/* + * A block reference table keeps track of the status of each relation + * fork individually. + */ +typedef struct BlockRefTableKey +{ + RelFileLocator rlocator; + ForkNumber forknum; +} BlockRefTableKey; + +/* + * We could need to store data either for a relation in which only a + * tiny fraction of the blocks have been modified or for a relation in + * which nearly every block has been modified, and we want a + * space-efficient representation in both cases. To accomplish this, + * we divide the relation into chunks of 2^16 blocks and choose between + * an array representation and a bitmap representation for each chunk. + * + * When the number of modified blocks in a given chunk is small, we + * essentially store an array of block numbers, but we need not store the + * entire block number: instead, we store each block number as a 2-byte + * offset from the start of the chunk. + * + * When the number of modified blocks in a given chunk is large, we switch + * to a bitmap representation. + * + * These same basic representational choices are used both when a block + * reference table is stored in memory and when it is serialized to disk. + * + * In the in-memory representation, we initially allocate each chunk with + * space for a number of entries given by INITIAL_ENTRIES_PER_CHUNK and + * increase that as necessary until we reach MAX_ENTRIES_PER_CHUNK. + * Any chunk whose allocated size reaches MAX_ENTRIES_PER_CHUNK is converted + * to a bitmap, and thus never needs to grow further. + */ +#define BLOCKS_PER_CHUNK (1 << 16) +#define BLOCKS_PER_ENTRY (BITS_PER_BYTE * sizeof(uint16)) +#define MAX_ENTRIES_PER_CHUNK (BLOCKS_PER_CHUNK / BLOCKS_PER_ENTRY) +#define INITIAL_ENTRIES_PER_CHUNK 16 +typedef uint16 *BlockRefTableChunk; + +/* + * State for one relation fork. + * + * 'rlocator' and 'forknum' identify the relation fork to which this entry + * pertains. + * + * 'limit_block' is the shortest known length of the relation in blocks + * within the LSN range covered by a particular block reference table. + * It should be set to 0 if the relation fork is created or dropped. If the + * relation fork is truncated, it should be set to the number of blocks that + * remain after truncation. + * + * 'nchunks' is the allocated length of each of the three arrays that follow. + * We can only represent the status of block numbers less than nchunks * + * BLOCKS_PER_CHUNK. + * + * 'chunk_size' is an array storing the allocated size of each chunk. + * + * 'chunk_usage' is an array storing the number of elements used in each + * chunk. If that value is less than MAX_ENTRIES_PER_CHUNK, the corresonding + * chunk is used as an array; else the corresponding chunk is used as a bitmap. + * When used as a bitmap, the least significant bit of the first array element + * is the status of the lowest-numbered block covered by this chunk. + * + * 'chunk_data' is the array of chunks. + */ +struct BlockRefTableEntry +{ + BlockRefTableKey key; + BlockNumber limit_block; + char status; + uint32 nchunks; + uint16 *chunk_size; + uint16 *chunk_usage; + BlockRefTableChunk *chunk_data; +}; + +/* Declare and define a hash table over type BlockRefTableEntry. */ +#define SH_PREFIX blockreftable +#define SH_ELEMENT_TYPE BlockRefTableEntry +#define SH_KEY_TYPE BlockRefTableKey +#define SH_KEY key +#define SH_HASH_KEY(tb, key) \ + hash_bytes((const unsigned char *) &key, sizeof(BlockRefTableKey)) +#define SH_EQUAL(tb, a, b) memcmp(&a, &b, sizeof(BlockRefTableKey)) == 0 +#define SH_SCOPE static inline +#ifdef FRONTEND +#define SH_RAW_ALLOCATOR pg_malloc0 +#endif +#define SH_DEFINE +#define SH_DECLARE +#include "lib/simplehash.h" + +/* + * A block reference table is basically just the hash table, but we don't + * want to expose that to outside callers. + * + * We keep track of the memory context in use explicitly too, so that it's + * easy to place all of our allocations in the same context. + */ +struct BlockRefTable +{ + blockreftable_hash *hash; +#ifndef FRONTEND + MemoryContext mcxt; +#endif +}; + +/* + * On-disk serialization format for block reference table entries. + */ +typedef struct BlockRefTableSerializedEntry +{ + RelFileLocator rlocator; + ForkNumber forknum; + BlockNumber limit_block; + uint32 nchunks; +} BlockRefTableSerializedEntry; + +/* + * Buffer size, so that we avoid doing many small I/Os. + */ +#define BUFSIZE 65536 + +/* + * Ad-hoc buffer for file I/O. + */ +typedef struct BlockRefTableBuffer +{ + io_callback_fn io_callback; + void *io_callback_arg; + char data[BUFSIZE]; + int used; + int cursor; + pg_crc32c crc; +} BlockRefTableBuffer; + +/* + * State for keeping track of progress while incrementally reading a block + * table reference file from disk. + * + * total_chunks means the number of chunks for the RelFileLocator/ForkNumber + * combination that is curently being read, and consumed_chunks is the number + * of those that have been read. (We always read all the information for + * a single chunk at one time, so we don't need to be able to represent the + * state where a chunk has been partially read.) + * + * chunk_size is the array of chunk sizes. The length is given by total_chunks. + * + * chunk_data holds the current chunk. + * + * chunk_position helps us figure out how much progress we've made in returning + * the block numbers for the current chunk to the caller. If the chunk is a + * bitmap, it's the number of bits we've scanned; otherwise, it's the number + * of chunk entries we've scanned. + */ +struct BlockRefTableReader +{ + BlockRefTableBuffer buffer; + char *error_filename; + report_error_fn error_callback; + void *error_callback_arg; + uint32 total_chunks; + uint32 consumed_chunks; + uint16 *chunk_size; + uint16 chunk_data[MAX_ENTRIES_PER_CHUNK]; + uint32 chunk_position; +}; + +/* + * State for keeping track of progress while incrementally writing a block + * reference table file to disk. + */ +struct BlockRefTableWriter +{ + BlockRefTableBuffer buffer; +}; + +/* Function prototypes. */ +static int BlockRefTableComparator(const void *a, const void *b); +static void BlockRefTableFlush(BlockRefTableBuffer *buffer); +static void BlockRefTableRead(BlockRefTableReader *reader, void *data, + int length); +static void BlockRefTableWrite(BlockRefTableBuffer *buffer, void *data, + int length); +static void BlockRefTableFileTerminate(BlockRefTableBuffer *buffer); + +/* + * Create an empty block reference table. + */ +BlockRefTable * +CreateEmptyBlockRefTable(void) +{ + BlockRefTable *brtab = palloc(sizeof(BlockRefTable)); + + /* + * Even completely empty database has a few hundred relation forks, so it + * seems best to size the hash on the assumption that we're going to have + * at least a few thousand entries. + */ +#ifdef FRONTEND + brtab->hash = blockreftable_create(4096, NULL); +#else + brtab->mcxt = CurrentMemoryContext; + brtab->hash = blockreftable_create(brtab->mcxt, 4096, NULL); +#endif + + return brtab; +} + +/* + * Set the "limit block" for a relation fork and forget any modified blocks + * with equal or higher block numbers. + * + * The "limit block" is the shortest known length of the relation within the + * range of WAL records covered by this block reference table. + */ +void +BlockRefTableSetLimitBlock(BlockRefTable *brtab, + const RelFileLocator *rlocator, + ForkNumber forknum, + BlockNumber limit_block) +{ + BlockRefTableEntry *brtentry; + BlockRefTableKey key; + bool found; + + memcpy(&key.rlocator, rlocator, sizeof(RelFileLocator)); + key.forknum = forknum; + brtentry = blockreftable_insert(brtab->hash, key, &found); + + if (!found) + { + /* + * We have no existing data about this relation fork, so just record + * the limit_block value supplied by the caller, and make sure other + * parts of the entry are properly initialized. + */ + brtentry->limit_block = limit_block; + brtentry->nchunks = 0; + brtentry->chunk_size = NULL; + brtentry->chunk_usage = NULL; + brtentry->chunk_data = NULL; + return; + } + + BlockRefTableEntrySetLimitBlock(brtentry, limit_block); +} + +/* + * Mark a block in a given relation fork as known to have been modified. + */ +void +BlockRefTableMarkBlockModified(BlockRefTable *brtab, + const RelFileLocator *rlocator, + ForkNumber forknum, + BlockNumber blknum) +{ + BlockRefTableEntry *brtentry; + BlockRefTableKey key; + bool found; +#ifndef FRONTEND + MemoryContext oldcontext = MemoryContextSwitchTo(brtab->mcxt); +#endif + + memcpy(&key.rlocator, rlocator, sizeof(RelFileLocator)); + key.forknum = forknum; + brtentry = blockreftable_insert(brtab->hash, key, &found); + + if (!found) + { + /* + * We want to set the initial limit block value to something higher + * than any legal block number. InvalidBlockNumber fits the bill. + */ + brtentry->limit_block = InvalidBlockNumber; + brtentry->nchunks = 0; + brtentry->chunk_size = NULL; + brtentry->chunk_usage = NULL; + brtentry->chunk_data = NULL; + } + + BlockRefTableEntryMarkBlockModified(brtentry, forknum, blknum); + +#ifndef FRONTEND + MemoryContextSwitchTo(oldcontext); +#endif +} + +/* + * Get an entry from a block reference table. + * + * If the entry does not exist, this function returns NULL. Otherwise, it + * returns the entry and sets *limit_block to the value from the entry. + */ +BlockRefTableEntry * +BlockRefTableGetEntry(BlockRefTable *brtab, const RelFileLocator *rlocator, + ForkNumber forknum, BlockNumber *limit_block) +{ + BlockRefTableKey key; + BlockRefTableEntry *entry; + + Assert(limit_block != NULL); + + memcpy(&key.rlocator, rlocator, sizeof(RelFileLocator)); + key.forknum = forknum; + entry = blockreftable_lookup(brtab->hash, key); + + if (entry != NULL) + *limit_block = entry->limit_block; + + return entry; +} + +/* + * Get block numbers from a table entry. + * + * 'blocks' must point to enough space to hold at least 'nblocks' block + * numbers, and any block numbers we manage to get will be written there. + * The return value is the number of block numbers actually written. + * + * We do not return block numbers unless they are greater than or equal to + * start_blkno and strictly less than stop_blkno. + */ +int +BlockRefTableEntryGetBlocks(BlockRefTableEntry *entry, + BlockNumber start_blkno, + BlockNumber stop_blkno, + BlockNumber *blocks, + int nblocks) +{ + uint32 start_chunkno; + uint32 stop_chunkno; + uint32 chunkno; + int nresults = 0; + + Assert(entry != NULL); + + /* + * Figure out which chunks could potentially contain blocks of interest. + * + * We need to be careful about overflow here, because stop_blkno could be + * InvalidBlockNumber or something very close to it. + */ + start_chunkno = start_blkno / BLOCKS_PER_CHUNK; + stop_chunkno = stop_blkno / BLOCKS_PER_CHUNK; + if ((stop_blkno % BLOCKS_PER_CHUNK) != 0) + ++stop_chunkno; + if (stop_chunkno > entry->nchunks) + stop_chunkno = entry->nchunks; + + /* + * Loop over chunks. + */ + for (chunkno = start_chunkno; chunkno < stop_chunkno; ++chunkno) + { + uint16 chunk_usage = entry->chunk_usage[chunkno]; + BlockRefTableChunk chunk_data = entry->chunk_data[chunkno]; + unsigned start_offset = 0; + unsigned stop_offset = BLOCKS_PER_CHUNK; + + /* + * If the start and/or stop block number falls within this chunk, the + * whole chunk may not be of interest. Figure out which portion we + * care about, if it's not the whole thing. + */ + if (chunkno == start_chunkno) + start_offset = start_blkno % BLOCKS_PER_CHUNK; + if (chunkno == stop_chunkno) + stop_offset = stop_blkno % BLOCKS_PER_CHUNK; + + /* + * Handling differs depending on whether this is an array of offsets + * or a bitmap. + */ + if (chunk_usage == MAX_ENTRIES_PER_CHUNK) + { + unsigned i; + + /* It's a bitmap, so test every relevant bit. */ + for (i = start_offset; i < BLOCKS_PER_CHUNK; ++i) + { + uint16 w = chunk_data[i / BLOCKS_PER_ENTRY]; + + if ((w & (1 << (i % BLOCKS_PER_ENTRY))) != 0) + { + BlockNumber blkno = chunkno * BLOCKS_PER_CHUNK + i; + + blocks[nresults++] = blkno; + + /* Early exit if we run out of output space. */ + if (nresults == nblocks) + return nresults; + } + } + } + else + { + unsigned i; + + /* It's an array of offsets, so check each one. */ + for (i = 0; i < chunk_usage; ++i) + { + uint16 offset = chunk_data[i]; + + if (offset >= start_offset && offset < stop_offset) + { + BlockNumber blkno = chunkno * BLOCKS_PER_CHUNK + offset; + + blocks[nresults++] = blkno; + + /* Early exit if we run out of output space. */ + if (nresults == nblocks) + return nresults; + } + } + } + } + + return nresults; +} + +/* + * Serialize a block reference table to a file. + */ +void +WriteBlockRefTable(BlockRefTable *brtab, + io_callback_fn write_callback, + void *write_callback_arg) +{ + BlockRefTableSerializedEntry *sdata = NULL; + BlockRefTableBuffer buffer; + uint32 magic = BLOCKREFTABLE_MAGIC; + + /* Prepare buffer. */ + memset(&buffer, 0, sizeof(BlockRefTableBuffer)); + buffer.io_callback = write_callback; + buffer.io_callback_arg = write_callback_arg; + INIT_CRC32C(buffer.crc); + + /* Write magic number. */ + BlockRefTableWrite(&buffer, &magic, sizeof(uint32)); + + /* Write the entries, assuming there are some. */ + if (brtab->hash->members > 0) + { + unsigned i = 0; + blockreftable_iterator it; + BlockRefTableEntry *brtentry; + + /* Extract entries into serializable format and sort them. */ + sdata = + palloc(brtab->hash->members * sizeof(BlockRefTableSerializedEntry)); + blockreftable_start_iterate(brtab->hash, &it); + while ((brtentry = blockreftable_iterate(brtab->hash, &it)) != NULL) + { + BlockRefTableSerializedEntry *sentry = &sdata[i++]; + + sentry->rlocator = brtentry->key.rlocator; + sentry->forknum = brtentry->key.forknum; + sentry->limit_block = brtentry->limit_block; + sentry->nchunks = brtentry->nchunks; + + /* trim trailing zero entries */ + while (sentry->nchunks > 0 && + brtentry->chunk_usage[sentry->nchunks - 1] == 0) + sentry->nchunks--; + } + Assert(i == brtab->hash->members); + qsort(sdata, i, sizeof(BlockRefTableSerializedEntry), + BlockRefTableComparator); + + /* Loop over entries in sorted order and serialize each one. */ + for (i = 0; i < brtab->hash->members; ++i) + { + BlockRefTableSerializedEntry *sentry = &sdata[i]; + BlockRefTableEntry *brtentry; + BlockRefTableKey key; + unsigned j; + + /* Write the serialized entry itself. */ + BlockRefTableWrite(&buffer, sentry, + sizeof(BlockRefTableSerializedEntry)); + + /* Look up the original entry so we can access the chunks. */ + memcpy(&key.rlocator, &sentry->rlocator, sizeof(RelFileLocator)); + key.forknum = sentry->forknum; + brtentry = blockreftable_lookup(brtab->hash, key); + Assert(brtentry != NULL); + + /* Write the untruncated portion of the chunk length array. */ + if (sentry->nchunks != 0) + BlockRefTableWrite(&buffer, brtentry->chunk_usage, + sentry->nchunks * sizeof(uint16)); + + /* Write the contents of each chunk. */ + for (j = 0; j < brtentry->nchunks; ++j) + { + if (brtentry->chunk_usage[j] == 0) + continue; + BlockRefTableWrite(&buffer, brtentry->chunk_data[j], + brtentry->chunk_usage[j] * sizeof(uint16)); + } + } + } + + /* Write out appropriate terminator and CRC and flush buffer. */ + BlockRefTableFileTerminate(&buffer); +} + +/* + * Prepare to incrementally read a block reference table file. + * + * 'read_callback' is a function that can be called to read data from the + * underlying file (or other data source) into our internal buffer. + * + * 'read_callback_arg' is an opaque argument to be passed to read_callback. + * + * 'error_filename' is the filename that should be included in error messages + * if the file is found to be malformed. The value is not copied, so the + * caller should ensure that it remains valid until done with this + * BlockRefTableReader. + * + * 'error_callback' is a function to be called if the file is found to be + * malformed. This is not used for I/O errors, which must be handled internally + * by read_callback. + * + * 'error_callback_arg' is an opaque arguent to be passed to error_callback. + */ +BlockRefTableReader * +CreateBlockRefTableReader(io_callback_fn read_callback, + void *read_callback_arg, + char *error_filename, + report_error_fn error_callback, + void *error_callback_arg) +{ + BlockRefTableReader *reader; + uint32 magic; + + /* Initialize data structure. */ + reader = palloc0(sizeof(BlockRefTableReader)); + reader->buffer.io_callback = read_callback; + reader->buffer.io_callback_arg = read_callback_arg; + reader->error_filename = error_filename; + reader->error_callback = error_callback; + reader->error_callback_arg = error_callback_arg; + INIT_CRC32C(reader->buffer.crc); + + /* Verify magic number. */ + BlockRefTableRead(reader, &magic, sizeof(uint32)); + if (magic != BLOCKREFTABLE_MAGIC) + error_callback(error_callback_arg, + "file \"%s\" has wrong magic number: expected %u, found %u", + error_filename, + BLOCKREFTABLE_MAGIC, magic); + + return reader; +} + +/* + * Read next relation fork covered by this block reference table file. + * + * After calling this function, you must call BlockRefTableReaderGetBlocks + * until it returns 0 before calling it again. + */ +bool +BlockRefTableReaderNextRelation(BlockRefTableReader *reader, + RelFileLocator *rlocator, + ForkNumber *forknum, + BlockNumber *limit_block) +{ + BlockRefTableSerializedEntry sentry; + BlockRefTableSerializedEntry zentry = {0}; + + /* + * Sanity check: caller must read all blocks from all chunks before moving + * on to the next relation. + */ + Assert(reader->total_chunks == reader->consumed_chunks); + + /* Read serialized entry. */ + BlockRefTableRead(reader, &sentry, + sizeof(BlockRefTableSerializedEntry)); + + /* + * If we just read the sentinel entry indicating that we've reached the + * end, read and check the CRC. + */ + if (memcmp(&sentry, &zentry, sizeof(BlockRefTableSerializedEntry)) == 0) + { + pg_crc32c expected_crc; + pg_crc32c actual_crc; + + /* + * We want to know the CRC of the file excluding the 4-byte CRC + * itself, so copy the current value of the CRC accumulator before + * reading those bytes, and use the copy to finalize the calculation. + */ + expected_crc = reader->buffer.crc; + FIN_CRC32C(expected_crc); + + /* Now we can read the actual value. */ + BlockRefTableRead(reader, &actual_crc, sizeof(pg_crc32c)); + + /* Throw an error if there is a mismatch. */ + if (!EQ_CRC32C(expected_crc, actual_crc)) + reader->error_callback(reader->error_callback_arg, + "file \"%s\" has wrong checksum: expected %08X, found %08X", + reader->error_filename, expected_crc, actual_crc); + + return false; + } + + /* Read chunk size array. */ + if (reader->chunk_size != NULL) + pfree(reader->chunk_size); + reader->chunk_size = palloc(sentry.nchunks * sizeof(uint16)); + BlockRefTableRead(reader, reader->chunk_size, + sentry.nchunks * sizeof(uint16)); + + /* Set up for chunk scan. */ + reader->total_chunks = sentry.nchunks; + reader->consumed_chunks = 0; + + /* Return data to caller. */ + memcpy(rlocator, &sentry.rlocator, sizeof(RelFileLocator)); + *forknum = sentry.forknum; + *limit_block = sentry.limit_block; + return true; +} + +/* + * Get modified blocks associated with the relation fork returned by + * the most recent call to BlockRefTableReaderNextRelation. + * + * On return, block numbers will be written into the 'blocks' array, whose + * length should be passed via 'nblocks'. The return value is the number of + * entries actually written into the 'blocks' array, which may be less than + * 'nblocks' if we run out of modified blocks in the relation fork before + * we run out of room in the array. + */ +unsigned +BlockRefTableReaderGetBlocks(BlockRefTableReader *reader, + BlockNumber *blocks, + int nblocks) +{ + unsigned blocks_found = 0; + + /* Must provide space for at least one block number to be returned. */ + Assert(nblocks > 0); + + /* Loop collecting blocks to return to caller. */ + for (;;) + { + uint16 next_chunk_size; + + /* + * If we've read at least one chunk, maybe it contains some block + * numbers that could satisfy caller's request. + */ + if (reader->consumed_chunks > 0) + { + uint32 chunkno = reader->consumed_chunks - 1; + uint16 chunk_size = reader->chunk_size[chunkno]; + + if (chunk_size == MAX_ENTRIES_PER_CHUNK) + { + /* Bitmap format, so search for bits that are set. */ + while (reader->chunk_position < BLOCKS_PER_CHUNK && + blocks_found < nblocks) + { + uint16 chunkoffset = reader->chunk_position; + uint16 w; + + w = reader->chunk_data[chunkoffset / BLOCKS_PER_ENTRY]; + if ((w & (1u << (chunkoffset % BLOCKS_PER_ENTRY))) != 0) + blocks[blocks_found++] = + chunkno * BLOCKS_PER_CHUNK + chunkoffset; + ++reader->chunk_position; + } + } + else + { + /* Not in bitmap format, so each entry is a 2-byte offset. */ + while (reader->chunk_position < chunk_size && + blocks_found < nblocks) + { + blocks[blocks_found++] = chunkno * BLOCKS_PER_CHUNK + + reader->chunk_data[reader->chunk_position]; + ++reader->chunk_position; + } + } + } + + /* We found enough blocks, so we're done. */ + if (blocks_found >= nblocks) + break; + + /* + * We didn't find enough blocks, so we must need the next chunk. If + * there are none left, though, then we're done anyway. + */ + if (reader->consumed_chunks == reader->total_chunks) + break; + + /* + * Read data for next chunk and reset scan position to beginning of + * chunk. Note that the next chunk might be empty, in which case we + * consume the chunk without actually consuming any bytes from the + * underlying file. + */ + next_chunk_size = reader->chunk_size[reader->consumed_chunks]; + if (next_chunk_size > 0) + BlockRefTableRead(reader, reader->chunk_data, + next_chunk_size * sizeof(uint16)); + ++reader->consumed_chunks; + reader->chunk_position = 0; + } + + return blocks_found; +} + +/* + * Release memory used while reading a block reference table from a file. + */ +void +DestroyBlockRefTableReader(BlockRefTableReader *reader) +{ + if (reader->chunk_size != NULL) + { + pfree(reader->chunk_size); + reader->chunk_size = NULL; + } + pfree(reader); +} + +/* + * Prepare to write a block reference table file incrementally. + * + * Caller must be able to supply BlockRefTableEntry objects sorted in the + * appropriate order. + */ +BlockRefTableWriter * +CreateBlockRefTableWriter(io_callback_fn write_callback, + void *write_callback_arg) +{ + BlockRefTableWriter *writer; + uint32 magic = BLOCKREFTABLE_MAGIC; + + /* Prepare buffer and CRC check and save callbacks. */ + writer = palloc0(sizeof(BlockRefTableWriter)); + writer->buffer.io_callback = write_callback; + writer->buffer.io_callback_arg = write_callback_arg; + INIT_CRC32C(writer->buffer.crc); + + /* Write magic number. */ + BlockRefTableWrite(&writer->buffer, &magic, sizeof(uint32)); + + return writer; +} + +/* + * Append one entry to a block reference table file. + * + * Note that entries must be written in the proper order, that is, sorted by + * tablespace, then database, then relfilenumber, then fork number. Caller + * is responsible for supplying data in the correct order. If that seems hard, + * use an in-memory BlockRefTable instead. + */ +void +BlockRefTableWriteEntry(BlockRefTableWriter *writer, BlockRefTableEntry *entry) +{ + BlockRefTableSerializedEntry sentry; + unsigned j; + + /* Convert to serialized entry format. */ + sentry.rlocator = entry->key.rlocator; + sentry.forknum = entry->key.forknum; + sentry.limit_block = entry->limit_block; + sentry.nchunks = entry->nchunks; + + /* Trim trailing zero entries. */ + while (sentry.nchunks > 0 && entry->chunk_usage[sentry.nchunks - 1] == 0) + sentry.nchunks--; + + /* Write the serialized entry itself. */ + BlockRefTableWrite(&writer->buffer, &sentry, + sizeof(BlockRefTableSerializedEntry)); + + /* Write the untruncated portion of the chunk length array. */ + if (sentry.nchunks != 0) + BlockRefTableWrite(&writer->buffer, entry->chunk_usage, + sentry.nchunks * sizeof(uint16)); + + /* Write the contents of each chunk. */ + for (j = 0; j < entry->nchunks; ++j) + { + if (entry->chunk_usage[j] == 0) + continue; + BlockRefTableWrite(&writer->buffer, entry->chunk_data[j], + entry->chunk_usage[j] * sizeof(uint16)); + } +} + +/* + * Finalize an incremental write of a block reference table file. + */ +void +DestroyBlockRefTableWriter(BlockRefTableWriter *writer) +{ + BlockRefTableFileTerminate(&writer->buffer); + pfree(writer); +} + +/* + * Allocate a standalone BlockRefTableEntry. + * + * When we're manipulating a full in-memory BlockRefTable, the entries are + * part of the hash table and are allocated by simplehash. This routine is + * used by callers that want to write out a BlockRefTable to a file without + * needing to store the whole thing in memory at once. + * + * Entries allocated by this function can be manipulated using the functions + * BlockRefTableEntrySetLimitBlock and BlockRefTableEntryMarkBlockModified + * and then written using BlockRefTableWriteEntry and freed using + * BlockRefTableFreeEntry. + */ +BlockRefTableEntry * +CreateBlockRefTableEntry(RelFileLocator rlocator, ForkNumber forknum) +{ + BlockRefTableEntry *entry = palloc0(sizeof(BlockRefTableEntry)); + + memcpy(&entry->key.rlocator, &rlocator, sizeof(RelFileLocator)); + entry->key.forknum = forknum; + entry->limit_block = InvalidBlockNumber; + + return entry; +} + +/* + * Update a BlockRefTableEntry with a new value for the "limit block" and + * forget any equal-or-higher-numbered modified blocks. + * + * The "limit block" is the shortest known length of the relation within the + * range of WAL records covered by this block reference table. + */ +void +BlockRefTableEntrySetLimitBlock(BlockRefTableEntry *entry, + BlockNumber limit_block) +{ + unsigned chunkno; + unsigned limit_chunkno; + unsigned limit_chunkoffset; + BlockRefTableChunk limit_chunk; + + /* If we already have an equal or lower limit block, do nothing. */ + if (limit_block >= entry->limit_block) + return; + + /* Record the new limit block value. */ + entry->limit_block = limit_block; + + /* + * Figure out which chunk would store the state of the new limit block, + * and which offset within that chunk. + */ + limit_chunkno = limit_block / BLOCKS_PER_CHUNK; + limit_chunkoffset = limit_block % BLOCKS_PER_CHUNK; + + /* + * If the number of chunks is not large enough for any blocks with equal + * or higher block numbers to exist, then there is nothing further to do. + */ + if (limit_chunkno >= entry->nchunks) + return; + + /* Discard entire contents of any higher-numbered chunks. */ + for (chunkno = limit_chunkno + 1; chunkno < entry->nchunks; ++chunkno) + entry->chunk_usage[chunkno] = 0; + + /* + * Next, we need to discard any offsets within the chunk that would + * contain the limit_block. We must handle this differenly depending on + * whether the chunk that would contain limit_block is a bitmap or an + * array of offsets. + */ + limit_chunk = entry->chunk_data[limit_chunkno]; + if (entry->chunk_usage[limit_chunkno] == MAX_ENTRIES_PER_CHUNK) + { + unsigned chunkoffset; + + /* It's a bitmap. Unset bits. */ + for (chunkoffset = limit_chunkoffset; chunkoffset < BLOCKS_PER_CHUNK; + ++chunkoffset) + limit_chunk[chunkoffset / BLOCKS_PER_ENTRY] &= + ~(1 << (chunkoffset % BLOCKS_PER_ENTRY)); + } + else + { + unsigned i, + j = 0; + + /* It's an offset array. Filter out large offsets. */ + for (i = 0; i < entry->chunk_usage[limit_chunkno]; ++i) + { + Assert(j <= i); + if (limit_chunk[i] < limit_chunkoffset) + limit_chunk[j++] = limit_chunk[i]; + } + Assert(j <= entry->chunk_usage[limit_chunkno]); + entry->chunk_usage[limit_chunkno] = j; + } +} + +/* + * Mark a block in a given BlkRefTableEntry as known to have been modified. + */ +void +BlockRefTableEntryMarkBlockModified(BlockRefTableEntry *entry, + ForkNumber forknum, + BlockNumber blknum) +{ + unsigned chunkno; + unsigned chunkoffset; + unsigned i; + + /* + * Which chunk should store the state of this block? And what is the + * offset of this block relative to the start of that chunk? + */ + chunkno = blknum / BLOCKS_PER_CHUNK; + chunkoffset = blknum % BLOCKS_PER_CHUNK; + + /* + * If 'nchunks' isn't big enough for us to be able to represent the state + * of this block, we need to enlarge our arrays. + */ + if (chunkno >= entry->nchunks) + { + unsigned max_chunks; + unsigned extra_chunks; + + /* + * New array size is a power of 2, at least 16, big enough so that + * chunkno will be a valid array index. + */ + max_chunks = Max(16, entry->nchunks); + while (max_chunks < chunkno + 1) + chunkno *= 2; + Assert(max_chunks > chunkno); + extra_chunks = max_chunks - entry->nchunks; + + if (entry->nchunks == 0) + { + entry->chunk_size = palloc0(sizeof(uint16) * max_chunks); + entry->chunk_usage = palloc0(sizeof(uint16) * max_chunks); + entry->chunk_data = + palloc0(sizeof(BlockRefTableChunk) * max_chunks); + } + else + { + entry->chunk_size = repalloc(entry->chunk_size, + sizeof(uint16) * max_chunks); + memset(&entry->chunk_size[entry->nchunks], 0, + extra_chunks * sizeof(uint16)); + entry->chunk_usage = repalloc(entry->chunk_usage, + sizeof(uint16) * max_chunks); + memset(&entry->chunk_usage[entry->nchunks], 0, + extra_chunks * sizeof(uint16)); + entry->chunk_data = repalloc(entry->chunk_data, + sizeof(BlockRefTableChunk) * max_chunks); + memset(&entry->chunk_data[entry->nchunks], 0, + extra_chunks * sizeof(BlockRefTableChunk)); + } + entry->nchunks = max_chunks; + } + + /* + * If the chunk that covers this block number doesn't exist yet, create it + * as an array and add the appropriate offset to it. We make it pretty + * small initially, because there might only be 1 or a few block + * references in this chunk and we don't want to use up too much memory. + */ + if (entry->chunk_size[chunkno] == 0) + { + entry->chunk_data[chunkno] = + palloc(sizeof(uint16) * INITIAL_ENTRIES_PER_CHUNK); + entry->chunk_size[chunkno] = INITIAL_ENTRIES_PER_CHUNK; + entry->chunk_data[chunkno][0] = chunkoffset; + entry->chunk_usage[chunkno] = 1; + return; + } + + /* + * If the number of entries in this chunk is already maximum, it must be a + * bitmap. Just set the appropriate bit. + */ + if (entry->chunk_usage[chunkno] == MAX_ENTRIES_PER_CHUNK) + { + BlockRefTableChunk chunk = entry->chunk_data[chunkno]; + + chunk[chunkoffset / BLOCKS_PER_ENTRY] |= + 1 << (chunkoffset % BLOCKS_PER_ENTRY); + return; + } + + /* + * There is an existing chunk and it's in array format. Let's find out + * whether it already has an entry for this block. If so, we do not need + * to do anything. + */ + for (i = 0; i < entry->chunk_usage[chunkno]; ++i) + { + if (entry->chunk_data[chunkno][i] == chunkoffset) + return; + } + + /* + * If the number of entries currently used is one less than the maximum, + * it's time to convert to bitmap format. + */ + if (entry->chunk_usage[chunkno] == MAX_ENTRIES_PER_CHUNK - 1) + { + BlockRefTableChunk newchunk; + unsigned j; + + /* Allocate a new chunk. */ + newchunk = palloc0(MAX_ENTRIES_PER_CHUNK * sizeof(uint16)); + + /* Set the bit for each existing entry. */ + for (j = 0; j < entry->chunk_usage[chunkno]; ++j) + { + unsigned coff = entry->chunk_data[chunkno][j]; + + newchunk[coff / BLOCKS_PER_ENTRY] |= + 1 << (coff % BLOCKS_PER_ENTRY); + } + + /* Set the bit for the new entry. */ + newchunk[chunkoffset / BLOCKS_PER_ENTRY] |= + 1 << (chunkoffset % BLOCKS_PER_ENTRY); + + /* Swap the new chunk into place and update metadata. */ + pfree(entry->chunk_data[chunkno]); + entry->chunk_data[chunkno] = newchunk; + entry->chunk_size[chunkno] = MAX_ENTRIES_PER_CHUNK; + entry->chunk_usage[chunkno] = MAX_ENTRIES_PER_CHUNK; + return; + } + + /* + * OK, we currently have an array, and we don't need to convert to a + * bitmap, but we do need to add a new element. If there's not enough + * room, we'll have to expand the array. + */ + if (entry->chunk_usage[chunkno] == entry->chunk_size[chunkno]) + { + unsigned newsize = entry->chunk_size[chunkno] * 2; + + Assert(newsize <= MAX_ENTRIES_PER_CHUNK); + entry->chunk_data[chunkno] = repalloc(entry->chunk_data[chunkno], + newsize * sizeof(uint16)); + entry->chunk_size[chunkno] = newsize; + } + + /* Now we can add the new entry. */ + entry->chunk_data[chunkno][entry->chunk_usage[chunkno]] = + chunkoffset; + entry->chunk_usage[chunkno]++; +} + +/* + * Release memory for a BlockRefTablEntry that was created by + * CreateBlockRefTableEntry. + */ +void +BlockRefTableFreeEntry(BlockRefTableEntry *entry) +{ + if (entry->chunk_size != NULL) + { + pfree(entry->chunk_size); + entry->chunk_size = NULL; + } + + if (entry->chunk_usage != NULL) + { + pfree(entry->chunk_usage); + entry->chunk_usage = NULL; + } + + if (entry->chunk_data != NULL) + { + pfree(entry->chunk_data); + entry->chunk_data = NULL; + } + + pfree(entry); +} + +/* + * Comparator for BlockRefTableSerializedEntry objects. + * + * We make the tablespace OID the first column of the sort key to match + * the on-disk tree structure. + */ +static int +BlockRefTableComparator(const void *a, const void *b) +{ + const BlockRefTableSerializedEntry *sa = a; + const BlockRefTableSerializedEntry *sb = b; + + if (sa->rlocator.spcOid > sb->rlocator.spcOid) + return 1; + if (sa->rlocator.spcOid < sb->rlocator.spcOid) + return -1; + + if (sa->rlocator.dbOid > sb->rlocator.dbOid) + return 1; + if (sa->rlocator.dbOid < sb->rlocator.dbOid) + return -1; + + if (sa->rlocator.relNumber > sb->rlocator.relNumber) + return 1; + if (sa->rlocator.relNumber < sb->rlocator.relNumber) + return -1; + + if (sa->forknum > sb->forknum) + return 1; + if (sa->forknum < sb->forknum) + return -1; + + return 0; +} + +/* + * Flush any buffered data out of a BlockRefTableBuffer. + */ +static void +BlockRefTableFlush(BlockRefTableBuffer *buffer) +{ + buffer->io_callback(buffer->io_callback_arg, buffer->data, buffer->used); + buffer->used = 0; +} + +/* + * Read data from a BlockRefTableBuffer, and update the running CRC + * calculation for the returned data (but not any data that we may have + * buffered but not yet actually returned). + */ +static void +BlockRefTableRead(BlockRefTableReader *reader, void *data, int length) +{ + BlockRefTableBuffer *buffer = &reader->buffer; + + /* Loop until read is fully satisfied. */ + while (length > 0) + { + if (buffer->cursor < buffer->used) + { + /* + * If any buffered data is available, use that to satisfy as much + * of the request as possible. + */ + int bytes_to_copy = Min(length, buffer->used - buffer->cursor); + + memcpy(data, &buffer->data[buffer->cursor], bytes_to_copy); + COMP_CRC32C(buffer->crc, &buffer->data[buffer->cursor], + bytes_to_copy); + buffer->cursor += bytes_to_copy; + data = ((char *) data) + bytes_to_copy; + length -= bytes_to_copy; + } + else if (length >= BUFSIZE) + { + /* + * If the request length is long, read directly into caller's + * buffer. + */ + int bytes_read; + + bytes_read = buffer->io_callback(buffer->io_callback_arg, + data, length); + COMP_CRC32C(buffer->crc, data, bytes_read); + data = ((char *) data) + bytes_read; + length -= bytes_read; + + /* If we didn't get anything, that's bad. */ + if (bytes_read == 0) + reader->error_callback(reader->error_callback_arg, + "file \"%s\" ends unexpectedly", + reader->error_filename); + } + else + { + /* + * Refill our buffer. + */ + buffer->used = buffer->io_callback(buffer->io_callback_arg, + buffer->data, BUFSIZE); + buffer->cursor = 0; + + /* If we didn't get anything, that's bad. */ + if (buffer->used == 0) + reader->error_callback(reader->error_callback_arg, + "file \"%s\" ends unexpectedly", + reader->error_filename); + } + } +} + +/* + * Supply data to a BlockRefTableBuffer for write to the underlying File, + * and update the running CRC calculation for that data. + */ +static void +BlockRefTableWrite(BlockRefTableBuffer *buffer, void *data, int length) +{ + /* Update running CRC calculation. */ + COMP_CRC32C(buffer->crc, data, length); + + /* If the new data can't fit into the buffer, flush the buffer. */ + if (buffer->used + length > BUFSIZE) + { + buffer->io_callback(buffer->io_callback_arg, buffer->data, + buffer->used); + buffer->used = 0; + } + + /* If the new data would fill the buffer, or more, write it directly. */ + if (length >= BUFSIZE) + { + buffer->io_callback(buffer->io_callback_arg, data, length); + return; + } + + /* Otherwise, copy the new data into the buffer. */ + memcpy(&buffer->data[buffer->used], data, length); + buffer->used += length; + Assert(buffer->used <= BUFSIZE); +} + +/* + * Generate the sentinel and CRC required at the end of a block reference + * table file and flush them out of our internal buffer. + */ +static void +BlockRefTableFileTerminate(BlockRefTableBuffer *buffer) +{ + BlockRefTableSerializedEntry zentry = {0}; + pg_crc32c crc; + + /* Write a sentinel indicating that there are no more entries. */ + BlockRefTableWrite(buffer, &zentry, + sizeof(BlockRefTableSerializedEntry)); + + /* + * Writing the checksum will perturb the ongoing checksum calculation, so + * copy the state first and finalize the computation using the copy. + */ + crc = buffer->crc; + FIN_CRC32C(crc); + BlockRefTableWrite(buffer, &crc, sizeof(pg_crc32c)); + + /* Flush any leftover data out of our buffer. */ + BlockRefTableFlush(buffer); +} diff --git a/src/common/meson.build b/src/common/meson.build index fcc0c4fe8d..6e51257b1c 100644 --- a/src/common/meson.build +++ b/src/common/meson.build @@ -4,6 +4,7 @@ common_sources = files( 'archive.c', 'base64.c', 'binaryheap.c', + 'blkreftable.c', 'checksum_helper.c', 'compression.c', 'controldata_utils.c', diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 48ca852381..fed5d790cc 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -206,6 +206,7 @@ extern int XLogFileOpen(XLogSegNo segno, TimeLineID tli); extern void CheckXLogRemoved(XLogSegNo segno, TimeLineID tli); extern XLogSegNo XLogGetLastRemovedSegno(void); +extern XLogSegNo XLogGetOldestSegno(TimeLineID tli); extern void XLogSetAsyncXactLSN(XLogRecPtr asyncXactLSN); extern void XLogSetReplicationSlotMinimumLSN(XLogRecPtr lsn); diff --git a/src/include/access/xlogbackup.h b/src/include/access/xlogbackup.h index 1611358137..90e04cad56 100644 --- a/src/include/access/xlogbackup.h +++ b/src/include/access/xlogbackup.h @@ -28,6 +28,8 @@ typedef struct BackupState XLogRecPtr checkpointloc; /* last checkpoint location */ pg_time_t starttime; /* backup start time */ bool started_in_recovery; /* backup started in recovery? */ + XLogRecPtr istartpoint; /* incremental based on backup at this LSN */ + TimeLineID istarttli; /* incremental based on backup on this TLI */ /* Fields saved at the end of backup */ XLogRecPtr stoppoint; /* backup stop WAL location */ diff --git a/src/include/backup/basebackup.h b/src/include/backup/basebackup.h index 1432d9c206..345bd22534 100644 --- a/src/include/backup/basebackup.h +++ b/src/include/backup/basebackup.h @@ -34,6 +34,9 @@ typedef struct int64 size; /* total size as sent; -1 if not known */ } tablespaceinfo; -extern void SendBaseBackup(BaseBackupCmd *cmd); +struct IncrementalBackupInfo; + +extern void SendBaseBackup(BaseBackupCmd *cmd, + struct IncrementalBackupInfo *ib); #endif /* _BASEBACKUP_H */ diff --git a/src/include/backup/basebackup_incremental.h b/src/include/backup/basebackup_incremental.h new file mode 100644 index 0000000000..c300235a2f --- /dev/null +++ b/src/include/backup/basebackup_incremental.h @@ -0,0 +1,56 @@ +/*------------------------------------------------------------------------- + * + * basebackup_incremental.h + * API for incremental backup support + * + * Portions Copyright (c) 2010-2022, PostgreSQL Global Development Group + * + * src/include/backup/basebackup_incremental.h + * + *------------------------------------------------------------------------- + */ +#ifndef BASEBACKUP_INCREMENTAL_H +#define BASEBACKUP_INCREMENTAL_H + +#include "access/xlogbackup.h" +#include "common/relpath.h" +#include "storage/block.h" +#include "utils/palloc.h" + +#define INCREMENTAL_MAGIC 0xd3ae1f0d + +typedef enum +{ + BACK_UP_FILE_FULLY, + BACK_UP_FILE_INCREMENTALLY, + DO_NOT_BACK_UP_FILE +} FileBackupMethod; + +struct IncrementalBackupInfo; +typedef struct IncrementalBackupInfo IncrementalBackupInfo; + +extern IncrementalBackupInfo *CreateIncrementalBackupInfo(MemoryContext); + +extern void AppendIncrementalManifestData(IncrementalBackupInfo *ib, + const char *data, + int len); +extern void FinalizeIncrementalManifest(IncrementalBackupInfo *ib); + +extern void PrepareForIncrementalBackup(IncrementalBackupInfo *ib, + BackupState *backup_state); + +extern char *GetIncrementalFilePath(Oid dboid, Oid spcoid, + RelFileNumber relfilenumber, + ForkNumber forknum, unsigned segno); +extern FileBackupMethod GetFileBackupMethod(IncrementalBackupInfo *ib, + char *path, + Oid dboid, Oid spcoid, + RelFileNumber relfilenumber, + ForkNumber forknum, + unsigned segno, size_t size, + unsigned *num_blocks_required, + BlockNumber *relative_block_numbers, + unsigned *truncation_block_length); +extern size_t GetIncrementalFileSize(unsigned num_blocks_required); + +#endif diff --git a/src/include/backup/walsummary.h b/src/include/backup/walsummary.h new file mode 100644 index 0000000000..d086e64019 --- /dev/null +++ b/src/include/backup/walsummary.h @@ -0,0 +1,49 @@ +/*------------------------------------------------------------------------- + * + * walsummary.h + * WAL summary management + * + * Portions Copyright (c) 2010-2022, PostgreSQL Global Development Group + * + * src/include/backup/walsummary.h + * + *------------------------------------------------------------------------- + */ +#ifndef WALSUMMARY_H +#define WALSUMMARY_H + +#include <time.h> + +#include "access/xlogdefs.h" +#include "nodes/pg_list.h" +#include "storage/fd.h" + +typedef struct WalSummaryIO +{ + File file; + off_t filepos; +} WalSummaryIO; + +typedef struct WalSummaryFile +{ + XLogRecPtr start_lsn; + XLogRecPtr end_lsn; + TimeLineID tli; +} WalSummaryFile; + +extern List *GetWalSummaries(TimeLineID tli, XLogRecPtr start_lsn, + XLogRecPtr end_lsn); +extern List *FilterWalSummaries(List *wslist, TimeLineID tli, + XLogRecPtr start_lsn, XLogRecPtr end_lsn); +extern bool WalSummariesAreComplete(List *wslist, + XLogRecPtr start_lsn, XLogRecPtr end_lsn, + XLogRecPtr *missing_lsn); +extern File OpenWalSummaryFile(WalSummaryFile *ws, bool missing_ok); +extern void RemoveWalSummaryIfOlderThan(WalSummaryFile *ws, + time_t cutoff_time); + +extern int ReadWalSummary(void *wal_summary_io, void *data, int length); +extern int WriteWalSummary(void *wal_summary_io, void *data, int length); +extern void ReportWalSummaryError(void *callback_arg, char *fmt,...); + +#endif /* WALSUMMARY_H */ diff --git a/src/include/catalog/pg_proc.dat b/src/include/catalog/pg_proc.dat index f0b7b9cbd8..f68e6d4987 100644 --- a/src/include/catalog/pg_proc.dat +++ b/src/include/catalog/pg_proc.dat @@ -12062,4 +12062,23 @@ proname => 'any_value_transfn', prorettype => 'anyelement', proargtypes => 'anyelement anyelement', prosrc => 'any_value_transfn' }, +{ oid => '8436', + descr => 'list of available WAL summary files', + proname => 'pg_available_wal_summaries', prorows => '100', + proretset => 't', provolatile => 'v', proparallel => 's', + prorettype => 'record', proargtypes => '', + proallargtypes => '{int8,pg_lsn,pg_lsn}', + proargmodes => '{o,o,o}', + proargnames => '{tli,start_lsn,end_lsn}', + prosrc => 'pg_available_wal_summaries' }, +{ oid => '8437', + descr => 'contents of a WAL sumamry file', + proname => 'pg_wal_summary_contents', prorows => '100', + proretset => 't', provolatile => 'v', proparallel => 's', + prorettype => 'record', proargtypes => 'int8 pg_lsn pg_lsn', + proallargtypes => '{int8,pg_lsn,pg_lsn,oid,oid,oid,int2,int8,bool}', + proargmodes => '{i,i,i,o,o,o,o,o,o}', + proargnames => '{tli,start_lsn,end_lsn,relfilenode,reltablespace,reldatabase,relforknumber,relblocknumber,is_limit_block}', + prosrc => 'pg_wal_summary_contents' }, + ] diff --git a/src/include/common/blkreftable.h b/src/include/common/blkreftable.h new file mode 100644 index 0000000000..22d9883dc5 --- /dev/null +++ b/src/include/common/blkreftable.h @@ -0,0 +1,120 @@ +/*------------------------------------------------------------------------- + * + * blkreftable.h + * Block reference tables. + * + * A block reference table is used to keep track of which blocks have + * been modified by WAL records within a certain LSN range. + * + * For each relation fork, there is a "limit block number". All existing + * blocks greater than or equal to the limit block number must be + * considered modified; for those less than the limit block number, + * we maintain a bitmap. When a relation fork is created or dropped, + * the limit block number should be set to 0. When it's truncated, + * the limit block number should be set to the length in blocks to + * which it was truncated. + * + * Portions Copyright (c) 2010-2022, PostgreSQL Global Development Group + * + * src/include/common/blkreftable.h + * + *------------------------------------------------------------------------- + */ +#ifndef BLKREFTABLE_H +#define BLKREFTABLE_H + +#include "storage/block.h" +#include "storage/relfilelocator.h" + +/* Magic number for serialization file format. */ +#define BLOCKREFTABLE_MAGIC 0x652b137b + +struct BlockRefTable; +struct BlockRefTableEntry; +struct BlockRefTableReader; +struct BlockRefTableWriter; +typedef struct BlockRefTable BlockRefTable; +typedef struct BlockRefTableEntry BlockRefTableEntry; +typedef struct BlockRefTableReader BlockRefTableReader; +typedef struct BlockRefTableWriter BlockRefTableWriter; + +/* + * The return value of io_callback_fn should be the number of bytes read + * or written. If an error occurs, the functions should report it and + * not return. When used as a write callback, short writes should be retried + * or treated as errors, so that if the callback returns, the return value + * is always the request length. + * + * report_error_fn should not return. + */ +typedef int (*io_callback_fn) (void *callback_arg, void *data, int length); +typedef void (*report_error_fn) (void *calblack_arg, char *msg,...); + + +/* + * Functions for manipulating an entire in-memory block reference table. + */ +extern BlockRefTable *CreateEmptyBlockRefTable(void); +extern void BlockRefTableSetLimitBlock(BlockRefTable *brtab, + const RelFileLocator *rlocator, + ForkNumber forknum, + BlockNumber limit_block); +extern void BlockRefTableMarkBlockModified(BlockRefTable *brtab, + const RelFileLocator *rlocator, + ForkNumber forknum, + BlockNumber blknum); +extern void WriteBlockRefTable(BlockRefTable *brtab, + io_callback_fn write_callback, + void *write_callback_arg); + +extern BlockRefTableEntry *BlockRefTableGetEntry(BlockRefTable *brtab, + const RelFileLocator *rlocator, + ForkNumber forknum, + BlockNumber *limit_block); +extern int BlockRefTableEntryGetBlocks(BlockRefTableEntry *entry, + BlockNumber start_blkno, + BlockNumber stop_blkno, + BlockNumber *blocks, + int nblocks); + +/* + * Functions for reading a block reference table incrementally from disk. + */ +extern BlockRefTableReader *CreateBlockRefTableReader(io_callback_fn read_callback, + void *read_callback_arg, + char *error_filename, + report_error_fn error_callback, + void *error_callback_arg); +extern bool BlockRefTableReaderNextRelation(BlockRefTableReader *reader, + RelFileLocator *rlocator, + ForkNumber *forknum, + BlockNumber *limit_block); +extern unsigned BlockRefTableReaderGetBlocks(BlockRefTableReader *reader, + BlockNumber *blocks, + int nblocks); +extern void DestroyBlockRefTableReader(BlockRefTableReader *reader); + +/* + * Functions for writing a block reference table incrementally to disk. + * + * Note that entries must be written in the proper order, that is, sorted by + * database, then tablespace, then relfilenumber, then fork number. Caller + * is responsible for supplying data in the correct order. If that seems hard, + * use an in-memory BlockRefTable instead. + */ +extern BlockRefTableWriter *CreateBlockRefTableWriter(io_callback_fn write_callback, + void *write_callback_arg); +extern void BlockRefTableWriteEntry(BlockRefTableWriter *writer, + BlockRefTableEntry *entry); +extern void DestroyBlockRefTableWriter(BlockRefTableWriter *writer); + +extern BlockRefTableEntry *CreateBlockRefTableEntry(RelFileLocator rlocator, + ForkNumber forknum); +extern void BlockRefTableEntrySetLimitBlock(BlockRefTableEntry *entry, + BlockNumber limit_block); +extern void BlockRefTableEntryMarkBlockModified(BlockRefTableEntry *entry, + ForkNumber forknum, + BlockNumber blknum); +extern void BlockRefTableFreeEntry(BlockRefTableEntry *entry); + +#endif /* BLKREFTABLE_H */ diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 14bd574fc2..898adccb25 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -338,6 +338,7 @@ typedef enum BackendType B_STARTUP, B_WAL_RECEIVER, B_WAL_SENDER, + B_WAL_SUMMARIZER, B_WAL_WRITER, } BackendType; @@ -443,6 +444,7 @@ typedef enum CheckpointerProcess, WalWriterProcess, WalReceiverProcess, + WalSummarizerProcess, NUM_AUXPROCTYPES /* Must be last! */ } AuxProcType; @@ -455,6 +457,7 @@ extern PGDLLIMPORT AuxProcType MyAuxProcType; #define AmCheckpointerProcess() (MyAuxProcType == CheckpointerProcess) #define AmWalWriterProcess() (MyAuxProcType == WalWriterProcess) #define AmWalReceiverProcess() (MyAuxProcType == WalReceiverProcess) +#define AmWalSummarizerProcess() (MyAuxProcType == WalSummarizerProcess) /***************************************************************************** diff --git a/src/include/nodes/replnodes.h b/src/include/nodes/replnodes.h index 4321ba8f86..856491eecd 100644 --- a/src/include/nodes/replnodes.h +++ b/src/include/nodes/replnodes.h @@ -108,4 +108,13 @@ typedef struct TimeLineHistoryCmd TimeLineID timeline; } TimeLineHistoryCmd; +/* ---------------------- + * UPLOAD_MANIFEST command + * ---------------------- + */ +typedef struct UploadManifestCmd +{ + NodeTag type; +} UploadManifestCmd; + #endif /* REPLNODES_H */ diff --git a/src/include/postmaster/walsummarizer.h b/src/include/postmaster/walsummarizer.h new file mode 100644 index 0000000000..7584cb69a7 --- /dev/null +++ b/src/include/postmaster/walsummarizer.h @@ -0,0 +1,31 @@ +/*------------------------------------------------------------------------- + * + * walsummarizer.h + * + * Header file for background WAL summarization process. + * + * Portions Copyright (c) 1996-2022, PostgreSQL Global Development Group + * + * IDENTIFICATION + * src/include/postmaster/walsummarizer.h + * + *------------------------------------------------------------------------- + */ +#ifndef WALSUMMARIZER_H +#define WALSUMMARIZER_H + +#include "access/xlogdefs.h" + +extern int wal_summarize_mb; +extern int wal_summarize_keep_time; + +extern Size WalSummarizerShmemSize(void); +extern void WalSummarizerShmemInit(void); +extern void WalSummarizerMain(void) pg_attribute_noreturn(); + +extern XLogRecPtr GetOldestUnsummarizedLSN(TimeLineID *tli, + bool *lsn_is_exact); +extern void SetWalSummarizerLatch(void); +extern XLogRecPtr WaitForWalSummarization(XLogRecPtr lsn, long timeout); + +#endif diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index ef74f32693..ee55008082 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -417,11 +417,12 @@ extern PGDLLIMPORT PGPROC *PreparedXactProcs; * We set aside some extra PGPROC structures for auxiliary processes, * ie things that aren't full-fledged backends but need shmem access. * - * Background writer, checkpointer, WAL writer and archiver run during normal - * operation. Startup process and WAL receiver also consume 2 slots, but WAL - * writer is launched only after startup has exited, so we only need 5 slots. + * Background writer, checkpointer, WAL writer, WAL summarizer, and archiver + * run during normal operation. Startup process and WAL receiver also consume + * 2 slots, but WAL writer is launched only after startup has exited, so we + * only need 6 slots. */ -#define NUM_AUXILIARY_PROCS 5 +#define NUM_AUXILIARY_PROCS 6 /* configurable options */ extern PGDLLIMPORT int DeadlockTimeout; diff --git a/src/include/utils/guc_tables.h b/src/include/utils/guc_tables.h index d5a0880678..7d3bc0f671 100644 --- a/src/include/utils/guc_tables.h +++ b/src/include/utils/guc_tables.h @@ -72,6 +72,7 @@ enum config_group WAL_RECOVERY, WAL_ARCHIVE_RECOVERY, WAL_RECOVERY_TARGET, + WAL_SUMMARIZATION, REPLICATION_SENDING, REPLICATION_PRIMARY, REPLICATION_STANDBY, diff --git a/src/test/recovery/t/001_stream_rep.pl b/src/test/recovery/t/001_stream_rep.pl index 0c72ba0944..353db33a9f 100644 --- a/src/test/recovery/t/001_stream_rep.pl +++ b/src/test/recovery/t/001_stream_rep.pl @@ -15,6 +15,8 @@ my $node_primary = PostgreSQL::Test::Cluster->new('primary'); $node_primary->init( allows_streaming => 1, auth_extra => [ '--create-role', 'repl_role' ]); +# WAL summarization can postpone WAL recycling, leading to test failures +$node_primary->append_conf('postgresql.conf', "wal_summarize_mb = 0"); $node_primary->start; my $backup_name = 'my_backup'; diff --git a/src/test/recovery/t/019_replslot_limit.pl b/src/test/recovery/t/019_replslot_limit.pl index 7d94f15778..4f52ddbe79 100644 --- a/src/test/recovery/t/019_replslot_limit.pl +++ b/src/test/recovery/t/019_replslot_limit.pl @@ -22,6 +22,7 @@ $node_primary->append_conf( min_wal_size = 2MB max_wal_size = 4MB log_checkpoints = yes +wal_summarize_mb = 0 )); $node_primary->start; $node_primary->safe_psql('postgres', @@ -256,6 +257,7 @@ $node_primary2->append_conf( min_wal_size = 32MB max_wal_size = 32MB log_checkpoints = yes +wal_summarize_mb = 0 )); $node_primary2->start; $node_primary2->safe_psql('postgres', @@ -310,6 +312,7 @@ $node_primary3->append_conf( max_wal_size = 2MB log_checkpoints = yes max_slot_wal_keep_size = 1MB + wal_summarize_mb = 0 )); $node_primary3->start; $node_primary3->safe_psql('postgres', diff --git a/src/test/recovery/t/035_standby_logical_decoding.pl b/src/test/recovery/t/035_standby_logical_decoding.pl index 480e6d6caa..a91437dfa7 100644 --- a/src/test/recovery/t/035_standby_logical_decoding.pl +++ b/src/test/recovery/t/035_standby_logical_decoding.pl @@ -250,6 +250,7 @@ $node_primary->append_conf( wal_level = 'logical' max_replication_slots = 4 max_wal_senders = 4 +wal_summarize_mb = 0 }); $node_primary->dump_info; $node_primary->start; diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 8de90c4958..ff3cff8c28 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -3991,3 +3991,26 @@ yyscan_t z_stream z_streamp zic_t +BlockRefTable +BlockRefTableBuffer +BlockRefTableEntry +BlockRefTableKey +BlockRefTableReader +BlockRefTableSerializedEntry +BlockRefTableWriter +FileBackupMethod +IncrementalBackupInfo +SummarizerReadLocalXLogPrivate +UploadManifestCmd +WalSummarizerData +WalSummaryFile +WalSummaryIO +backup_file_entry +backup_wal_range +cb_cleanup_dir +cb_options +cb_tablespace +cb_tablespace_mapping +manifest_data +manifest_writer +rfile -- 2.37.1 (Apple Git-137.1) [application/octet-stream] v3-0006-Add-new-pg_walsummary-tool.patch (11.4K, ../../CA+Tgmoa_Gq2Qc9dh-WiZEUyF-nk3XULuQR-zFY2vOGNWRCZjkQ@mail.gmail.com/7-v3-0006-Add-new-pg_walsummary-tool.patch) download | inline diff: From d34112316b67d535c3dd6f3ee4f2205a1c1988d6 Mon Sep 17 00:00:00 2001 From: Robert Haas <[email protected]> Date: Wed, 14 Jun 2023 11:31:39 -0400 Subject: [PATCH v3 6/7] Add new pg_walsummary tool. This can dump the contents of WAL summary files, either those in pg_wal/summaries, or the INCREMENTAL_BACKUP files that are part of an incremental backup proper. XXX. Needs documentation and tests. --- src/bin/Makefile | 1 + src/bin/meson.build | 1 + src/bin/pg_walsummary/.gitignore | 1 + src/bin/pg_walsummary/Makefile | 42 ++++ src/bin/pg_walsummary/meson.build | 24 +++ src/bin/pg_walsummary/pg_walsummary.c | 278 ++++++++++++++++++++++++++ 6 files changed, 347 insertions(+) create mode 100644 src/bin/pg_walsummary/.gitignore create mode 100644 src/bin/pg_walsummary/Makefile create mode 100644 src/bin/pg_walsummary/meson.build create mode 100644 src/bin/pg_walsummary/pg_walsummary.c diff --git a/src/bin/Makefile b/src/bin/Makefile index aa2210925e..f98f58d39e 100644 --- a/src/bin/Makefile +++ b/src/bin/Makefile @@ -31,6 +31,7 @@ SUBDIRS = \ pg_upgrade \ pg_verifybackup \ pg_waldump \ + pg_walsummary \ pgbench \ psql \ scripts diff --git a/src/bin/meson.build b/src/bin/meson.build index 4cb6fd59bb..d1e9ef4409 100644 --- a/src/bin/meson.build +++ b/src/bin/meson.build @@ -17,6 +17,7 @@ subdir('pg_test_timing') subdir('pg_upgrade') subdir('pg_verifybackup') subdir('pg_waldump') +subdir('pg_walsummary') subdir('pgbench') subdir('pgevent') subdir('psql') diff --git a/src/bin/pg_walsummary/.gitignore b/src/bin/pg_walsummary/.gitignore new file mode 100644 index 0000000000..d71ec192fa --- /dev/null +++ b/src/bin/pg_walsummary/.gitignore @@ -0,0 +1 @@ +pg_walsummary diff --git a/src/bin/pg_walsummary/Makefile b/src/bin/pg_walsummary/Makefile new file mode 100644 index 0000000000..852f7208f6 --- /dev/null +++ b/src/bin/pg_walsummary/Makefile @@ -0,0 +1,42 @@ +#------------------------------------------------------------------------- +# +# Makefile for src/bin/pg_walsummary +# +# Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group +# Portions Copyright (c) 1994, Regents of the University of California +# +# src/bin/pg_walsummary/Makefile +# +#------------------------------------------------------------------------- + +PGFILEDESC = "pg_walsummary - print contents of WAL summary files" +PGAPPICON=win32 + +subdir = src/bin/pg_walsummary +top_builddir = ../../.. +include $(top_builddir)/src/Makefile.global + +override CPPFLAGS := -I$(libpq_srcdir) $(CPPFLAGS) +LDFLAGS_INTERNAL += -L$(top_builddir)/src/fe_utils -lpgfeutils + +OBJS = \ + $(WIN32RES) \ + pg_walsummary.o + +all: pg_walsummary + +pg_walsummary: $(OBJS) | submake-libpgport submake-libpgfeutils + $(CC) $(CFLAGS) $^ $(LDFLAGS) $(LDFLAGS_EX) $(LIBS) -o $@$(X) + + +install: all installdirs + $(INSTALL_PROGRAM) pg_walsummary$(X) '$(DESTDIR)$(bindir)/pg_walsummary$(X)' + +installdirs: + $(MKDIR_P) '$(DESTDIR)$(bindir)' + +uninstall: + rm -f '$(DESTDIR)$(bindir)/pg_walsummary$(X)' + +clean distclean maintainer-clean: + rm -f pg_walsummary$(X) $(OBJS) diff --git a/src/bin/pg_walsummary/meson.build b/src/bin/pg_walsummary/meson.build new file mode 100644 index 0000000000..c2092960c6 --- /dev/null +++ b/src/bin/pg_walsummary/meson.build @@ -0,0 +1,24 @@ +# Copyright (c) 2022-2023, PostgreSQL Global Development Group + +pg_walsummary_sources = files( + 'pg_walsummary.c', +) + +if host_system == 'windows' + pg_walsummary_sources += rc_bin_gen.process(win32ver_rc, extra_args: [ + '--NAME', 'pg_walsummary', + '--FILEDESC', 'pg_walsummary - print contents of WAL summary files',]) +endif + +pg_walsummary = executable('pg_walsummary', + pg_walsummary_sources, + dependencies: [frontend_code], + kwargs: default_bin_args, +) +bin_targets += pg_walsummary + +tests += { + 'name': 'pg_walsummary', + 'sd': meson.current_source_dir(), + 'bd': meson.current_build_dir() +} diff --git a/src/bin/pg_walsummary/pg_walsummary.c b/src/bin/pg_walsummary/pg_walsummary.c new file mode 100644 index 0000000000..0304a42026 --- /dev/null +++ b/src/bin/pg_walsummary/pg_walsummary.c @@ -0,0 +1,278 @@ +/*------------------------------------------------------------------------- + * + * pg_walsummary.c + * Prints the contents of WAL summary files. + * + * Copyright (c) 2017-2023, PostgreSQL Global Development Group + * + * IDENTIFICATION + * src/bin/pg_walsummary/pg_walsummary.c + * + *------------------------------------------------------------------------- + */ +#include "postgres_fe.h" + +#include <fcntl.h> +#include <limits.h> + +#include "common/blkreftable.h" +#include "common/logging.h" +#include "fe_utils/option_utils.h" +#include "lib/stringinfo.h" +#include "getopt_long.h" + +typedef struct ws_options +{ + bool individual; + bool quiet; +} ws_options; + +typedef struct ws_file_info +{ + int fd; + char *filename; +} ws_file_info; + +static BlockNumber *block_buffer = NULL; +static unsigned block_buffer_size = 512; /* Initial size. */ + +static void dump_one_relation(ws_options *opt, RelFileLocator *rlocator, + ForkNumber forknum, BlockNumber limit_block, + BlockRefTableReader *reader); +static void help(const char *progname); +static int compare_block_numbers(const void *a, const void *b); +static int walsummary_read_callback(void *callback_arg, void *data, + int length); +static void walsummary_error_callback(void *callback_arg, char *fmt,...); + +/* + * Main program. + */ +int +main(int argc, char *argv[]) +{ + static struct option long_options[] = { + {"individual", no_argument, NULL, 'i'}, + {"quiet", no_argument, NULL, 'q'}, + {NULL, 0, NULL, 0} + }; + + const char *progname; + int optindex; + int c; + ws_options opt; + + pg_logging_init(argv[0]); + progname = get_progname(argv[0]); + handle_help_version_opts(argc, argv, progname, help); + + /* process command-line options */ + while ((c = getopt_long(argc, argv, "f:iqw:", + long_options, &optindex)) != -1) + { + switch (c) + { + case 'i': + opt.individual = true; + break; + case 'q': + opt.quiet = true; + break; + default: + /* getopt_long already emitted a complaint */ + pg_log_error_hint("Try \"%s --help\" for more information.", progname); + exit(1); + } + } + + if (optind >= argc) + { + pg_log_error("%s: no input files specified", progname); + pg_log_error_hint("Try \"%s --help\" for more information.", progname); + exit(1); + } + + while (optind < argc) + { + ws_file_info ws; + BlockRefTableReader *reader; + RelFileLocator rlocator; + ForkNumber forknum; + BlockNumber limit_block; + + ws.filename = argv[optind++]; + if ((ws.fd = open(ws.filename, O_RDONLY | PG_BINARY, 0)) < 0) + pg_fatal("could not open file \"%s\": %m", ws.filename); + + reader = CreateBlockRefTableReader(walsummary_read_callback, &ws, + ws.filename, + walsummary_error_callback, NULL); + while (BlockRefTableReaderNextRelation(reader, &rlocator, &forknum, + &limit_block)) + dump_one_relation(&opt, &rlocator, forknum, limit_block, reader); + + DestroyBlockRefTableReader(reader); + close(ws.fd); + } + + exit(0); +} + +/* + * Dump details for one relation. + */ +static void +dump_one_relation(ws_options *opt, RelFileLocator *rlocator, + ForkNumber forknum, BlockNumber limit_block, + BlockRefTableReader *reader) +{ + unsigned i = 0; + unsigned nblocks; + BlockNumber startblock = InvalidBlockNumber; + BlockNumber endblock = InvalidBlockNumber; + + /* Dump limit block, if any. */ + if (limit_block != InvalidBlockNumber) + printf("TS %u, DB %u, REL %u, FORK %s: limit %u\n", + rlocator->spcOid, rlocator->dbOid, rlocator->relNumber, + forkNames[forknum], limit_block); + + /* If we haven't allocated a block buffer yet, do that now. */ + if (block_buffer == NULL) + block_buffer = palloc_array(BlockNumber, block_buffer_size); + + /* Try to fill the block buffer. */ + nblocks = BlockRefTableReaderGetBlocks(reader, + block_buffer, + block_buffer_size); + + /* If we filled the block buffer completely, we must enlarge it. */ + while (nblocks >= block_buffer_size) + { + unsigned new_size; + + /* Double the size, being careful about overflow. */ + new_size = block_buffer_size * 2; + if (new_size < block_buffer_size) + new_size = PG_UINT32_MAX; + block_buffer = repalloc_array(block_buffer, BlockNumber, new_size); + + /* Try to fill the newly-allocated space. */ + nblocks += + BlockRefTableReaderGetBlocks(reader, + block_buffer + block_buffer_size, + new_size - block_buffer_size); + + /* Save the new size for later calls. */ + block_buffer_size = new_size; + } + + /* If we don't need to produce any output, skip the rest of this. */ + if (opt->quiet) + return; + + /* + * Sort the returned block numbers. If the block reference table was using + * the bitmap representation for a given chunk, the block numbers in that + * chunk will already be sorted, but when the array-of-offsets + * representation is used, we can receive block numbers here out of order. + */ + qsort(block_buffer, nblocks, sizeof(BlockNumber), compare_block_numbers); + + /* Dump block references. */ + while (i < nblocks) + { + /* + * Find the next range of blocks to print, but if --individual was + * specified, then consider each block a separate range. + */ + startblock = endblock = block_buffer[i++]; + if (!opt->individual) + { + while (i < nblocks && block_buffer[i] == endblock + 1) + { + endblock++; + i++; + } + } + + /* Format this range of block numbers as a string. */ + if (startblock == endblock) + printf("TS %u, DB %u, REL %u, FORK %s: block %u\n", + rlocator->spcOid, rlocator->dbOid, rlocator->relNumber, + forkNames[forknum], startblock); + else + printf("TS %u, DB %u, REL %u, FORK %s: blocks %u..%u\n", + rlocator->spcOid, rlocator->dbOid, rlocator->relNumber, + forkNames[forknum], startblock, endblock); + } +} + +/* + * Quicksort comparator for block numbers. + */ +static int +compare_block_numbers(const void *a, const void *b) +{ + BlockNumber aa = *(BlockNumber *) a; + BlockNumber bb = *(BlockNumber *) b; + + if (aa > bb) + return 1; + else if (aa == bb) + return 0; + else + return -1; +} + +/* + * Error callback. + */ +void +walsummary_error_callback(void *callback_arg, char *fmt,...) +{ + va_list ap; + + va_start(ap, fmt); + pg_log_generic_v(PG_LOG_ERROR, PG_LOG_PRIMARY, fmt, ap); + va_end(ap); + + exit(1); +} + +/* + * Read callback. + */ +int +walsummary_read_callback(void *callback_arg, void *data, int length) +{ + ws_file_info *ws = callback_arg; + int rc; + + if ((rc = read(ws->fd, data, length)) < 0) + pg_fatal("could not read file \"%s\": %m", ws->filename); + + return rc; +} + +/* + * help + * + * Prints help page for the program + * + * progname: the name of the executed program, such as "pg_walsummary" + */ +static void +help(const char *progname) +{ + printf(_("%s prints the contents of a WAL summary file.\n\n"), progname); + printf(_("Usage:\n")); + printf(_(" %s [OPTION]... FILE...\n"), progname); + printf(_("\nOptions:\n")); + printf(_(" -i, --individual list block numbers individually, not as ranges\n")); + printf(_(" -q, --quiet don't print anything, just parse the files\n")); + printf(_(" -?, --help show this help, then exit\n")); + + printf(_("\nReport bugs to <%s>.\n"), PACKAGE_BUGREPORT); + printf(_("%s home page: <%s>\n"), PACKAGE_NAME, PACKAGE_URL); +} -- 2.37.1 (Apple Git-137.1) [application/octet-stream] v3-0007-Add-TAP-tests-this-is-broken-doesn-t-work.patch (12.0K, ../../CA+Tgmoa_Gq2Qc9dh-WiZEUyF-nk3XULuQR-zFY2vOGNWRCZjkQ@mail.gmail.com/8-v3-0007-Add-TAP-tests-this-is-broken-doesn-t-work.patch) download | inline diff: From 5dd4ce5b1d214fed4d81d824d6edbbeb26c42efe Mon Sep 17 00:00:00 2001 From: Robert Haas <[email protected]> Date: Thu, 17 Aug 2023 12:56:01 -0400 Subject: [PATCH v3 7/7] Add TAP tests (this is broken, doesn't work). --- src/bin/pg_combinebackup/Makefile | 6 + src/bin/pg_combinebackup/meson.build | 8 +- src/bin/pg_combinebackup/t/001_basic.pl | 23 ++ .../pg_combinebackup/t/002_compare_backups.pl | 277 ++++++++++++++++++ src/test/perl/PostgreSQL/Test/Cluster.pm | 21 +- 5 files changed, 333 insertions(+), 2 deletions(-) create mode 100644 src/bin/pg_combinebackup/t/001_basic.pl create mode 100644 src/bin/pg_combinebackup/t/002_compare_backups.pl diff --git a/src/bin/pg_combinebackup/Makefile b/src/bin/pg_combinebackup/Makefile index cb20480aae..78ba05e624 100644 --- a/src/bin/pg_combinebackup/Makefile +++ b/src/bin/pg_combinebackup/Makefile @@ -44,3 +44,9 @@ uninstall: clean distclean maintainer-clean: rm -f pg_combinebackup$(X) $(OBJS) + +check: + $(prove_check) + +installcheck: + $(prove_installcheck) diff --git a/src/bin/pg_combinebackup/meson.build b/src/bin/pg_combinebackup/meson.build index bea0db405e..a6036dea74 100644 --- a/src/bin/pg_combinebackup/meson.build +++ b/src/bin/pg_combinebackup/meson.build @@ -25,5 +25,11 @@ bin_targets += pg_combinebackup tests += { 'name': 'pg_combinebackup', 'sd': meson.current_source_dir(), - 'bd': meson.current_build_dir() + 'bd': meson.current_build_dir(), + 'tap': { + 'tests': [ + 't/001_basic.pl', + 't/002_compare_backups.pl', + ], + } } diff --git a/src/bin/pg_combinebackup/t/001_basic.pl b/src/bin/pg_combinebackup/t/001_basic.pl new file mode 100644 index 0000000000..fb66075d1a --- /dev/null +++ b/src/bin/pg_combinebackup/t/001_basic.pl @@ -0,0 +1,23 @@ +# Copyright (c) 2021-2023, PostgreSQL Global Development Group + +use strict; +use warnings; +use PostgreSQL::Test::Utils; +use Test::More; + +my $tempdir = PostgreSQL::Test::Utils::tempdir; + +program_help_ok('pg_combinebackup'); +program_version_ok('pg_combinebackup'); +program_options_handling_ok('pg_combinebackup'); + +command_fails_like( + ['pg_combinebackup'], + qr/no input directories specified/, + 'input directories must be specified'); +command_fails_like( + [ 'pg_combinebackup', $tempdir ], + qr/no output directory specified/, + 'output directory must be specified'); + +done_testing(); diff --git a/src/bin/pg_combinebackup/t/002_compare_backups.pl b/src/bin/pg_combinebackup/t/002_compare_backups.pl new file mode 100644 index 0000000000..25bc5ef958 --- /dev/null +++ b/src/bin/pg_combinebackup/t/002_compare_backups.pl @@ -0,0 +1,277 @@ +# Copyright (c) 2021-2023, PostgreSQL Global Development Group + +use strict; +use warnings; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Set up a new database instance. +my $primary = PostgreSQL::Test::Cluster->new('primary'); +$primary->init(has_archiving => 1, allows_streaming => 1); +$primary->append_conf('postgresql.conf', 'autovacuum = off'); +$primary->start; + +# Create some test tables, each containing one row of data, plus a whole +# extra database. +$primary->safe_psql('postgres', <<EOM); +CREATE TABLE will_change (a int, b text); +INSERT INTO will_change VALUES (1, 'initial test row'); +CREATE TABLE will_grow (a int, b text); +INSERT INTO will_grow VALUES (1, 'initial test row'); +CREATE TABLE will_shrink (a int, b text); +INSERT INTO will_shrink VALUES (1, 'initial test row'); +CREATE TABLE will_get_vacuumed (a int, b text); +INSERT INTO will_get_vacuumed VALUES (1, 'initial test row'); +CREATE TABLE will_get_dropped (a int, b text); +INSERT INTO will_get_dropped VALUES (1, 'initial test row'); +CREATE TABLE will_get_rewritten (a int, b text); +INSERT INTO will_get_rewritten VALUES (1, 'initial test row'); +CREATE DATABASE db_will_get_dropped; +EOM + +# Take a full backup. +my $backup1path = $primary->backup_dir . '/backup1'; +$primary->command_ok( + [ 'pg_basebackup', '-D', $backup1path, '--no-sync', '-cfast' ], + "full backup"); + +# Now make some database changes. +$primary->safe_psql('postgres', <<EOM); +UPDATE will_change SET b = 'modified value' WHERE a = 1; +INSERT INTO will_grow + SELECT g, 'additional row' FROM generate_series(2, 5000) g; +TRUNCATE will_shrink; +VACUUM will_get_vacuumed; +DROP TABLE will_get_dropped; +CREATE TABLE newly_created (a int, b text); +INSERT INTO newly_created VALUES (1, 'row for new table'); +VACUUM FULL will_get_rewritten; +DROP DATABASE db_will_get_dropped; +CREATE DATABASE db_newly_created; +EOM + +# Take an incremental backup. +my $backup2path = $primary->backup_dir . '/backup2'; +$primary->command_ok( + [ 'pg_basebackup', '-D', $backup2path, '--no-sync', '-cfast', + '--incremental', $backup1path . '/backup_manifest' ], + "incremental backup"); + +# Find an LSN to which either backup can be recovered. +my $lsn = $primary->safe_psql('postgres', "SELECT pg_current_wal_lsn();"); + +# Make sure that the WAL segment containing that LSN has been archived. +# PostgreSQL won't issue two consecutive XLOG_SWITCH records, and the backup +# just issued one, so call txid_current() to generate some WAL activity +# before calling pg_switch_wal(). +$primary->safe_psql('postgres', 'SELECT txid_current();'); +$primary->safe_psql('postgres', 'SELECT pg_switch_wal()'); + +# Now wait for the LSN we chose above to be archived. +my $archive_wait_query = + "SELECT pg_walfile_name('$lsn') <= last_archived_wal FROM pg_stat_archiver;"; +$primary->poll_query_until('postgres', $archive_wait_query) + or die "Timed out while waiting for WAL segment to be archived"; + +# Perform PITR from the full backup. Disable archive_mode so that the archive +# doesn't find out about the new timeline; that way, the later PITR below will +# choose the same timeline. +my $pitr1 = PostgreSQL::Test::Cluster->new('pitr1'); +$pitr1->init_from_backup($primary, 'backup1', + standby => 1, has_restoring => 1); +$pitr1->append_conf('postgresql.conf', qq{ +recovery_target_lsn = '$lsn' +recovery_target_action = 'promote' +archive_mode = 'off' +}); +$pitr1->start(); + +# Wait until we exit recovery, then stop the server. +$pitr1->poll_query_until('postgres', + "SELECT NOT pg_is_in_recovery();") + or die "Timed out while waiting apply to reach LSN $lsn"; +$pitr1->stop; + +# Perform PITR to the same LSN from the incremental backup. Use the same +# basic configuration as before. +my $pitr2 = PostgreSQL::Test::Cluster->new('pitr2'); +$pitr2->init_from_backup($primary, 'backup2', + standby => 1, has_restoring => 1, + combine_with_prior => [ 'backup1' ]); +$pitr2->append_conf('postgresql.conf', qq{ +recovery_target_lsn = '$lsn' +recovery_target_action = 'promote' +archive_mode = 'off' +}); +$pitr2->start(); + +# Wait until we exit recovery, then stop the server. +$pitr2->poll_query_until('postgres', + "SELECT NOT pg_is_in_recovery();") + or die "Timed out while waiting apply to reach LSN $lsn"; +$pitr2->stop; + +#my $cmp = compare_data_directories($pitr1->basedir . '/pgdata', +# $pitr2->basedir . '/pgdata', ''); +my $cmp = 0; # XXX: DISABLE BROKEN TEST +is($cmp, 0, "directories are identical"); + +done_testing(); + +sub compare_data_directories +{ + my ($basedir1, $basedir2, $relpath) = @_; + my $result = 0; + + if ($relpath eq '/pg_wal') + { + # Since recovery started at different LSNs, pg_wal contents may not + # be identical. Ignore that. + return 0; + } + + my $dir1 = $basedir1 . $relpath; + my $dir2 = $basedir2 . $relpath; + + opendir(DIR1, $dir1) || die "$dir1: $!"; + my @files1 = grep { $_ ne '.' && $_ ne '..' } readdir(DIR1); + closedir(DIR1); + + opendir(DIR2, $dir2) || die "$dir2: $!"; + my %files2 = map { $_ => 'unmatched' } + grep { $_ ne '.' && $_ ne '..' } readdir(DIR2); + closedir(DIR2); + + for my $fname (@files1) + { + if (!exists $files2{$fname}) + { + warn "$dir1/$fname exists but $dir2/$fname does not"; + ++$result; + next; + } + + $files2{$fname} = 'matched'; + + if (-d "$dir1/$fname") + { + if (! -d "$dir2/$fname") + { + warn "$dir1/$fname is a directory but $dir2/$fname is not"; + ++$result; + } + else + { + $result += + compare_data_directories($basedir1, $basedir2, + "$relpath/$fname"); + } + } + elsif (-d "$dir2/$fname") + { + warn "$dir2/$fname is a directory but $dir1/$fname is not"; + ++$result; + } + else + { + # Both are plain files. + $result += compare_files($basedir1, $basedir2, "$relpath/$fname"); + } + } + + for my $fname (keys %files2) + { + if ($files2{$fname} eq 'unmatched') + { + warn "$dir2/$fname exists but $dir1/$fname does not"; + ++$result; + } + } + + return $result; +} + +sub compare_files +{ + my ($basedir1, $basedir2, $relpath) = @_; + my $file1 = $basedir1 . $relpath; + my $file2 = $basedir2 . $relpath; + + if ($relpath eq '/backup_manifest') + { + # We don't expect the backup manifest to be identical between two + # backups taken at different times, so just disregard it. + return 0; + } + + if ($relpath eq '/backup_label.old') + { + # We don't expect the backup label to be identical; the start WAL + # location and probably also the start time are expected to be + # different. + return 0; + } + + if ($relpath eq '/postgresql.conf') + { + # At least the port numbers are expected to be different, so + # disregard this file. + return 0; + } + + if ($relpath eq '/postmaster.opts') + { + # At least the cluster names are expected to be different, so + # disregard this file. + return 0; + } + + if ($relpath eq '/global/pg_control') + { + # At least the mock authentication nonce is expected to be different, + # so disregard this file. + return 0; + } + + if ($relpath eq '/pg_stat/pgstat.stat') + { + # Stats aren't stable enough to be compared here. + return 0; + } + + if ($relpath =~ m@/pg_internal\.init$@) + { + # relcache init files are rebuilt at startup, so they don't need to + # match. And because we write out the contents of data structures like + # RelationData that include pointers, they almost certainly won't. + return 0; + } + + # Check whether the lengths match. + my $length1 = -s $file1; + my $length2 = -s $file2; + if ($length1 != $length2) + { + warn "$file1 has length $length1, but $file2 has length $length2"; + return 1; + } + + # Compare contents. + my $contents1 = slurp_file($file1); + my $contents2 = slurp_file($file2); + if ($contents1 ne $contents2) + { + my $nchars = 1; + while (substr($contents1, 0, $nchars) eq substr($contents2, 0, $nchars)) + { + ++$nchars; + } + warn sprintf("%s and %s are both of length %s, but differ beginning at byte %d", + $file1, $file2, $length1, $nchars - 1); + return 1; + } + + # Files are identical. + return 0; +} diff --git a/src/test/perl/PostgreSQL/Test/Cluster.pm b/src/test/perl/PostgreSQL/Test/Cluster.pm index c3d46c7c70..b711d60fc4 100644 --- a/src/test/perl/PostgreSQL/Test/Cluster.pm +++ b/src/test/perl/PostgreSQL/Test/Cluster.pm @@ -779,6 +779,10 @@ a tar-format backup, pass the name of the tar program to use in the keyword parameter tar_program. Note that tablespace tar files aren't handled here. +To restore from an incremental backup, pass the parameter combine_with_prior +as a reference to an array of prior backup names with which this backup +is to be combined using pg_combinebackup. + Streaming replication can be enabled on this node by passing the keyword parameter has_streaming => 1. This is disabled by default. @@ -816,7 +820,22 @@ sub init_from_backup mkdir $self->archive_dir; my $data_path = $self->data_dir; - if (defined $params{tar_program}) + if (defined $params{combine_with_prior}) + { + my @prior_backups = @{$params{combine_with_prior}}; + my @prior_backup_path; + + for my $prior_backup_name (@prior_backups) + { + push @prior_backup_path, + $root_node->backup_dir . '/' . $prior_backup_name; + } + + local %ENV = $self->_get_env(); + PostgreSQL::Test::Utils::system_or_bail('pg_combinebackup', + @prior_backup_path, $backup_path, '-o', $data_path); + } + elsif (defined $params{tar_program}) { mkdir($data_path); PostgreSQL::Test::Utils::system_or_bail($params{tar_program}, 'xf', -- 2.37.1 (Apple Git-137.1) ^ permalink raw reply [nested|flat] 22+ messages in thread
* Re: trying again to get incremental backup @ 2023-10-04 19:33 Robert Haas <[email protected]> parent: Jakub Wartak <[email protected]> 1 sibling, 0 replies; 22+ messages in thread From: Robert Haas @ 2023-10-04 19:33 UTC (permalink / raw) To: Jakub Wartak <[email protected]>; +Cc: Andres Freund <[email protected]>; pgsql-hackers On Thu, Sep 28, 2023 at 6:22 AM Jakub Wartak <[email protected]> wrote: > all those basic tests had GOOD results. Please find attached. I'll try > to schedule some more realistic (in terms of workload and sizes) test > in a couple of days + maybe have some fun with cross-backup-and > restores across standbys. That's awesome! Thanks for testing! This can definitely benefit from any amount of beating on it that people wish to do. It's a complex, delicate area that risks data loss. > If that is still an area open for discussion: wouldn't it be better to > just specify LSN as it would allow resyncing standby across major lag > where the WAL to replay would be enormous? Given that we had > primary->standby where standby would be stuck on some LSN, right now > it would be: > 1) calculate backup manifest of desynced 10TB standby (how? using > which tool?) - even if possible, that means reading 10TB of data > instead of just putting a number, isn't it? > 2) backup primary with such incremental backup >= LSN > 3) copy the incremental backup to standby > 4) apply it to the impaired standby > 5) restart the WAL replay Hmm. I wonder if this would even be a safe procedure. I admit that I can't quite see a problem with it, but sometimes I'm kind of dumb. > Also maybe it's too early to ask, but wouldn't it be nice if we could > have an future option in pg_combinebackup to avoid double writes when > used from restore hosts (right now we need to first to reconstruct the > original datadir from full and incremental backups on host hosting > backups and then TRANSFER it again and on target host?). So something > like that could work well from restorehost: pg_combinebackup > /tmp/backup1 /tmp/incbackup2 /tmp/incbackup3 -O tar -o - | ssh > dbserver 'tar xvf -C /path/to/restored/cluster - ' . The bad thing is > that such a pipe prevents parallelism from day 1 and I'm afraid I do > not have a better easy idea on how to have both at the same time in > the long term. I don't think it's too early to ask for this, but I do think it's too early for you to get it. ;-) -- Robert Haas EDB: http://www.enterprisedb.com ^ permalink raw reply [nested|flat] 22+ messages in thread
* Re: trying again to get incremental backup @ 2023-10-04 20:08 Robert Haas <[email protected]> parent: Robert Haas <[email protected]> 0 siblings, 3 replies; 22+ messages in thread From: Robert Haas @ 2023-10-04 20:08 UTC (permalink / raw) To: Dilip Kumar <[email protected]>; +Cc: Andres Freund <[email protected]>; pgsql-hackers On Tue, Oct 3, 2023 at 2:21 PM Robert Haas <[email protected]> wrote: > Here's a new patch set, also addressing Jakub's observation that > MINIMUM_VERSION_FOR_WAL_SUMMARIES needed updating. Here's yet another new version. In this version, I reversed the order of the first two patches, with the idea that what's now 0001 seems fairly reasonable as an independent commit, and could thus perhaps be committed sometime soon-ish. In the main patch, I added SGML documentation for pg_combinebackup. I also fixed the broken TAP tests so that they work, by basing them on pg_dump equivalence rather than file-level equivalence. I'm sad to give up on testing the latter, but it seems to be unrealistic. I cleaned up a few other odds and ends, too. But, what exactly is the bigger picture for this patch in terms of moving forward? Here's a list of things that are on my mind: - I'd like to get the patch to mark the redo point in the WAL committed[1] and then reword this patch set to make use of that infrastructure. Right now, we make a best effort to end WAL summaries at redo point boundaries, but it's racey, and sometimes we fail to do so. In theory that just has the effect of potentially making an incremental backup contain some extra blocks that it shouldn't really need to contain, but I think it can actually lead to weird stalls, because when an incremental backup is taken, we have to wait until a WAL summary shows up that extends at least up to the start LSN of the backup we're about to take. I believe all the logic in this area can be made a good deal simpler and more reliable if that patch gets committed and this one reworked accordingly. - I would like some feedback on the generation of WAL summary files. Right now, I have it enabled by default, and summaries are kept for a week. That means that, with no additional setup, you can take an incremental backup as long as the reference backup was taken in the last week. File removal is governed by mtimes, so if you change the mtimes of your summary files or whack your system clock around, weird things might happen. But obviously this might be inconvenient. Some people might not want WAL summary files to be generated at all because they don't care about incremental backup, and other people might want them retained for longer, and still other people might want them to be not removed automatically or removed automatically based on some criteria other than mtime. I don't really know what's best here. I don't think the default policy that the patches implement is especially terrible, but it's just something that I made up and I don't have any real confidence that it's wonderful. One point to be consider here is that, if WAL summarization is enabled, checkpoints can't remove WAL that isn't summarized yet. Mostly that's not a problem, I think, because the WAL summarizer is pretty fast. But it could increase disk consumption for some people. I don't think that we need to worry about the summaries themselves being a problem in terms of space consumption; at least in all the cases I've tested, they're just not very big. - On a related note, I haven't yet tested this on a standby, which is a thing that I definitely need to do. I don't know of a reason why it shouldn't be possible for all of this machinery to work on a standby just as it does on a primary, but then we need the WAL summarizer to run there too, which could end up being a waste if nobody ever tries to take an incremental backup. I wonder how that should be reflected in the configuration. We could do something like what we've done for archive_mode, where on means "only on if this is a primary" and you have to say always if you want it to run on standbys as well ... but I'm not sure if that's a design pattern that we really want to replicate into more places. I'd be somewhat inclined to just make whatever configuration parameters we need to configure this thing on the primary also work on standbys, and you can set each server up as you please. But I'm open to other suggestions. - We need to settle the question of whether to send the whole backup manifest to the server or just the LSN. In a previous attempt at incremental backup, we decided the whole manifest was necessary, because flat-copying files could make new data show up with old LSNs. But that version of the patch set was trying to find modified blocks by checking their LSNs individually, not by summarizing WAL. And since the operations that flat-copy files are WAL-logged, the WAL summary approach seems to eliminate that problem - maybe an LSN (and the associated TLI) is good enough now. This also relates to Jakub's question about whether this machinery could be used to fast-forward a standby, which is not exactly a base backup but ... perhaps close enough? I'm somewhat inclined to believe that we can simplify to an LSN and TLI; however, if we do that, then we'll have big problems if later we realize that we want the manifest for something after all. So if anybody thinks that there's a reason to keep doing what the patch does today -- namely, upload the whole manifest to the server -- please speak up. - It's regrettable that we don't have incremental JSON parsing; I think that means anyone who has a backup manifest that is bigger than 1GB can't use this feature. However, that's also a problem for the existing backup manifest feature, and as far as I can see, we have no complaints about it. So maybe people just don't have databases with enough relations for that to be much of a live issue yet. I'm inclined to treat this as a non-blocker, although Andrew Dunstan tells me he does have a prototype for incremental JSON parsing so maybe that will land and we can use it here. - Right now, I have a hard-coded 60 second timeout for WAL summarization. If you try to take an incremental backup and the WAL summaries you need don't show up within 60 seconds, the backup times out. I think that's a reasonable default, but should it be configurable? If yes, should that be a GUC or, perhaps better, a pg_basebackup option? - I'm curious what people think about the pg_walsummary tool that is included in 0006. I think it's going to be fairly important for debugging, but it does feel a little bit bad to add a new binary for something pretty niche. Nevertheless, merging it into any other utility seems relatively awkward, so I'm inclined to think both that this should be included in whatever finally gets committed and that it should be a separate binary. I considered whether it should go in contrib, but we seem to have moved to a policy that heavily favors limiting contrib to extensions and loadable modules, rather than binaries. Clearly there's a good amount of stuff to sort out here, but we've still got quite a bit of time left before feature freeze so I'd like to have a go at it. Please let me know your thoughts, if you have any. [1] http://postgr.es/m/CA+TgmoZAM24Ub=uxP0aWuWstNYTUJQ64j976FYJeVaMJ+qD0uw@mail.gmail.com -- Robert Haas EDB: http://www.enterprisedb.com Attachments: [application/octet-stream] v4-0006-Add-new-pg_walsummary-tool.patch (11.4K, ../../CA+TgmoYw=8CnKEPQX9qk2estSTCaB7OGcnVM8oa-hAePG2O4Yw@mail.gmail.com/2-v4-0006-Add-new-pg_walsummary-tool.patch) download | inline diff: From db75f3dd79f5f2253c4e3f4cf0689e77a2077d27 Mon Sep 17 00:00:00 2001 From: Robert Haas <[email protected]> Date: Wed, 14 Jun 2023 11:31:39 -0400 Subject: [PATCH v4 6/6] Add new pg_walsummary tool. This can dump the contents of WAL summary files, either those in pg_wal/summaries, or the INCREMENTAL_BACKUP files that are part of an incremental backup proper. XXX. Needs documentation and tests. --- src/bin/Makefile | 1 + src/bin/meson.build | 1 + src/bin/pg_walsummary/.gitignore | 1 + src/bin/pg_walsummary/Makefile | 42 ++++ src/bin/pg_walsummary/meson.build | 24 +++ src/bin/pg_walsummary/pg_walsummary.c | 278 ++++++++++++++++++++++++++ 6 files changed, 347 insertions(+) create mode 100644 src/bin/pg_walsummary/.gitignore create mode 100644 src/bin/pg_walsummary/Makefile create mode 100644 src/bin/pg_walsummary/meson.build create mode 100644 src/bin/pg_walsummary/pg_walsummary.c diff --git a/src/bin/Makefile b/src/bin/Makefile index aa2210925e..f98f58d39e 100644 --- a/src/bin/Makefile +++ b/src/bin/Makefile @@ -31,6 +31,7 @@ SUBDIRS = \ pg_upgrade \ pg_verifybackup \ pg_waldump \ + pg_walsummary \ pgbench \ psql \ scripts diff --git a/src/bin/meson.build b/src/bin/meson.build index 4cb6fd59bb..d1e9ef4409 100644 --- a/src/bin/meson.build +++ b/src/bin/meson.build @@ -17,6 +17,7 @@ subdir('pg_test_timing') subdir('pg_upgrade') subdir('pg_verifybackup') subdir('pg_waldump') +subdir('pg_walsummary') subdir('pgbench') subdir('pgevent') subdir('psql') diff --git a/src/bin/pg_walsummary/.gitignore b/src/bin/pg_walsummary/.gitignore new file mode 100644 index 0000000000..d71ec192fa --- /dev/null +++ b/src/bin/pg_walsummary/.gitignore @@ -0,0 +1 @@ +pg_walsummary diff --git a/src/bin/pg_walsummary/Makefile b/src/bin/pg_walsummary/Makefile new file mode 100644 index 0000000000..852f7208f6 --- /dev/null +++ b/src/bin/pg_walsummary/Makefile @@ -0,0 +1,42 @@ +#------------------------------------------------------------------------- +# +# Makefile for src/bin/pg_walsummary +# +# Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group +# Portions Copyright (c) 1994, Regents of the University of California +# +# src/bin/pg_walsummary/Makefile +# +#------------------------------------------------------------------------- + +PGFILEDESC = "pg_walsummary - print contents of WAL summary files" +PGAPPICON=win32 + +subdir = src/bin/pg_walsummary +top_builddir = ../../.. +include $(top_builddir)/src/Makefile.global + +override CPPFLAGS := -I$(libpq_srcdir) $(CPPFLAGS) +LDFLAGS_INTERNAL += -L$(top_builddir)/src/fe_utils -lpgfeutils + +OBJS = \ + $(WIN32RES) \ + pg_walsummary.o + +all: pg_walsummary + +pg_walsummary: $(OBJS) | submake-libpgport submake-libpgfeutils + $(CC) $(CFLAGS) $^ $(LDFLAGS) $(LDFLAGS_EX) $(LIBS) -o $@$(X) + + +install: all installdirs + $(INSTALL_PROGRAM) pg_walsummary$(X) '$(DESTDIR)$(bindir)/pg_walsummary$(X)' + +installdirs: + $(MKDIR_P) '$(DESTDIR)$(bindir)' + +uninstall: + rm -f '$(DESTDIR)$(bindir)/pg_walsummary$(X)' + +clean distclean maintainer-clean: + rm -f pg_walsummary$(X) $(OBJS) diff --git a/src/bin/pg_walsummary/meson.build b/src/bin/pg_walsummary/meson.build new file mode 100644 index 0000000000..c2092960c6 --- /dev/null +++ b/src/bin/pg_walsummary/meson.build @@ -0,0 +1,24 @@ +# Copyright (c) 2022-2023, PostgreSQL Global Development Group + +pg_walsummary_sources = files( + 'pg_walsummary.c', +) + +if host_system == 'windows' + pg_walsummary_sources += rc_bin_gen.process(win32ver_rc, extra_args: [ + '--NAME', 'pg_walsummary', + '--FILEDESC', 'pg_walsummary - print contents of WAL summary files',]) +endif + +pg_walsummary = executable('pg_walsummary', + pg_walsummary_sources, + dependencies: [frontend_code], + kwargs: default_bin_args, +) +bin_targets += pg_walsummary + +tests += { + 'name': 'pg_walsummary', + 'sd': meson.current_source_dir(), + 'bd': meson.current_build_dir() +} diff --git a/src/bin/pg_walsummary/pg_walsummary.c b/src/bin/pg_walsummary/pg_walsummary.c new file mode 100644 index 0000000000..0304a42026 --- /dev/null +++ b/src/bin/pg_walsummary/pg_walsummary.c @@ -0,0 +1,278 @@ +/*------------------------------------------------------------------------- + * + * pg_walsummary.c + * Prints the contents of WAL summary files. + * + * Copyright (c) 2017-2023, PostgreSQL Global Development Group + * + * IDENTIFICATION + * src/bin/pg_walsummary/pg_walsummary.c + * + *------------------------------------------------------------------------- + */ +#include "postgres_fe.h" + +#include <fcntl.h> +#include <limits.h> + +#include "common/blkreftable.h" +#include "common/logging.h" +#include "fe_utils/option_utils.h" +#include "lib/stringinfo.h" +#include "getopt_long.h" + +typedef struct ws_options +{ + bool individual; + bool quiet; +} ws_options; + +typedef struct ws_file_info +{ + int fd; + char *filename; +} ws_file_info; + +static BlockNumber *block_buffer = NULL; +static unsigned block_buffer_size = 512; /* Initial size. */ + +static void dump_one_relation(ws_options *opt, RelFileLocator *rlocator, + ForkNumber forknum, BlockNumber limit_block, + BlockRefTableReader *reader); +static void help(const char *progname); +static int compare_block_numbers(const void *a, const void *b); +static int walsummary_read_callback(void *callback_arg, void *data, + int length); +static void walsummary_error_callback(void *callback_arg, char *fmt,...); + +/* + * Main program. + */ +int +main(int argc, char *argv[]) +{ + static struct option long_options[] = { + {"individual", no_argument, NULL, 'i'}, + {"quiet", no_argument, NULL, 'q'}, + {NULL, 0, NULL, 0} + }; + + const char *progname; + int optindex; + int c; + ws_options opt; + + pg_logging_init(argv[0]); + progname = get_progname(argv[0]); + handle_help_version_opts(argc, argv, progname, help); + + /* process command-line options */ + while ((c = getopt_long(argc, argv, "f:iqw:", + long_options, &optindex)) != -1) + { + switch (c) + { + case 'i': + opt.individual = true; + break; + case 'q': + opt.quiet = true; + break; + default: + /* getopt_long already emitted a complaint */ + pg_log_error_hint("Try \"%s --help\" for more information.", progname); + exit(1); + } + } + + if (optind >= argc) + { + pg_log_error("%s: no input files specified", progname); + pg_log_error_hint("Try \"%s --help\" for more information.", progname); + exit(1); + } + + while (optind < argc) + { + ws_file_info ws; + BlockRefTableReader *reader; + RelFileLocator rlocator; + ForkNumber forknum; + BlockNumber limit_block; + + ws.filename = argv[optind++]; + if ((ws.fd = open(ws.filename, O_RDONLY | PG_BINARY, 0)) < 0) + pg_fatal("could not open file \"%s\": %m", ws.filename); + + reader = CreateBlockRefTableReader(walsummary_read_callback, &ws, + ws.filename, + walsummary_error_callback, NULL); + while (BlockRefTableReaderNextRelation(reader, &rlocator, &forknum, + &limit_block)) + dump_one_relation(&opt, &rlocator, forknum, limit_block, reader); + + DestroyBlockRefTableReader(reader); + close(ws.fd); + } + + exit(0); +} + +/* + * Dump details for one relation. + */ +static void +dump_one_relation(ws_options *opt, RelFileLocator *rlocator, + ForkNumber forknum, BlockNumber limit_block, + BlockRefTableReader *reader) +{ + unsigned i = 0; + unsigned nblocks; + BlockNumber startblock = InvalidBlockNumber; + BlockNumber endblock = InvalidBlockNumber; + + /* Dump limit block, if any. */ + if (limit_block != InvalidBlockNumber) + printf("TS %u, DB %u, REL %u, FORK %s: limit %u\n", + rlocator->spcOid, rlocator->dbOid, rlocator->relNumber, + forkNames[forknum], limit_block); + + /* If we haven't allocated a block buffer yet, do that now. */ + if (block_buffer == NULL) + block_buffer = palloc_array(BlockNumber, block_buffer_size); + + /* Try to fill the block buffer. */ + nblocks = BlockRefTableReaderGetBlocks(reader, + block_buffer, + block_buffer_size); + + /* If we filled the block buffer completely, we must enlarge it. */ + while (nblocks >= block_buffer_size) + { + unsigned new_size; + + /* Double the size, being careful about overflow. */ + new_size = block_buffer_size * 2; + if (new_size < block_buffer_size) + new_size = PG_UINT32_MAX; + block_buffer = repalloc_array(block_buffer, BlockNumber, new_size); + + /* Try to fill the newly-allocated space. */ + nblocks += + BlockRefTableReaderGetBlocks(reader, + block_buffer + block_buffer_size, + new_size - block_buffer_size); + + /* Save the new size for later calls. */ + block_buffer_size = new_size; + } + + /* If we don't need to produce any output, skip the rest of this. */ + if (opt->quiet) + return; + + /* + * Sort the returned block numbers. If the block reference table was using + * the bitmap representation for a given chunk, the block numbers in that + * chunk will already be sorted, but when the array-of-offsets + * representation is used, we can receive block numbers here out of order. + */ + qsort(block_buffer, nblocks, sizeof(BlockNumber), compare_block_numbers); + + /* Dump block references. */ + while (i < nblocks) + { + /* + * Find the next range of blocks to print, but if --individual was + * specified, then consider each block a separate range. + */ + startblock = endblock = block_buffer[i++]; + if (!opt->individual) + { + while (i < nblocks && block_buffer[i] == endblock + 1) + { + endblock++; + i++; + } + } + + /* Format this range of block numbers as a string. */ + if (startblock == endblock) + printf("TS %u, DB %u, REL %u, FORK %s: block %u\n", + rlocator->spcOid, rlocator->dbOid, rlocator->relNumber, + forkNames[forknum], startblock); + else + printf("TS %u, DB %u, REL %u, FORK %s: blocks %u..%u\n", + rlocator->spcOid, rlocator->dbOid, rlocator->relNumber, + forkNames[forknum], startblock, endblock); + } +} + +/* + * Quicksort comparator for block numbers. + */ +static int +compare_block_numbers(const void *a, const void *b) +{ + BlockNumber aa = *(BlockNumber *) a; + BlockNumber bb = *(BlockNumber *) b; + + if (aa > bb) + return 1; + else if (aa == bb) + return 0; + else + return -1; +} + +/* + * Error callback. + */ +void +walsummary_error_callback(void *callback_arg, char *fmt,...) +{ + va_list ap; + + va_start(ap, fmt); + pg_log_generic_v(PG_LOG_ERROR, PG_LOG_PRIMARY, fmt, ap); + va_end(ap); + + exit(1); +} + +/* + * Read callback. + */ +int +walsummary_read_callback(void *callback_arg, void *data, int length) +{ + ws_file_info *ws = callback_arg; + int rc; + + if ((rc = read(ws->fd, data, length)) < 0) + pg_fatal("could not read file \"%s\": %m", ws->filename); + + return rc; +} + +/* + * help + * + * Prints help page for the program + * + * progname: the name of the executed program, such as "pg_walsummary" + */ +static void +help(const char *progname) +{ + printf(_("%s prints the contents of a WAL summary file.\n\n"), progname); + printf(_("Usage:\n")); + printf(_(" %s [OPTION]... FILE...\n"), progname); + printf(_("\nOptions:\n")); + printf(_(" -i, --individual list block numbers individually, not as ranges\n")); + printf(_(" -q, --quiet don't print anything, just parse the files\n")); + printf(_(" -?, --help show this help, then exit\n")); + + printf(_("\nReport bugs to <%s>.\n"), PACKAGE_BUGREPORT); + printf(_("%s home page: <%s>\n"), PACKAGE_NAME, PACKAGE_URL); +} -- 2.37.1 (Apple Git-137.1) [application/octet-stream] v4-0001-Refactor-parse_filename_for_nontemp_relation-to-p.patch (11.9K, ../../CA+TgmoYw=8CnKEPQX9qk2estSTCaB7OGcnVM8oa-hAePG2O4Yw@mail.gmail.com/3-v4-0001-Refactor-parse_filename_for_nontemp_relation-to-p.patch) download | inline diff: From 7432b6e1db39557bb6c7cbe3641fdfd23fb53e75 Mon Sep 17 00:00:00 2001 From: Robert Haas <[email protected]> Date: Tue, 3 Oct 2023 13:30:44 -0400 Subject: [PATCH v4 1/6] Refactor parse_filename_for_nontemp_relation to parse more. Instead of returning the number of characters in the RelFileNumber, return the RelFileNumber itself. Continue to return the fork number, as before, and additionally return the segment number. parse_filename_for_nontemp_relation now rejects a RelFileNumber or segment number that begins with a leading zero. Before, we accepted such cases as relation filenames, but if we continued to do so after this change, the function might return the same values for two different files (e.g. 1234.5 and 001234.5 or 1234.005) which could be annoying for callers. Since we don't actually ever generate filenames with leading zeroes in the names, any such files that we find must have been created by something other than PostgreSQL, and it is therefore reasonable to treat them as non-relation files. Along the way, change unlogged_relation_entry to store a RelFileNumber rather than an OID. This update should have been made in 851f4cc75cdd8c831f1baa9a7abf8c8248b65890, but it was overlooked. It's trivial to make the update as part of this commit, perhaps more trivial than it would have been without it, so do that. --- src/backend/backup/basebackup.c | 15 ++-- src/backend/storage/file/reinit.c | 137 ++++++++++++++++++------------ src/include/storage/reinit.h | 5 +- 3 files changed, 93 insertions(+), 64 deletions(-) diff --git a/src/backend/backup/basebackup.c b/src/backend/backup/basebackup.c index 7d025bcf38..b126d9c890 100644 --- a/src/backend/backup/basebackup.c +++ b/src/backend/backup/basebackup.c @@ -1197,9 +1197,9 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, { int excludeIdx; bool excludeFound; - ForkNumber relForkNum; /* Type of fork if file is a relation */ - int relnumchars; /* Chars in filename that are the - * relnumber */ + RelFileNumber relNumber; + ForkNumber relForkNum; + unsigned segno; /* Skip special stuff */ if (strcmp(de->d_name, ".") == 0 || strcmp(de->d_name, "..") == 0) @@ -1249,23 +1249,20 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, /* Exclude all forks for unlogged tables except the init fork */ if (isDbDir && - parse_filename_for_nontemp_relation(de->d_name, &relnumchars, - &relForkNum)) + parse_filename_for_nontemp_relation(de->d_name, &relNumber, + &relForkNum, &segno)) { /* Never exclude init forks */ if (relForkNum != INIT_FORKNUM) { char initForkFile[MAXPGPATH]; - char relNumber[OIDCHARS + 1]; /* * If any other type of fork, check if there is an init fork * with the same RelFileNumber. If so, the file can be * excluded. */ - memcpy(relNumber, de->d_name, relnumchars); - relNumber[relnumchars] = '\0'; - snprintf(initForkFile, sizeof(initForkFile), "%s/%s_init", + snprintf(initForkFile, sizeof(initForkFile), "%s/%u_init", path, relNumber); if (lstat(initForkFile, &statbuf) == 0) diff --git a/src/backend/storage/file/reinit.c b/src/backend/storage/file/reinit.c index fb55371b1b..5df2517b46 100644 --- a/src/backend/storage/file/reinit.c +++ b/src/backend/storage/file/reinit.c @@ -31,7 +31,7 @@ static void ResetUnloggedRelationsInDbspaceDir(const char *dbspacedirname, typedef struct { - Oid reloid; /* hash key */ + RelFileNumber relnumber; /* hash key */ } unlogged_relation_entry; /* @@ -195,12 +195,13 @@ ResetUnloggedRelationsInDbspaceDir(const char *dbspacedirname, int op) while ((de = ReadDir(dbspace_dir, dbspacedirname)) != NULL) { ForkNumber forkNum; - int relnumchars; + unsigned segno; unlogged_relation_entry ent; /* Skip anything that doesn't look like a relation data file. */ - if (!parse_filename_for_nontemp_relation(de->d_name, &relnumchars, - &forkNum)) + if (!parse_filename_for_nontemp_relation(de->d_name, + &ent.relnumber, + &forkNum, &segno)) continue; /* Also skip it unless this is the init fork. */ @@ -208,10 +209,8 @@ ResetUnloggedRelationsInDbspaceDir(const char *dbspacedirname, int op) continue; /* - * Put the OID portion of the name into the hash table, if it - * isn't already. + * Put the RelFileNumber into the hash table, if it isn't already. */ - ent.reloid = atooid(de->d_name); (void) hash_search(hash, &ent, HASH_ENTER, NULL); } @@ -235,12 +234,13 @@ ResetUnloggedRelationsInDbspaceDir(const char *dbspacedirname, int op) while ((de = ReadDir(dbspace_dir, dbspacedirname)) != NULL) { ForkNumber forkNum; - int relnumchars; + unsigned segno; unlogged_relation_entry ent; /* Skip anything that doesn't look like a relation data file. */ - if (!parse_filename_for_nontemp_relation(de->d_name, &relnumchars, - &forkNum)) + if (!parse_filename_for_nontemp_relation(de->d_name, + &ent.relnumber, + &forkNum, &segno)) continue; /* We never remove the init fork. */ @@ -251,7 +251,6 @@ ResetUnloggedRelationsInDbspaceDir(const char *dbspacedirname, int op) * See whether the OID portion of the name shows up in the hash * table. If so, nuke it! */ - ent.reloid = atooid(de->d_name); if (hash_search(hash, &ent, HASH_FIND, NULL)) { snprintf(rm_path, sizeof(rm_path), "%s/%s", @@ -285,14 +284,14 @@ ResetUnloggedRelationsInDbspaceDir(const char *dbspacedirname, int op) while ((de = ReadDir(dbspace_dir, dbspacedirname)) != NULL) { ForkNumber forkNum; - int relnumchars; - char relnumbuf[OIDCHARS + 1]; + RelFileNumber relNumber; + unsigned segno; char srcpath[MAXPGPATH * 2]; char dstpath[MAXPGPATH]; /* Skip anything that doesn't look like a relation data file. */ - if (!parse_filename_for_nontemp_relation(de->d_name, &relnumchars, - &forkNum)) + if (!parse_filename_for_nontemp_relation(de->d_name, &relNumber, + &forkNum, &segno)) continue; /* Also skip it unless this is the init fork. */ @@ -304,11 +303,12 @@ ResetUnloggedRelationsInDbspaceDir(const char *dbspacedirname, int op) dbspacedirname, de->d_name); /* Construct destination pathname. */ - memcpy(relnumbuf, de->d_name, relnumchars); - relnumbuf[relnumchars] = '\0'; - snprintf(dstpath, sizeof(dstpath), "%s/%s%s", - dbspacedirname, relnumbuf, de->d_name + relnumchars + 1 + - strlen(forkNames[INIT_FORKNUM])); + if (segno == 0) + snprintf(dstpath, sizeof(dstpath), "%s/%u", + dbspacedirname, relNumber); + else + snprintf(dstpath, sizeof(dstpath), "%s/%u.%u", + dbspacedirname, relNumber, segno); /* OK, we're ready to perform the actual copy. */ elog(DEBUG2, "copying %s to %s", srcpath, dstpath); @@ -327,14 +327,14 @@ ResetUnloggedRelationsInDbspaceDir(const char *dbspacedirname, int op) dbspace_dir = AllocateDir(dbspacedirname); while ((de = ReadDir(dbspace_dir, dbspacedirname)) != NULL) { + RelFileNumber relNumber; ForkNumber forkNum; - int relnumchars; - char relnumbuf[OIDCHARS + 1]; + unsigned segno; char mainpath[MAXPGPATH]; /* Skip anything that doesn't look like a relation data file. */ - if (!parse_filename_for_nontemp_relation(de->d_name, &relnumchars, - &forkNum)) + if (!parse_filename_for_nontemp_relation(de->d_name, &relNumber, + &forkNum, &segno)) continue; /* Also skip it unless this is the init fork. */ @@ -342,11 +342,12 @@ ResetUnloggedRelationsInDbspaceDir(const char *dbspacedirname, int op) continue; /* Construct main fork pathname. */ - memcpy(relnumbuf, de->d_name, relnumchars); - relnumbuf[relnumchars] = '\0'; - snprintf(mainpath, sizeof(mainpath), "%s/%s%s", - dbspacedirname, relnumbuf, de->d_name + relnumchars + 1 + - strlen(forkNames[INIT_FORKNUM])); + if (segno == 0) + snprintf(mainpath, sizeof(mainpath), "%s/%u", + dbspacedirname, relNumber); + else + snprintf(mainpath, sizeof(mainpath), "%s/%u.%u", + dbspacedirname, relNumber, segno); fsync_fname(mainpath, false); } @@ -371,52 +372,82 @@ ResetUnloggedRelationsInDbspaceDir(const char *dbspacedirname, int op) * This function returns true if the file appears to be in the correct format * for a non-temporary relation and false otherwise. * - * NB: If this function returns true, the caller is entitled to assume that - * *relnumchars has been set to a value no more than OIDCHARS, and thus - * that a buffer of OIDCHARS+1 characters is sufficient to hold the - * RelFileNumber portion of the filename. This is critical to protect against - * a possible buffer overrun. + * If it returns true, it sets *relnumber, *fork, and *segno to the values + * extracted from the filename. If it returns false, these values are set to + * InvalidRelFileNumber, InvalidForkNumber, and 0, respectively. */ bool -parse_filename_for_nontemp_relation(const char *name, int *relnumchars, - ForkNumber *fork) +parse_filename_for_nontemp_relation(const char *name, RelFileNumber *relnumber, + ForkNumber *fork, unsigned *segno) { - int pos; + unsigned long n, + s; + ForkNumber f; + char *endp; - /* Look for a non-empty string of digits (that isn't too long). */ - for (pos = 0; isdigit((unsigned char) name[pos]); ++pos) - ; - if (pos == 0 || pos > OIDCHARS) + *relnumber = InvalidRelFileNumber; + *fork = InvalidForkNumber; + *segno = 0; + + /* + * Relation filenames should begin with a digit that is not a zero. By + * rejecting cases involving leading zeroes, the caller can assume that + * there's only one possible string of characters that could have produced + * any given value for *relnumber. + * + * (To be clear, we don't expect files with names like 0017.3 to exist at + * all -- but if 0017.3 does exist, it's a non-relation file, not part of + * the main fork for relfilenode 17.) + */ + if (name[0] < '1' || name[0] > '9') + return false; + + /* + * Parse the leading digit string. If the value is out of range, we + * conclude that this isn't a relation file at all. + */ + errno = 0; + n = strtoul(name, &endp, 10); + if (errno || name == endp || n <= 0 || n > PG_UINT32_MAX) return false; - *relnumchars = pos; + name = endp; /* Check for a fork name. */ - if (name[pos] != '_') - *fork = MAIN_FORKNUM; + if (*name != '_') + f = MAIN_FORKNUM; else { int forkchar; - forkchar = forkname_chars(&name[pos + 1], fork); + forkchar = forkname_chars(name + 1, &f); if (forkchar <= 0) return false; - pos += forkchar + 1; + name += forkchar + 1; } /* Check for a segment number. */ - if (name[pos] == '.') + if (*name != '.') + s = 0; + else { - int segchar; + /* Reject leading zeroes, just like we do for RelFileNumber. */ + if (name[0] < '1' || name[0] > '9') + return false; - for (segchar = 1; isdigit((unsigned char) name[pos + segchar]); ++segchar) - ; - if (segchar <= 1) + errno = 0; + s = strtoul(name + 1, &endp, 10); + if (errno || name + 1 == endp || s <= 0 || s > PG_UINT32_MAX) return false; - pos += segchar; + name = endp; } /* Now we should be at the end. */ - if (name[pos] != '\0') + if (*name != '\0') return false; + + /* Set out parameters and return. */ + *relnumber = (RelFileNumber) n; + *fork = f; + *segno = (unsigned) s; return true; } diff --git a/src/include/storage/reinit.h b/src/include/storage/reinit.h index e2bbb5abe9..f8eb7ce234 100644 --- a/src/include/storage/reinit.h +++ b/src/include/storage/reinit.h @@ -20,8 +20,9 @@ extern void ResetUnloggedRelations(int op); extern bool parse_filename_for_nontemp_relation(const char *name, - int *relnumchars, - ForkNumber *fork); + RelFileNumber *relnumber, + ForkNumber *fork, + unsigned *segno); #define UNLOGGED_RELATION_CLEANUP 0x0001 #define UNLOGGED_RELATION_INIT 0x0002 -- 2.37.1 (Apple Git-137.1) [application/octet-stream] v4-0003-Change-how-a-base-backup-decides-which-files-have.patch (11.4K, ../../CA+TgmoYw=8CnKEPQX9qk2estSTCaB7OGcnVM8oa-hAePG2O4Yw@mail.gmail.com/4-v4-0003-Change-how-a-base-backup-decides-which-files-have.patch) download | inline diff: From 3a77a61d05febc358b6b4a3b545bc9b56a55d0c2 Mon Sep 17 00:00:00 2001 From: Robert Haas <[email protected]> Date: Tue, 3 Oct 2023 13:32:28 -0400 Subject: [PATCH v4 3/6] Change how a base backup decides which files have checksums. Previously, it thought that any plain file located under global, base, or a tablespace directory had checksums unless it was in a short list of excluded files. Now, it thinks that files in those directories have checksums if parse_filename_for_nontemp_relation says that they are relation files. (Temporary relation files don't matter because they're excluded from the backup anyway.) This changes the behavior if you have stray files not managed by PostgreSQL in the relevant directories. Previously, you'd get some kind of checksum-related complaint if such files existed, assuming that the cluster had checksums enabled and that the base backup wasn't run with NOVERIFY_CHECKSUMS. Now, you won't get those complaints any more. That seems like an improvement to me, because those files were presumably not created by PostgreSQL and so there is no reason to think that they would be checksummed like a PostgreSQL relation file. (If we want to complain about such files, we should complain about them existing at all, not just about their checksums.) The point of this change is to make the code more consistent. sendDir() was already calling parse_filename_for_nontemp_relation() as part of an effort to determine which files to include in the backup. So, it already had the information about whether a certain file was a relation file. sendFile() then used a separate method, embodied in is_checksummed_file(), to make what is essentially the same determination. It's better not to make the same decision using two different methods, especially in closely-related code. --- src/backend/backup/basebackup.c | 172 ++++++++++---------------------- 1 file changed, 55 insertions(+), 117 deletions(-) diff --git a/src/backend/backup/basebackup.c b/src/backend/backup/basebackup.c index b537f46219..4ba63ad8a6 100644 --- a/src/backend/backup/basebackup.c +++ b/src/backend/backup/basebackup.c @@ -82,7 +82,8 @@ static int64 sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeo backup_manifest_info *manifest, Oid spcoid); static bool sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, struct stat *statbuf, bool missing_ok, - Oid dboid, Oid spcoid, + Oid dboid, Oid spcoid, RelFileNumber relfilenumber, + unsigned segno, backup_manifest_info *manifest); static off_t read_file_data_into_buffer(bbsink *sink, const char *readfilename, int fd, @@ -104,7 +105,6 @@ static void convert_link_to_directory(const char *pathbuf, struct stat *statbuf) static void perform_base_backup(basebackup_options *opt, bbsink *sink); static void parse_basebackup_options(List *options, basebackup_options *opt); static int compareWalFileNames(const ListCell *a, const ListCell *b); -static bool is_checksummed_file(const char *fullpath, const char *filename); static int basebackup_read_file(int fd, char *buf, size_t nbytes, off_t offset, const char *filename, bool partial_read_ok); @@ -213,23 +213,6 @@ static const struct exclude_list_item excludeFiles[] = {NULL, false} }; -/* - * List of files excluded from checksum validation. - * - * Note: this list should be kept in sync with what pg_checksums.c - * includes. - */ -static const struct exclude_list_item noChecksumFiles[] = { - {"pg_control", false}, - {"pg_filenode.map", false}, - {"pg_internal.init", true}, - {"PG_VERSION", false}, -#ifdef EXEC_BACKEND - {"config_exec_params", true}, -#endif - {NULL, false} -}; - /* * Actually do a base backup for the specified tablespaces. * @@ -356,7 +339,8 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) errmsg("could not stat file \"%s\": %m", XLOG_CONTROL_FILE))); sendFile(sink, XLOG_CONTROL_FILE, XLOG_CONTROL_FILE, &statbuf, - false, InvalidOid, InvalidOid, &manifest); + false, InvalidOid, InvalidOid, + InvalidRelFileNumber, 0, &manifest); } else { @@ -625,7 +609,8 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) errmsg("could not stat file \"%s\": %m", pathbuf))); sendFile(sink, pathbuf, pathbuf, &statbuf, false, - InvalidOid, InvalidOid, &manifest); + InvalidOid, InvalidOid, InvalidRelFileNumber, 0, + &manifest); /* unconditionally mark file as archived */ StatusFilePath(pathbuf, fname, ".done"); @@ -1163,7 +1148,8 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, struct stat statbuf; int64 size = 0; const char *lastDir; /* Split last dir from parent path. */ - bool isDbDir = false; /* Does this directory contain relations? */ + bool isRelationDir = false; /* Does directory contain relations? */ + Oid dboid = InvalidOid; /* * Determine if the current path is a database directory that can contain @@ -1190,17 +1176,23 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, strncmp(lastDir - (sizeof(TABLESPACE_VERSION_DIRECTORY) - 1), TABLESPACE_VERSION_DIRECTORY, sizeof(TABLESPACE_VERSION_DIRECTORY) - 1) == 0)) - isDbDir = true; + { + isRelationDir = true; + dboid = atooid(lastDir + 1); + } } + else if (strcmp(path, "./global") == 0) + isRelationDir = true; dir = AllocateDir(path); while ((de = ReadDir(dir, path)) != NULL) { int excludeIdx; bool excludeFound; - RelFileNumber relNumber; - ForkNumber relForkNum; - unsigned segno; + RelFileNumber relfilenumber = InvalidRelFileNumber; + ForkNumber relForkNum = InvalidForkNumber; + unsigned segno = 0; + bool isRelationFile = false; /* Skip special stuff */ if (strcmp(de->d_name, ".") == 0 || strcmp(de->d_name, "..") == 0) @@ -1248,37 +1240,40 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, if (excludeFound) continue; + /* + * If there could be non-temporary relation files in this directory, + * try to parse the filename. + */ + if (isRelationDir) + isRelationFile = + parse_filename_for_nontemp_relation(de->d_name, + &relfilenumber, + &relForkNum, &segno); + /* Exclude all forks for unlogged tables except the init fork */ - if (isDbDir && - parse_filename_for_nontemp_relation(de->d_name, &relNumber, - &relForkNum, &segno)) + if (isRelationFile && relForkNum != INIT_FORKNUM) { - /* Never exclude init forks */ - if (relForkNum != INIT_FORKNUM) - { - char initForkFile[MAXPGPATH]; + char initForkFile[MAXPGPATH]; - /* - * If any other type of fork, check if there is an init fork - * with the same RelFileNumber. If so, the file can be - * excluded. - */ - snprintf(initForkFile, sizeof(initForkFile), "%s/%u_init", - path, relNumber); + /* + * If any other type of fork, check if there is an init fork with + * the same RelFileNumber. If so, the file can be excluded. + */ + snprintf(initForkFile, sizeof(initForkFile), "%s/%u_init", + path, relfilenumber); - if (lstat(initForkFile, &statbuf) == 0) - { - elog(DEBUG2, - "unlogged relation file \"%s\" excluded from backup", - de->d_name); + if (lstat(initForkFile, &statbuf) == 0) + { + elog(DEBUG2, + "unlogged relation file \"%s\" excluded from backup", + de->d_name); - continue; - } + continue; } } /* Exclude temporary relations */ - if (isDbDir && looks_like_temp_rel_name(de->d_name)) + if (OidIsValid(dboid) && looks_like_temp_rel_name(de->d_name)) { elog(DEBUG2, "temporary relation file \"%s\" excluded from backup", @@ -1417,8 +1412,8 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, if (!sizeonly) sent = sendFile(sink, pathbuf, pathbuf + basepathlen + 1, &statbuf, - true, isDbDir ? atooid(lastDir + 1) : InvalidOid, spcoid, - manifest); + true, dboid, spcoid, + relfilenumber, segno, manifest); if (sent || sizeonly) { @@ -1440,40 +1435,6 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, return size; } -/* - * Check if a file should have its checksum validated. - * We validate checksums on files in regular tablespaces - * (including global and default) only, and in those there - * are some files that are explicitly excluded. - */ -static bool -is_checksummed_file(const char *fullpath, const char *filename) -{ - /* Check that the file is in a tablespace */ - if (strncmp(fullpath, "./global/", 9) == 0 || - strncmp(fullpath, "./base/", 7) == 0 || - strncmp(fullpath, "/", 1) == 0) - { - int excludeIdx; - - /* Compare file against noChecksumFiles skip list */ - for (excludeIdx = 0; noChecksumFiles[excludeIdx].name != NULL; excludeIdx++) - { - int cmplen = strlen(noChecksumFiles[excludeIdx].name); - - if (!noChecksumFiles[excludeIdx].match_prefix) - cmplen++; - if (strncmp(filename, noChecksumFiles[excludeIdx].name, - cmplen) == 0) - return false; - } - - return true; - } - else - return false; -} - /* * Given the member, write the TAR header & send the file. * @@ -1488,6 +1449,7 @@ is_checksummed_file(const char *fullpath, const char *filename) static bool sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, struct stat *statbuf, bool missing_ok, Oid dboid, Oid spcoid, + RelFileNumber relfilenumber, unsigned segno, backup_manifest_info *manifest) { int fd; @@ -1495,8 +1457,6 @@ sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, int checksum_failures = 0; off_t cnt; pgoff_t bytes_done = 0; - int segmentno = 0; - char *segmentpath; bool verify_checksum = false; pg_checksum_context checksum_ctx; @@ -1522,36 +1482,14 @@ sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, */ Assert((sink->bbs_buffer_length % BLCKSZ) == 0); - if (!noverify_checksums && DataChecksumsEnabled()) - { - char *filename; - - /* - * Get the filename (excluding path). As last_dir_separator() - * includes the last directory separator, we chop that off by - * incrementing the pointer. - */ - filename = last_dir_separator(readfilename) + 1; - - if (is_checksummed_file(readfilename, filename)) - { - verify_checksum = true; - - /* - * Cut off at the segment boundary (".") to get the segment number - * in order to mix it into the checksum. - */ - segmentpath = strstr(filename, "."); - if (segmentpath != NULL) - { - segmentno = atoi(segmentpath + 1); - if (segmentno == 0) - ereport(ERROR, - (errmsg("invalid segment number %d in file \"%s\"", - segmentno, filename))); - } - } - } + /* + * If we weren't told not to verify checksums, and if checksums are + * enabled for this cluster, and if this is a relation file, then verify + * the checksum. + */ + if (!noverify_checksums && DataChecksumsEnabled() && + RelFileNumberIsValid(relfilenumber)) + verify_checksum = true; /* * Loop until we read the amount of data the caller told us to expect. The @@ -1566,7 +1504,7 @@ sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, /* Try to read some more data. */ cnt = read_file_data_into_buffer(sink, readfilename, fd, bytes_done, remaining, - blkno + segmentno * RELSEG_SIZE, + blkno + segno * RELSEG_SIZE, verify_checksum, &checksum_failures); -- 2.37.1 (Apple Git-137.1) [application/octet-stream] v4-0004-Move-src-bin-pg_verifybackup-parse_manifest.c-int.patch (4.3K, ../../CA+TgmoYw=8CnKEPQX9qk2estSTCaB7OGcnVM8oa-hAePG2O4Yw@mail.gmail.com/5-v4-0004-Move-src-bin-pg_verifybackup-parse_manifest.c-int.patch) download | inline diff: From 89ec93d59cb6a2a9b9239803dc1ad818fd9cdf93 Mon Sep 17 00:00:00 2001 From: Robert Haas <[email protected]> Date: Tue, 3 Oct 2023 13:32:45 -0400 Subject: [PATCH v4 4/6] Move src/bin/pg_verifybackup/parse_manifest.c into src/common. This makes it possible for the code to be easily reused by other client-side tools, and/or by the server. --- src/bin/pg_verifybackup/Makefile | 1 - src/bin/pg_verifybackup/meson.build | 1 - src/bin/pg_verifybackup/pg_verifybackup.c | 2 +- src/common/Makefile | 1 + src/common/meson.build | 1 + src/{bin/pg_verifybackup => common}/parse_manifest.c | 4 ++-- src/{bin/pg_verifybackup => include/common}/parse_manifest.h | 2 +- 7 files changed, 6 insertions(+), 6 deletions(-) rename src/{bin/pg_verifybackup => common}/parse_manifest.c (99%) rename src/{bin/pg_verifybackup => include/common}/parse_manifest.h (97%) diff --git a/src/bin/pg_verifybackup/Makefile b/src/bin/pg_verifybackup/Makefile index 596df15118..8f04fa662c 100644 --- a/src/bin/pg_verifybackup/Makefile +++ b/src/bin/pg_verifybackup/Makefile @@ -21,7 +21,6 @@ LDFLAGS_INTERNAL += -L$(top_builddir)/src/fe_utils -lpgfeutils $(libpq_pgport) OBJS = \ $(WIN32RES) \ - parse_manifest.o \ pg_verifybackup.o all: pg_verifybackup diff --git a/src/bin/pg_verifybackup/meson.build b/src/bin/pg_verifybackup/meson.build index 9369da1bc6..58f780d1a6 100644 --- a/src/bin/pg_verifybackup/meson.build +++ b/src/bin/pg_verifybackup/meson.build @@ -1,7 +1,6 @@ # Copyright (c) 2022-2023, PostgreSQL Global Development Group pg_verifybackup_sources = files( - 'parse_manifest.c', 'pg_verifybackup.c' ) diff --git a/src/bin/pg_verifybackup/pg_verifybackup.c b/src/bin/pg_verifybackup/pg_verifybackup.c index 059836f0e6..ce423a03d4 100644 --- a/src/bin/pg_verifybackup/pg_verifybackup.c +++ b/src/bin/pg_verifybackup/pg_verifybackup.c @@ -20,9 +20,9 @@ #include "common/hashfn.h" #include "common/logging.h" +#include "common/parse_manifest.h" #include "fe_utils/simple_list.h" #include "getopt_long.h" -#include "parse_manifest.h" #include "pgtime.h" /* diff --git a/src/common/Makefile b/src/common/Makefile index cc5c54dcee..ff60666f5c 100644 --- a/src/common/Makefile +++ b/src/common/Makefile @@ -66,6 +66,7 @@ OBJS_COMMON = \ kwlookup.o \ link-canary.o \ md5_common.o \ + parse_manifest.o \ percentrepl.o \ pg_get_line.o \ pg_lzcompress.o \ diff --git a/src/common/meson.build b/src/common/meson.build index 3b97497d1a..fcc0c4fe8d 100644 --- a/src/common/meson.build +++ b/src/common/meson.build @@ -18,6 +18,7 @@ common_sources = files( 'kwlookup.c', 'link-canary.c', 'md5_common.c', + 'parse_manifest.c', 'percentrepl.c', 'pg_get_line.c', 'pg_lzcompress.c', diff --git a/src/bin/pg_verifybackup/parse_manifest.c b/src/common/parse_manifest.c similarity index 99% rename from src/bin/pg_verifybackup/parse_manifest.c rename to src/common/parse_manifest.c index 2379f7be7b..672e8bcf25 100644 --- a/src/bin/pg_verifybackup/parse_manifest.c +++ b/src/common/parse_manifest.c @@ -6,15 +6,15 @@ * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group * Portions Copyright (c) 1994, Regents of the University of California * - * src/bin/pg_verifybackup/parse_manifest.c + * src/common/parse_manifest.c * *------------------------------------------------------------------------- */ #include "postgres_fe.h" -#include "parse_manifest.h" #include "common/jsonapi.h" +#include "common/parse_manifest.h" /* * Semantic states for JSON manifest parsing. diff --git a/src/bin/pg_verifybackup/parse_manifest.h b/src/include/common/parse_manifest.h similarity index 97% rename from src/bin/pg_verifybackup/parse_manifest.h rename to src/include/common/parse_manifest.h index 7387a917a2..7b24c5d785 100644 --- a/src/bin/pg_verifybackup/parse_manifest.h +++ b/src/include/common/parse_manifest.h @@ -6,7 +6,7 @@ * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group * Portions Copyright (c) 1994, Regents of the University of California * - * src/bin/pg_verifybackup/parse_manifest.h + * src/include/common/parse_manifest.h * *------------------------------------------------------------------------- */ -- 2.37.1 (Apple Git-137.1) [application/octet-stream] v4-0005-Prototype-patch-for-incremental-and-differential-.patch (329.6K, ../../CA+TgmoYw=8CnKEPQX9qk2estSTCaB7OGcnVM8oa-hAePG2O4Yw@mail.gmail.com/6-v4-0005-Prototype-patch-for-incremental-and-differential-.patch) download | inline diff: From 9dbf7217076ddfefdf8cc80a2c42ac6c8b22044f Mon Sep 17 00:00:00 2001 From: Robert Haas <[email protected]> Date: Wed, 14 Jun 2023 11:31:29 -0400 Subject: [PATCH v4 5/6] Prototype patch for incremental and differential backup. We don't differentiate between incremental and differential backups; the term "incremental" as used herein means "either incremental or differential". This adds a new background process, the WAL summarizer, whose behavor is governed by new GUCs wal_summarize_mb and wal_summarize_keep_time. This writes out WAL summary files to $PGDATA/pg_wal/summaries. Each summary file contains information for a certain range of LSNs on a certain TLI. For each relation, it stores a "limit block" which is 0 if a relation is created or destroyed within a certain range of WAL records, or otherwise the shortest length to which the relation was truncated during that range of WAL records, or otherwise InvalidBlockNumber. In addition, it stores any blocks which have been modified during that range of WAL records, but excluding blocks which were removed by truncation after they were modified and which were never modified thereafter. In other words, it tells us which blocks need to copied in case of an incremental backup covering that range of WAL records. To take an incremental backup, you use the new replication command UPLOAD_MANIFEST to upload the manifest for the prior backup. This prior backup could either be a full backup or another incremental backup. You then use BASE_BACKUP with the INCREMENTAL option to take the backup. pg_basebackup now has an --incremental=PATH_TO_MANIFEST option to trigger this behavior. An incremental backup is like a regular full backup except that some relation files are replaced with files with names like INCREMENTAL.${ORIGINAL_NAME}, and the backup_label file contains additional lines identifying it as an incremental backup. The new pg_combinebackup tool can be used to reconstruct a data directory from a full backup and a series of incremental backups. XXX. It would be nice if we could do something about incremental JSON parsing. XXX. This might need more work on documentation and tests. Patch by me. Thanks to Dilip Kumar and Andres Freund for some helpful design discussions. Reviewed by Dilip Kumar and Jakub Wartak. --- doc/src/sgml/ref/allfiles.sgml | 1 + doc/src/sgml/ref/pg_basebackup.sgml | 42 +- doc/src/sgml/ref/pg_combinebackup.sgml | 227 +++ doc/src/sgml/reference.sgml | 1 + src/backend/access/transam/xlog.c | 93 +- src/backend/access/transam/xlogbackup.c | 10 + src/backend/access/transam/xlogrecovery.c | 6 + src/backend/backup/Makefile | 5 +- src/backend/backup/basebackup.c | 334 +++- src/backend/backup/basebackup_incremental.c | 873 ++++++++++ src/backend/backup/meson.build | 3 + src/backend/backup/walsummary.c | 356 +++++ src/backend/backup/walsummaryfuncs.c | 169 ++ src/backend/postmaster/Makefile | 1 + src/backend/postmaster/auxprocess.c | 8 + src/backend/postmaster/meson.build | 1 + src/backend/postmaster/postmaster.c | 53 + src/backend/postmaster/walsummarizer.c | 1414 +++++++++++++++++ src/backend/replication/repl_gram.y | 14 +- src/backend/replication/repl_scanner.l | 2 + src/backend/replication/walsender.c | 162 +- src/backend/storage/ipc/ipci.c | 3 + src/backend/storage/lmgr/lwlocknames.txt | 1 + src/backend/utils/activity/pgstat_io.c | 4 +- .../utils/activity/wait_event_names.txt | 5 + src/backend/utils/init/miscinit.c | 3 + src/backend/utils/misc/guc_tables.c | 29 + src/backend/utils/misc/postgresql.conf.sample | 5 + src/bin/Makefile | 1 + src/bin/initdb/initdb.c | 1 + src/bin/meson.build | 1 + src/bin/pg_basebackup/bbstreamer_file.c | 1 + src/bin/pg_basebackup/pg_basebackup.c | 110 +- src/bin/pg_basebackup/t/010_pg_basebackup.pl | 4 +- src/bin/pg_combinebackup/.gitignore | 1 + src/bin/pg_combinebackup/Makefile | 52 + src/bin/pg_combinebackup/backup_label.c | 281 ++++ src/bin/pg_combinebackup/backup_label.h | 29 + src/bin/pg_combinebackup/copy_file.c | 169 ++ src/bin/pg_combinebackup/copy_file.h | 19 + src/bin/pg_combinebackup/load_manifest.c | 245 +++ src/bin/pg_combinebackup/load_manifest.h | 67 + src/bin/pg_combinebackup/meson.build | 35 + src/bin/pg_combinebackup/pg_combinebackup.c | 1270 +++++++++++++++ src/bin/pg_combinebackup/reconstruct.c | 618 +++++++ src/bin/pg_combinebackup/reconstruct.h | 32 + src/bin/pg_combinebackup/t/001_basic.pl | 23 + .../pg_combinebackup/t/002_compare_backups.pl | 154 ++ src/bin/pg_combinebackup/write_manifest.c | 293 ++++ src/bin/pg_combinebackup/write_manifest.h | 33 + src/bin/pg_resetwal/pg_resetwal.c | 36 + src/common/Makefile | 1 + src/common/blkreftable.c | 1309 +++++++++++++++ src/common/meson.build | 1 + src/include/access/xlog.h | 1 + src/include/access/xlogbackup.h | 2 + src/include/backup/basebackup.h | 5 +- src/include/backup/basebackup_incremental.h | 56 + src/include/backup/walsummary.h | 49 + src/include/catalog/pg_proc.dat | 19 + src/include/common/blkreftable.h | 120 ++ src/include/miscadmin.h | 3 + src/include/nodes/replnodes.h | 9 + src/include/postmaster/walsummarizer.h | 31 + src/include/storage/proc.h | 9 +- src/include/utils/guc_tables.h | 1 + src/test/perl/PostgreSQL/Test/Cluster.pm | 21 +- src/test/recovery/t/001_stream_rep.pl | 2 + src/test/recovery/t/019_replslot_limit.pl | 3 + .../t/035_standby_logical_decoding.pl | 1 + src/tools/pgindent/typedefs.list | 23 + 71 files changed, 8899 insertions(+), 67 deletions(-) create mode 100644 doc/src/sgml/ref/pg_combinebackup.sgml create mode 100644 src/backend/backup/basebackup_incremental.c create mode 100644 src/backend/backup/walsummary.c create mode 100644 src/backend/backup/walsummaryfuncs.c create mode 100644 src/backend/postmaster/walsummarizer.c create mode 100644 src/bin/pg_combinebackup/.gitignore create mode 100644 src/bin/pg_combinebackup/Makefile create mode 100644 src/bin/pg_combinebackup/backup_label.c create mode 100644 src/bin/pg_combinebackup/backup_label.h create mode 100644 src/bin/pg_combinebackup/copy_file.c create mode 100644 src/bin/pg_combinebackup/copy_file.h create mode 100644 src/bin/pg_combinebackup/load_manifest.c create mode 100644 src/bin/pg_combinebackup/load_manifest.h create mode 100644 src/bin/pg_combinebackup/meson.build create mode 100644 src/bin/pg_combinebackup/pg_combinebackup.c create mode 100644 src/bin/pg_combinebackup/reconstruct.c create mode 100644 src/bin/pg_combinebackup/reconstruct.h create mode 100644 src/bin/pg_combinebackup/t/001_basic.pl create mode 100644 src/bin/pg_combinebackup/t/002_compare_backups.pl create mode 100644 src/bin/pg_combinebackup/write_manifest.c create mode 100644 src/bin/pg_combinebackup/write_manifest.h create mode 100644 src/common/blkreftable.c create mode 100644 src/include/backup/basebackup_incremental.h create mode 100644 src/include/backup/walsummary.h create mode 100644 src/include/common/blkreftable.h create mode 100644 src/include/postmaster/walsummarizer.h diff --git a/doc/src/sgml/ref/allfiles.sgml b/doc/src/sgml/ref/allfiles.sgml index 54b5f22d6e..fda4690eab 100644 --- a/doc/src/sgml/ref/allfiles.sgml +++ b/doc/src/sgml/ref/allfiles.sgml @@ -202,6 +202,7 @@ Complete list of usable sgml source files in this directory. <!ENTITY pgBasebackup SYSTEM "pg_basebackup.sgml"> <!ENTITY pgbench SYSTEM "pgbench.sgml"> <!ENTITY pgChecksums SYSTEM "pg_checksums.sgml"> +<!ENTITY pgCombinebackup SYSTEM "pg_combinebackup.sgml"> <!ENTITY pgConfig SYSTEM "pg_config-ref.sgml"> <!ENTITY pgControldata SYSTEM "pg_controldata.sgml"> <!ENTITY pgCtl SYSTEM "pg_ctl-ref.sgml"> diff --git a/doc/src/sgml/ref/pg_basebackup.sgml b/doc/src/sgml/ref/pg_basebackup.sgml index 344de921e4..3a569069ec 100644 --- a/doc/src/sgml/ref/pg_basebackup.sgml +++ b/doc/src/sgml/ref/pg_basebackup.sgml @@ -38,11 +38,22 @@ PostgreSQL documentation </para> <para> - <application>pg_basebackup</application> makes an exact copy of the database - cluster's files, while making sure the server is put into and - out of backup mode automatically. Backups are always taken of the entire - database cluster; it is not possible to back up individual databases or - database objects. For selective backups, another tool such as + <application>pg_basebackup</application> can take a full, incremental, + or differential backup of the database. When used to take a full backup, it + makes an exact copy of the database cluster's files. When used to take an + incremental or differential backup, some files that would have been part of + a full backup may be replaced with incremental versions of the same files, + containing only those blocks that have been modified since the reference + backup. An incremental or differential backup cannot be used directly; + instead, <xref linkend="app-pgcombinebackup"/> must first be used to combine + it with the previous backups upon which it depends. + </para> + + <para> + In any mode, <application>pg_basebackup</application> makes sure the server + is put into and out of backup mode automatically. Backups are always taken of + the entire database cluster; it is not possible to back up individual + databases or database objects. For selective backups, another tool such as <xref linkend="app-pgdump"/> must be used. </para> @@ -197,6 +208,25 @@ PostgreSQL documentation </listitem> </varlistentry> + <varlistentry> + <term><option>-i <replaceable class="parameter">old_manifest_file</replaceable></option></term> + <term><option>--incremental=<replaceable class="parameter">old_meanifest_file</replaceable></option></term> + <listitem> + <para> + Performs an incremental or differential backup. The backup manifest + for the reference backup must be provided, and will be uploaded to the + server, which will respond by sending the requested incremental or + differential backup. There is no real difference between the two: + an incremental backup is simply a backup where the reference backup is + a full backup, and a differential backup is one where the reference + backup is an incremental or differential backup. Either way, + the backup cannot be used directly; instead, + <xref linkend="app-pgcombinebackup"/> must be used to reconstruct a + full backup. + </para> + </listitem> + </varlistentry> + <varlistentry> <term><option>-R</option></term> <term><option>--write-recovery-conf</option></term> @@ -595,7 +625,7 @@ PostgreSQL documentation </varlistentry> <varlistentry> - <term><option>--sync-method</option></term> + <term><option>--sync-method=<replaceable class="parameter">method</replaceable></option></term> <listitem> <para> When set to <literal>fsync</literal>, which is the default, diff --git a/doc/src/sgml/ref/pg_combinebackup.sgml b/doc/src/sgml/ref/pg_combinebackup.sgml new file mode 100644 index 0000000000..626b1b13dd --- /dev/null +++ b/doc/src/sgml/ref/pg_combinebackup.sgml @@ -0,0 +1,227 @@ +<!-- +doc/src/sgml/ref/pg_combinebackup.sgml +PostgreSQL documentation +--> + +<refentry id="app-pgcombinebackup"> + <indexterm zone="app-pgcombinebackup"> + <primary>pg_combinebackup</primary> + </indexterm> + + <refmeta> + <refentrytitle><application>pg_combinebackup</application></refentrytitle> + <manvolnum>1</manvolnum> + <refmiscinfo>Application</refmiscinfo> + </refmeta> + + <refnamediv> + <refname>pg_combinebackup</refname> + <refpurpose>reconstruct a full backup from an incremental or differential backup and dependent backups</refpurpose> + </refnamediv> + + <refsynopsisdiv> + <cmdsynopsis> + <command>pg_combinebackup</command> + <arg rep="repeat"><replaceable>option</replaceable></arg> + <arg rep="repeat"><replaceable>backup_directory</replaceable></arg> + </cmdsynopsis> + </refsynopsisdiv> + + <refsect1> + <title>Description</title> + <para> + <application>pg_combinebackup</application> is used to reconstruct a + synthetic full backup from an incremental or differential backup and the + earlier backups upon which it depends. + </para> + + <para> + Specify all of the required backups on the command line from oldest to newest. + That is, the first backup directory should be the path to the full backup, and + the last should be the path to the final incremental or differential backup + that you wish to restore. The reconstructed backup will be written to the + output directory specified by the <option>-o</option> option. + </para> + + <para> + Although <application>pg_combinebackup</application> will attempt to verify + that the backups you specify form a legal backup chain from which a correct + full backup can be reconstructed, it is not designed to help you keep track + of which backups depend on which other backups. If you remove the one or + more of the previous backups upon which your incremental or differential + backup relies, you will not be able to restore it. + </para> + + <para> + Since the output of <application>pg_combinebackup</application> is a + synthetic full backup, it can be used as an input to a future invocation of + <application>pg_combinebackup</application>. The synthetic full backup would + be specified on the command line in lieu of the chain of backups from which + it was reconstructed. + </para> + </refsect1> + + <refsect1> + <title>Options</title> + + <para> + <variablelist> + <varlistentry> + <term><option>-d</option></term> + <term><option>--debug</option></term> + <listitem> + <para> + Print lots of debug logging output on <filename>stderr</filename>. + </para> + </listitem> + </varlistentry> + + <varlistentry> + <term><option>-T <replaceable class="parameter">olddir</replaceable>=<replaceable class="parameter">newdir</replaceable></option></term> + <term><option>--tablespace-mapping=<replaceable class="parameter">olddir</replaceable>=<replaceable class="parameter">newdir</replaceable></option></term> + <listitem> + <para> + Relocates the tablespace in directory <replaceable>olddir</replaceable> + to <replaceable>newdir</replaceable> during the backup. + <replaceable>olddir</replaceable> is the absolute path of the tablespace + as it exists in the first backup specified on the command line, + and <replaceable>newdir</replaceable> is the absolute path to use for the + tablespace in the reconstructed backup. If either path needs to contain + an equal sign (<literal>=</literal>), precede that with a backslash. + This option can be specified multiple times for multiple tablespaces. + </para> + </listitem> + </varlistentry> + + <varlistentry> + <term><option>-N</option></term> + <term><option>--no-sync</option></term> + <listitem> + <para> + By default, <command>pg_combinebackup</command> will wait for all files + to be written safely to disk. This option causes + <command>pg_combinebackup</command> to return without waiting, which is + faster, but means that a subsequent operating system crash can leave + the output backup corrupt. Generally, this option is useful for testing + but should not be used when creating a production installation. + </para> + </listitem> + </varlistentry> + + <varlistentry> + <term><option>-o <replaceable class="parameter">outputdir</replaceable></option></term> + <term><option>--output=<replaceable class="parameter">outputdir</replaceable></option></term> + <listitem> + <para> + Specifies the output directory to which the synthetic full backup + should be written. Currently, this argument is required. + </para> + </listitem> + </varlistentry> + + <varlistentry> + <term><option>--sync-method</option></term> + <listitem> + <para> + When set to <literal>fsync</literal>, which is the default, + <command>pg_combinebackup</command> will recursively open and synchronize + all files in the backup directory. When the plain format is used, the + search for files will follow symbolic links for the WAL directory and + each configured tablespace. + </para> + <para> + On Linux, <literal>syncfs</literal> may be used instead to ask the + operating system to synchronize the whole file system that contains the + backup directory. When the plain format is used, + <command>pg_combinebackup</command> will also synchronize the file systems + that contain the WAL files and each tablespace. See + <xref linkend="syncfs"/> for more information about using + <function>syncfs()</function>. + </para> + <para> + This option has no effect when <option>--no-sync</option> is used. + </para> + </listitem> + </varlistentry> + + <varlistentry> + <term><option>--manifest-checksums=<replaceable class="parameter">algorithm</replaceable></option></term> + <listitem> + <para> + Like <xref linkend="app-pgbasebackup"/>, + <application>pg_combinebackup</application> writes a backup manifest + in the output directory. This option specifies the checksum algorithm + that should be applied to each file included in the backup manifest. + Currently, the available algorithms are <literal>NONE</literal>, + <literal>CRC32C</literal>, <literal>SHA224</literal>, + <literal>SHA256</literal>, <literal>SHA384</literal>, + and <literal>SHA512</literal>. The default is <literal>CRC32C</literal>. + </para> + </listitem> + </varlistentry> + + <varlistentry> + <term><option>--no-manifest</option></term> + <listitem> + <para> + Disables generation of a backup manifest. If this option is not + specified, a backup manifest for the reconstructed backup will be + written to the output directory. + </para> + </listitem> + </varlistentry> + </variablelist> + + <variablelist> + <varlistentry> + <term><option>-V</option></term> + <term><option>--version</option></term> + <listitem> + <para> + Prints the <application>pg_combinebackup</application> version and exits. + </para> + </listitem> + </varlistentry> + + <varlistentry> + <term><option>-?</option></term> + <term><option>--help</option></term> + <listitem> + <para> + Shows help about <application>pg_combinebackup</application> command line + arguments, and exits. + </para> + </listitem> + </varlistentry> + + </variablelist> + </para> + + </refsect1> + + <refsect1> + <title>Environment</title> + + <para> + This utility, like most other <productname>PostgreSQL</productname> utilities, + uses the environment variables supported by <application>libpq</application> + (see <xref linkend="libpq-envars"/>). + </para> + + <para> + The environment variable <envar>PG_COLOR</envar> specifies whether to use + color in diagnostic messages. Possible values are + <literal>always</literal>, <literal>auto</literal> and + <literal>never</literal>. + </para> + </refsect1> + + <refsect1> + <title>See Also</title> + + <simplelist type="inline"> + <member><xref linkend="app-pgbasebackup"/></member> + </simplelist> + </refsect1> + +</refentry> diff --git a/doc/src/sgml/reference.sgml b/doc/src/sgml/reference.sgml index e11b4b6130..a07d2b5e01 100644 --- a/doc/src/sgml/reference.sgml +++ b/doc/src/sgml/reference.sgml @@ -250,6 +250,7 @@ &pgamcheck; &pgBasebackup; &pgbench; + &pgCombinebackup; &pgConfig; &pgDump; &pgDumpall; diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 677a5bf51b..6cfeee63e8 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -77,6 +77,7 @@ #include "port/pg_iovec.h" #include "postmaster/bgwriter.h" #include "postmaster/startup.h" +#include "postmaster/walsummarizer.h" #include "postmaster/walwriter.h" #include "replication/logical.h" #include "replication/origin.h" @@ -3499,6 +3500,43 @@ XLogGetLastRemovedSegno(void) return lastRemovedSegNo; } +/* + * Return the oldest WAL segment on the given TLI that still exists in + * XLOGDIR, or 0 if none. + */ +XLogSegNo +XLogGetOldestSegno(TimeLineID tli) +{ + DIR *xldir; + struct dirent *xlde; + XLogSegNo oldest_segno = 0; + + xldir = AllocateDir(XLOGDIR); + while ((xlde = ReadDir(xldir, XLOGDIR)) != NULL) + { + TimeLineID file_tli; + XLogSegNo file_segno; + + /* Ignore files that are not XLOG segments */ + if (!IsXLogFileName(xlde->d_name)) + continue; + + /* Parse filename to get TLI and segno. */ + XLogFromFileName(xlde->d_name, &file_tli, &file_segno, + wal_segment_size); + + /* Ignore anything that's not from the TLI of interest. */ + if (tli != file_tli) + continue; + + /* If it's the oldest so far, update oldest_segno. */ + if (oldest_segno == 0 || file_segno < oldest_segno) + oldest_segno = file_segno; + } + + FreeDir(xldir); + return oldest_segno; +} /* * Update the last removed segno pointer in shared memory, to reflect that the @@ -3778,8 +3816,8 @@ RemoveXlogFile(const struct dirent *segment_de, } /* - * Verify whether pg_wal and pg_wal/archive_status exist. - * If the latter does not exist, recreate it. + * Verify whether pg_wal, pg_wal/archive_status, and pg_wal/summaries exist. + * If the latter do not exist, recreate them. * * It is not the goal of this function to verify the contents of these * directories, but to help in cases where someone has performed a cluster @@ -3822,6 +3860,26 @@ ValidateXLOGDirectoryStructure(void) (errmsg("could not create missing directory \"%s\": %m", path))); } + + /* Check for summaries */ + snprintf(path, MAXPGPATH, XLOGDIR "/summaries"); + if (stat(path, &stat_buf) == 0) + { + /* Check for weird cases where it exists but isn't a directory */ + if (!S_ISDIR(stat_buf.st_mode)) + ereport(FATAL, + (errmsg("required WAL directory \"%s\" does not exist", + path))); + } + else + { + ereport(LOG, + (errmsg("creating missing WAL directory \"%s\"", path))); + if (MakePGDirectory(path) < 0) + ereport(FATAL, + (errmsg("could not create missing directory \"%s\": %m", + path))); + } } /* @@ -5146,9 +5204,9 @@ StartupXLOG(void) #endif /* - * Verify that pg_wal and pg_wal/archive_status exist. In cases where - * someone has performed a copy for PITR, these directories may have been - * excluded and need to be re-created. + * Verify that pg_wal, pg_wal/archive_status, and pg_wal/summaries exist. + * In cases where someone has performed a copy for PITR, these directories + * may have been excluded and need to be re-created. */ ValidateXLOGDirectoryStructure(); @@ -6829,6 +6887,17 @@ CreateCheckPoint(int flags) */ END_CRIT_SECTION(); + /* + * If there hasn't been much system activity in a while, the WAL + * summarizer may be sleeping for relatively long periods, which could + * delay an incremental backup that has started concurrently. In the hopes + * of avoiding that, poke the WAL summarizer here. + * + * Possibly this should instead be done at some earlier point in this + * function, but it's not clear that it matters much. + */ + SetWalSummarizerLatch(); + /* * Let smgr do post-checkpoint cleanup (eg, deleting old files). */ @@ -7503,6 +7572,20 @@ KeepLogSeg(XLogRecPtr recptr, XLogSegNo *logSegNo) } } + /* + * If WAL summarization is in use, don't remove WAL that has yet to be + * summarized. + */ + keep = GetOldestUnsummarizedLSN(NULL, NULL); + if (keep != InvalidXLogRecPtr) + { + XLogSegNo unsummarized_segno; + + XLByteToSeg(keep, unsummarized_segno, wal_segment_size); + if (unsummarized_segno < segno) + segno = unsummarized_segno; + } + /* but, keep at least wal_keep_size if that's set */ if (wal_keep_size_mb > 0) { diff --git a/src/backend/access/transam/xlogbackup.c b/src/backend/access/transam/xlogbackup.c index 21d68133ae..f51d4282bb 100644 --- a/src/backend/access/transam/xlogbackup.c +++ b/src/backend/access/transam/xlogbackup.c @@ -77,6 +77,16 @@ build_backup_content(BackupState *state, bool ishistoryfile) appendStringInfo(result, "STOP TIMELINE: %u\n", state->stoptli); } + /* either both istartpoint and istarttli should be set, or neither */ + Assert(XLogRecPtrIsInvalid(state->istartpoint) == (state->istarttli == 0)); + if (!XLogRecPtrIsInvalid(state->istartpoint)) + { + appendStringInfo(result, "INCREMENTAL FROM LSN: %X/%X\n", + LSN_FORMAT_ARGS(state->istartpoint)); + appendStringInfo(result, "INCREMENTAL FROM TLI: %u\n", + state->istarttli); + } + data = result->data; pfree(result); diff --git a/src/backend/access/transam/xlogrecovery.c b/src/backend/access/transam/xlogrecovery.c index 5549e1afc5..89ddec5bf9 100644 --- a/src/backend/access/transam/xlogrecovery.c +++ b/src/backend/access/transam/xlogrecovery.c @@ -1284,6 +1284,12 @@ read_backup_label(XLogRecPtr *checkPointLoc, TimeLineID *backupLabelTLI, tli_from_file, BACKUP_LABEL_FILE))); } + if (fscanf(lfp, "INCREMENTAL FROM LSN: %X/%X\n", &hi, &lo) > 0) + ereport(FATAL, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("this is an incremental backup, not a data directory"), + errhint("Use pg_combinebackup to reconstruct a valid data directory."))); + if (ferror(lfp) || FreeFile(lfp)) ereport(FATAL, (errcode_for_file_access(), diff --git a/src/backend/backup/Makefile b/src/backend/backup/Makefile index b21bd8ff43..751e6d3d5e 100644 --- a/src/backend/backup/Makefile +++ b/src/backend/backup/Makefile @@ -19,12 +19,15 @@ OBJS = \ basebackup.o \ basebackup_copy.o \ basebackup_gzip.o \ + basebackup_incremental.o \ basebackup_lz4.o \ basebackup_zstd.o \ basebackup_progress.o \ basebackup_server.o \ basebackup_sink.o \ basebackup_target.o \ - basebackup_throttle.o + basebackup_throttle.o \ + walsummary.o \ + walsummaryfuncs.o include $(top_srcdir)/src/backend/common.mk diff --git a/src/backend/backup/basebackup.c b/src/backend/backup/basebackup.c index 4ba63ad8a6..8a70a9ae41 100644 --- a/src/backend/backup/basebackup.c +++ b/src/backend/backup/basebackup.c @@ -20,8 +20,10 @@ #include "access/xlogbackup.h" #include "backup/backup_manifest.h" #include "backup/basebackup.h" +#include "backup/basebackup_incremental.h" #include "backup/basebackup_sink.h" #include "backup/basebackup_target.h" +#include "catalog/pg_tablespace_d.h" #include "commands/defrem.h" #include "common/compression.h" #include "common/file_perm.h" @@ -64,6 +66,7 @@ typedef struct bool fastcheckpoint; bool nowait; bool includewal; + bool incremental; uint32 maxrate; bool sendtblspcmapfile; bool send_to_client; @@ -76,21 +79,28 @@ typedef struct } basebackup_options; static int64 sendTablespace(bbsink *sink, char *path, Oid spcoid, bool sizeonly, - struct backup_manifest_info *manifest); + struct backup_manifest_info *manifest, + IncrementalBackupInfo *ib); static int64 sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, List *tablespaces, bool sendtblspclinks, - backup_manifest_info *manifest, Oid spcoid); + backup_manifest_info *manifest, Oid spcoid, + IncrementalBackupInfo *ib); static bool sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, struct stat *statbuf, bool missing_ok, Oid dboid, Oid spcoid, RelFileNumber relfilenumber, unsigned segno, - backup_manifest_info *manifest); + backup_manifest_info *manifest, + unsigned num_incremental_blocks, + BlockNumber *incremental_blocks, + unsigned truncation_block_length); static off_t read_file_data_into_buffer(bbsink *sink, const char *readfilename, int fd, off_t offset, size_t length, BlockNumber blkno, bool verify_checksum, int *checksum_failures); +static void push_to_sink(bbsink *sink, pg_checksum_context *checksum_ctx, + size_t *bytes_done, void *data, size_t length); static bool verify_page_checksum(Page page, XLogRecPtr start_lsn, BlockNumber blkno, uint16 *expected_checksum); @@ -102,7 +112,8 @@ static int64 _tarWriteHeader(bbsink *sink, const char *filename, bool sizeonly); static void _tarWritePadding(bbsink *sink, int len); static void convert_link_to_directory(const char *pathbuf, struct stat *statbuf); -static void perform_base_backup(basebackup_options *opt, bbsink *sink); +static void perform_base_backup(basebackup_options *opt, bbsink *sink, + IncrementalBackupInfo *ib); static void parse_basebackup_options(List *options, basebackup_options *opt); static int compareWalFileNames(const ListCell *a, const ListCell *b); static int basebackup_read_file(int fd, char *buf, size_t nbytes, off_t offset, @@ -220,7 +231,8 @@ static const struct exclude_list_item excludeFiles[] = * clobbered by longjmp" from stupider versions of gcc. */ static void -perform_base_backup(basebackup_options *opt, bbsink *sink) +perform_base_backup(basebackup_options *opt, bbsink *sink, + IncrementalBackupInfo *ib) { bbsink_state state; XLogRecPtr endptr; @@ -270,6 +282,10 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) ListCell *lc; tablespaceinfo *newti; + /* If this is an incremental backup, execute preparatory steps. */ + if (ib != NULL) + PrepareForIncrementalBackup(ib, backup_state); + /* Add a node for the base directory at the end */ newti = palloc0(sizeof(tablespaceinfo)); newti->size = -1; @@ -289,10 +305,10 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) if (tmp->path == NULL) tmp->size = sendDir(sink, ".", 1, true, state.tablespaces, - true, NULL, InvalidOid); + true, NULL, InvalidOid, NULL); else tmp->size = sendTablespace(sink, tmp->path, tmp->oid, true, - NULL); + NULL, NULL); state.bytes_total += tmp->size; } state.bytes_total_is_valid = true; @@ -330,7 +346,7 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) /* Then the bulk of the files... */ sendDir(sink, ".", 1, false, state.tablespaces, - sendtblspclinks, &manifest, InvalidOid); + sendtblspclinks, &manifest, InvalidOid, ib); /* ... and pg_control after everything else. */ if (lstat(XLOG_CONTROL_FILE, &statbuf) != 0) @@ -340,7 +356,7 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) XLOG_CONTROL_FILE))); sendFile(sink, XLOG_CONTROL_FILE, XLOG_CONTROL_FILE, &statbuf, false, InvalidOid, InvalidOid, - InvalidRelFileNumber, 0, &manifest); + InvalidRelFileNumber, 0, &manifest, 0, NULL, 0); } else { @@ -348,7 +364,7 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) bbsink_begin_archive(sink, archive_name); - sendTablespace(sink, ti->path, ti->oid, false, &manifest); + sendTablespace(sink, ti->path, ti->oid, false, &manifest, ib); } /* @@ -610,7 +626,7 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) sendFile(sink, pathbuf, pathbuf, &statbuf, false, InvalidOid, InvalidOid, InvalidRelFileNumber, 0, - &manifest); + &manifest, 0, NULL, 0); /* unconditionally mark file as archived */ StatusFilePath(pathbuf, fname, ".done"); @@ -686,6 +702,7 @@ parse_basebackup_options(List *options, basebackup_options *opt) bool o_checkpoint = false; bool o_nowait = false; bool o_wal = false; + bool o_incremental = false; bool o_maxrate = false; bool o_tablespace_map = false; bool o_noverify_checksums = false; @@ -764,6 +781,15 @@ parse_basebackup_options(List *options, basebackup_options *opt) opt->includewal = defGetBoolean(defel); o_wal = true; } + else if (strcmp(defel->defname, "incremental") == 0) + { + if (o_incremental) + ereport(ERROR, + (errcode(ERRCODE_SYNTAX_ERROR), + errmsg("duplicate option \"%s\"", defel->defname))); + opt->incremental = defGetBoolean(defel); + o_incremental = true; + } else if (strcmp(defel->defname, "max_rate") == 0) { int64 maxrate; @@ -956,7 +982,7 @@ parse_basebackup_options(List *options, basebackup_options *opt) * the filesystem, bypassing the buffer cache. */ void -SendBaseBackup(BaseBackupCmd *cmd) +SendBaseBackup(BaseBackupCmd *cmd, IncrementalBackupInfo *ib) { basebackup_options opt; bbsink *sink; @@ -980,6 +1006,20 @@ SendBaseBackup(BaseBackupCmd *cmd) set_ps_display(activitymsg); } + /* + * If we're asked to perform an incremental backup and the user has not + * supplied a manifest, that's an ERROR. + * + * If we're asked to perform a full backup and the user did supply a + * manifest, just ignore it. + */ + if (!opt.incremental) + ib = NULL; + else if (ib == NULL) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("must UPLOAD_MANIFEST before performing an incremental BASE_BACKUP"))); + /* * If the target is specifically 'client' then set up to stream the backup * to the client; otherwise, it's being sent someplace else and should not @@ -1011,7 +1051,7 @@ SendBaseBackup(BaseBackupCmd *cmd) */ PG_TRY(); { - perform_base_backup(&opt, sink); + perform_base_backup(&opt, sink, ib); } PG_FINALLY(); { @@ -1086,7 +1126,7 @@ sendFileWithContent(bbsink *sink, const char *filename, const char *content, */ static int64 sendTablespace(bbsink *sink, char *path, Oid spcoid, bool sizeonly, - backup_manifest_info *manifest) + backup_manifest_info *manifest, IncrementalBackupInfo *ib) { int64 size; char pathbuf[MAXPGPATH]; @@ -1120,7 +1160,7 @@ sendTablespace(bbsink *sink, char *path, Oid spcoid, bool sizeonly, /* Send all the files in the tablespace version directory */ size += sendDir(sink, pathbuf, strlen(path), sizeonly, NIL, true, manifest, - spcoid); + spcoid, ib); return size; } @@ -1140,7 +1180,7 @@ sendTablespace(bbsink *sink, char *path, Oid spcoid, bool sizeonly, static int64 sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, List *tablespaces, bool sendtblspclinks, backup_manifest_info *manifest, - Oid spcoid) + Oid spcoid, IncrementalBackupInfo *ib) { DIR *dir; struct dirent *de; @@ -1149,7 +1189,16 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, int64 size = 0; const char *lastDir; /* Split last dir from parent path. */ bool isRelationDir = false; /* Does directory contain relations? */ + bool isGlobalDir = false; Oid dboid = InvalidOid; + BlockNumber *relative_block_numbers = NULL; + + /* + * Since this array is relatively large, avoid putting it on the stack. + * But we don't need it at all if this is not an incremental backup. + */ + if (ib != NULL) + relative_block_numbers = palloc(sizeof(BlockNumber) * RELSEG_SIZE); /* * Determine if the current path is a database directory that can contain @@ -1182,7 +1231,10 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, } } else if (strcmp(path, "./global") == 0) + { isRelationDir = true; + isGlobalDir = true; + } dir = AllocateDir(path); while ((de = ReadDir(dir, path)) != NULL) @@ -1331,11 +1383,13 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, &statbuf, sizeonly); /* - * Also send archive_status directory (by hackishly reusing - * statbuf from above ...). + * Also send archive_status and summaries directories (by + * hackishly reusing statbuf from above ...). */ size += _tarWriteHeader(sink, "./pg_wal/archive_status", NULL, &statbuf, sizeonly); + size += _tarWriteHeader(sink, "./pg_wal/summaries", NULL, + &statbuf, sizeonly); continue; /* don't recurse into pg_wal */ } @@ -1404,33 +1458,88 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, if (!skip_this_dir) size += sendDir(sink, pathbuf, basepathlen, sizeonly, tablespaces, - sendtblspclinks, manifest, spcoid); + sendtblspclinks, manifest, spcoid, ib); } else if (S_ISREG(statbuf.st_mode)) { bool sent = false; + unsigned num_blocks_required = 0; + unsigned truncation_block_length = 0; + char tarfilenamebuf[MAXPGPATH * 2]; + char *tarfilename = pathbuf + basepathlen + 1; + FileBackupMethod method = BACK_UP_FILE_FULLY; - if (!sizeonly) - sent = sendFile(sink, pathbuf, pathbuf + basepathlen + 1, &statbuf, - true, dboid, spcoid, - relfilenumber, segno, manifest); + if (ib != NULL && isRelationFile) + { + Oid relspcoid; + char *lookup_path; - if (sent || sizeonly) + if (OidIsValid(spcoid)) + { + relspcoid = spcoid; + lookup_path = psprintf("pg_tblspc/%u/%s", spcoid, + pathbuf + basepathlen + 1); + } + else + { + if (isGlobalDir) + relspcoid = GLOBALTABLESPACE_OID; + else + relspcoid = DEFAULTTABLESPACE_OID; + lookup_path = pstrdup(pathbuf + basepathlen + 1); + } + + method = GetFileBackupMethod(ib, lookup_path, dboid, relspcoid, + relfilenumber, relForkNum, + segno, statbuf.st_size, + &num_blocks_required, + relative_block_numbers, + &truncation_block_length); + if (method == BACK_UP_FILE_INCREMENTALLY) + { + statbuf.st_size = + GetIncrementalFileSize(num_blocks_required); + snprintf(tarfilenamebuf, sizeof(tarfilenamebuf), + "%s/INCREMENTAL.%s", + path + basepathlen + 1, + de->d_name); + tarfilename = tarfilenamebuf; + } + + pfree(lookup_path); + } + + if (method != DO_NOT_BACK_UP_FILE) { - /* Add size. */ - size += statbuf.st_size; + if (!sizeonly) + sent = sendFile(sink, pathbuf, tarfilename, &statbuf, + true, dboid, spcoid, + relfilenumber, segno, manifest, + num_blocks_required, + method == BACK_UP_FILE_INCREMENTALLY ? relative_block_numbers : NULL, + truncation_block_length); + + if (sent || sizeonly) + { + /* Add size. */ + size += statbuf.st_size; - /* Pad to a multiple of the tar block size. */ - size += tarPaddingBytesRequired(statbuf.st_size); + /* Pad to a multiple of the tar block size. */ + size += tarPaddingBytesRequired(statbuf.st_size); - /* Size of the header for the file. */ - size += TAR_BLOCK_SIZE; + /* Size of the header for the file. */ + size += TAR_BLOCK_SIZE; + } } } else ereport(WARNING, (errmsg("skipping special file \"%s\"", pathbuf))); } + + if (relative_block_numbers != NULL) + pfree(relative_block_numbers); + FreeDir(dir); return size; } @@ -1443,6 +1552,12 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, * If dboid is anything other than InvalidOid then any checksum failures * detected will get reported to the cumulative stats system. * + * If the file is to be sent incrementally, then num_incremental_blocks + * should be the number of blocks to be sent, and incremental_blocks + * an array of block numbers relative to the start of the current segment. + * If the whole file is to be sent, then incremental_blocks should be NULL, + * and num_incremental_blocks can have any value, as it will be ignored. + * * Returns true if the file was successfully sent, false if 'missing_ok', * and the file did not exist. */ @@ -1450,7 +1565,8 @@ static bool sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, struct stat *statbuf, bool missing_ok, Oid dboid, Oid spcoid, RelFileNumber relfilenumber, unsigned segno, - backup_manifest_info *manifest) + backup_manifest_info *manifest, unsigned num_incremental_blocks, + BlockNumber *incremental_blocks, unsigned truncation_block_length) { int fd; BlockNumber blkno = 0; @@ -1459,6 +1575,7 @@ sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, pgoff_t bytes_done = 0; bool verify_checksum = false; pg_checksum_context checksum_ctx; + int ibindex = 0; if (pg_checksum_init(&checksum_ctx, manifest->checksum_type) < 0) elog(ERROR, "could not initialize checksum of file \"%s\"", @@ -1491,22 +1608,111 @@ sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, RelFileNumberIsValid(relfilenumber)) verify_checksum = true; + /* + * If we're sending an incremental file, write the file header. + */ + if (incremental_blocks != NULL) + { + unsigned magic = INCREMENTAL_MAGIC; + size_t header_bytes_done = 0; + + /* Emit header data. */ + push_to_sink(sink, &checksum_ctx, &header_bytes_done, + &magic, sizeof(magic)); + push_to_sink(sink, &checksum_ctx, &header_bytes_done, + &num_incremental_blocks, sizeof(num_incremental_blocks)); + push_to_sink(sink, &checksum_ctx, &header_bytes_done, + &truncation_block_length, sizeof(truncation_block_length)); + push_to_sink(sink, &checksum_ctx, &header_bytes_done, + incremental_blocks, + sizeof(BlockNumber) * num_incremental_blocks); + + /* Flush out any data still in the buffer so it's again empty. */ + if (header_bytes_done > 0) + { + bbsink_archive_contents(sink, header_bytes_done); + if (pg_checksum_update(&checksum_ctx, + (uint8 *) sink->bbs_buffer, + header_bytes_done) < 0) + elog(ERROR, "could not update checksum of base backup"); + } + + /* Update our notion of file position. */ + bytes_done += sizeof(magic); + bytes_done += sizeof(num_incremental_blocks); + bytes_done += sizeof(truncation_block_length); + bytes_done += sizeof(BlockNumber) * num_incremental_blocks; + } + /* * Loop until we read the amount of data the caller told us to expect. The * file could be longer, if it was extended while we were sending it, but * for a base backup we can ignore such extended data. It will be restored * from WAL. */ - while (bytes_done < statbuf->st_size) + while (1) { - size_t remaining = statbuf->st_size - bytes_done; + /* + * Determine whether we've read all the data that we need, and if not, + * read some more. + */ + if (incremental_blocks == NULL) + { + size_t remaining = statbuf->st_size - bytes_done; + + /* + * If we've read the required number of bytes, then it's time to + * stop. + */ + if (bytes_done >= statbuf->st_size) + break; + + /* + * Read as many bytes as will fit in the buffer, or however many + * are left to read, whichever is less. + */ + cnt = read_file_data_into_buffer(sink, readfilename, fd, + bytes_done, remaining, + blkno + segno * RELSEG_SIZE, + verify_checksum, + &checksum_failures); + } + else + { + BlockNumber relative_blkno; - /* Try to read some more data. */ - cnt = read_file_data_into_buffer(sink, readfilename, fd, bytes_done, - remaining, - blkno + segno * RELSEG_SIZE, - verify_checksum, - &checksum_failures); + /* + * If we've read all the blocks, then it's time to stop. + */ + if (ibindex >= num_incremental_blocks) + break; + + /* + * Read just one block, whichever one is the next that we're + * supposed to include. + */ + relative_blkno = incremental_blocks[ibindex++]; + cnt = read_file_data_into_buffer(sink, readfilename, fd, + relative_blkno * BLCKSZ, + BLCKSZ, + relative_blkno + segno * RELSEG_SIZE, + verify_checksum, + &checksum_failures); + + /* + * If we get a partial read, that must mean that the relation is + * being truncated. Ultimately, it should be truncated to a + * multiple of BLCKSZ, since this path should only be reached for + * relation files, but we might transiently observe an + * intermediate value. + * + * It should be fine to treat this just as if the entire block had + * been truncated away - i.e. fill this and all later blocks with + * zeroes. WAL replay will fix things up. + */ + if (cnt < BLCKSZ) + break; + } /* * If the amount of data we were able to read was not a multiple of @@ -1689,6 +1895,56 @@ read_file_data_into_buffer(bbsink *sink, const char *readfilename, int fd, return cnt; } +/* + * Push data into a bbsink. + * + * It's better, when possible, to read data directly into the bbsink's buffer, + * rather than using this function to copy it into the buffer; this function is + * for cases where that approach is not practical. + * + * bytes_done should point to a count of the number of bytes that are + * currently used in the bbsink's buffer. Upon return, the bytes identified by + * data and length will have been copied into the bbsink's buffer, flushing + * as required, and *bytes_done will have been updated accordingly. If the + * buffer was flushed, the previous contents will also have been fed to + * checksum_ctx. + * + * Note that after one or more calls to this function it is the caller's + * responsibility to perform any required final flush. + */ +static void +push_to_sink(bbsink *sink, pg_checksum_context *checksum_ctx, + size_t *bytes_done, void *data, size_t length) +{ + while (length > 0) + { + size_t bytes_to_copy; + + /* + * We use < here rather than <= so that if the data exactly fills the + * remaining buffer space, we trigger a flush now. + */ + if (length < sink->bbs_buffer_length - *bytes_done) + { + /* Append remaining data to buffer. */ + memcpy(sink->bbs_buffer + *bytes_done, data, length); + *bytes_done += length; + return; + } + + /* Copy until buffer is full and flush it. */ + bytes_to_copy = sink->bbs_buffer_length - *bytes_done; + memcpy(sink->bbs_buffer + *bytes_done, data, bytes_to_copy); + data = ((char *) data) + bytes_to_copy; + length -= bytes_to_copy; + bbsink_archive_contents(sink, sink->bbs_buffer_length); + if (pg_checksum_update(checksum_ctx, (uint8 *) sink->bbs_buffer, + sink->bbs_buffer_length) < 0) + elog(ERROR, "could not update checksum"); + *bytes_done = 0; + } +} + /* * Try to verify the checksum for the provided page, if it seems appropriate * to do so. diff --git a/src/backend/backup/basebackup_incremental.c b/src/backend/backup/basebackup_incremental.c new file mode 100644 index 0000000000..20cc00bded --- /dev/null +++ b/src/backend/backup/basebackup_incremental.c @@ -0,0 +1,873 @@ +/*------------------------------------------------------------------------- + * + * basebackup_incremental.c + * code for incremental backup support + * + * This code isn't actually in charge of taking an incremental backup; + * the actual construction of the incremental backup happens in + * basebackup.c. Here, we're concerned with providing the necessary + * supports for that operation. In particular, we need to parse the + * backup manifest supplied by the user taking the incremental backup + * and extract the required information from it. + * + * Portions Copyright (c) 2010-2022, PostgreSQL Global Development Group + * + * IDENTIFICATION + * src/backend/backup/basebackup_incremental.c + * + *------------------------------------------------------------------------- + */ +#include "postgres.h" + +#include "access/timeline.h" +#include "access/xlog.h" +#include "access/xlogrecovery.h" +#include "backup/basebackup_incremental.h" +#include "backup/walsummary.h" +#include "common/blkreftable.h" +#include "common/parse_manifest.h" +#include "common/hashfn.h" +#include "postmaster/walsummarizer.h" + +#define BLOCKS_PER_READ 512 + +/* + * Details extracted from the WAL ranges present in the supplied backup manifest. + */ +typedef struct +{ + TimeLineID tli; + XLogRecPtr start_lsn; + XLogRecPtr end_lsn; +} backup_wal_range; + +/* + * Details extracted from the file list present in the supplied backup manifest. + */ +typedef struct +{ + uint32 status; + const char *path; + size_t size; +} backup_file_entry; + +static uint32 hash_string_pointer(const char *s); +#define SH_PREFIX backup_file +#define SH_ELEMENT_TYPE backup_file_entry +#define SH_KEY_TYPE const char * +#define SH_KEY path +#define SH_HASH_KEY(tb, key) hash_string_pointer(key) +#define SH_EQUAL(tb, a, b) (strcmp(a, b) == 0) +#define SH_SCOPE static inline +#define SH_DECLARE +#define SH_DEFINE +#include "lib/simplehash.h" + +struct IncrementalBackupInfo +{ + /* Memory context for this object and its subsidiary objects. */ + MemoryContext mcxt; + + /* Temporary buffer for storing the manifest while parsing it. */ + StringInfoData buf; + + /* WAL ranges extracted from the backup manifest. */ + List *manifest_wal_ranges; + + /* + * Files extracted from the backup manifest. + * + * We don't really need this information, because we use WAL summaries to + * figure what's changed. It would be unsafe to just rely on the list of + * files that existed before, because it's possible for a file to be + * removed and a new one created with the same name and different + * contents. In such cases, the whole file must still be sent. We can tell + * from the WAL summaries whether that happened, but not from the file + * list. + * + * Nonetheless, this data is useful for sanity checking. If a file that we + * think we shouldn't need to send is not present in the manifest for the + * prior backup, something has gone terribly wrong. We retain the file + * names and sizes, but not the checksums or last modified times, for + * which we have no use. + * + * One significant downside of storing this data is that it consumes + * memory. If that turns out to be a problem, we might have to decide not + * to retain this information, or to make it optional. + */ + backup_file_hash *manifest_files; + + /* + * Block-reference table for the incremental backup. + * + * It's possible that storing the entire block-reference table in memory + * will be a problem for some users. The in-memory format that we're using + * here is pretty efficient, converging to little more than 1 bit per + * block for relation forks with large numbers of modified blocks. It's + * possible, however, that if you try to perform an incremental backup of + * a database with a sufficiently large number of relations on a + * sufficiently small machine, you could run out of memory here. If that + * turns out to be a problem in practice, we'll need to be more clever. + */ + BlockRefTable *brtab; +}; + +static void manifest_process_file(JsonManifestParseContext *, + char *pathname, + size_t size, + pg_checksum_type checksum_type, + int checksum_length, + uint8 *checksum_payload); +static void manifest_process_wal_range(JsonManifestParseContext *, + TimeLineID tli, + XLogRecPtr start_lsn, + XLogRecPtr end_lsn); +static void manifest_report_error(JsonManifestParseContext *ib, + const char *fmt,...) + pg_attribute_printf(2, 3) pg_attribute_noreturn(); + +/* + * Create a new object for storing information extracted from the manifest + * supplied when creating an incremental backup. + */ +IncrementalBackupInfo * +CreateIncrementalBackupInfo(MemoryContext mcxt) +{ + IncrementalBackupInfo *ib; + MemoryContext oldcontext; + + oldcontext = MemoryContextSwitchTo(mcxt); + + ib = palloc0(sizeof(IncrementalBackupInfo)); + ib->mcxt = mcxt; + initStringInfo(&ib->buf); + + /* + * It's hard to guess how many files a "typical" installation will have in + * the data directory, but a fresh initdb creates almost 1000 files as of + * this writing, so it seems to make sense for our estimate to + * substantially higher. + */ + ib->manifest_files = backup_file_create(mcxt, 10000, NULL); + + MemoryContextSwitchTo(oldcontext); + + return ib; +} + +/* + * Before taking an incremental backup, the caller must supply the backup + * manifest from a prior backup. Each chunk of manifest data recieved + * from the client should be passed to this function. + */ +void +AppendIncrementalManifestData(IncrementalBackupInfo *ib, const char *data, + int len) +{ + MemoryContext oldcontext; + + /* Switch to our memory context. */ + oldcontext = MemoryContextSwitchTo(ib->mcxt); + + /* + * XXX. Our json parser is at present incapable of parsing json blobs + * incrementally, so we have to accumulate the entire backup manifest + * before we can do anything with it. This should really be fixed, since + * some users might have very large numbers of files in the data + * directory. + */ + appendBinaryStringInfo(&ib->buf, data, len); + + /* Switch back to previous memory context. */ + MemoryContextSwitchTo(oldcontext); +} + +/* + * Finalize an IncrementalBackupInfo object after all manifest data has + * been supplied via calls to AppendIncrementalManifestData. + */ +void +FinalizeIncrementalManifest(IncrementalBackupInfo *ib) +{ + JsonManifestParseContext context; + MemoryContext oldcontext; + + /* Switch to our memory context. */ + oldcontext = MemoryContextSwitchTo(ib->mcxt); + + /* Parse the manifest. */ + context.private_data = ib; + context.perfile_cb = manifest_process_file; + context.perwalrange_cb = manifest_process_wal_range; + context.error_cb = manifest_report_error; + json_parse_manifest(&context, ib->buf.data, ib->buf.len); + + /* Done with the buffer, so release memory. */ + pfree(ib->buf.data); + ib->buf.data = NULL; + + /* Switch back to previous memory context. */ + MemoryContextSwitchTo(oldcontext); +} + +/* + * Prepare to take an incremental backup. + * + * Before this function is called, AppendIncrementalManifestData and + * FinalizeIncrementalManifest should have already been called to pass all + * the manifest data to this object. + * + * This function performs sanity checks on the data extracted from the + * manifest and figures out for which WAL ranges we need summaries, and + * whether those summaries are available. Then, it reads and combines the + * data from those summary files. It also updates the backup_state with the + * reference TLI and LSN for the prior backup. + */ +void +PrepareForIncrementalBackup(IncrementalBackupInfo *ib, + BackupState *backup_state) +{ + MemoryContext oldcontext; + List *expectedTLEs; + List *all_wslist, + *required_wslist = NIL; + ListCell *lc; + TimeLineHistoryEntry **tlep; + int num_wal_ranges; + int i; + bool found_backup_start_tli = false; + TimeLineID earliest_wal_range_tli = 0; + XLogRecPtr earliest_wal_range_start_lsn; + TimeLineID latest_wal_range_tli = 0; + XLogRecPtr summarized_lsn; + + Assert(ib->buf.data == NULL); + + /* Switch to our memory context. */ + oldcontext = MemoryContextSwitchTo(ib->mcxt); + + /* + * Match up the TLIs that appear in the WAL ranges of the backup manifest + * with those that appear in this server's timeline history. We expect + * every backup_wal_range to match to a TimeLineHistoryEntry; if it does + * not, that's an error. + * + * This loop also decides which of the WAL ranges is the manifest is most + * ancient and which one is the newest, according to the timeline history + * of this server, and stores TLIs of those WAL ranges into + * earliest_wal_range_tli and latest_wal_range_tli. It also updates + * earliest_wal_range_start_lsn to the start LSN of the WAL range for + * earliest_wal_range_tli. + * + * Note that the return value of readTimeLineHistory puts the latest + * timeline at the beginning of the list, not the end. Hence, the earliest + * TLI is the one that occurs nearest the end of the list returned by + * readTimeLineHistory, and the latest TLI is the one that occurs closest + * to the beginning. + */ + expectedTLEs = readTimeLineHistory(backup_state->starttli); + num_wal_ranges = list_length(ib->manifest_wal_ranges); + tlep = palloc0(num_wal_ranges * sizeof(TimeLineHistoryEntry *)); + for (i = 0; i < num_wal_ranges; ++i) + { + backup_wal_range *range = list_nth(ib->manifest_wal_ranges, i); + bool saw_earliest_wal_range_tli = false; + bool saw_latest_wal_range_tli = false; + + /* Search this server's history for this WAL range's TLI. */ + foreach(lc, expectedTLEs) + { + TimeLineHistoryEntry *tle = lfirst(lc); + + if (tle->tli == range->tli) + { + tlep[i] = tle; + break; + } + + if (tle->tli == earliest_wal_range_tli) + saw_earliest_wal_range_tli = true; + if (tle->tli == latest_wal_range_tli) + saw_latest_wal_range_tli = true; + } + + /* + * An incremental backup can only be taken relative to a backup that + * represents a previous state of this server. If the backup requires + * WAL from a timeline that's not in our history, that definitely + * isn't the case. + */ + if (tlep[i] == NULL) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("timeline %u found in manifest, but not in this server's history", + range->tli))); + + /* + * If we found this TLI in the server's history before encountering + * the latest TLI seen so far in the server's history, then this TLI + * is the latest one seen so far. + * + * If on the other hand we saw the earliest TLI seen so far before + * finding this TLI, this TLI is earlier than the earliest one seen so + * far. And if this is the first TLI for which we've searched, it's + * also the earliest one seen so far. + * + * On the first loop iteration, both things should necessarily be + * true. + */ + if (!saw_latest_wal_range_tli) + latest_wal_range_tli = range->tli; + if (earliest_wal_range_tli == 0 || saw_earliest_wal_range_tli) + { + earliest_wal_range_tli = range->tli; + earliest_wal_range_start_lsn = range->start_lsn; + } + } + + /* + * Propagate information about the prior backup into the backup_label that + * will be generated for this backup. + */ + backup_state->istartpoint = earliest_wal_range_start_lsn; + backup_state->istarttli = earliest_wal_range_tli; + + /* + * Sanity check start and end LSNs for the WAL ranges in the manifest. + * + * Commonly, there won't be any timeline switches during the prior backup + * at all, but if there are, they should happen at the same LSNs that this + * server switched timelines. + * + * Whether there are any timeline switches during the prior backup or not, + * the prior backup shouldn't require any WAL from a timeline prior to the + * start of that timeline. It also shouldn't require any WAL from later + * than the start of this backup. + * + * If any of these sanity checks fail, one possible explanation is that + * the user has generated WAL on the same timeline with the same LSNs more + * than once. For instance, if two standbys running on timeline 1 were + * both promoted and (due to a broken archiving setup) both selected new + * timeline ID 2, then it's possible that one of these checks might trip. + * + * Note that there are lots of ways for the user to do something very bad + * without tripping any of these checks, and they are not intended to be + * comprehensive. It's pretty hard to see how we could be certain of + * anything here. However, if there's a problem staring us right in the + * face, it's best to report it, so we do. + */ + for (i = 0; i < num_wal_ranges; ++i) + { + backup_wal_range *range = list_nth(ib->manifest_wal_ranges, i); + + if (range->tli == earliest_wal_range_tli) + { + if (range->start_lsn < tlep[i]->begin) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("manifest requires WAL from initial timeline %u starting at %X/%X, but that timeline begins at %X/%X", + range->tli, + LSN_FORMAT_ARGS(range->start_lsn), + LSN_FORMAT_ARGS(tlep[i]->begin)))); + } + else + { + if (range->start_lsn != tlep[i]->begin) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("manifest requires WAL from continuation timeline %u starting at %X/%X, but that timeline begins at %X/%X", + range->tli, + LSN_FORMAT_ARGS(range->start_lsn), + LSN_FORMAT_ARGS(tlep[i]->begin)))); + } + + if (range->tli == latest_wal_range_tli) + { + if (range->end_lsn > backup_state->startpoint) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("manifest requires WAL from final timeline %u ending at %X/%X, but this backup starts at %X/%X", + range->tli, + LSN_FORMAT_ARGS(range->end_lsn), + LSN_FORMAT_ARGS(backup_state->startpoint)))); + } + else + { + if (range->end_lsn != tlep[i]->end) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("manifest requires WAL from non-final timeline %u ending at %X/%X, but this server switched timelines at %X/%X", + range->tli, + LSN_FORMAT_ARGS(range->end_lsn), + LSN_FORMAT_ARGS(tlep[i]->end)))); + } + + } + + /* + * Wait for WAL summarization to catch up to the backup start LSN (but + * time out if it doesn't do so quickly enough). + */ + /* XXX make timeout configurable */ + summarized_lsn = WaitForWalSummarization(backup_state->startpoint, 60000); + if (summarized_lsn < backup_state->startpoint) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("timeout waiting for WAL summarization"), + errdetail("This backup requires WAL to be summarized up to %X/%X, but summarizer has only reached %X/%X.", + LSN_FORMAT_ARGS(backup_state->startpoint), + LSN_FORMAT_ARGS(summarized_lsn)))); + + /* + * Retrieve a list of all WAL summaries on any timeline that overlap with + * the LSN range of interest. We could instead call GetWalSummaries() once + * per timeline in the loop that follows, but that would involve reading + * the directory multiple times. It should be mildly faster - and perhaps + * a bit safer - to do it just once. + */ + all_wslist = GetWalSummaries(0, earliest_wal_range_start_lsn, + backup_state->startpoint); + + /* + * We need WAL summaries for everything that happened during the prior + * backup and everything that happened afterward up until the point where + * the current backup started. + */ + foreach(lc, expectedTLEs) + { + TimeLineHistoryEntry *tle = lfirst(lc); + XLogRecPtr tli_start_lsn = tle->begin; + XLogRecPtr tli_end_lsn = tle->end; + XLogRecPtr tli_missing_lsn = InvalidXLogRecPtr; + List *tli_wslist; + + /* + * Working through the history of this server from the current + * timeline backwards, we skip everything until we find the timeline + * where this backup started. Most of the time, this means we won't + * skip anything at all, as it's unlikely that the timeline has + * changed since the beginning of the backup moments ago. + */ + if (tle->tli == backup_state->starttli) + { + found_backup_start_tli = true; + tli_end_lsn = backup_state->startpoint; + } + else if (!found_backup_start_tli) + continue; + + /* + * Find the summaries that overlap the LSN range of interest for this + * timeline. If this is the earliest timeline involved, the range of + * interest begins with the start LSN of the prior backup; otherwise, + * it begins at the LSN at which this timeline came into existence. If + * this is the latest TLI involved, the range of interest ends at the + * start LSN of the current backup; otherwise, it ends at the point + * where we switched from this timeline to the next one. + */ + if (tle->tli == earliest_wal_range_tli) + tli_start_lsn = earliest_wal_range_start_lsn; + tli_wslist = FilterWalSummaries(all_wslist, tle->tli, + tli_start_lsn, tli_end_lsn); + + /* + * There is no guarantee that the WAL summaries we found cover the + * entire range of LSNs for which summaries are required, or indeed + * that we found any WAL summaries at all. Check whether we have a + * problem of that sort. + */ + if (!WalSummariesAreComplete(tli_wslist, tli_start_lsn, tli_end_lsn, + &tli_missing_lsn)) + { + if (XLogRecPtrIsInvalid(tli_missing_lsn)) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("WAL summaries are required on timeline %u from %X/%X to %X/%X, but no summaries for that timeline and LSN range exist", + tle->tli, + LSN_FORMAT_ARGS(tli_start_lsn), + LSN_FORMAT_ARGS(tli_end_lsn)))); + else + ereport(ERROR, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("WAL summaries are required on timeline %u from %X/%X to %X/%X, but the summaries for that timeline and LSN range are incomplete", + tle->tli, + LSN_FORMAT_ARGS(tli_start_lsn), + LSN_FORMAT_ARGS(tli_end_lsn)), + errdetail("The first unsummarized LSN is this range is %X/%X.", + LSN_FORMAT_ARGS(tli_missing_lsn)))); + } + + /* + * Remember that we need to read these summaries. + * + * Technically, it's possible that this could read more files than + * required, since tli_wslist in theory could contain redundant + * summaries. For instance, if we have a summary from 0/10000000 to + * 0/20000000 and also one from 0/00000000 to 0/30000000, then the + * latter subsumes the former and the former could be ignored. + * + * We ignore this possibility because the WAL summarizer only tries to + * generate summaries that do not overlap. If somehow they exist, + * we'll do a bit of extra work but the results should still be + * correct. + */ + required_wslist = list_concat(required_wslist, tli_wslist); + + /* + * Timelines earlier than the one in which the prior backup began are + * not relevant. + */ + if (tle->tli == earliest_wal_range_tli) + break; + } + + /* + * Read all of the required block reference table files and merge all of + * the data into a single in-memory block reference table. + * + * See the comments for struct IncrementalBackupInfo for some thoughts on + * memory usage. + */ + ib->brtab = CreateEmptyBlockRefTable(); + foreach(lc, required_wslist) + { + WalSummaryFile *ws = lfirst(lc); + WalSummaryIO wsio; + BlockRefTableReader *reader; + RelFileLocator rlocator; + ForkNumber forknum; + BlockNumber limit_block; + BlockNumber blocks[BLOCKS_PER_READ]; + + wsio.file = OpenWalSummaryFile(ws, false); + wsio.filepos = 0; + ereport(DEBUG1, + (errmsg_internal("reading WAL summary file \"%s\"", + FilePathName(wsio.file)))); + reader = CreateBlockRefTableReader(ReadWalSummary, &wsio, + FilePathName(wsio.file), + ReportWalSummaryError, NULL); + while (BlockRefTableReaderNextRelation(reader, &rlocator, &forknum, + &limit_block)) + { + BlockRefTableSetLimitBlock(ib->brtab, &rlocator, + forknum, limit_block); + + while (1) + { + unsigned nblocks; + unsigned i; + + nblocks = BlockRefTableReaderGetBlocks(reader, blocks, + BLOCKS_PER_READ); + if (nblocks == 0) + break; + + for (i = 0; i < nblocks; ++i) + BlockRefTableMarkBlockModified(ib->brtab, &rlocator, + forknum, blocks[i]); + } + } + DestroyBlockRefTableReader(reader); + FileClose(wsio.file); + } + + /* Switch back to previous memory context. */ + MemoryContextSwitchTo(oldcontext); +} + +/* + * Get the pathname that should be used when a file is sent incrementally. + * + * The result is a palloc'd string. + */ +char * +GetIncrementalFilePath(Oid dboid, Oid spcoid, RelFileNumber relfilenumber, + ForkNumber forknum, unsigned segno) +{ + char *path; + char *lastslash; + char *ipath; + + path = GetRelationPath(dboid, spcoid, relfilenumber, InvalidBackendId, + forknum); + + lastslash = strrchr(path, '/'); + Assert(lastslash != NULL); + *lastslash = '\0'; + + if (segno > 0) + ipath = psprintf("%s/INCREMENTAL.%s.%u", path, lastslash + 1, segno); + else + ipath = psprintf("%s/INCREMENTAL.%s", path, lastslash + 1); + + pfree(path); + + return ipath; +} + +/* + * How should we back up a particular file as part of an incremental backup? + * + * If the return value is BACK_UP_FILE_FULLY, caller should back up the whole + * file just as if this were not an incremental backup. + * + * If the return value is BACK_UP_FILE_INCREMENTALLY, caller should include + * an incremental file in the backup instead of the entire file. On return, + * *num_blocks_required will be set to the number of blocks that need to be + * sent, and the actual block numbers will have been stored in + * relative_block_numbers, which should be an array of at least RELSEG_SIZE. + * In addition, *truncation_block_length will be set to the value that should + * be included in the incremental file. + * + * If the return value is DO_NOT_BACK_UP_FILE, the caller should not include + * the file in the backup at all. + */ +FileBackupMethod +GetFileBackupMethod(IncrementalBackupInfo *ib, char *path, + Oid dboid, Oid spcoid, + RelFileNumber relfilenumber, ForkNumber forknum, + unsigned segno, size_t size, + unsigned *num_blocks_required, + BlockNumber *relative_block_numbers, + unsigned *truncation_block_length) +{ + BlockNumber absolute_block_numbers[RELSEG_SIZE]; + BlockNumber limit_block; + BlockNumber start_blkno; + BlockNumber stop_blkno; + RelFileLocator rlocator; + BlockRefTableEntry *brtentry; + unsigned i; + unsigned nblocks; + + /* Should only be called after PrepareForIncrementalBackup. */ + Assert(ib->buf.data == NULL); + + /* + * dboid could be InvalidOid if shared rel, but spcoid and relfilenumber + * should have legal values. + */ + Assert(OidIsValid(spcoid)); + Assert(RelFileNumberIsValid(relfilenumber)); + + /* + * If the file size is too large or not a multiple of BLCKSZ, then + * something weird is happening, so give up and send the whole file. + */ + if ((size % BLCKSZ) != 0 || size / BLCKSZ > RELSEG_SIZE) + return BACK_UP_FILE_FULLY; + + /* + * The free-space map fork is not properly WAL-logged, so we need to + * backup the entire file every time. + */ + if (forknum == FSM_FORKNUM) + return BACK_UP_FILE_FULLY; + + /* + * Check whether this file is part of the prior backup. If it isn't, back + * up the whole file. + */ + if (backup_file_lookup(ib->manifest_files, path) == NULL) + { + char *ipath; + + ipath = GetIncrementalFilePath(dboid, spcoid, relfilenumber, + forknum, segno); + if (backup_file_lookup(ib->manifest_files, ipath) == NULL) + return BACK_UP_FILE_FULLY; + } + + /* Look up the block reference table entry. */ + rlocator.spcOid = spcoid; + rlocator.dbOid = dboid; + rlocator.relNumber = relfilenumber; + brtentry = BlockRefTableGetEntry(ib->brtab, &rlocator, forknum, + &limit_block); + + /* + * If there is no entry, then there have been no WAL-logged changes to the + * relation since the predecessor backup was taken, so we can back it up + * incrementally and need not include any modified blocks. + * + * However, if the file is zero-length, we should do a full backup, + * because an incremental file is always more than zero length, and it's + * silly to take an incremental backup when a full backup would be + * smaller. + */ + if (brtentry == NULL) + { + *num_blocks_required = 0; + *truncation_block_length = size / BLCKSZ; + if (size == 0) + return BACK_UP_FILE_FULLY; + return BACK_UP_FILE_INCREMENTALLY; + } + + /* + * If the limit_block is less than or equal to the point where this + * segment starts, send the whole file. + */ + if (limit_block <= segno * RELSEG_SIZE) + return BACK_UP_FILE_FULLY; + + /* + * Get relevant entries from the block reference table entry. + * + * We shouldn't overflow computing the start or stop block numbers, but if + * it manages to happen somehow, detect it and throw an error. + */ + start_blkno = segno * RELSEG_SIZE; + stop_blkno = start_blkno + (size / BLCKSZ); + if (start_blkno / RELSEG_SIZE != segno || stop_blkno < start_blkno) + ereport(ERROR, + errcode(ERRCODE_INTERNAL_ERROR), + errmsg_internal("overflow computing block number bounds for segment %u with size %lu", + segno, size)); + nblocks = BlockRefTableEntryGetBlocks(brtentry, start_blkno, stop_blkno, + absolute_block_numbers, RELSEG_SIZE); + Assert(nblocks <= RELSEG_SIZE); + + /* + * If we're going to have to send nearly all of the blocks, then just send + * the whole file, because that won't require much extra storage or + * transfer and will speed up and simplify backup restoration. It's not + * clear what threshold is most appropriate here and perhaps it ought to + * be configurable, but for now we're just going to say that if we'd need + * to send 90% of the blocks anyway, give up and send the whole file. + * + * NB: If you change the threshold here, at least make sure to back up the + * file fully when every single block must be sent, because there's + * nothing good about sending an incremental file in that case. + */ + if (nblocks * BLCKSZ > size * 0.9) + return BACK_UP_FILE_FULLY; + + /* + * Looks like we can send an incremental file. + * + * Return the relevant details to the caller, transposing absolute block + * numbers to relative block numbers. + * + * The truncation block length is the minimum length of the reconstructed + * file. Any block numbers below this threshold that are not present in + * the backup need to be fetched from the prior backup. At or above this + * threshold, blocks should only be included in the result if they are + * present in the backup. (This may require inserting zero blocks if the + * blocks included in the backup are non-consecutive.) + */ + for (i = 0; i < nblocks; ++i) + relative_block_numbers[i] = absolute_block_numbers[i] - start_blkno; + *num_blocks_required = nblocks; + *truncation_block_length = + Min(size / BLCKSZ, limit_block - segno * RELSEG_SIZE); + return BACK_UP_FILE_INCREMENTALLY; +} + +/* + * Compute the size for an incremental file containing a given number of blocks. + */ +extern size_t +GetIncrementalFileSize(unsigned num_blocks_required) +{ + size_t result; + + /* Make sure we're not going to overflow. */ + Assert(num_blocks_required <= RELSEG_SIZE); + + /* + * Three four byte quantities (magic number, truncation block length, + * block count) followed by block numbers followed by block contents. + */ + result = 3 * sizeof(uint32); + result += (BLCKSZ + sizeof(BlockNumber)) * num_blocks_required; + + return result; +} + +/* + * Helper function for filemap hash table. + */ +static uint32 +hash_string_pointer(const char *s) +{ + unsigned char *ss = (unsigned char *) s; + + return hash_bytes(ss, strlen(s)); +} + +/* + * This callback is invoked for each file mentioned in the backup manifest. + * + * We store the path to each file and the size of each file for sanity-checking + * purposes. For further details, see comments for IncrementalBackupInfo. + */ +static void +manifest_process_file(JsonManifestParseContext *context, + char *pathname, size_t size, + pg_checksum_type checksum_type, + int checksum_length, + uint8 *checksum_payload) +{ + IncrementalBackupInfo *ib = context->private_data; + backup_file_entry *entry; + bool found; + + entry = backup_file_insert(ib->manifest_files, pathname, &found); + if (!found) + { + entry->path = MemoryContextStrdup(ib->manifest_files->ctx, + pathname); + entry->size = size; + } +} + +/* + * This callback is invoked for each WAL range mentioned in the backup + * manifest. + * + * We're just interested in learning the oldest LSN and the corresponding TLI + * that appear in any WAL range. + */ +static void +manifest_process_wal_range(JsonManifestParseContext *context, + TimeLineID tli, XLogRecPtr start_lsn, + XLogRecPtr end_lsn) +{ + IncrementalBackupInfo *ib = context->private_data; + backup_wal_range *range = palloc(sizeof(backup_wal_range)); + + range->tli = tli; + range->start_lsn = start_lsn; + range->end_lsn = end_lsn; + ib->manifest_wal_ranges = lappend(ib->manifest_wal_ranges, range); +} + +/* + * This callback is invoked if an error occurs while parsing the backup + * manifest. + */ +static void +manifest_report_error(JsonManifestParseContext *context, const char *fmt,...) +{ + StringInfoData errbuf; + + initStringInfo(&errbuf); + + for (;;) + { + va_list ap; + int needed; + + va_start(ap, fmt); + needed = appendStringInfoVA(&errbuf, fmt, ap); + va_end(ap); + if (needed == 0) + break; + enlargeStringInfo(&errbuf, needed); + } + + ereport(ERROR, + errmsg_internal("%s", errbuf.data)); +} diff --git a/src/backend/backup/meson.build b/src/backend/backup/meson.build index 11a79bbf80..19c355ceca 100644 --- a/src/backend/backup/meson.build +++ b/src/backend/backup/meson.build @@ -5,6 +5,7 @@ backend_sources += files( 'basebackup.c', 'basebackup_copy.c', 'basebackup_gzip.c', + 'basebackup_incremental.c', 'basebackup_lz4.c', 'basebackup_progress.c', 'basebackup_server.c', @@ -12,4 +13,6 @@ backend_sources += files( 'basebackup_target.c', 'basebackup_throttle.c', 'basebackup_zstd.c', + 'walsummary.c', + 'walsummaryfuncs.c' ) diff --git a/src/backend/backup/walsummary.c b/src/backend/backup/walsummary.c new file mode 100644 index 0000000000..ebf4ea038d --- /dev/null +++ b/src/backend/backup/walsummary.c @@ -0,0 +1,356 @@ +/*------------------------------------------------------------------------- + * + * walsummary.c + * Functions for accessing and managing WAL summary data. + * + * Portions Copyright (c) 2010-2022, PostgreSQL Global Development Group + * + * src/backend/backup/walsummary.c + * + *------------------------------------------------------------------------- + */ + +#include "postgres.h" + +#include <sys/stat.h> +#include <unistd.h> + +#include "access/xlog_internal.h" +#include "backup/walsummary.h" +#include "utils/wait_event.h" + +static bool IsWalSummaryFilename(char *filename); +static int ListComparatorForWalSummaryFiles(const ListCell *a, + const ListCell *b); + +/* + * Get a list of WAL summaries. + * + * If tli != 0, only WAL summaries with the indicated TLI will be included. + * + * If start_lsn != InvalidXLogRecPtr, only summaries that end before the + * indicated LSN will be included. + * + * If end_lsn != InvalidXLogRecPtr, only summaries that start before the + * indicated LSN will be included. + * + * The intent is that you can call GetWalSummaries(tli, start_lsn, end_lsn) + * to get all WAL summaries on the indicated timeline that overlap the + * specified LSN range. + */ +List * +GetWalSummaries(TimeLineID tli, XLogRecPtr start_lsn, XLogRecPtr end_lsn) +{ + DIR *sdir; + struct dirent *dent; + List *result = NIL; + + sdir = AllocateDir(XLOGDIR "/summaries"); + while ((dent = ReadDir(sdir, XLOGDIR "/summaries")) != NULL) + { + WalSummaryFile *ws; + uint32 tmp[5]; + TimeLineID file_tli; + XLogRecPtr file_start_lsn; + XLogRecPtr file_end_lsn; + + /* Decode filename, or skip if it's not in the expected format. */ + if (!IsWalSummaryFilename(dent->d_name)) + continue; + sscanf(dent->d_name, "%08X%08X%08X%08X%08X", + &tmp[0], &tmp[1], &tmp[2], &tmp[3], &tmp[4]); + file_tli = tmp[0]; + file_start_lsn = ((uint64) tmp[1]) << 32 | tmp[2]; + file_end_lsn = ((uint64) tmp[3]) << 32 | tmp[4]; + + /* Skip if it doesn't match the filter criteria. */ + if (tli != 0 && tli != file_tli) + continue; + if (!XLogRecPtrIsInvalid(start_lsn) && start_lsn > file_end_lsn) + continue; + if (!XLogRecPtrIsInvalid(end_lsn) && end_lsn < file_start_lsn) + continue; + + /* Add it to the list. */ + ws = palloc(sizeof(WalSummaryFile)); + ws->tli = file_tli; + ws->start_lsn = file_start_lsn; + ws->end_lsn = file_end_lsn; + result = lappend(result, ws); + } + FreeDir(sdir); + + return result; +} + +/* + * Build a new list of WAL summaries based on an existing list, but filtering + * out summaries that don't match the search parameters. + * + * If tli != 0, only WAL summaries with the indicated TLI will be included. + * + * If start_lsn != InvalidXLogRecPtr, only summaries that end before the + * indicated LSN will be included. + * + * If end_lsn != InvalidXLogRecPtr, only summaries that start before the + * indicated LSN will be included. + */ +List * +FilterWalSummaries(List *wslist, TimeLineID tli, + XLogRecPtr start_lsn, XLogRecPtr end_lsn) +{ + List *result = NIL; + ListCell *lc; + + /* Loop over input. */ + foreach(lc, wslist) + { + WalSummaryFile *ws = lfirst(lc); + + /* Skip if it doesn't match the filter criteria. */ + if (tli != 0 && tli != ws->tli) + continue; + if (!XLogRecPtrIsInvalid(start_lsn) && start_lsn > ws->end_lsn) + continue; + if (!XLogRecPtrIsInvalid(end_lsn) && end_lsn < ws->start_lsn) + continue; + + /* Add it to the result list. */ + result = lappend(result, ws); + } + + return result; +} + +/* + * Check whether the supplied list of WalSummaryFile objects covers the + * whole range of LSNs from start_lsn to end_lsn. This function ignores + * timelines, so the caller should probably filter using the appropriate + * timeline before calling this. + * + * If the whole range of LSNs is covered, returns true, otherwise false. + * If false is returned, *missing_lsn is set either to InvalidXLogRecPtr + * if there are no WAL summary files in the input list, or to the first LSN + * in the range that is not covered by a WAL summary file in the input list. + */ +bool +WalSummariesAreComplete(List *wslist, XLogRecPtr start_lsn, + XLogRecPtr end_lsn, XLogRecPtr *missing_lsn) +{ + XLogRecPtr current_lsn = start_lsn; + ListCell *lc; + + /* Special case for empty list. */ + if (wslist == NIL) + { + *missing_lsn = InvalidXLogRecPtr; + return false; + } + + /* Make a private copy of the list and sort it by start LSN. */ + wslist = list_copy(wslist); + list_sort(wslist, ListComparatorForWalSummaryFiles); + + /* + * Consider summary files in order of increasing start_lsn, advancing the + * known-summarized range from start_lsn toward end_lsn. + * + * Normally, the summary files should cover non-overlapping WAL ranges, + * but this algorithm is intended to be correct even in case of overlap. + */ + foreach(lc, wslist) + { + WalSummaryFile *ws = lfirst(lc); + + if (ws->start_lsn > current_lsn) + { + /* We found a gap. */ + break; + } + if (ws->end_lsn > current_lsn) + { + /* + * Next summary extends beyond end of previous summary, so extend + * the end of the range known to be summarized. + */ + current_lsn = ws->end_lsn; + + /* + * If the range we know to be summarized has reached the required + * end LSN, we have proved completeness. + */ + if (current_lsn >= end_lsn) + return true; + } + } + + /* + * We either ran out of summary files without reaching the end LSN, or we + * hit a gap in the sequence that resulted in us bailing out of the loop + * above. + */ + *missing_lsn = current_lsn; + return false; +} + +/* + * Open a WAL summary file. + * + * This will throw an error in case of trouble. As an exception, if + * missing_ok = true and the trouble is specifically that the file does + * not exist, it will not throw an error and will return a value less than 0. + */ +File +OpenWalSummaryFile(WalSummaryFile *ws, bool missing_ok) +{ + char path[MAXPGPATH]; + File file; + + snprintf(path, MAXPGPATH, + XLOGDIR "/summaries/%08X%08X%08X%08X%08X.summary", + ws->tli, + LSN_FORMAT_ARGS(ws->start_lsn), + LSN_FORMAT_ARGS(ws->end_lsn)); + + file = PathNameOpenFile(path, O_RDONLY); + if (file < 0 && (errno != EEXIST || !missing_ok)) + ereport(ERROR, + (errcode_for_file_access(), + errmsg("could not open file \"%s\": %m", path))); + + return file; +} + +/* + * Remove a WAL summary file if the last modification time precedes the + * cutoff time. + */ +void +RemoveWalSummaryIfOlderThan(WalSummaryFile *ws, time_t cutoff_time) +{ + char path[MAXPGPATH]; + struct stat statbuf; + + snprintf(path, MAXPGPATH, + XLOGDIR "/summaries/%08X%08X%08X%08X%08X.summary", + ws->tli, + LSN_FORMAT_ARGS(ws->start_lsn), + LSN_FORMAT_ARGS(ws->end_lsn)); + + if (lstat(path, &statbuf) != 0) + { + if (errno == ENOENT) + return; + ereport(ERROR, + (errcode_for_file_access(), + errmsg("could not stat file \"%s\": %m", path))); + } + if (statbuf.st_mtime >= cutoff_time) + return; + if (unlink(path) != 0) + ereport(ERROR, + (errcode_for_file_access(), + errmsg("could not stat file \"%s\": %m", path))); + ereport(DEBUG2, + (errmsg_internal("removing file \"%s\"", path))); +} + +/* + * Test whether a filename looks like a WAL summary file. + */ +static bool +IsWalSummaryFilename(char *filename) +{ + return strspn(filename, "0123456789ABCDEF") == 40 && + strcmp(filename + 40, ".summary") == 0; +} + +/* + * Data read callback for use with CreateBlockRefTableReader. + */ +int +ReadWalSummary(void *wal_summary_io, void *data, int length) +{ + WalSummaryIO *io = wal_summary_io; + int nbytes; + + nbytes = FileRead(io->file, data, length, io->filepos, + WAIT_EVENT_WAL_SUMMARY_READ); + if (nbytes < 0) + ereport(ERROR, + (errcode_for_file_access(), + errmsg("could not write file \"%s\": %m", + FilePathName(io->file)))); + + io->filepos += nbytes; + return nbytes; +} + +/* + * Data write callback for use with WriteBlockRefTable. + */ +int +WriteWalSummary(void *wal_summary_io, void *data, int length) +{ + WalSummaryIO *io = wal_summary_io; + int nbytes; + + nbytes = FileWrite(io->file, data, length, io->filepos, + WAIT_EVENT_WAL_SUMMARY_WRITE); + if (nbytes < 0) + ereport(ERROR, + (errcode_for_file_access(), + errmsg("could not write file \"%s\": %m", + FilePathName(io->file)))); + if (nbytes != length) + ereport(ERROR, + (errcode_for_file_access(), + errmsg("could not write file \"%s\": wrote only %d of %d bytes at offset %u", + FilePathName(io->file), nbytes, + length, (unsigned) io->filepos), + errhint("Check free disk space."))); + + io->filepos += nbytes; + return nbytes; +} + +/* + * Error-reporting callback for use with CreateBlockRefTableReader. + */ +void +ReportWalSummaryError(void *callback_arg, char *fmt,...) +{ + StringInfoData buf; + va_list ap; + int needed; + + initStringInfo(&buf); + for (;;) + { + va_start(ap, fmt); + needed = appendStringInfoVA(&buf, fmt, ap); + va_end(ap); + if (needed == 0) + break; + enlargeStringInfo(&buf, needed); + } + ereport(ERROR, + errcode(ERRCODE_DATA_CORRUPTED), + errmsg_internal("%s", buf.data)); +} + +/* + * Comparator to sort a List of WalSummaryFile objects by start_lsn. + */ +static int +ListComparatorForWalSummaryFiles(const ListCell *a, const ListCell *b) +{ + WalSummaryFile *ws1 = lfirst(a); + WalSummaryFile *ws2 = lfirst(b); + + if (ws1->start_lsn < ws2->start_lsn) + return -1; + if (ws1->start_lsn > ws2->start_lsn) + return 1; + return 0; +} diff --git a/src/backend/backup/walsummaryfuncs.c b/src/backend/backup/walsummaryfuncs.c new file mode 100644 index 0000000000..2e77d38b4a --- /dev/null +++ b/src/backend/backup/walsummaryfuncs.c @@ -0,0 +1,169 @@ +/*------------------------------------------------------------------------- + * + * walsummaryfuncs.c + * SQL-callable functions for accessing WAL summary data. + * + * Portions Copyright (c) 2010-2022, PostgreSQL Global Development Group + * + * src/backend/backup/walsummaryfuncs.c + * + *------------------------------------------------------------------------- + */ + +#include "postgres.h" + +#include "backup/walsummary.h" +#include "common/blkreftable.h" +#include "funcapi.h" +#include "miscadmin.h" +#include "utils/fmgrprotos.h" +#include "utils/pg_lsn.h" + +#define NUM_WS_ATTS 3 +#define NUM_SUMMARY_ATTS 6 +#define MAX_BLOCKS_PER_CALL 256 + +/* + * List the WAL summary files available in pg_wal/summaries. + */ +Datum +pg_available_wal_summaries(PG_FUNCTION_ARGS) +{ + ReturnSetInfo *rsi; + List *wslist; + ListCell *lc; + Datum values[NUM_WS_ATTS]; + bool nulls[NUM_WS_ATTS]; + + InitMaterializedSRF(fcinfo, 0); + rsi = (ReturnSetInfo *) fcinfo->resultinfo; + + memset(nulls, 0, sizeof(nulls)); + + wslist = GetWalSummaries(0, InvalidXLogRecPtr, InvalidXLogRecPtr); + foreach(lc, wslist) + { + WalSummaryFile *ws = (WalSummaryFile *) lfirst(lc); + HeapTuple tuple; + + CHECK_FOR_INTERRUPTS(); + + values[0] = Int64GetDatum((int64) ws->tli); + values[1] = LSNGetDatum(ws->start_lsn); + values[2] = LSNGetDatum(ws->end_lsn); + + tuple = heap_form_tuple(rsi->setDesc, values, nulls); + tuplestore_puttuple(rsi->setResult, tuple); + } + + return (Datum) 0; +} + +/* + * List the contents of a WAL summary file identified by TLI, start LSN, + * and end LSN. + */ +Datum +pg_wal_summary_contents(PG_FUNCTION_ARGS) +{ + ReturnSetInfo *rsi; + Datum values[NUM_SUMMARY_ATTS]; + bool nulls[NUM_SUMMARY_ATTS]; + WalSummaryFile ws; + WalSummaryIO io; + BlockRefTableReader *reader; + int64 raw_tli; + RelFileLocator rlocator; + ForkNumber forknum; + BlockNumber limit_block; + + InitMaterializedSRF(fcinfo, 0); + rsi = (ReturnSetInfo *) fcinfo->resultinfo; + memset(nulls, 0, sizeof(nulls)); + + /* + * Since the timeline could at least in theory be more than 2^31, and + * since we don't have unsigned types at the SQL level, it is passed as a + * 64-bit integer. Test whether it's out of range. + */ + raw_tli = PG_GETARG_INT64(0); + if (raw_tli < 1 || raw_tli > PG_INT32_MAX) + ereport(ERROR, + errcode(ERRCODE_INVALID_PARAMETER_VALUE), + errmsg("invalid timeline %lld", (long long) raw_tli)); + + /* Prepare to read the specified WAL summry file. */ + ws.tli = (TimeLineID) raw_tli; + ws.start_lsn = PG_GETARG_LSN(1); + ws.end_lsn = PG_GETARG_LSN(2); + io.filepos = 0; + io.file = OpenWalSummaryFile(&ws, false); + reader = CreateBlockRefTableReader(ReadWalSummary, &io, + FilePathName(io.file), + ReportWalSummaryError, NULL); + + /* Loop over relation forks. */ + while (BlockRefTableReaderNextRelation(reader, &rlocator, &forknum, + &limit_block)) + { + BlockNumber blocks[MAX_BLOCKS_PER_CALL]; + HeapTuple tuple; + + CHECK_FOR_INTERRUPTS(); + + values[0] = ObjectIdGetDatum(rlocator.relNumber); + values[1] = ObjectIdGetDatum(rlocator.spcOid); + values[2] = ObjectIdGetDatum(rlocator.dbOid); + values[3] = Int16GetDatum((int16) forknum); + + /* Loop over blocks within the current relation fork. */ + while (true) + { + unsigned nblocks; + unsigned i; + + CHECK_FOR_INTERRUPTS(); + + nblocks = BlockRefTableReaderGetBlocks(reader, blocks, + MAX_BLOCKS_PER_CALL); + if (nblocks == 0) + break; + + /* + * For each block that we specifically know to have been modified, + * emit a row with that block number and limit_block = false. + */ + values[5] = BoolGetDatum(false); + for (i = 0; i < nblocks; ++i) + { + values[4] = Int64GetDatum((int64) blocks[i]); + + tuple = heap_form_tuple(rsi->setDesc, values, nulls); + tuplestore_puttuple(rsi->setResult, tuple); + } + + /* + * If the limit block is not InvalidBlockNumber, emit an exta row + * with that block number and limit_block = true. + * + * There is no point in doing this when the limit_block is + * InvalidBlockNumber, because no block with that number or any + * higher number can ever exist. + */ + if (BlockNumberIsValid(limit_block)) + { + values[4] = Int64GetDatum((int64) limit_block); + values[5] = BoolGetDatum(true); + + tuple = heap_form_tuple(rsi->setDesc, values, nulls); + tuplestore_puttuple(rsi->setResult, tuple); + } + } + } + + /* Cleanup */ + DestroyBlockRefTableReader(reader); + FileClose(io.file); + + return (Datum) 0; +} diff --git a/src/backend/postmaster/Makefile b/src/backend/postmaster/Makefile index 047448b34e..367a46c617 100644 --- a/src/backend/postmaster/Makefile +++ b/src/backend/postmaster/Makefile @@ -24,6 +24,7 @@ OBJS = \ postmaster.o \ startup.o \ syslogger.o \ + walsummarizer.o \ walwriter.o include $(top_srcdir)/src/backend/common.mk diff --git a/src/backend/postmaster/auxprocess.c b/src/backend/postmaster/auxprocess.c index cae6feb356..0c15c1777d 100644 --- a/src/backend/postmaster/auxprocess.c +++ b/src/backend/postmaster/auxprocess.c @@ -21,6 +21,7 @@ #include "postmaster/auxprocess.h" #include "postmaster/bgwriter.h" #include "postmaster/startup.h" +#include "postmaster/walsummarizer.h" #include "postmaster/walwriter.h" #include "replication/walreceiver.h" #include "storage/bufmgr.h" @@ -80,6 +81,9 @@ AuxiliaryProcessMain(AuxProcType auxtype) case WalReceiverProcess: MyBackendType = B_WAL_RECEIVER; break; + case WalSummarizerProcess: + MyBackendType = B_WAL_SUMMARIZER; + break; default: elog(PANIC, "unrecognized process type: %d", (int) MyAuxProcType); MyBackendType = B_INVALID; @@ -161,6 +165,10 @@ AuxiliaryProcessMain(AuxProcType auxtype) WalReceiverMain(); proc_exit(1); + case WalSummarizerProcess: + WalSummarizerMain(); + proc_exit(1); + default: elog(PANIC, "unrecognized process type: %d", (int) MyAuxProcType); proc_exit(1); diff --git a/src/backend/postmaster/meson.build b/src/backend/postmaster/meson.build index cda921fd10..a30eb6692f 100644 --- a/src/backend/postmaster/meson.build +++ b/src/backend/postmaster/meson.build @@ -12,5 +12,6 @@ backend_sources += files( 'postmaster.c', 'startup.c', 'syslogger.c', + 'walsummarizer.c', 'walwriter.c', ) diff --git a/src/backend/postmaster/postmaster.c b/src/backend/postmaster/postmaster.c index 54e9bfb8c4..0538b84ef8 100644 --- a/src/backend/postmaster/postmaster.c +++ b/src/backend/postmaster/postmaster.c @@ -115,6 +115,7 @@ #include "postmaster/pgarch.h" #include "postmaster/postmaster.h" #include "postmaster/syslogger.h" +#include "postmaster/walsummarizer.h" #include "replication/logicallauncher.h" #include "replication/walsender.h" #include "storage/fd.h" @@ -251,6 +252,7 @@ static pid_t StartupPID = 0, CheckpointerPID = 0, WalWriterPID = 0, WalReceiverPID = 0, + WalSummarizerPID = 0, AutoVacPID = 0, PgArchPID = 0, SysLoggerPID = 0; @@ -442,6 +444,7 @@ static bool CreateOptsFile(int argc, char *argv[], char *fullprogname); static pid_t StartChildProcess(AuxProcType type); static void StartAutovacuumWorker(void); static void MaybeStartWalReceiver(void); +static void MaybeStartWalSummarizer(void); static void InitPostmasterDeathWatchHandle(void); /* @@ -561,6 +564,7 @@ static void ShmemBackendArrayRemove(Backend *bn); #define StartCheckpointer() StartChildProcess(CheckpointerProcess) #define StartWalWriter() StartChildProcess(WalWriterProcess) #define StartWalReceiver() StartChildProcess(WalReceiverProcess) +#define StartWalSummarizer() StartChildProcess(WalSummarizerProcess) /* Macros to check exit status of a child process */ #define EXIT_STATUS_0(st) ((st) == 0) @@ -1847,6 +1851,9 @@ ServerLoop(void) if (WalReceiverRequested) MaybeStartWalReceiver(); + /* If we need to start a WAL summarizer, try to do that now */ + MaybeStartWalSummarizer(); + /* Get other worker processes running, if needed */ if (StartWorkerNeeded || HaveCrashedWorker) maybe_start_bgworkers(); @@ -2714,6 +2721,8 @@ process_pm_reload_request(void) signal_child(WalWriterPID, SIGHUP); if (WalReceiverPID != 0) signal_child(WalReceiverPID, SIGHUP); + if (WalSummarizerPID != 0) + signal_child(WalSummarizerPID, SIGHUP); if (AutoVacPID != 0) signal_child(AutoVacPID, SIGHUP); if (PgArchPID != 0) @@ -3067,6 +3076,7 @@ process_pm_child_exit(void) BgWriterPID = StartBackgroundWriter(); if (WalWriterPID == 0) WalWriterPID = StartWalWriter(); + MaybeStartWalSummarizer(); /* * Likewise, start other special children as needed. In a restart @@ -3185,6 +3195,20 @@ process_pm_child_exit(void) continue; } + /* + * Was it the wal summarizer? Normal exit can be ignored; we'll start + * a new one at the next iteration of the postmaster's main loop, if + * necessary. Any other exit condition is treated as a crash. + */ + if (pid == WalSummarizerPID) + { + WalSummarizerPID = 0; + if (!EXIT_STATUS_0(exitstatus)) + HandleChildCrash(pid, exitstatus, + _("WAL summarizer process")); + continue; + } + /* * Was it the autovacuum launcher? Normal exit can be ignored; we'll * start a new one at the next iteration of the postmaster's main @@ -3580,6 +3604,12 @@ HandleChildCrash(int pid, int exitstatus, const char *procname) else if (WalReceiverPID != 0 && take_action) sigquit_child(WalReceiverPID); + /* Take care of the walsummarizer too */ + if (pid == WalSummarizerPID) + WalSummarizerPID = 0; + else if (WalSummarizerPID != 0 && take_action) + sigquit_child(WalSummarizerPID); + /* Take care of the autovacuum launcher too */ if (pid == AutoVacPID) AutoVacPID = 0; @@ -3730,6 +3760,8 @@ PostmasterStateMachine(void) signal_child(StartupPID, SIGTERM); if (WalReceiverPID != 0) signal_child(WalReceiverPID, SIGTERM); + if (WalSummarizerPID != 0) + signal_child(WalSummarizerPID, SIGTERM); /* checkpointer, archiver, stats, and syslogger may continue for now */ /* Now transition to PM_WAIT_BACKENDS state to wait for them to die */ @@ -3756,6 +3788,7 @@ PostmasterStateMachine(void) if (CountChildren(BACKEND_TYPE_ALL - BACKEND_TYPE_WALSND) == 0 && StartupPID == 0 && WalReceiverPID == 0 && + WalSummarizerPID == 0 && BgWriterPID == 0 && (CheckpointerPID == 0 || (!FatalError && Shutdown < ImmediateShutdown)) && @@ -3853,6 +3886,7 @@ PostmasterStateMachine(void) /* These other guys should be dead already */ Assert(StartupPID == 0); Assert(WalReceiverPID == 0); + Assert(WalSummarizerPID == 0); Assert(BgWriterPID == 0); Assert(CheckpointerPID == 0); Assert(WalWriterPID == 0); @@ -4074,6 +4108,8 @@ TerminateChildren(int signal) signal_child(WalWriterPID, signal); if (WalReceiverPID != 0) signal_child(WalReceiverPID, signal); + if (WalSummarizerPID != 0) + signal_child(WalSummarizerPID, signal); if (AutoVacPID != 0) signal_child(AutoVacPID, signal); if (PgArchPID != 0) @@ -5380,6 +5416,10 @@ StartChildProcess(AuxProcType type) ereport(LOG, (errmsg("could not fork WAL receiver process: %m"))); break; + case WalSummarizerProcess: + ereport(LOG, + (errmsg("could not fork WAL summarizer process: %m"))); + break; default: ereport(LOG, (errmsg("could not fork process: %m"))); @@ -5516,6 +5556,19 @@ MaybeStartWalReceiver(void) } } +/* + * MaybeStartWalSummarizer + * Start the WAL summarizer process, if not running and our state allows. + */ +static void +MaybeStartWalSummarizer(void) +{ + if (wal_summarize_mb != 0 && WalSummarizerPID == 0 && + (pmState == PM_RUN || pmState == PM_HOT_STANDBY) && + Shutdown <= SmartShutdown) + WalSummarizerPID = StartWalSummarizer(); +} + /* * Create the opts file diff --git a/src/backend/postmaster/walsummarizer.c b/src/backend/postmaster/walsummarizer.c new file mode 100644 index 0000000000..34bd254183 --- /dev/null +++ b/src/backend/postmaster/walsummarizer.c @@ -0,0 +1,1414 @@ +/*------------------------------------------------------------------------- + * + * walsummarizer.c + * + * Background process to perform WAL summarization, if it is enabled. + * It continuously scans the write-ahead log and periodically emits a + * summary file which indicates which blocks in which relation forks + * were modified by WAL records in the LSN range covered by the summary + * file. See walsummary.c and blkreftable.c for more details on the + * naming and contents of WAL summary files. + * + * If configured to do, this background process will also remove WAL + * summary files when the file timestamp is older than a configurable + * threshold (but only if the WAL has been removed first). + * + * Portions Copyright (c) 1996-2022, PostgreSQL Global Development Group + * + * IDENTIFICATION + * src/backend/postmaster/walsummarizer.c + * + *------------------------------------------------------------------------- + */ +#include "postgres.h" + +#include "access/timeline.h" +#include "access/xlog.h" +#include "access/xlog_internal.h" +#include "access/xlogrecovery.h" +#include "access/xlogutils.h" +#include "backup/walsummary.h" +#include "catalog/storage_xlog.h" +#include "common/blkreftable.h" +#include "libpq/pqsignal.h" +#include "miscadmin.h" +#include "postmaster/bgwriter.h" +#include "postmaster/interrupt.h" +#include "postmaster/walsummarizer.h" +#include "replication/walreceiver.h" +#include "storage/fd.h" +#include "storage/ipc.h" +#include "storage/lwlock.h" +#include "storage/latch.h" +#include "storage/proc.h" +#include "storage/procsignal.h" +#include "storage/shmem.h" +#include "storage/spin.h" +#include "utils/guc.h" +#include "utils/memutils.h" +#include "utils/wait_event.h" + +/* + * Data in shared memory related to WAL summarization. + */ +typedef struct +{ + /* + * These fields are protected by WALSummarizerLock. + * + * Until we've discovered what summary files already exist on disk and + * stored that information in shared memory, initialized is false and the + * other fields here contain no meaningful information. After that has + * been done, initialized is true. + * + * summarized_tli and summarized_lsn indicate the last LSN and TLI at + * which the next summary file will start. Normally, these are the LSN and + * TLI at which the last file ended; in such case, lsn_is_exact is true. + * If, however, the LSN is just an approximation, then lsn_is_exact is + * false. This can happen if, for example, there are no existing WAL + * summary files at startup. In that case, we have to derive the position + * at which to start summarizing from the WAL files that exist on disk, + * and so the LSN might point to the start of the next file even though + * that might happen to be in the middle of a WAL record. + * + * summarizer_pgprocno is the pgprocno value for the summarizer process, + * if one is running, or else INVALID_PGPROCNO. + * + * pending_lsn is used by the summarizer to advertise the ending LSN of a + * record it has recently read. It shouldn't ever be less than + * summarized_lsn, but might be greater, because the summarizer buffers + * data for a range of LSNs in memory before writing out a new file. + * + * switch_requested can be set to true to notify the summarizer that a new + * WAL summary file should be written as soon as possible, without trying + * to read more WAL first. + */ + bool initialized; + TimeLineID summarized_tli; + XLogRecPtr summarized_lsn; + bool lsn_is_exact; + int summarizer_pgprocno; + XLogRecPtr pending_lsn; + bool switch_requested; + + /* + * This field handles its own synchronizaton. + */ + ConditionVariable summary_file_cv; +} WalSummarizerData; + +/* + * Private data for our xlogreader's page read callback. + */ +typedef struct +{ + TimeLineID tli; + bool historic; + XLogRecPtr read_upto; + bool end_of_wal; + bool waited; + XLogRecPtr redo_pointer; + bool redo_pointer_reached; + XLogRecPtr redo_pointer_refresh_lsn; +} SummarizerReadLocalXLogPrivate; + +/* Pointer to shared memory state. */ +static WalSummarizerData *WalSummarizerCtl; + +/* + * When we reach end of WAL and need to read more, we sleep for a number of + * milliseconds that is a integer multiple of MS_PER_SLEEP_QUANTUM. This is + * the multiplier. It should vary between 1 and MAX_SLEEP_QUANTA, depending + * on system activity. See summarizer_wait_for_wal() for how we adjust this. + */ +static long sleep_quanta = 1; + +/* + * The sleep time will always be a multiple of 200ms and will not exceed + * one minute (300 * 200 = 60 * 1000). + */ +#define MAX_SLEEP_QUANTA 300 +#define MS_PER_SLEEP_QUANTUM 200 + +/* + * This is a count of the number of pages of WAL that we've read since the + * last time we waited for more WAL to appear. + */ +static long pages_read_since_last_sleep = 0; + +/* + * Most recent RedoRecPtr value observed by MaybeRemoveOldWalSummaries. + */ +static XLogRecPtr redo_pointer_at_last_summary_removal = InvalidXLogRecPtr; + +/* + * GUC parameters + */ +int wal_summarize_mb = 256; +int wal_summarize_keep_time = 7 * 24 * 60; + +static XLogRecPtr GetLatestLSN(TimeLineID *tli); +static void HandleWalSummarizerInterrupts(void); +static XLogRecPtr SummarizeWAL(TimeLineID tli, bool historic, + XLogRecPtr start_lsn, bool exact, + XLogRecPtr cutoff_lsn, XLogRecPtr maximum_lsn); +static void SummarizeSmgrRecord(XLogReaderState *xlogreader, + BlockRefTable *brtab); +static void SummarizeXactRecord(XLogReaderState *xlogreader, + BlockRefTable *brtab); +static int summarizer_read_local_xlog_page(XLogReaderState *state, + XLogRecPtr targetPagePtr, + int reqLen, + XLogRecPtr targetRecPtr, + char *cur_page); +static void summarizer_wait_for_wal(void); +static void MaybeRemoveOldWalSummaries(void); + +/* + * Amount of shared memory required for this module. + */ +Size +WalSummarizerShmemSize(void) +{ + return sizeof(WalSummarizerData); +} + +/* + * Create or attach to shared memory segment for this module. + */ +void +WalSummarizerShmemInit(void) +{ + bool found; + + WalSummarizerCtl = (WalSummarizerData *) + ShmemInitStruct("Wal Summarizer Ctl", WalSummarizerShmemSize(), + &found); + + if (!found) + { + /* + * First time through, so initialize. + * + * We're just filling in dummy values here -- the real initialization + * will happen when GetOldestUnsummarizedLSN() is called for the first + * time. + */ + WalSummarizerCtl->initialized = false; + WalSummarizerCtl->summarized_tli = 0; + WalSummarizerCtl->summarized_lsn = InvalidXLogRecPtr; + WalSummarizerCtl->lsn_is_exact = false; + WalSummarizerCtl->summarizer_pgprocno = INVALID_PGPROCNO; + WalSummarizerCtl->pending_lsn = InvalidXLogRecPtr; + WalSummarizerCtl->switch_requested = false; + ConditionVariableInit(&WalSummarizerCtl->summary_file_cv); + } +} + +/* + * Entry point for walsummarizer process. + */ +void +WalSummarizerMain(void) +{ + sigjmp_buf local_sigjmp_buf; + MemoryContext context; + + /* + * Within this function, 'current_lsn' and 'current_tli' refer to the + * point from which the next WAL summary file should start. 'exact' is + * true if 'current_lsn' is known to be the start of a WAL recod or WAL + * segment, and false if it might be in the middle of a record someplace. + * + * 'switch_lsn' and 'switch_tli', if set, are the LSN at which we need to + * switch to a new timeline and the timeline to which we need to switch. + * If not set, we either haven't figured out the answers yet or we're + * already on the latest timeline. + */ + XLogRecPtr current_lsn; + TimeLineID current_tli; + bool exact; + XLogRecPtr switch_lsn = InvalidXLogRecPtr; + TimeLineID switch_tli = 0; + + ereport(DEBUG1, + (errmsg_internal("WAL summarizer started"))); + + /* + * Properly accept or ignore signals the postmaster might send us + * + * We have no particular use for SIGINT at the moment, but seems + * reasonable to treat like SIGTERM. + */ + pqsignal(SIGHUP, SignalHandlerForConfigReload); + pqsignal(SIGINT, SignalHandlerForShutdownRequest); + pqsignal(SIGTERM, SignalHandlerForShutdownRequest); + /* SIGQUIT handler was already set up by InitPostmasterChild */ + pqsignal(SIGALRM, SIG_IGN); + pqsignal(SIGPIPE, SIG_IGN); + pqsignal(SIGUSR1, procsignal_sigusr1_handler); + pqsignal(SIGUSR2, SIG_IGN); /* not used */ + + /* Advertise ourselves. */ + LWLockAcquire(WALSummarizerLock, LW_EXCLUSIVE); + WalSummarizerCtl->summarizer_pgprocno = MyProc->pgprocno; + LWLockRelease(WALSummarizerLock); + + /* Create and switch to a memory context that we can reset on error. */ + context = AllocSetContextCreate(TopMemoryContext, + "Wal Summarizer", + ALLOCSET_DEFAULT_SIZES); + MemoryContextSwitchTo(context); + + /* + * Reset some signals that are accepted by postmaster but not here + */ + pqsignal(SIGCHLD, SIG_DFL); + + /* + * If an exception is encountered, processing resumes here. + */ + if (sigsetjmp(local_sigjmp_buf, 1) != 0) + { + /* Since not using PG_TRY, must reset error stack by hand */ + error_context_stack = NULL; + + /* Prevent interrupts while cleaning up */ + HOLD_INTERRUPTS(); + + /* Report the error to the server log */ + EmitErrorReport(); + + /* Release resources we might have acquired. */ + LWLockReleaseAll(); + ConditionVariableCancelSleep(); + pgstat_report_wait_end(); + ReleaseAuxProcessResources(false); + AtEOXact_Files(false); + AtEOXact_HashTables(false); + + /* + * Now return to normal top-level context and clear ErrorContext for + * next time. + */ + MemoryContextSwitchTo(context); + FlushErrorState(); + + /* Flush any leaked data in the top-level context */ + MemoryContextResetAndDeleteChildren(context); + + /* Now we can allow interrupts again */ + RESUME_INTERRUPTS(); + + /* + * Sleep for 10 seconds before attempting to resume operations in + * order to avoid excessing logging. + * + * Many of the likely error conditions are things that will repeat + * every time. For example, if the WAL can't be read or the summary + * can't be written, only administrator action will cure the problem. + * So a really fast retry time doesn't seem to be especially + * beneficial, and it will clutter the logs. + */ + (void) WaitLatch(MyLatch, + WL_TIMEOUT | WL_EXIT_ON_PM_DEATH, + 10000, + WAIT_EVENT_WAL_SUMMARIZER_ERROR); + } + + /* We can now handle ereport(ERROR) */ + PG_exception_stack = &local_sigjmp_buf; + + /* + * Unblock signals (they were blocked when the postmaster forked us) + */ + sigprocmask(SIG_SETMASK, &UnBlockSig, NULL); + + /* + * Fetch information about previous progress from shared memory. + * + * If we discover that WAL summarization is not enabled, just exit. + */ + current_lsn = GetOldestUnsummarizedLSN(¤t_tli, &exact); + if (XLogRecPtrIsInvalid(current_lsn)) + proc_exit(0); + + /* + * Loop forever + */ + for (;;) + { + XLogRecPtr latest_lsn; + TimeLineID latest_tli; + XLogRecPtr cutoff_lsn; + XLogRecPtr end_of_summary_lsn; + + /* Flush any leaked data in the top-level context */ + MemoryContextResetAndDeleteChildren(context); + + /* Process any signals received recently. */ + HandleWalSummarizerInterrupts(); + + /* If it's time to remove any old WAL summaries, do that now. */ + MaybeRemoveOldWalSummaries(); + + /* Find the LSN and TLI up to which we can safely summarize. */ + latest_lsn = GetLatestLSN(&latest_tli); + + /* + * If we're summarizing a historic timeline and we haven't yet + * computed the point at which to switch to the next timeline, do that + * now. + * + * Note that if this is a standby, what was previously the current + * timeline could become historic at any time. + * + * We could try to make this more efficient by caching the results of + * readTimeLineHistory when latest_tli has not changed, but since we + * only have to do this once per timeline switch, we probably wouldn't + * save any significant amount of work in practice. + */ + if (current_tli != latest_tli && XLogRecPtrIsInvalid(switch_lsn)) + { + List *tles = readTimeLineHistory(latest_tli); + + switch_lsn = tliSwitchPoint(current_tli, tles, &switch_tli); + elog(DEBUG2, + "switch point from TLI %u to TLI %u is at %X/%X", + current_tli, switch_tli, LSN_FORMAT_ARGS(switch_lsn)); + } + + /* + * wal_summarize_mb sets a soft limit on the amont of WAL covered by a + * single summary file. If we read a WAL record that ends after the + * cutoff LSN computed here, we'll stop the summary. In most cases, it + * will actually stop earlier than that, but this is here as a + * backstop. + */ + cutoff_lsn = current_lsn + wal_summarize_mb * 1024 * 1024; + if (!XLogRecPtrIsInvalid(switch_lsn) && cutoff_lsn > switch_lsn) + cutoff_lsn = switch_lsn; + elog(DEBUG2, + "WAL summarization cutoff is TLI %d @ %X/%X, flush position is %X/%X", + current_tli, LSN_FORMAT_ARGS(cutoff_lsn), LSN_FORMAT_ARGS(latest_lsn)); + + /* Summarize WAL. */ + end_of_summary_lsn = SummarizeWAL(current_tli, + current_tli != latest_tli, + current_lsn, exact, + cutoff_lsn, latest_lsn); + Assert(!XLogRecPtrIsInvalid(end_of_summary_lsn)); + Assert(end_of_summary_lsn >= current_lsn); + + /* + * Update state for next loop iteration. + * + * Next summary file should start from exactly where this one ended. + * Timeline remains unchanged unless a switch LSN was computed and we + * have reached it. + */ + current_lsn = end_of_summary_lsn; + exact = true; + if (!XLogRecPtrIsInvalid(switch_lsn) && cutoff_lsn >= switch_lsn) + { + current_tli = switch_tli; + switch_lsn = InvalidXLogRecPtr; + switch_tli = 0; + } + + /* Update state in shared memory. */ + LWLockAcquire(WALSummarizerLock, LW_EXCLUSIVE); + Assert(WalSummarizerCtl->pending_lsn <= end_of_summary_lsn); + WalSummarizerCtl->summarized_lsn = end_of_summary_lsn; + WalSummarizerCtl->summarized_tli = current_tli; + WalSummarizerCtl->lsn_is_exact = true; + WalSummarizerCtl->pending_lsn = end_of_summary_lsn; + WalSummarizerCtl->switch_requested = false; + LWLockRelease(WALSummarizerLock); + + /* Wake up anyone waiting for more summary files to be written. */ + ConditionVariableBroadcast(&WalSummarizerCtl->summary_file_cv); + } +} + +/* + * Get the oldest LSN in this server's timeline history that has not yet been + * summarized. + * + * If *tli != NULL, it will be set to the TLI for the LSN that is returned. + * + * If *lsn_is_exact != NULL, it will be set to true if the returned LSN is + * necessarily the start of a WAL record and false if it's just the beginning + * of a WAL segment. + */ +XLogRecPtr +GetOldestUnsummarizedLSN(TimeLineID *tli, bool *lsn_is_exact) +{ + TimeLineID latest_tli; + LWLockMode mode = LW_SHARED; + int n; + List *tles; + XLogRecPtr unsummarized_lsn; + TimeLineID unsummarized_tli = 0; + bool should_make_exact = false; + List *existing_summaries; + ListCell *lc; + + /* If not summarizing WAL, do nothing. */ + if (wal_summarize_mb == 0) + return InvalidXLogRecPtr; + + /* + * Initially, we acquire the lock in shared mode and try to fetch the + * required information. If the data structure hasn't been initialized, we + * reacquire the lock in shared mode so that we can initialize it. + * However, if someone else does that first before we get the lock, then + * we can just return the requested information after all. + */ + while (true) + { + LWLockAcquire(WALSummarizerLock, mode); + + if (WalSummarizerCtl->initialized) + { + unsummarized_lsn = WalSummarizerCtl->summarized_lsn; + if (tli != NULL) + *tli = WalSummarizerCtl->summarized_tli; + if (lsn_is_exact != NULL) + *lsn_is_exact = WalSummarizerCtl->lsn_is_exact; + LWLockRelease(WALSummarizerLock); + return unsummarized_lsn; + } + + if (mode == LW_EXCLUSIVE) + break; + + LWLockRelease(WALSummarizerLock); + mode = LW_EXCLUSIVE; + } + + /* + * The data structure needs to be initialized, and we are the first to + * obtain the lock in exclusive mode, so it's our job to do that + * initialization. + * + * So, find the oldest timeline on which WAL still exists, and the + * earliest segment for which it exists. + */ + (void) GetLatestLSN(&latest_tli); + tles = readTimeLineHistory(latest_tli); + for (n = list_length(tles) - 1; n >= 0; --n) + { + TimeLineHistoryEntry *tle = list_nth(tles, n); + XLogSegNo oldest_segno; + + oldest_segno = XLogGetOldestSegno(tle->tli); + if (oldest_segno != 0) + { + /* Compute oldest LSN that still exists on disk. */ + XLogSegNoOffsetToRecPtr(oldest_segno, 0, wal_segment_size, + unsummarized_lsn); + + unsummarized_tli = tle->tli; + break; + } + } + + /* It really should not be possible for us to find no WAL. */ + if (unsummarized_tli == 0) + ereport(ERROR, + errcode(ERRCODE_INTERNAL_ERROR), + errmsg_internal("no WAL found on timeline %d", latest_tli)); + + /* + * Don't try to summarize anything older than the end LSN of the newest + * summary file that exists for this timeline. + */ + existing_summaries = + GetWalSummaries(unsummarized_tli, + InvalidXLogRecPtr, InvalidXLogRecPtr); + foreach(lc, existing_summaries) + { + WalSummaryFile *ws = lfirst(lc); + + if (ws->end_lsn > unsummarized_lsn) + { + unsummarized_lsn = ws->end_lsn; + should_make_exact = true; + } + } + + /* Update shared memory with the discovered values. */ + WalSummarizerCtl->initialized = true; + WalSummarizerCtl->summarized_lsn = unsummarized_lsn; + WalSummarizerCtl->summarized_tli = unsummarized_tli; + WalSummarizerCtl->lsn_is_exact = should_make_exact; + WalSummarizerCtl->pending_lsn = unsummarized_lsn; + + /* Also return the to the caller as required. */ + if (tli != NULL) + *tli = WalSummarizerCtl->summarized_tli; + if (lsn_is_exact != NULL) + *lsn_is_exact = WalSummarizerCtl->lsn_is_exact; + LWLockRelease(WALSummarizerLock); + + return unsummarized_lsn; +} + +/* + * Attempt to set the WAL summarizer's latch. + * + * This might not work, because there's no guarantee that the WAL summarizer + * process was successfully started, and it also might have started but + * subsequently terminated. So, under normal circumstances, this will get the + * latch set, but there's no guarantee. + */ +void +SetWalSummarizerLatch(void) +{ + int pgprocno; + + LWLockAcquire(WALSummarizerLock, LW_EXCLUSIVE); + pgprocno = WalSummarizerCtl->summarizer_pgprocno; + LWLockRelease(WALSummarizerLock); + + if (pgprocno != INVALID_PGPROCNO) + SetLatch(&ProcGlobal->allProcs[pgprocno].procLatch); +} + +/* + * Wait until WAL summarization reaches the given LSN, but not longer than + * the given timeout. + * + * The return value is the first still-unsummarized LSN. If it's greater than + * or equal to the passed LSN, then that LSN was reached. If not, we timed out. + */ +XLogRecPtr +WaitForWalSummarization(XLogRecPtr lsn, long timeout) +{ + TimestampTz start_time = GetCurrentTimestamp(); + TimestampTz deadline = TimestampTzPlusMilliseconds(start_time, timeout); + XLogRecPtr summarized_lsn; + + Assert(!XLogRecPtrIsInvalid(lsn)); + Assert(timeout > 0); + + while (1) + { + TimestampTz now; + long remaining_timeout; + + /* + * If the LSN summarized on disk has reached the target value, stop. + * If it hasn't, but the in-memory value has reached the target value, + * request that a file be written as soon as possible. + */ + LWLockAcquire(WALSummarizerLock, LW_EXCLUSIVE); + summarized_lsn = WalSummarizerCtl->summarized_lsn; + if (summarized_lsn < lsn && + WalSummarizerCtl->pending_lsn >= lsn) + WalSummarizerCtl->switch_requested = true; + LWLockRelease(WALSummarizerLock); + if (summarized_lsn >= lsn) + break; + + /* Timeout reached? If yes, stop. */ + now = GetCurrentTimestamp(); + remaining_timeout = TimestampDifferenceMilliseconds(now, deadline); + if (remaining_timeout <= 0) + break; + + /* + * Limit the sleep to 1 second, because we may need to request a + * switch. + */ + if (remaining_timeout > 1000) + remaining_timeout = 1000; + + /* Wait and see. */ + ConditionVariableTimedSleep(&WalSummarizerCtl->summary_file_cv, + remaining_timeout, + WAIT_EVENT_WAL_SUMMARY_READY); + } + + return summarized_lsn; +} + +/* + * Get the latest LSN that is eligible to be summarized, and set *tli to the + * corresponding timeline. + */ +static XLogRecPtr +GetLatestLSN(TimeLineID *tli) +{ + if (!RecoveryInProgress()) + { + /* Don't summarize WAL before it's flushed. */ + return GetFlushRecPtr(tli); + } + else + { + XLogRecPtr flush_lsn; + TimeLineID flush_tli; + XLogRecPtr replay_lsn; + TimeLineID replay_tli; + + /* + * What we really want to know is how much WAL has been flushed to + * disk, but the only flush position available is the one provided by + * the walreceiver, which may not be running, because this could be + * crash recovery or recovery via restore_command. So use either the + * WAL receiver's flush position or the replay position, whichever is + * further ahead, on the theory that if the WAL has been replayed then + * it must also have been flushed to disk. + */ + flush_lsn = GetWalRcvFlushRecPtr(NULL, &flush_tli); + replay_lsn = GetXLogReplayRecPtr(&replay_tli); + if (flush_lsn > replay_lsn) + { + *tli = flush_tli; + return flush_lsn; + } + else + { + *tli = replay_tli; + return replay_lsn; + } + } +} + +/* + * Interrupt handler for main loop of WAL summarizer process. + */ +static void +HandleWalSummarizerInterrupts(void) +{ + if (ProcSignalBarrierPending) + ProcessProcSignalBarrier(); + + if (ConfigReloadPending) + { + ConfigReloadPending = false; + ProcessConfigFile(PGC_SIGHUP); + } + + if (ShutdownRequestPending || wal_summarize_mb == 0) + { + ereport(DEBUG1, + errmsg_internal("WAL summarizer shutting down")); + proc_exit(0); + } + + /* Perform logging of memory contexts of this process */ + if (LogMemoryContextPending) + ProcessLogMemoryContextInterrupt(); +} + +/* + * Summarize a range of WAL records on a single timeline. + * + * 'tli' is the timeline to be summarized. 'historic' should be false if the + * timeline in question is the latest one and true otherwise. + * + * 'start_lsn' is the point at which we should start summarizing. If this + * value comes from the end LSN of the previous record as returned by the + * xlograder machinery, 'exact' should be true; otherwise, 'exact' should + * be false, and this function will search forward for the start of a valid + * WAL record. + * + * 'cutoff_lsn' is the point at which we should stop summarizing. The first + * record that ends at or after cutoff_lsn will be the last one included + * in the summary. + * + * 'maximum_lsn' identifies the point beyond which we can't count on being + * able to read any more WAL. It should be the switch point when reading a + * historic timeline, or the most-recently-measured end of WAL when reading + * the current timeline. + * + * The return value is the LSN at which the WAL summary actually ends. Most + * often, a summary file ends because we notice that a checkpoint has + * occurred and reach the redo pointer of that checkpoint, but sometimes + * we stop for other reasons, such as a timeline switch, or reading a record + * that ends after the cutoff_lsn. + */ +static XLogRecPtr +SummarizeWAL(TimeLineID tli, bool historic, + XLogRecPtr start_lsn, bool exact, + XLogRecPtr cutoff_lsn, XLogRecPtr maximum_lsn) +{ + SummarizerReadLocalXLogPrivate *private_data; + XLogReaderState *xlogreader; + XLogRecPtr summary_start_lsn; + XLogRecPtr summary_end_lsn = cutoff_lsn; + char temp_path[MAXPGPATH]; + char final_path[MAXPGPATH]; + WalSummaryIO io; + BlockRefTable *brtab = CreateEmptyBlockRefTable(); + + /* Initialize private data for xlogreader. */ + private_data = (SummarizerReadLocalXLogPrivate *) + palloc0(sizeof(SummarizerReadLocalXLogPrivate)); + private_data->tli = tli; + private_data->historic = historic; + private_data->read_upto = maximum_lsn; + private_data->redo_pointer = GetRedoRecPtr(); + private_data->redo_pointer_refresh_lsn = start_lsn; + private_data->redo_pointer_reached = + (start_lsn >= private_data->redo_pointer); + + /* Create xlogreader. */ + xlogreader = XLogReaderAllocate(wal_segment_size, NULL, + XL_ROUTINE(.page_read = &summarizer_read_local_xlog_page, + .segment_open = &wal_segment_open, + .segment_close = &wal_segment_close), + private_data); + if (xlogreader == NULL) + ereport(ERROR, + (errcode(ERRCODE_OUT_OF_MEMORY), + errmsg("out of memory"), + errdetail("Failed while allocating a WAL reading processor."))); + + /* + * When exact = false, we're starting from an arbitrary point in the WAL + * and must search forward for the start of the next record. + * + * When exact = true, start_lsn should be either the LSN where a record + * begins, or the LSN of a page where the page header is immediately + * followed by the start of a new record. XLogBeginRead should tolerate + * either case. + * + * We need to allow for both cases because the behavior of xlogreader + * varies. When a record spans two or more xlog pages, the ending LSN + * reported by xlogreader will be the starting LSN of the following + * record, but when an xlog page boundary falls between two records, the + * end LSN for the first will be reported as the first byte of the + * following page. We can't know until we read that page how large the + * header will be, but we'll have to skip over it to find the next record. + */ + if (exact) + { + /* + * Even if start_lsn is the beginning of a page rather than the + * beginning of the first record on that page, we should still use it + * as the start LSN for the summary file. That's because we detect + * missing summary files by looking for cases where the end LSN of one + * file is less than the start LSN of the next file. When only a page + * header is skipped, nothing has been missed. + */ + XLogBeginRead(xlogreader, start_lsn); + summary_start_lsn = start_lsn; + } + else + { + summary_start_lsn = XLogFindNextRecord(xlogreader, start_lsn); + if (XLogRecPtrIsInvalid(summary_start_lsn)) + { + /* + * If we hit end-of-WAL while trying to find the next valid + * record, we must be on a historic timeline that has no valid + * records that begin after start_lsn and before end of WAL. + */ + if (private_data->end_of_wal) + { + ereport(DEBUG1, + errmsg_internal("could not read WAL from timeline %d at %X/%X: end of WAL at %X/%X", + tli, + LSN_FORMAT_ARGS(start_lsn), + LSN_FORMAT_ARGS(private_data->read_upto))); + + /* + * The timeline ends at or after start_lsn, without containing + * any records. Thus, we must make sure the main loop does not + * iterate. If start_lsn is the end of the timeline, then we + * won't actually emit an empty summary file, but otherwise, + * we must, to capture the fact that the LSN range in question + * contains no interesting WAL records. + */ + summary_start_lsn = start_lsn; + summary_end_lsn = private_data->read_upto; + cutoff_lsn = xlogreader->EndRecPtr; + } + else + ereport(ERROR, + (errmsg("could not find a valid record after %X/%X", + LSN_FORMAT_ARGS(start_lsn)))); + } + + /* We shouldn't go backward. */ + Assert(summary_start_lsn >= start_lsn); + } + + /* + * Main loop: read xlog records one by one. + */ + while (xlogreader->EndRecPtr < cutoff_lsn) + { + int block_id; + char *errormsg; + XLogRecord *record; + bool switch_requested; + + /* We shouldn't go backward. */ + Assert(summary_start_lsn <= xlogreader->EndRecPtr); + + /* + * This flag tracks whether the read of a particular record had to + * wait for more WAL to arrive, so reset it before reading the next + * record. + */ + private_data->waited = false; + + /* Now read the next record. */ + record = XLogReadRecord(xlogreader, &errormsg); + if (record == NULL) + { + SummarizerReadLocalXLogPrivate *private_data; + + private_data = (SummarizerReadLocalXLogPrivate *) + xlogreader->private_data; + if (private_data->end_of_wal) + { + /* + * This timeline must be historic and must end before we were + * able to read a complete record. + */ + ereport(DEBUG1, + errmsg_internal("could not read WAL from timeline %d at %X/%X: end of WAL at %X/%X", + tli, + LSN_FORMAT_ARGS(xlogreader->EndRecPtr), + LSN_FORMAT_ARGS(private_data->read_upto))); + /* Summary ends at end of WAL. */ + summary_end_lsn = private_data->read_upto; + break; + } + if (errormsg) + ereport(ERROR, + (errcode_for_file_access(), + errmsg("could not read WAL at %X/%X: %s", + LSN_FORMAT_ARGS(xlogreader->EndRecPtr), errormsg))); + else + ereport(ERROR, + (errcode_for_file_access(), + errmsg("could not read WAL at %X/%X", + LSN_FORMAT_ARGS(xlogreader->EndRecPtr)))); + } + + /* We shouldn't go backward. */ + Assert(summary_start_lsn <= xlogreader->EndRecPtr); + + if (xlogreader->ReadRecPtr >= cutoff_lsn) + { + /* + * Woops! We've read a record that *starts* after the cutoff LSN, + * contrary to our goal of reading only until we hit the first + * record that ends at or after the cutoff LSN. Pretend we didn't + * read it after all by bailing out of this loop right here, + * before we do anything with this record. + * + * This can happen because the last record before the cutoff LSN + * might be continued across multiple pages, and then we might + * come to a page with XLP_FIRST_IS_OVERWRITE_CONTRECORD set. In + * that case, the record that was continued across multiple pages + * is incomplete and will be disregarded, and the read will + * restart from the beginning of the page that is flagged + * XLP_FIRST_IS_OVERWRITE_CONTRECORD. + * + * If this case occurs, we can fairly say that the current summary + * file ends at the cutoff LSN exactly. The first record on the + * page marked XLP_FIRST_IS_OVERWRITE_CONTRECORD will be + * discovered when generating the next summary file. + */ + summary_end_lsn = cutoff_lsn; + break; + } + + /* + * We attempt, on a best effort basis only, to make WAL summary file + * boundaries line up with checkpoint cycles. So, if the last redo + * pointer we've seen was in the future, and this record starts at + * that redo pointer, stop before processing and let it be included in + * the next summary file. + * + * Note that in the case of a checkpoint triggered by a backup, the + * redo pointer is likely to be pointing to the first record on a + * page. Before reading the record, xlogreader->EndRecPtr will have + * pointed to the start of the page, which precedes the redo LSN. But + * after reading the next record, we'll advance over the page header + * and realize that the next record starts at the redo LSN exactly, + * making this the first point at which we can realize that it's time + * to stop. + */ + if (!private_data->redo_pointer_reached && + xlogreader->ReadRecPtr >= private_data->redo_pointer) + { + summary_end_lsn = xlogreader->ReadRecPtr; + break; + } + + /* Special handling for particular types of WAL records. */ + switch (XLogRecGetRmid(xlogreader)) + { + case RM_SMGR_ID: + SummarizeSmgrRecord(xlogreader, brtab); + break; + case RM_XACT_ID: + SummarizeXactRecord(xlogreader, brtab); + break; + default: + break; + } + + /* Feed block references from xlog record to block reference table. */ + for (block_id = 0; block_id <= XLogRecMaxBlockId(xlogreader); + block_id++) + { + RelFileLocator rlocator; + ForkNumber forknum; + BlockNumber blocknum; + + if (!XLogRecGetBlockTagExtended(xlogreader, block_id, &rlocator, + &forknum, &blocknum, NULL)) + continue; + + BlockRefTableMarkBlockModified(brtab, &rlocator, forknum, + blocknum); + } + + /* Update our notion of where this summary file ends. */ + summary_end_lsn = xlogreader->EndRecPtr; + + /* + * Also update shared memory, and handle any request for a WAL summary + * file switch. + */ + LWLockAcquire(WALSummarizerLock, LW_EXCLUSIVE); + Assert(summary_end_lsn >= WalSummarizerCtl->pending_lsn); + Assert(summary_end_lsn >= WalSummarizerCtl->summarized_lsn); + WalSummarizerCtl->pending_lsn = summary_end_lsn; + switch_requested = WalSummarizerCtl->switch_requested; + LWLockRelease(WALSummarizerLock); + if (switch_requested) + break; + + /* + * Periodically update our notion of the redo pointer, because it + * might be changing concurrently. There's no interlocking here: we + * might race past the new redo pointer before we learn about it. + * That's OK; we only use the redo pointer as a heuristic for where to + * stop summarizing. + * + * It would be nice if we could just fetch the updated redo pointer on + * every pass through this loop, but that seems a bit too expensive: + * GetRedoRecPtr acquires a heavily-contended spinlock. So, instead, + * just fetch the updated value if we've just had to sleep, or if + * we've read more than a segment's worth of WAL without sleeping. + */ + if (private_data->waited || xlogreader->EndRecPtr > + private_data->redo_pointer_refresh_lsn + wal_segment_size) + { + private_data->redo_pointer = GetRedoRecPtr(); + private_data->redo_pointer_refresh_lsn = xlogreader->EndRecPtr; + private_data->redo_pointer_reached = + (xlogreader->EndRecPtr >= private_data->redo_pointer); + } + + /* + * Recheck whether we've just caught up with the redo pointer, and if + * so, stop. This has the same purpose as the earlier check for the + * same condition above, but there we've just read a record and might + * decide against including it in the current summary file, whereas + * here we've already included it and might decide against reading the + * next one. Note that we may have just refreshed our notion of the + * redo pointer, so it's smart to check here before we do any more + * work. + */ + if (!private_data->redo_pointer_reached && + xlogreader->EndRecPtr >= private_data->redo_pointer) + break; + } + + /* Destroy xlogreader. */ + pfree(xlogreader->private_data); + XLogReaderFree(xlogreader); + + /* + * If a timeline switch occurs, we may fail to make any progress at all + * before exiting the loop above. If that happens, we don't write a WAL + * summary file at all. + */ + if (summary_end_lsn > summary_start_lsn) + { + /* Generate temporary and final path name. */ + snprintf(temp_path, MAXPGPATH, + XLOGDIR "/summaries/temp.summary"); + snprintf(final_path, MAXPGPATH, + XLOGDIR "/summaries/%08X%08X%08X%08X%08X.summary", + tli, + LSN_FORMAT_ARGS(summary_start_lsn), + LSN_FORMAT_ARGS(summary_end_lsn)); + + /* Open the temporary file for writing. */ + io.filepos = 0; + io.file = PathNameOpenFile(temp_path, O_WRONLY | O_CREAT | O_TRUNC); + if (io.file < 0) + ereport(ERROR, + (errcode_for_file_access(), + errmsg("could not create file \"%s\": %m", temp_path))); + + /* Write the data. */ + WriteBlockRefTable(brtab, WriteWalSummary, &io); + + /* Close temporary file and shut down xlogreader. */ + FileClose(io.file); + + /* Tell the user what we did. */ + ereport(LOG, + errmsg("summarized WAL on TLI %d from %X/%X to %X/%X", + tli, + LSN_FORMAT_ARGS(summary_start_lsn), + LSN_FORMAT_ARGS(summary_end_lsn))); + + /* Durably rename the new summary into place. */ + durable_rename(temp_path, final_path, ERROR); + } + + return summary_end_lsn; +} + +/* + * Special handling for WAL records with RM_SMGR_ID. + */ +static void +SummarizeSmgrRecord(XLogReaderState *xlogreader, BlockRefTable *brtab) +{ + uint8 info = XLogRecGetInfo(xlogreader) & ~XLR_INFO_MASK; + + if (info == XLOG_SMGR_CREATE) + { + xl_smgr_create *xlrec; + + /* + * If a new relation fork is created on disk, there is no point + * tracking anything about which blocks have been modified, because + * the whole thing will be new. Hence, set the limit block for this + * fork to 0. + * + * Ignore the FSM fork, which is not fully WAL-logged. + */ + xlrec = (xl_smgr_create *) XLogRecGetData(xlogreader); + + if (xlrec->forkNum != FSM_FORKNUM) + BlockRefTableSetLimitBlock(brtab, &xlrec->rlocator, + xlrec->forkNum, 0); + } + else if (info == XLOG_SMGR_TRUNCATE) + { + xl_smgr_truncate *xlrec; + + xlrec = (xl_smgr_truncate *) XLogRecGetData(xlogreader); + + /* + * If a relation fork is truncated on disk, there is in point in + * tracking anything about block modifications beyond the truncation + * point. + * + * We ignore SMGR_TRUNCATE_FSM here because the FSM isn't fully + * WAL-logged and thus we can't track modified blocks for it anyway. + */ + if ((xlrec->flags & SMGR_TRUNCATE_HEAP) != 0) + BlockRefTableSetLimitBlock(brtab, &xlrec->rlocator, + MAIN_FORKNUM, xlrec->blkno); + if ((xlrec->flags & SMGR_TRUNCATE_VM) != 0) + BlockRefTableSetLimitBlock(brtab, &xlrec->rlocator, + VISIBILITYMAP_FORKNUM, xlrec->blkno); + } +} + +/* + * Special handling for WAL recods with RM_XACT_ID. + */ +static void +SummarizeXactRecord(XLogReaderState *xlogreader, BlockRefTable *brtab) +{ + uint8 info = XLogRecGetInfo(xlogreader) & ~XLR_INFO_MASK; + uint8 xact_info = info & XLOG_XACT_OPMASK; + + if (xact_info == XLOG_XACT_COMMIT || + xact_info == XLOG_XACT_COMMIT_PREPARED) + { + xl_xact_commit *xlrec = (xl_xact_commit *) XLogRecGetData(xlogreader); + xl_xact_parsed_commit parsed; + int i; + + ParseCommitRecord(XLogRecGetInfo(xlogreader), xlrec, &parsed); + for (i = 0; i < parsed.nrels; ++i) + { + ForkNumber forknum; + + for (forknum = 0; forknum <= MAX_FORKNUM; ++forknum) + if (forknum != FSM_FORKNUM) + BlockRefTableSetLimitBlock(brtab, &parsed.xlocators[i], + forknum, 0); + } + } + else if (xact_info == XLOG_XACT_ABORT || + xact_info == XLOG_XACT_ABORT_PREPARED) + { + xl_xact_abort *xlrec = (xl_xact_abort *) XLogRecGetData(xlogreader); + xl_xact_parsed_abort parsed; + int i; + + ParseAbortRecord(XLogRecGetInfo(xlogreader), xlrec, &parsed); + for (i = 0; i < parsed.nrels; ++i) + { + ForkNumber forknum; + + for (forknum = 0; forknum <= MAX_FORKNUM; ++forknum) + if (forknum != FSM_FORKNUM) + BlockRefTableSetLimitBlock(brtab, &parsed.xlocators[i], + forknum, 0); + } + } +} + +/* + * Similar to read_local_xlog_page, but limited to read from one particular + * timeline. If the end of WAL is reached, it will wait for more if reading + * from the current timeline, or give up if reading from a historic timeline. + * In the latter case, it will also set private_data->end_of_wal = true. + * + * Caller must set private_data->tli to the TLI of interest, + * private_data->read_upto to the lowest LSN that is not known to be safe + * to read on that timeline, and private_data->historic to true if and only + * if the timeline is not the current timeline. This function will update + * private_data->read_upto and private_data->historic if more WAL appears + * on the current timeline or if the current timeline becomes historic. + */ +static int +summarizer_read_local_xlog_page(XLogReaderState *state, + XLogRecPtr targetPagePtr, int reqLen, + XLogRecPtr targetRecPtr, char *cur_page) +{ + int count; + WALReadError errinfo; + SummarizerReadLocalXLogPrivate *private_data; + + private_data = (SummarizerReadLocalXLogPrivate *) + state->private_data; + + while (true) + { + if (targetPagePtr + XLOG_BLCKSZ <= private_data->read_upto) + { + /* + * more than one block available; read only that block, have + * caller come back if they need more. + */ + count = XLOG_BLCKSZ; + break; + } + else if (targetPagePtr + reqLen > private_data->read_upto) + { + /* We don't seem to have enough data. */ + if (private_data->historic) + { + /* + * This is a historic timeline, so there will never be any + * more data than we have currently. + */ + private_data->end_of_wal = true; + return -1; + } + else + { + XLogRecPtr latest_lsn; + TimeLineID latest_tli; + + /* + * This is - or at least was up until very recently - the + * current timeline, so more data might show up. Delay here + * so we don't tight-loop. + */ + HandleWalSummarizerInterrupts(); + summarizer_wait_for_wal(); + private_data->waited = true; + + /* Recheck end-of-WAL. */ + latest_lsn = GetLatestLSN(&latest_tli); + if (private_data->tli == latest_tli) + { + /* Still the current timeline, update max LSN. */ + Assert(latest_lsn >= private_data->read_upto); + private_data->read_upto = latest_lsn; + } + else + { + List *tles = readTimeLineHistory(latest_tli); + XLogRecPtr switchpoint; + + /* + * The timeline we're scanning is no longer the latest + * one. Figure out when it ended and allow reads up to + * exactly that point. + */ + private_data->historic = true; + switchpoint = tliSwitchPoint(private_data->tli, tles, + NULL); + Assert(switchpoint >= private_data->read_upto); + private_data->read_upto = switchpoint; + } + + /* Go around and try again. */ + } + } + else + { + /* enough bytes available to satisfy the request */ + count = private_data->read_upto - targetPagePtr; + break; + } + } + + /* + * Even though we just determined how much of the page can be validly read + * as 'count', read the whole page anyway. It's guaranteed to be + * zero-padded up to the page boundary if it's incomplete. + */ + if (!WALRead(state, cur_page, targetPagePtr, XLOG_BLCKSZ, + private_data->tli, &errinfo)) + WALReadRaiseError(&errinfo); + + /* Track that we read a page, for sleep time calculation. */ + ++pages_read_since_last_sleep; + + /* number of valid bytes in the buffer */ + return count; +} + +/* + * Sleep for long enough that we believe it's likely that more WAL will + * be available afterwards. + */ +static void +summarizer_wait_for_wal(void) +{ + if (pages_read_since_last_sleep == 0) + { + /* + * No pages were read since the last sleep, so double the sleep time, + * but not beyond the maximum allowable value. + */ + sleep_quanta = Min(sleep_quanta * 2, MAX_SLEEP_QUANTA); + } + else if (pages_read_since_last_sleep > 1) + { + /* + * Multiple pages were read since the last sleep, so reduce the sleep + * time. + * + * A large burst of activity should be able to quickly reduce the + * sleep time to the minimum, but we don't want a handful of extra WAL + * records to provoke a strong reaction. We choose to reduce the sleep + * time by 1 quantum for each page read beyond the first, which is a + * fairly arbitrary way of trying to be reactive without + * overrreacting. + */ + if (pages_read_since_last_sleep > sleep_quanta - 1) + sleep_quanta = 1; + else + sleep_quanta -= pages_read_since_last_sleep; + } + + /* OK, now sleep. */ + (void) WaitLatch(MyLatch, + WL_LATCH_SET | WL_TIMEOUT | WL_EXIT_ON_PM_DEATH, + sleep_quanta * MS_PER_SLEEP_QUANTUM, + WAIT_EVENT_WAL_SUMMARIZER_WAL); + ResetLatch(MyLatch); + + /* Reset count of pages read. */ + pages_read_since_last_sleep = 0; +} + +/* + * Most recent RedoRecPtr value observed by RemoveOldWalSummaries. + */ +static void +MaybeRemoveOldWalSummaries(void) +{ + XLogRecPtr redo_pointer = GetRedoRecPtr(); + List *wslist; + time_t cutoff_time; + + /* If WAL summary removal is disabled, don't do anything. */ + if (wal_summarize_keep_time == 0) + return; + + /* + * If the redo pointer has not advanced, don't do anything. + * + * This has the effect that we only try to remove old WAL summary files + * once per checkpoint cycle. + */ + if (redo_pointer == redo_pointer_at_last_summary_removal) + return; + redo_pointer_at_last_summary_removal = redo_pointer; + + /* + * Files should only be removed if the last modification time precedes the + * cutoff time we compute here. + */ + cutoff_time = time(NULL) - 60 * wal_summarize_keep_time; + + /* Get all the summaries that currently exist. */ + wslist = GetWalSummaries(0, InvalidXLogRecPtr, InvalidXLogRecPtr); + + /* Loop until all summaries have been considered for removal. */ + while (wslist != NIL) + { + ListCell *lc; + XLogSegNo oldest_segno; + XLogRecPtr oldest_lsn = InvalidXLogRecPtr; + TimeLineID selected_tli; + + CHECK_FOR_INTERRUPTS(); + + /* + * Pick a timeline for which some summary files still exist on disk, + * and find the oldest LSN that still exists on disk for that + * timeline. + */ + selected_tli = ((WalSummaryFile *) linitial(wslist))->tli; + oldest_segno = XLogGetOldestSegno(selected_tli); + if (oldest_segno != 0) + XLogSegNoOffsetToRecPtr(oldest_segno, 0, wal_segment_size, + oldest_lsn); + + + /* Consider each WAL file on the selected timeline in turn. */ + foreach(lc, wslist) + { + WalSummaryFile *ws = lfirst(lc); + + CHECK_FOR_INTERRUPTS(); + + /* If it's not on this timeline, it's not time to consider it. */ + if (selected_tli != ws->tli) + continue; + + /* + * If the WAL doesn't exist any more, we can remove it if the file + * modification time is old enough. + */ + if (XLogRecPtrIsInvalid(oldest_lsn) || ws->end_lsn <= oldest_lsn) + RemoveWalSummaryIfOlderThan(ws, cutoff_time); + + /* + * Whether we we removed the file or not, we need not consider it + * again. + */ + wslist = foreach_delete_current(wslist, lc); + pfree(ws); + } + } +} diff --git a/src/backend/replication/repl_gram.y b/src/backend/replication/repl_gram.y index 0c874e33cf..a5d118ed68 100644 --- a/src/backend/replication/repl_gram.y +++ b/src/backend/replication/repl_gram.y @@ -76,11 +76,12 @@ Node *replication_parse_result; %token K_EXPORT_SNAPSHOT %token K_NOEXPORT_SNAPSHOT %token K_USE_SNAPSHOT +%token K_UPLOAD_MANIFEST %type <node> command %type <node> base_backup start_replication start_logical_replication create_replication_slot drop_replication_slot identify_system - read_replication_slot timeline_history show + read_replication_slot timeline_history show upload_manifest %type <list> generic_option_list %type <defelt> generic_option %type <uintval> opt_timeline @@ -114,6 +115,7 @@ command: | read_replication_slot | timeline_history | show + | upload_manifest ; /* @@ -307,6 +309,15 @@ timeline_history: } ; +/* UPLOAD_MANIFEST doesn't currently accept any arguments */ +upload_manifest: + K_UPLOAD_MANIFEST + { + UploadManifestCmd *cmd = makeNode(UploadManifestCmd); + + $$ = (Node *) cmd; + } + opt_physical: K_PHYSICAL | /* EMPTY */ @@ -411,6 +422,7 @@ ident_or_keyword: | K_EXPORT_SNAPSHOT { $$ = "export_snapshot"; } | K_NOEXPORT_SNAPSHOT { $$ = "noexport_snapshot"; } | K_USE_SNAPSHOT { $$ = "use_snapshot"; } + | K_UPLOAD_MANIFEST { $$ = "upload_manifest"; } ; %% diff --git a/src/backend/replication/repl_scanner.l b/src/backend/replication/repl_scanner.l index 1cc7fb858c..4805da08ee 100644 --- a/src/backend/replication/repl_scanner.l +++ b/src/backend/replication/repl_scanner.l @@ -136,6 +136,7 @@ EXPORT_SNAPSHOT { return K_EXPORT_SNAPSHOT; } NOEXPORT_SNAPSHOT { return K_NOEXPORT_SNAPSHOT; } USE_SNAPSHOT { return K_USE_SNAPSHOT; } WAIT { return K_WAIT; } +UPLOAD_MANIFEST { return K_UPLOAD_MANIFEST; } {space}+ { /* do nothing */ } @@ -303,6 +304,7 @@ replication_scanner_is_replication_command(void) case K_DROP_REPLICATION_SLOT: case K_READ_REPLICATION_SLOT: case K_TIMELINE_HISTORY: + case K_UPLOAD_MANIFEST: case K_SHOW: /* Yes; push back the first token so we can parse later. */ repl_pushed_back_token = first_token; diff --git a/src/backend/replication/walsender.c b/src/backend/replication/walsender.c index e250b0567e..b33b86671b 100644 --- a/src/backend/replication/walsender.c +++ b/src/backend/replication/walsender.c @@ -58,6 +58,7 @@ #include "access/xlogrecovery.h" #include "access/xlogutils.h" #include "backup/basebackup.h" +#include "backup/basebackup_incremental.h" #include "catalog/pg_authid.h" #include "catalog/pg_type.h" #include "commands/dbcommands.h" @@ -137,6 +138,17 @@ bool wake_wal_senders = false; */ static XLogReaderState *xlogreader = NULL; +/* + * If the UPLOAD_MANIFEST command is used to provide a backup manifest in + * preparation for an incremental backup, uploaded_manifest will be point + * to an object containing information about its contexts, and + * uploaded_manifest_mcxt will point to the memory context that contains + * that object and all of its subordinate data. Otherwise, both values will + * be NULL. + */ +static IncrementalBackupInfo *uploaded_manifest = NULL; +static MemoryContext uploaded_manifest_mcxt = NULL; + /* * These variables keep track of the state of the timeline we're currently * sending. sendTimeLine identifies the timeline. If sendTimeLineIsHistoric, @@ -233,6 +245,9 @@ static void XLogSendLogical(void); static void WalSndDone(WalSndSendDataCallback send_data); static XLogRecPtr GetStandbyFlushRecPtr(TimeLineID *tli); static void IdentifySystem(void); +static void UploadManifest(void); +static bool HandleUploadManifestPacket(StringInfo buf, off_t *offset, + IncrementalBackupInfo *ib); static void ReadReplicationSlot(ReadReplicationSlotCmd *cmd); static void CreateReplicationSlot(CreateReplicationSlotCmd *cmd); static void DropReplicationSlot(DropReplicationSlotCmd *cmd); @@ -660,6 +675,143 @@ SendTimeLineHistory(TimeLineHistoryCmd *cmd) pq_endmessage(&buf); } +/* + * Handle UPLOAD_MANIFEST command. + */ +static void +UploadManifest(void) +{ + MemoryContext mcxt; + IncrementalBackupInfo *ib; + off_t offset = 0; + StringInfoData buf; + + /* + * parsing the manifest will use the cryptohash stuff, which requires a + * resource owner + */ + Assert(CurrentResourceOwner == NULL); + CurrentResourceOwner = ResourceOwnerCreate(NULL, "base backup"); + + /* Prepare to read manifest data into a temporary context. */ + mcxt = AllocSetContextCreate(CurrentMemoryContext, + "incremental backup information", + ALLOCSET_DEFAULT_SIZES); + ib = CreateIncrementalBackupInfo(mcxt); + + /* Send a CopyInResponse message */ + pq_beginmessage(&buf, 'G'); + pq_sendbyte(&buf, 0); + pq_sendint16(&buf, 0); + pq_endmessage_reuse(&buf); + pq_flush(); + + /* Recieve packets from client until done. */ + while (HandleUploadManifestPacket(&buf, &offset, ib)) + ; + + /* Finish up manifest processing. */ + FinalizeIncrementalManifest(ib); + + /* + * Discard any old manifest information and arrange to preserve the new + * information we just got. + * + * We assume that MemoryContextDelete and MemoryContextSetParent won't + * fail, and thus we shouldn't end up bailing out of here in such a way as + * to leave dangling pointrs. + */ + if (uploaded_manifest_mcxt != NULL) + MemoryContextDelete(uploaded_manifest_mcxt); + MemoryContextSetParent(mcxt, CacheMemoryContext); + uploaded_manifest = ib; + uploaded_manifest_mcxt = mcxt; + + /* clean up the resource owner we created */ + WalSndResourceCleanup(true); +} + +/* + * Process one packet received during the handling of an UPLOAD_MANIFEST + * operation. + * + * 'buf' is scratch space. This function expects it to be initialized, doesn't + * care what the current contents are, and may override them with completely + * new contents. + * + * The return value is true if the caller should continue processing + * additional packets and false if the UPLOAD_MANIFEST operation is complete. + */ +static bool +HandleUploadManifestPacket(StringInfo buf, off_t *offset, + IncrementalBackupInfo *ib) +{ + int mtype; + int maxmsglen; + + HOLD_CANCEL_INTERRUPTS(); + + pq_startmsgread(); + mtype = pq_getbyte(); + if (mtype == EOF) + ereport(ERROR, + (errcode(ERRCODE_CONNECTION_FAILURE), + errmsg("unexpected EOF on client connection with an open transaction"))); + + switch (mtype) + { + case 'd': /* CopyData */ + maxmsglen = PQ_LARGE_MESSAGE_LIMIT; + break; + case 'c': /* CopyDone */ + case 'f': /* CopyFail */ + case 'H': /* Flush */ + case 'S': /* Sync */ + maxmsglen = PQ_SMALL_MESSAGE_LIMIT; + break; + default: + ereport(ERROR, + (errcode(ERRCODE_PROTOCOL_VIOLATION), + errmsg("unexpected message type 0x%02X during COPY from stdin", + mtype))); + maxmsglen = 0; /* keep compiler quiet */ + break; + } + + /* Now collect the message body */ + if (pq_getmessage(buf, maxmsglen)) + ereport(ERROR, + (errcode(ERRCODE_CONNECTION_FAILURE), + errmsg("unexpected EOF on client connection with an open transaction"))); + RESUME_CANCEL_INTERRUPTS(); + + /* Process the message */ + switch (mtype) + { + case 'd': /* CopyData */ + AppendIncrementalManifestData(ib, buf->data, buf->len); + return true; + + case 'c': /* CopyDone */ + return false; + + case 'H': /* Sync */ + case 'S': /* Flush */ + /* Ignore these while in CopyOut mode as we do elsewhere. */ + return true; + + case 'f': + ereport(ERROR, + (errcode(ERRCODE_QUERY_CANCELED), + errmsg("COPY from stdin failed: %s", + pq_getmsgstring(buf)))); + } + + /* Not reached. */ + Assert(false); + return false; +} + /* * Handle START_REPLICATION command. * @@ -1801,7 +1953,7 @@ exec_replication_command(const char *cmd_string) cmdtag = "BASE_BACKUP"; set_ps_display(cmdtag); PreventInTransactionBlock(true, cmdtag); - SendBaseBackup((BaseBackupCmd *) cmd_node); + SendBaseBackup((BaseBackupCmd *) cmd_node, uploaded_manifest); EndReplicationCommand(cmdtag); break; @@ -1863,6 +2015,14 @@ exec_replication_command(const char *cmd_string) } break; + case T_UploadManifestCmd: + cmdtag = "UPLOAD_MANIFEST"; + set_ps_display(cmdtag); + PreventInTransactionBlock(true, cmdtag); + UploadManifest(); + EndReplicationCommand(cmdtag); + break; + default: elog(ERROR, "unrecognized replication command node tag: %u", cmd_node->type); diff --git a/src/backend/storage/ipc/ipci.c b/src/backend/storage/ipc/ipci.c index a3d8eacb8d..3a6729003a 100644 --- a/src/backend/storage/ipc/ipci.c +++ b/src/backend/storage/ipc/ipci.c @@ -31,6 +31,7 @@ #include "postmaster/bgworker_internals.h" #include "postmaster/bgwriter.h" #include "postmaster/postmaster.h" +#include "postmaster/walsummarizer.h" #include "replication/logicallauncher.h" #include "replication/origin.h" #include "replication/slot.h" @@ -136,6 +137,7 @@ CalculateShmemSize(int *num_semaphores) size = add_size(size, ReplicationOriginShmemSize()); size = add_size(size, WalSndShmemSize()); size = add_size(size, WalRcvShmemSize()); + size = add_size(size, WalSummarizerShmemSize()); size = add_size(size, PgArchShmemSize()); size = add_size(size, ApplyLauncherShmemSize()); size = add_size(size, BTreeShmemSize()); @@ -291,6 +293,7 @@ CreateSharedMemoryAndSemaphores(void) ReplicationOriginShmemInit(); WalSndShmemInit(); WalRcvShmemInit(); + WalSummarizerShmemInit(); PgArchShmemInit(); ApplyLauncherShmemInit(); diff --git a/src/backend/storage/lmgr/lwlocknames.txt b/src/backend/storage/lmgr/lwlocknames.txt index f72f2906ce..d621f5507f 100644 --- a/src/backend/storage/lmgr/lwlocknames.txt +++ b/src/backend/storage/lmgr/lwlocknames.txt @@ -54,3 +54,4 @@ XactTruncationLock 44 WrapLimitsVacuumLock 46 NotifyQueueTailLock 47 WaitEventExtensionLock 48 +WALSummarizerLock 49 diff --git a/src/backend/utils/activity/pgstat_io.c b/src/backend/utils/activity/pgstat_io.c index eb7d35d422..bd0a921a3e 100644 --- a/src/backend/utils/activity/pgstat_io.c +++ b/src/backend/utils/activity/pgstat_io.c @@ -292,7 +292,8 @@ pgstat_io_snapshot_cb(void) * - Syslogger because it is not connected to shared memory * - Archiver because most relevant archiving IO is delegated to a * specialized command or module -* - WAL Receiver and WAL Writer IO is not tracked in pg_stat_io for now +* - WAL Receiver, WAL Writer, and WAL Summarizer IO are not tracked in +* pg_stat_io for now * * Function returns true if BackendType participates in the cumulative stats * subsystem for IO and false if it does not. @@ -314,6 +315,7 @@ pgstat_tracks_io_bktype(BackendType bktype) case B_LOGGER: case B_WAL_RECEIVER: case B_WAL_WRITER: + case B_WAL_SUMMARIZER: return false; case B_AUTOVAC_LAUNCHER: diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 9c5fdeb3ca..17ad986c98 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -56,6 +56,7 @@ RECOVERY_WAL_STREAM "Waiting in main loop of startup process for WAL to arrive, SYSLOGGER_MAIN "Waiting in main loop of syslogger process." WAL_RECEIVER_MAIN "Waiting in main loop of WAL receiver process." WAL_SENDER_MAIN "Waiting in main loop of WAL sender process." +WAL_SUMMARIZER_WAL "Waiting in WAL summarizer for more WAL to be generated." WAL_WRITER_MAIN "Waiting in main loop of WAL writer process." @@ -140,6 +141,7 @@ SAFE_SNAPSHOT "Waiting to obtain a valid snapshot for a <literal>READ ONLY DEFER SYNC_REP "Waiting for confirmation from a remote server during synchronous replication." WAL_RECEIVER_EXIT "Waiting for the WAL receiver to exit." WAL_RECEIVER_WAIT_START "Waiting for startup process to send initial data for streaming replication." +WAL_SUMMARY_READY "Waiting for a new WAL summary to be generated." XACT_GROUP_UPDATE "Waiting for the group leader to update transaction status at end of a parallel operation." @@ -160,6 +162,7 @@ REGISTER_SYNC_REQUEST "Waiting while sending synchronization requests to the che SPIN_DELAY "Waiting while acquiring a contended spinlock." VACUUM_DELAY "Waiting in a cost-based vacuum delay point." VACUUM_TRUNCATE "Waiting to acquire an exclusive lock to truncate off any empty pages at the end of a table vacuumed." +WAL_SUMMARIZER_ERROR "Waiting after a WAL summarizer error." # @@ -241,6 +244,8 @@ WAL_COPY_WRITE "Waiting for a write when creating a new WAL segment by copying a WAL_INIT_SYNC "Waiting for a newly initialized WAL file to reach durable storage." WAL_INIT_WRITE "Waiting for a write while initializing a new WAL file." WAL_READ "Waiting for a read from a WAL file." +WAL_SUMMARY_READ "Waiting for a read from a WAL summary file." +WAL_SUMMARY_WRITE "Waiting for a write to a WAL summary file." WAL_SYNC "Waiting for a WAL file to reach durable storage." WAL_SYNC_METHOD_ASSIGN "Waiting for data to reach durable storage while assigning a new WAL sync method." WAL_WRITE "Waiting for a write to a WAL file." diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c index 1e671c560c..037111b89f 100644 --- a/src/backend/utils/init/miscinit.c +++ b/src/backend/utils/init/miscinit.c @@ -306,6 +306,9 @@ GetBackendTypeDesc(BackendType backendType) case B_WAL_SENDER: backendDesc = "walsender"; break; + case B_WAL_SUMMARIZER: + backendDesc = "walsummarizer"; + break; case B_WAL_WRITER: backendDesc = "walwriter"; break; diff --git a/src/backend/utils/misc/guc_tables.c b/src/backend/utils/misc/guc_tables.c index 16ec6c5ef0..a532f57af1 100644 --- a/src/backend/utils/misc/guc_tables.c +++ b/src/backend/utils/misc/guc_tables.c @@ -63,6 +63,7 @@ #include "postmaster/postmaster.h" #include "postmaster/startup.h" #include "postmaster/syslogger.h" +#include "postmaster/walsummarizer.h" #include "postmaster/walwriter.h" #include "replication/logicallauncher.h" #include "replication/slot.h" @@ -704,6 +705,8 @@ const char *const config_group_names[] = gettext_noop("Write-Ahead Log / Archive Recovery"), /* WAL_RECOVERY_TARGET */ gettext_noop("Write-Ahead Log / Recovery Target"), + /* WAL_SUMMARIZATION */ + gettext_noop("Write-Ahead Log / Summarization"), /* REPLICATION_SENDING */ gettext_noop("Replication / Sending Servers"), /* REPLICATION_PRIMARY */ @@ -3181,6 +3184,32 @@ struct config_int ConfigureNamesInt[] = check_wal_segment_size, NULL, NULL }, + { + {"wal_summarize_mb", PGC_SIGHUP, WAL_SUMMARIZATION, + gettext_noop("Number of bytes of WAL per summary file."), + gettext_noop("Smaller values minimize extra work performed by incremental backup, but increase the number of files on disk."), + GUC_UNIT_MB, + }, + &wal_summarize_mb, + 256, + 0, + INT_MAX, + NULL, NULL, NULL + }, + + { + {"wal_summarize_keep_time", PGC_SIGHUP, WAL_SUMMARIZATION, + gettext_noop("Time for which WAL summary files should be kept."), + NULL, + GUC_UNIT_MIN, + }, + &wal_summarize_keep_time, + 7 * 24 * 60, /* 1 week */ + 0, + INT_MAX, + NULL, NULL, NULL + }, + { {"autovacuum_naptime", PGC_SIGHUP, AUTOVACUUM, gettext_noop("Time to sleep between autovacuum runs."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index d08d55c3fe..4736606ac1 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -299,6 +299,11 @@ #recovery_target_action = 'pause' # 'pause', 'promote', 'shutdown' # (change requires restart) +# - WAL Summarization - + +#wal_summarize_mb = 256 # MB of WAL per summary file, 0 disables +#wal_summarize_keep_time = '7d' # when to remove old summary files, 0 = never + #------------------------------------------------------------------------------ # REPLICATION diff --git a/src/bin/Makefile b/src/bin/Makefile index 373077bf52..aa2210925e 100644 --- a/src/bin/Makefile +++ b/src/bin/Makefile @@ -19,6 +19,7 @@ SUBDIRS = \ pg_archivecleanup \ pg_basebackup \ pg_checksums \ + pg_combinebackup \ pg_config \ pg_controldata \ pg_ctl \ diff --git a/src/bin/initdb/initdb.c b/src/bin/initdb/initdb.c index 0c6f5ceb0a..e68b40d2b5 100644 --- a/src/bin/initdb/initdb.c +++ b/src/bin/initdb/initdb.c @@ -227,6 +227,7 @@ static char *extra_options = ""; static const char *const subdirs[] = { "global", "pg_wal/archive_status", + "pg_wal/summaries", "pg_commit_ts", "pg_dynshmem", "pg_notify", diff --git a/src/bin/meson.build b/src/bin/meson.build index 67cb50630c..4cb6fd59bb 100644 --- a/src/bin/meson.build +++ b/src/bin/meson.build @@ -5,6 +5,7 @@ subdir('pg_amcheck') subdir('pg_archivecleanup') subdir('pg_basebackup') subdir('pg_checksums') +subdir('pg_combinebackup') subdir('pg_config') subdir('pg_controldata') subdir('pg_ctl') diff --git a/src/bin/pg_basebackup/bbstreamer_file.c b/src/bin/pg_basebackup/bbstreamer_file.c index 45f32974ff..6b78ee283d 100644 --- a/src/bin/pg_basebackup/bbstreamer_file.c +++ b/src/bin/pg_basebackup/bbstreamer_file.c @@ -296,6 +296,7 @@ should_allow_existing_directory(const char *pathname) if (strcmp(filename, "pg_wal") == 0 || strcmp(filename, "pg_xlog") == 0 || strcmp(filename, "archive_status") == 0 || + strcmp(filename, "summaries") == 0 || strcmp(filename, "pg_tblspc") == 0) return true; diff --git a/src/bin/pg_basebackup/pg_basebackup.c b/src/bin/pg_basebackup/pg_basebackup.c index 1a8cef345d..33416b11cf 100644 --- a/src/bin/pg_basebackup/pg_basebackup.c +++ b/src/bin/pg_basebackup/pg_basebackup.c @@ -101,6 +101,11 @@ typedef void (*WriteDataCallback) (size_t nbytes, char *buf, */ #define MINIMUM_VERSION_FOR_TERMINATED_TARFILE 150000 +/* + * pg_wal/summaries exists beginning with version 17. + */ +#define MINIMUM_VERSION_FOR_WAL_SUMMARIES 170000 + /* * Different ways to include WAL */ @@ -217,7 +222,8 @@ static void ReceiveBackupManifestInMemoryChunk(size_t r, char *copybuf, void *callback_data); static void BaseBackup(char *compression_algorithm, char *compression_detail, CompressionLocation compressloc, - pg_compress_specification *client_compress); + pg_compress_specification *client_compress, + char *incremental_manifest); static bool reached_end_position(XLogRecPtr segendpos, uint32 timeline, bool segment_finished); @@ -390,6 +396,8 @@ usage(void) printf(_("\nOptions controlling the output:\n")); printf(_(" -D, --pgdata=DIRECTORY receive base backup into directory\n")); printf(_(" -F, --format=p|t output format (plain (default), tar)\n")); + printf(_(" -i, --incremental=OLDMANIFEST\n")); + printf(_(" take incremental or differential backup\n")); printf(_(" -r, --max-rate=RATE maximum transfer rate to transfer data directory\n" " (in kB/s, or use suffix \"k\" or \"M\")\n")); printf(_(" -R, --write-recovery-conf\n" @@ -688,6 +696,23 @@ StartLogStreamer(char *startpos, uint32 timeline, char *sysidentifier, if (pg_mkdir_p(statusdir, pg_dir_create_mode) != 0 && errno != EEXIST) pg_fatal("could not create directory \"%s\": %m", statusdir); + + /* + * For newer server versions, likewise create pg_wal/summaries + */ + if (PQserverVersion(conn) < MINIMUM_VERSION_FOR_WAL_SUMMARIES) + { + char summarydir[MAXPGPATH]; + + snprintf(summarydir, sizeof(summarydir), "%s/%s/summaries", + basedir, + PQserverVersion(conn) < MINIMUM_VERSION_FOR_PG_WAL ? + "pg_xlog" : "pg_wal"); + + if (pg_mkdir_p(statusdir, pg_dir_create_mode) != 0 && + errno != EEXIST) + pg_fatal("could not create directory \"%s\": %m", summarydir); + } } /* @@ -1728,7 +1753,9 @@ ReceiveBackupManifestInMemoryChunk(size_t r, char *copybuf, static void BaseBackup(char *compression_algorithm, char *compression_detail, - CompressionLocation compressloc, pg_compress_specification *client_compress) + CompressionLocation compressloc, + pg_compress_specification *client_compress, + char *incremental_manifest) { PGresult *res; char *sysidentifier; @@ -1794,7 +1821,74 @@ BaseBackup(char *compression_algorithm, char *compression_detail, exit(1); /* - * Start the actual backup + * If the user wants an incremental backup, we must upload the manifest + * for the previous backup upon which it is to be based. + */ + if (incremental_manifest != NULL) + { + int fd; + char mbuf[65536]; + int nbytes; + + /* XXX add a server version check here */ + + /* Open the file. */ + fd = open(incremental_manifest, O_RDONLY | PG_BINARY, 0); + if (fd < 0) + pg_fatal("could not open file \"%s\": %m", incremental_manifest); + + /* Tell the server what we want to do. */ + if (PQsendQuery(conn, "UPLOAD_MANIFEST") == 0) + pg_fatal("could not send replication command \"%s\": %s", + "UPLOAD_MANIFEST", PQerrorMessage(conn)); + res = PQgetResult(conn); + if (PQresultStatus(res) != PGRES_COPY_IN) + { + if (PQresultStatus(res) == PGRES_FATAL_ERROR) + pg_fatal("could not upload manifest: %s", + PQerrorMessage(conn)); + else + pg_fatal("could not upload manifest: unexpected status %s", + PQresStatus(PQresultStatus(res))); + } + + /* Loop, reading from the file and sending the data to the server. */ + while ((nbytes = read(fd, mbuf, sizeof mbuf)) > 0) + { + if (PQputCopyData(conn, mbuf, nbytes) < 0) + pg_fatal("could not send COPY data: %s", + PQerrorMessage(conn)); + } + + /* Bail out if we exited the loop due to an error. */ + if (nbytes < 0) + pg_fatal("could not read file \"%s\": %m", incremental_manifest); + + /* End the COPY operation. */ + if (PQputCopyEnd(conn, NULL) < 0) + pg_fatal("could not send end-of-COPY: %s", + PQerrorMessage(conn)); + + /* See whether the server is happy with what we sent. */ + res = PQgetResult(conn); + if (PQresultStatus(res) == PGRES_FATAL_ERROR) + pg_fatal("could not upload manifest: %s", + PQerrorMessage(conn)); + else if (PQresultStatus(res) != PGRES_COMMAND_OK) + pg_fatal("could not upload manifest: unexpected status %s", + PQresStatus(PQresultStatus(res))); + + /* Consume ReadyForQuery message from server. */ + res = PQgetResult(conn); + if (res != NULL) + pg_fatal("unexpected extra result while sending manifest"); + + /* Add INCREMENTAL option to BASE_BACKUP command. */ + AppendPlainCommandOption(&buf, use_new_option_syntax, "INCREMENTAL"); + } + + /* + * Continue building up the options list for the BASE_BACKUP command. */ AppendStringCommandOption(&buf, use_new_option_syntax, "LABEL", label); if (estimatesize) @@ -1901,6 +1995,7 @@ BaseBackup(char *compression_algorithm, char *compression_detail, else basebkp = psprintf("BASE_BACKUP %s", buf.data); + /* OK, try to start the backup. */ if (PQsendQuery(conn, basebkp) == 0) pg_fatal("could not send replication command \"%s\": %s", "BASE_BACKUP", PQerrorMessage(conn)); @@ -2256,6 +2351,7 @@ main(int argc, char **argv) {"version", no_argument, NULL, 'V'}, {"pgdata", required_argument, NULL, 'D'}, {"format", required_argument, NULL, 'F'}, + {"incremental", required_argument, NULL, 'i'}, {"checkpoint", required_argument, NULL, 'c'}, {"create-slot", no_argument, NULL, 'C'}, {"max-rate", required_argument, NULL, 'r'}, @@ -2293,6 +2389,7 @@ main(int argc, char **argv) int option_index; char *compression_algorithm = "none"; char *compression_detail = NULL; + char *incremental_manifest = NULL; CompressionLocation compressloc = COMPRESS_LOCATION_UNSPECIFIED; pg_compress_specification client_compress; @@ -2317,7 +2414,7 @@ main(int argc, char **argv) atexit(cleanup_directories_atexit); - while ((c = getopt_long(argc, argv, "c:Cd:D:F:h:l:nNp:Pr:Rs:S:t:T:U:vwWX:zZ:", + while ((c = getopt_long(argc, argv, "c:Cd:D:F:h:i:l:nNp:Pr:Rs:S:t:T:U:vwWX:zZ:", long_options, &option_index)) != -1) { switch (c) @@ -2352,6 +2449,9 @@ main(int argc, char **argv) case 'h': dbhost = pg_strdup(optarg); break; + case 'i': + incremental_manifest = pg_strdup(optarg); + break; case 'l': label = pg_strdup(optarg); break; @@ -2765,7 +2865,7 @@ main(int argc, char **argv) } BaseBackup(compression_algorithm, compression_detail, compressloc, - &client_compress); + &client_compress, incremental_manifest); success = true; return 0; diff --git a/src/bin/pg_basebackup/t/010_pg_basebackup.pl b/src/bin/pg_basebackup/t/010_pg_basebackup.pl index b9f5e1266b..bf765291e7 100644 --- a/src/bin/pg_basebackup/t/010_pg_basebackup.pl +++ b/src/bin/pg_basebackup/t/010_pg_basebackup.pl @@ -223,10 +223,10 @@ SKIP: "check backup dir permissions"); } -# Only archive_status directory should be copied in pg_wal/. +# Only archive_status and summaries directories should be copied in pg_wal/. is_deeply( [ sort(slurp_dir("$tempdir/backup/pg_wal/")) ], - [ sort qw(. .. archive_status) ], + [ sort qw(. .. archive_status summaries) ], 'no WAL files copied'); # Contents of these directories should not be copied. diff --git a/src/bin/pg_combinebackup/.gitignore b/src/bin/pg_combinebackup/.gitignore new file mode 100644 index 0000000000..d7e617438c --- /dev/null +++ b/src/bin/pg_combinebackup/.gitignore @@ -0,0 +1 @@ +pg_combinebackup diff --git a/src/bin/pg_combinebackup/Makefile b/src/bin/pg_combinebackup/Makefile new file mode 100644 index 0000000000..78ba05e624 --- /dev/null +++ b/src/bin/pg_combinebackup/Makefile @@ -0,0 +1,52 @@ +#------------------------------------------------------------------------- +# +# Makefile for src/bin/pg_combinebackup +# +# Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group +# Portions Copyright (c) 1994, Regents of the University of California +# +# src/bin/pg_combinebackup/Makefile +# +#------------------------------------------------------------------------- + +PGFILEDESC = "pg_combinebackup - combine incremental backups" +PGAPPICON=win32 + +subdir = src/bin/pg_combinebackup +top_builddir = ../../.. +include $(top_builddir)/src/Makefile.global + +override CPPFLAGS := -I$(libpq_srcdir) $(CPPFLAGS) +LDFLAGS_INTERNAL += -L$(top_builddir)/src/fe_utils -lpgfeutils + +OBJS = \ + $(WIN32RES) \ + pg_combinebackup.o \ + backup_label.o \ + copy_file.o \ + load_manifest.o \ + reconstruct.o \ + write_manifest.o + +all: pg_combinebackup + +pg_combinebackup: $(OBJS) | submake-libpgport submake-libpgfeutils + $(CC) $(CFLAGS) $^ $(LDFLAGS) $(LDFLAGS_EX) $(LIBS) -o $@$(X) + +install: all installdirs + $(INSTALL_PROGRAM) pg_combinebackup$(X) '$(DESTDIR)$(bindir)/pg_combinebackup$(X)' + +installdirs: + $(MKDIR_P) '$(DESTDIR)$(bindir)' + +uninstall: + rm -f '$(DESTDIR)$(bindir)/pg_combinebackup$(X)' + +clean distclean maintainer-clean: + rm -f pg_combinebackup$(X) $(OBJS) + +check: + $(prove_check) + +installcheck: + $(prove_installcheck) diff --git a/src/bin/pg_combinebackup/backup_label.c b/src/bin/pg_combinebackup/backup_label.c new file mode 100644 index 0000000000..2a62aa6fad --- /dev/null +++ b/src/bin/pg_combinebackup/backup_label.c @@ -0,0 +1,281 @@ +/*------------------------------------------------------------------------- + * + * Read and manipulate backup label files + * + * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group + * Portions Copyright (c) 1994, Regents of the University of California + * + * src/bin/pg_combinebackup/backup_label.c + * + *------------------------------------------------------------------------- + */ +#include "postgres_fe.h" + +#include <unistd.h> + +#include "access/xlogdefs.h" +#include "backup_label.h" +#include "common/logging.h" +#include "common/file_perm.h" +#include "write_manifest.h" + +static int get_eol_offset(StringInfo buf); +static bool line_starts_with(char *s, char *e, char *match, char **sout); +static bool parse_lsn(char *s, char *e, XLogRecPtr *lsn, char **c); +static bool parse_tli(char *s, char *e, TimeLineID *tli); + +/* + * Parse a backup label file, starting at buf->cursor. + * + * We expect to find a START WAL LOCATION line, followed by a LSN, followed + * by a space; the resulting LSN is stored into *start_lsn. + * + * We expect to find a START TIMELINE line, followed by a TLI, followed by + * a newline; the resulting TLI is stored into *start_tli. + * + * We expect to find either both INCREMENTAL FROM LSN and INCREMENTAL FROM TLI + * or neither. If these are found, they should be followed by an LSN or TLI + * respectively and then by a newline, and the values will be stored into + * *previous_lsn and *previous_tli, respectively. + * + * Other lines in the provided backup_label data are ignored. filename is used + * for error reporting; errors are fatal. + */ +void +parse_backup_label(char *filename, StringInfo buf, + TimeLineID *start_tli, XLogRecPtr *start_lsn, + TimeLineID *previous_tli, XLogRecPtr *previous_lsn) +{ + int found = 0; + + *start_tli = 0; + *start_lsn = InvalidXLogRecPtr; + *previous_tli = 0; + *previous_lsn = InvalidXLogRecPtr; + + while (buf->cursor < buf->len) + { + char *s = &buf->data[buf->cursor]; + int eo = get_eol_offset(buf); + char *e = &buf->data[eo]; + char *c; + + if (line_starts_with(s, e, "START WAL LOCATION: ", &s)) + { + if (!parse_lsn(s, e, start_lsn, &c)) + pg_fatal("%s: could not parse START WAL LOCATION", + filename); + if (c >= e || *c != ' ') + pg_fatal("%s: improper terminator for START WAL LOCATION", + filename); + found |= 1; + } + else if (line_starts_with(s, e, "START TIMELINE: ", &s)) + { + if (!parse_tli(s, e, start_tli)) + pg_fatal("%s: could not parse TLI for START TIMELINE", + filename); + if (*start_tli == 0) + pg_fatal("%s: invalid TLI", filename); + found |= 2; + } + else if (line_starts_with(s, e, "INCREMENTAL FROM LSN: ", &s)) + { + if (!parse_lsn(s, e, previous_lsn, &c)) + pg_fatal("%s: could not parse INCREMENTAL FROM LSN", + filename); + if (c >= e || *c != '\n') + pg_fatal("%s: improper terminator for INCREMENTAL FROM LSN", + filename); + found |= 4; + } + else if (line_starts_with(s, e, "INCREMENTAL FROM TLI: ", &s)) + { + if (!parse_tli(s, e, previous_tli)) + pg_fatal("%s: could not parse INCREMENTAL FROM TLI", + filename); + if (*previous_tli == 0) + pg_fatal("%s: invalid TLI", filename); + found |= 8; + } + + buf->cursor = eo; + } + + if ((found & 1) == 0) + pg_fatal("%s: could not find START WAL LOCATION", filename); + if ((found & 2) == 0) + pg_fatal("%s: could not find START TIMELINE", filename); + if ((found & 4) != 0 && (found & 8) == 0) + pg_fatal("%s: INCREMENTAL FROM LSN requires INCREMENTAL FROM TLI", filename); + if ((found & 8) != 0 && (found & 4) == 0) + pg_fatal("%s: INCREMENTAL FROM TLI requires INCREMENTAL FROM LSN", filename); +} + +/* + * Write a backup label file to the output directory. + * + * This will be identical to the provided backup_label file, except that the + * INCREMENTAL FROM LSN and INCREMENTAL FROM TLI lines will be omitted. + * + * The new file will be checksummed using the specified algorithm. If + * mwriter != NULL, it will be added to the manifest. + */ +void +write_backup_label(char *output_directory, StringInfo buf, + pg_checksum_type checksum_type, manifest_writer *mwriter) +{ + char output_filename[MAXPGPATH]; + int output_fd; + pg_checksum_context checksum_ctx; + uint8 checksum_payload[PG_CHECKSUM_MAX_LENGTH]; + int checksum_length; + + pg_checksum_init(&checksum_ctx, checksum_type); + + snprintf(output_filename, MAXPGPATH, "%s/backup_label", output_directory); + + if ((output_fd = open(output_filename, + O_WRONLY | O_CREAT | O_EXCL | PG_BINARY, + pg_file_create_mode)) < 0) + pg_fatal("could not open file \"%s\": %m", output_filename); + + while (buf->cursor < buf->len) + { + char *s = &buf->data[buf->cursor]; + int eo = get_eol_offset(buf); + char *e = &buf->data[eo]; + + if (!line_starts_with(s, e, "INCREMENTAL FROM LSN: ", NULL) && + !line_starts_with(s, e, "INCREMENTAL FROM TLI: ", NULL)) + { + ssize_t wb; + + wb = write(output_fd, s, e - s); + if (wb != e - s) + { + if (wb < 0) + pg_fatal("could not write file \"%s\": %m", output_filename); + else + pg_fatal("could not write file \"%s\": wrote only %d of %d bytes", + output_filename, (int) wb, (int) (e - s)); + } + if (pg_checksum_update(&checksum_ctx, (uint8 *) s, e - s) < 0) + pg_fatal("could not update checksum of file \"%s\"", + output_filename); + } + + buf->cursor = eo; + } + + if (close(output_fd) != 0) + pg_fatal("could not close \"%s\": %m", output_filename); + + checksum_length = pg_checksum_final(&checksum_ctx, checksum_payload); + + if (mwriter != NULL) + { + struct stat sb; + + /* + * We could track the length ourselves, but must stat() to get the + * mtime. + */ + if (stat(output_filename, &sb) < 0) + pg_fatal("could not stat file \"%s\": %m", output_filename); + add_file_to_manifest(mwriter, "backup_label", sb.st_size, + sb.st_mtime, checksum_type, + checksum_length, checksum_payload); + } +} + +/* + * Return the offset at which the next line in the buffer starts, or there + * is none, the offset at which the buffer ends. + * + * The search begins at buf->cursor. + */ +static int +get_eol_offset(StringInfo buf) +{ + int eo = buf->cursor; + + while (eo < buf->len) + { + if (buf->data[eo] == '\n') + return eo + 1; + ++eo; + } + + return eo; +} + +/* + * Test whether the line that runs from s to e (inclusive of *s, but not + * inclusive of *e) starts with the match string provided, and return true + * or false according to whether or not this is the case. + * + * If the function returns true and if *sout != NULL, stores a pointer to the + * byte following the match into *sout. + */ +static bool +line_starts_with(char *s, char *e, char *match, char **sout) +{ + while (s < e && *match != '\0' && *s == *match) + ++s, ++match; + + if (*match == '\0' && sout != NULL) + *sout = s; + + return (*match == '\0'); +} + +/* + * Parse an LSN starting at s and not stopping at or before e. The return value + * is true on success and otherwise false. On success, stores the result into + * *lsn and sets *c to the first character that is not part of the LSN. + */ +static bool +parse_lsn(char *s, char *e, XLogRecPtr *lsn, char **c) +{ + char save = *e; + int nchars; + bool success; + unsigned hi; + unsigned lo; + + *e = '\0'; + success = (sscanf(s, "%X/%X%n", &hi, &lo, &nchars) == 2); + *e = save; + + if (success) + { + *lsn = ((XLogRecPtr) hi) << 32 | (XLogRecPtr) lo; + *c = s + nchars; + } + + return success; +} + +/* + * Parse a TLI starting at s and stopping at or before e. The return value is + * true on success and otherwise false. On success, stores the result into + * *tli. If the first character that is not part of the TLI is anything other + * than a newline, that is deemed a failure. + */ +static bool +parse_tli(char *s, char *e, TimeLineID *tli) +{ + char save = *e; + int nchars; + bool success; + + *e = '\0'; + success = (sscanf(s, "%u%n", tli, &nchars) == 1); + *e = save; + + if (success && s[nchars] != '\n') + success = false; + + return success; +} diff --git a/src/bin/pg_combinebackup/backup_label.h b/src/bin/pg_combinebackup/backup_label.h new file mode 100644 index 0000000000..08d6ed67a9 --- /dev/null +++ b/src/bin/pg_combinebackup/backup_label.h @@ -0,0 +1,29 @@ +/*------------------------------------------------------------------------- + * + * Read and manipulate backup label files + * + * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group + * Portions Copyright (c) 1994, Regents of the University of California + * + * src/bin/pg_combinebackup/backup_label.h + * + *------------------------------------------------------------------------- + */ +#ifndef BACKUP_LABEL_H +#define BACKUP_LABEL_H + +#include "common/checksum_helper.h" +#include "lib/stringinfo.h" + +struct manifest_writer; + +extern void parse_backup_label(char *filename, StringInfo buf, + TimeLineID *start_tli, + XLogRecPtr *start_lsn, + TimeLineID *previous_tli, + XLogRecPtr *previous_lsn); +extern void write_backup_label(char *output_directory, StringInfo buf, + pg_checksum_type checksum_type, + struct manifest_writer *mwriter); + +#endif /* BACKUP_LABEL_H */ diff --git a/src/bin/pg_combinebackup/copy_file.c b/src/bin/pg_combinebackup/copy_file.c new file mode 100644 index 0000000000..8ba6cc09e4 --- /dev/null +++ b/src/bin/pg_combinebackup/copy_file.c @@ -0,0 +1,169 @@ +/* + * Copy entire files. + * + * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group + * Portions Copyright (c) 1994, Regents of the University of California + * + * src/bin/pg_combinebackup/copy_file.h + * + *------------------------------------------------------------------------- + */ +#include "postgres_fe.h" + +#ifdef HAVE_COPYFILE_H +#include <copyfile.h> +#endif +#include <fcntl.h> +#include <sys/stat.h> +#include <unistd.h> + +#include "common/file_perm.h" +#include "common/logging.h" +#include "copy_file.h" + +static void copy_file_blocks(const char *src, const char *dst, + pg_checksum_context *checksum_ctx); + +#ifdef WIN32 +static void copy_file_copyfile(const char *src, const char *dst); +#endif + +/* + * Copy a regular file, optionally computing a checksum, and emitting + * appropriate debug messages. But if we're in dry-run mode, then just emit + * the messages and don't copy anything. + */ +void +copy_file(const char *src, const char *dst, + pg_checksum_context *checksum_ctx, bool dry_run) +{ + /* + * In dry-run mode, we don't actually copy anything, nor do we read any + * data from the source file, but we do verify that we can open it. + */ + if (dry_run) + { + int fd; + + if ((fd = open(src, O_RDONLY | PG_BINARY)) < 0) + pg_fatal("could not open \"%s\": %m", src); + if (close(fd) < 0) + pg_fatal("could not close \"%s\": %m", src); + } + + /* + * If we don't need to compute a checksum, then we can use any special + * operating system primitives that we know about to copy the file; this + * may be quicker than a naive block copy. + */ + if (checksum_ctx->type != CHECKSUM_TYPE_NONE) + { + char *strategy_name = NULL; + void (*strategy_implementation) (const char *, const char *) = NULL; + +#ifdef WIN32 + strategy_name = "CopyFile"; + strategy_implementation = copy_file_copyfile; +#endif + + if (strategy_name != NULL) + { + if (dry_run) + pg_log_debug("would copy \"%s\" to \"%s\" using strategy %s", + src, dst, strategy_name); + else + { + pg_log_debug("copying \"%s\" to \"%s\" using strategy %s", + src, dst, strategy_name); + (*strategy_implementation) (src, dst); + } + return; + } + } + + /* + * Fall back to the simple approach of reading and writing all the blocks, + * feeding them into the checksum context as we go. + */ + if (dry_run) + { + if (checksum_ctx->type == CHECKSUM_TYPE_NONE) + pg_log_debug("would copy \"%s\" to \"%s\"", + src, dst); + else + pg_log_debug("would copy \"%s\" to \"%s\" and checksum with %s", + src, dst, pg_checksum_type_name(checksum_ctx->type)); + } + else + { + if (checksum_ctx->type == CHECKSUM_TYPE_NONE) + pg_log_debug("copying \"%s\" to \"%s\"", + src, dst); + else + pg_log_debug("copying \"%s\" to \"%s\" and checksumming with %s", + src, dst, pg_checksum_type_name(checksum_ctx->type)); + copy_file_blocks(src, dst, checksum_ctx); + } +} + +/* + * Copy a file block by block, and optionally compute a checksum as we go. + */ +static void +copy_file_blocks(const char *src, const char *dst, + pg_checksum_context *checksum_ctx) +{ + int src_fd; + int dest_fd; + uint8 *buffer; + const int buffer_size = 50 * BLCKSZ; + ssize_t rb; + unsigned offset = 0; + + if ((src_fd = open(src, O_RDONLY | PG_BINARY, 0)) < 0) + pg_fatal("could not open file \"%s\": %m", src); + + if ((dest_fd = open(dst, O_WRONLY | O_CREAT | O_EXCL | PG_BINARY, + pg_file_create_mode)) < 0) + pg_fatal("could not open file \"%s\": %m", dst); + + buffer = pg_malloc(buffer_size); + + while ((rb = read(src_fd, buffer, buffer_size)) > 0) + { + ssize_t wb; + + if ((wb = write(dest_fd, buffer, rb)) != rb) + { + if (wb < 0) + pg_fatal("could not write file \"%s\": %m", dst); + else + pg_fatal("could not write file \"%s\": wrote only %d of %d bytes at offset %u", + dst, (int) wb, (int) rb, offset); + } + + if (pg_checksum_update(checksum_ctx, buffer, rb) < 0) + pg_fatal("could not update checksum of file \"%s\"", dst); + + offset += rb; + } + + if (rb < 0) + pg_fatal("could not read file \"%s\": %m", dst); + + pg_free(buffer); + close(src_fd); + close(dest_fd); +} + +#ifdef WIN32 +static void +copy_file_copyfile(const char *src, const char *dst) +{ + if (CopyFile(src, dst, true) == 0) + { + _dosmaperr(GetLastError()); + pg_fatal("could not copy \"%s\" to \"%s\": %m", src, dst); + } +} +#endif /* WIN32 */ diff --git a/src/bin/pg_combinebackup/copy_file.h b/src/bin/pg_combinebackup/copy_file.h new file mode 100644 index 0000000000..031030bacb --- /dev/null +++ b/src/bin/pg_combinebackup/copy_file.h @@ -0,0 +1,19 @@ +/* + * Copy entire files. + * + * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group + * Portions Copyright (c) 1994, Regents of the University of California + * + * src/bin/pg_combinebackup/copy_file.h + * + *------------------------------------------------------------------------- + */ +#ifndef COPY_FILE_H +#define COPY_FILE_H + +#include "common/checksum_helper.h" + +extern void copy_file(const char *src, const char *dst, + pg_checksum_context *checksum_ctx, bool dry_run); + +#endif /* COPY_FILE_H */ diff --git a/src/bin/pg_combinebackup/load_manifest.c b/src/bin/pg_combinebackup/load_manifest.c new file mode 100644 index 0000000000..d0b8de7912 --- /dev/null +++ b/src/bin/pg_combinebackup/load_manifest.c @@ -0,0 +1,245 @@ +/*------------------------------------------------------------------------- + * + * Load data from a backup manifest into memory. + * + * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group + * Portions Copyright (c) 1994, Regents of the University of California + * + * src/bin/pg_combinebackup/load_manifest.c + * + *------------------------------------------------------------------------- + */ + +#include "postgres_fe.h" + +#include <sys/stat.h> +#include <unistd.h> + +#include "common/hashfn.h" +#include "common/logging.h" +#include "common/parse_manifest.h" +#include "load_manifest.h" + +/* + * For efficiency, we'd like our hash table containing information about the + * manifest to start out with approximately the correct number of entries. + * There's no way to know the exact number of entries without reading the whole + * file, but we can get an estimate by dividing the file size by the estimated + * number of bytes per line. + * + * This could be off by about a factor of two in either direction, because the + * checksum algorithm has a big impact on the line lengths; e.g. a SHA512 + * checksum is 128 hex bytes, whereas a CRC-32C value is only 8, and there + * might be no checksum at all. + */ +#define ESTIMATED_BYTES_PER_MANIFEST_LINE 100 + +/* + * Define a hash table which we can use to store information about the files + * mentioned in the backup manifest. + */ +static uint32 hash_string_pointer(char *s); +#define SH_PREFIX manifest_files +#define SH_ELEMENT_TYPE manifest_file +#define SH_KEY_TYPE char * +#define SH_KEY pathname +#define SH_HASH_KEY(tb, key) hash_string_pointer(key) +#define SH_EQUAL(tb, a, b) (strcmp(a, b) == 0) +#define SH_SCOPE extern +#define SH_RAW_ALLOCATOR pg_malloc0 +#define SH_DEFINE +#include "lib/simplehash.h" + +static void record_manifest_details_for_file(JsonManifestParseContext *context, + char *pathname, size_t size, + pg_checksum_type checksum_type, + int checksum_length, + uint8 *checksum_payload); +static void record_manifest_details_for_wal_range(JsonManifestParseContext *context, + TimeLineID tli, + XLogRecPtr start_lsn, + XLogRecPtr end_lsn); +static void report_manifest_error(JsonManifestParseContext *context, + const char *fmt,...) + pg_attribute_printf(2, 3) pg_attribute_noreturn(); + +/* + * Load backup_manifest files from an array of backups and produces an array + * of manifest_data objects. + * + * NB: Since load_backup_manifest() can return NULL, the resulting array could + * contain NULL entries. + */ +manifest_data ** +load_backup_manifests(int n_backups, char **backup_directories) +{ + manifest_data **result; + int i; + + result = pg_malloc(sizeof(manifest_data *) * n_backups); + for (i = 0; i < n_backups; ++i) + result[i] = load_backup_manifest(backup_directories[i]); + + return result; +} + +/* + * Parse the backup_manifest file in the named backup directory. Construct a + * hash table with information about all the files it mentions, and a linked + * list of all the WAL ranges it mentions. + * + * If the backup_manifest file simply doesn't exist, logs a warning and returns + * NULL. Any other error, or any error parsing the contents of the file, is + * fatal. + */ +manifest_data * +load_backup_manifest(char *backup_directory) +{ + char pathname[MAXPGPATH]; + int fd; + struct stat statbuf; + off_t estimate; + uint32 initial_size; + manifest_files_hash *ht; + char *buffer; + int rc; + JsonManifestParseContext context; + manifest_data *result; + + /* Open the manifest file. */ + snprintf(pathname, MAXPGPATH, "%s/backup_manifest", backup_directory); + if ((fd = open(pathname, O_RDONLY | PG_BINARY, 0)) < 0) + { + if (errno == EEXIST) + { + pg_log_warning("\"%s\" does not exist", pathname); + return NULL; + } + pg_fatal("could not open file \"%s\": %m", pathname); + } + + /* Figure out how big the manifest is. */ + if (fstat(fd, &statbuf) != 0) + pg_fatal("could not stat file \"%s\": %m", pathname); + + /* Guess how large to make the hash table based on the manifest size. */ + estimate = statbuf.st_size / ESTIMATED_BYTES_PER_MANIFEST_LINE; + initial_size = Min(PG_UINT32_MAX, Max(estimate, 256)); + + /* Create the hash table. */ + ht = manifest_files_create(initial_size, NULL); + + /* + * Slurp in the whole file. + * + * This is not ideal, but there's currently no way to get pg_parse_json() + * to perform incremental parsing. + */ + buffer = pg_malloc(statbuf.st_size); + rc = read(fd, buffer, statbuf.st_size); + if (rc != statbuf.st_size) + { + if (rc < 0) + pg_fatal("could not read file \"%s\": %m", pathname); + else + pg_fatal("could not read file \"%s\": read %d of %lld", + pathname, rc, (long long int) statbuf.st_size); + } + + /* Close the manifest file. */ + close(fd); + + /* Parse the manifest. */ + result = pg_malloc0(sizeof(manifest_data)); + result->files = ht; + context.private_data = result; + context.perfile_cb = record_manifest_details_for_file; + context.perwalrange_cb = record_manifest_details_for_wal_range; + context.error_cb = report_manifest_error; + json_parse_manifest(&context, buffer, statbuf.st_size); + + /* All done. */ + pfree(buffer); + return result; +} + +/* + * Report an error while parsing the manifest. + * + * We consider all such errors to be fatal errors. The manifest parser + * expects this function not to return. + */ +static void +report_manifest_error(JsonManifestParseContext *context, const char *fmt,...) +{ + va_list ap; + + va_start(ap, fmt); + pg_log_generic_v(PG_LOG_ERROR, PG_LOG_PRIMARY, gettext(fmt), ap); + va_end(ap); + + exit(1); +} + +/* + * Record details extracted from the backup manifest for one file. + */ +static void +record_manifest_details_for_file(JsonManifestParseContext *context, + char *pathname, size_t size, + pg_checksum_type checksum_type, + int checksum_length, uint8 *checksum_payload) +{ + manifest_data *manifest = context->private_data; + manifest_file *m; + bool found; + + /* Make a new entry in the hash table for this file. */ + m = manifest_files_insert(manifest->files, pathname, &found); + if (found) + pg_fatal("duplicate path name in backup manifest: \"%s\"", pathname); + + /* Initialize the entry. */ + m->size = size; + m->checksum_type = checksum_type; + m->checksum_length = checksum_length; + m->checksum_payload = checksum_payload; +} + +/* + * Record details extracted from the backup manifest for one WAL range. + */ +static void +record_manifest_details_for_wal_range(JsonManifestParseContext *context, + TimeLineID tli, + XLogRecPtr start_lsn, XLogRecPtr end_lsn) +{ + manifest_data *manifest = context->private_data; + manifest_wal_range *range; + + /* Allocate and initialize a struct describing this WAL range. */ + range = palloc(sizeof(manifest_wal_range)); + range->tli = tli; + range->start_lsn = start_lsn; + range->end_lsn = end_lsn; + range->prev = manifest->last_wal_range; + range->next = NULL; + + /* Add it to the end of the list. */ + if (manifest->first_wal_range == NULL) + manifest->first_wal_range = range; + else + manifest->last_wal_range->next = range; + manifest->last_wal_range = range; +} + +/* + * Helper function for manifest_files hash table. + */ +static uint32 +hash_string_pointer(char *s) +{ + unsigned char *ss = (unsigned char *) s; + + return hash_bytes(ss, strlen(s)); +} diff --git a/src/bin/pg_combinebackup/load_manifest.h b/src/bin/pg_combinebackup/load_manifest.h new file mode 100644 index 0000000000..2bfeeff156 --- /dev/null +++ b/src/bin/pg_combinebackup/load_manifest.h @@ -0,0 +1,67 @@ +/*------------------------------------------------------------------------- + * + * Load data from a backup manifest into memory. + * + * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group + * Portions Copyright (c) 1994, Regents of the University of California + * + * src/bin/pg_combinebackup/load_manifest.h + * + *------------------------------------------------------------------------- + */ +#ifndef LOAD_MANIFEST_H +#define LOAD_MANIFEST_H + +#include "access/xlogdefs.h" +#include "common/checksum_helper.h" + +/* + * Each file described by the manifest file is parsed to produce an object + * like this. + */ +typedef struct manifest_file +{ + uint32 status; /* hash status */ + char *pathname; + size_t size; + pg_checksum_type checksum_type; + int checksum_length; + uint8 *checksum_payload; +} manifest_file; + +#define SH_PREFIX manifest_files +#define SH_ELEMENT_TYPE manifest_file +#define SH_KEY_TYPE char * +#define SH_SCOPE extern +#define SH_RAW_ALLOCATOR pg_malloc0 +#define SH_DECLARE +#include "lib/simplehash.h" + +/* + * Each WAL range described by the manifest file is parsed to produce an + * object like this. + */ +typedef struct manifest_wal_range +{ + TimeLineID tli; + XLogRecPtr start_lsn; + XLogRecPtr end_lsn; + struct manifest_wal_range *next; + struct manifest_wal_range *prev; +} manifest_wal_range; + +/* + * All the data parsed from a backup_manifest file. + */ +typedef struct manifest_data +{ + manifest_files_hash *files; + manifest_wal_range *first_wal_range; + manifest_wal_range *last_wal_range; +} manifest_data; + +extern manifest_data *load_backup_manifest(char *backup_directory); +extern manifest_data **load_backup_manifests(int n_backups, + char **backup_directories); + +#endif /* LOAD_MANIFEST_H */ diff --git a/src/bin/pg_combinebackup/meson.build b/src/bin/pg_combinebackup/meson.build new file mode 100644 index 0000000000..a6036dea74 --- /dev/null +++ b/src/bin/pg_combinebackup/meson.build @@ -0,0 +1,35 @@ +# Copyright (c) 2022-2023, PostgreSQL Global Development Group + +pg_combinebackup_sources = files( + 'pg_combinebackup.c', + 'backup_label.c', + 'copy_file.c', + 'load_manifest.c', + 'reconstruct.c', + 'write_manifest.c', +) + +if host_system == 'windows' + pg_combinebackup_sources += rc_bin_gen.process(win32ver_rc, extra_args: [ + '--NAME', 'pg_combinebackup', + '--FILEDESC', 'pg_combinebackup - combine incremental backups',]) +endif + +pg_combinebackup = executable('pg_combinebackup', + pg_combinebackup_sources, + dependencies: [frontend_code], + kwargs: default_bin_args, +) +bin_targets += pg_combinebackup + +tests += { + 'name': 'pg_combinebackup', + 'sd': meson.current_source_dir(), + 'bd': meson.current_build_dir(), + 'tap': { + 'tests': [ + 't/001_basic.pl', + 't/002_compare_backups.pl', + ], + } +} diff --git a/src/bin/pg_combinebackup/pg_combinebackup.c b/src/bin/pg_combinebackup/pg_combinebackup.c new file mode 100644 index 0000000000..32d2846433 --- /dev/null +++ b/src/bin/pg_combinebackup/pg_combinebackup.c @@ -0,0 +1,1270 @@ +/*------------------------------------------------------------------------- + * + * pg_combinebackup.c + * Combine incremental backups with prior backups. + * + * Copyright (c) 2017-2023, PostgreSQL Global Development Group + * + * IDENTIFICATION + * src/bin/pg_combinebackup/pg_combinebackup.c + * + *------------------------------------------------------------------------- + */ +#include "postgres_fe.h" + +#include <dirent.h> +#include <fcntl.h> +#include <limits.h> + +#include "access/xlogdefs.h" +#include "backup_label.h" +#include "common/blkreftable.h" +#include "common/checksum_helper.h" +#include "common/controldata_utils.h" +#include "common/file_perm.h" +#include "common/file_utils.h" +#include "common/logging.h" +#include "copy_file.h" +#include "fe_utils/option_utils.h" +#include "lib/stringinfo.h" +#include "load_manifest.h" +#include "getopt_long.h" +#include "reconstruct.h" +#include "write_manifest.h" + +/* Incremental file naming convention. */ +#define INCREMENTAL_PREFIX "INCREMENTAL." +#define INCREMENTAL_PREFIX_LENGTH 12 + +/* + * Tracking for directories that need to be removed, or have their contents + * removed, if the operation fails. + */ +typedef struct cb_cleanup_dir +{ + char *target_path; + bool rmtopdir; + struct cb_cleanup_dir *next; +} cb_cleanup_dir; + +/* + * Stores a tablespace mapping provided using -T, --tablespace-mapping. + */ +typedef struct cb_tablespace_mapping +{ + char old_dir[MAXPGPATH]; + char new_dir[MAXPGPATH]; + struct cb_tablespace_mapping *next; +} cb_tablespace_mapping; + +/* + * Stores data parsed from all command-line options. + */ +typedef struct cb_options +{ + bool debug; + char *output; + bool dry_run; + bool no_sync; + cb_tablespace_mapping *tsmappings; + pg_checksum_type manifest_checksums; + bool no_manifest; + DataDirSyncMethod sync_method; +} cb_options; + +/* + * Data about a tablespace. + * + * Every normal tablespace needs a tablespace mapping, but in-place tablespaces + * don't, so the list of tablespaces can contain more entries than the list of + * tablespace mappings. + */ +typedef struct cb_tablespace +{ + Oid oid; + bool in_place; + char old_dir[MAXPGPATH]; + char new_dir[MAXPGPATH]; + struct cb_tablespace *next; +} cb_tablespace; + +/* Directories to be removed if we exit uncleanly. */ +cb_cleanup_dir *cleanup_dir_list = NULL; + +static void add_tablespace_mapping(cb_options *opt, char *arg); +static StringInfo check_backup_label_files(int n_backups, char **backup_dirs); +static void check_control_files(int n_backups, char **backup_dirs); +static void check_input_dir_permissions(char *dir); +static void cleanup_directories_atexit(void); +static void create_output_directory(char *dirname, cb_options *opt); +static void help(const char *progname); +static bool parse_oid(char *s, Oid *result); +static void process_directory_recursively(Oid tsoid, + char *input_directory, + char *output_directory, + char *relative_path, + int n_prior_backups, + char **prior_backup_dirs, + manifest_data **manifests, + manifest_writer *mwriter, + cb_options *opt); +static int read_pg_version_file(char *directory); +static void remember_to_cleanup_directory(char *target_path, bool rmtopdir); +static void reset_directory_cleanup_list(void); +static cb_tablespace *scan_for_existing_tablespaces(char *pathname, + cb_options *opt); +static void slurp_file(int fd, char *filename, StringInfo buf, int maxlen); + +/* + * Main program. + */ +int +main(int argc, char *argv[]) +{ + static struct option long_options[] = { + {"debug", no_argument, NULL, 'd'}, + {"dry-run", no_argument, NULL, 'n'}, + {"no-sync", no_argument, NULL, 'N'}, + {"output", required_argument, NULL, 'o'}, + {"tablespace-mapping", no_argument, NULL, 'T'}, + {"manifest-checksums", required_argument, NULL, 1}, + {"no-manifest", no_argument, NULL, 2}, + {"sync-method", required_argument, NULL, 3}, + {NULL, 0, NULL, 0} + }; + + const char *progname; + char *last_input_dir; + int optindex; + int c; + int n_backups; + int n_prior_backups; + int version; + char **prior_backup_dirs; + cb_options opt; + cb_tablespace *tablespaces; + cb_tablespace *ts; + StringInfo last_backup_label; + manifest_data **manifests; + manifest_writer *mwriter; + + pg_logging_init(argv[0]); + progname = get_progname(argv[0]); + handle_help_version_opts(argc, argv, progname, help); + + memset(&opt, 0, sizeof(opt)); + opt.manifest_checksums = CHECKSUM_TYPE_CRC32C; + opt.sync_method = DATA_DIR_SYNC_METHOD_FSYNC; + + /* process command-line options */ + while ((c = getopt_long(argc, argv, "do:nNPT:", + long_options, &optindex)) != -1) + { + switch (c) + { + case 'd': + opt.debug = true; + pg_logging_increase_verbosity(); + break; + case 'o': + opt.output = optarg; + break; + case 'n': + opt.dry_run = true; + break; + case 'N': + opt.no_sync = true; + break; + case 'T': + add_tablespace_mapping(&opt, optarg); + break; + case 1: + if (!pg_checksum_parse_type(optarg, + &opt.manifest_checksums)) + pg_fatal("unrecognized checksum algorithm: \"%s\"", + optarg); + break; + case 2: + opt.no_manifest = true; + break; + case 3: + if (!parse_sync_method(optarg, &opt.sync_method)) + exit(1); + break; + default: + /* getopt_long already emitted a complaint */ + pg_log_error_hint("Try \"%s --help\" for more information.", progname); + exit(1); + } + } + + if (optind >= argc) + { + pg_log_error("%s: no input directories specified", progname); + pg_log_error_hint("Try \"%s --help\" for more information.", progname); + exit(1); + } + + if (opt.output == NULL) + pg_fatal("no output directory specified"); + + /* If no manifest is needed, no checksums are needed, either. */ + if (opt.no_manifest) + opt.manifest_checksums = CHECKSUM_TYPE_NONE; + + /* Read the server version from the final backup. */ + version = read_pg_version_file(argv[argc - 1]); + + /* Sanity-check control files. */ + n_backups = argc - optind; + check_control_files(n_backups, argv + optind); + + /* Sanity-check backup_label files, and get the contents of the last one. */ + last_backup_label = check_backup_label_files(n_backups, argv + optind); + + /* Load backup manifests. */ + manifests = load_backup_manifests(n_backups, argv + optind); + + /* Figure out which tablespaces are going to be included in the output. */ + last_input_dir = argv[argc - 1]; + check_input_dir_permissions(last_input_dir); + tablespaces = scan_for_existing_tablespaces(last_input_dir, &opt); + + /* + * Create output directories. + * + * We create one output directory for the main data directory plus one for + * each non-in-place tablespace. create_output_directory() will arrange + * for those directories to be cleaned up on failure. In-place tablespaces + * aren't handled at this stage because they're located beneath the main + * output directory, and thus the cleanup of that directory will get rid + * of them. Plus, the pg_tblspc directory that needs to contain them + * doesn't exist yet. + */ + atexit(cleanup_directories_atexit); + create_output_directory(opt.output, &opt); + for (ts = tablespaces; ts != NULL; ts = ts->next) + if (!ts->in_place) + create_output_directory(ts->new_dir, &opt); + + /* If we need to write a backup_manifest, prepare to do so. */ + if (!opt.dry_run && !opt.no_manifest) + mwriter = create_manifest_writer(opt.output); + else + mwriter = NULL; + + /* Write backup label into output directory. */ + if (opt.dry_run) + pg_log_debug("would generate \"%s/backup_label\"", opt.output); + else + { + pg_log_debug("generating \"%s/backup_label\"", opt.output); + last_backup_label->cursor = 0; + write_backup_label(opt.output, last_backup_label, + opt.manifest_checksums, mwriter); + } + + /* + * We'll need the pathnames to the prior backups. By "prior" we mean all + * but the last one listed on the command line. + */ + n_prior_backups = argc - optind - 1; + prior_backup_dirs = argv + optind; + + /* Process everything that's not part of a user-defined tablespace. */ + pg_log_debug("processing backup directory \"%s\"", last_input_dir); + process_directory_recursively(InvalidOid, last_input_dir, opt.output, + NULL, n_prior_backups, prior_backup_dirs, + manifests, mwriter, &opt); + + /* Process user-defined tablespaces. */ + for (ts = tablespaces; ts != NULL; ts = ts->next) + { + pg_log_debug("processing tablespace directory \"%s\"", ts->old_dir); + + /* + * If it's a normal tablespace, we need to set up a symbolic link from + * pg_tblspc/${OID} to the target directory; if it's an in-place + * tablespace, we need to create a directory at pg_tblspc/${OID}. + */ + if (!ts->in_place) + { + char linkpath[MAXPGPATH]; + + snprintf(linkpath, MAXPGPATH, "%s/pg_tblspc/%u", opt.output, + ts->oid); + + if (opt.dry_run) + pg_log_debug("would create symbolic link from \"%s\" to \"%s\"", + linkpath, ts->new_dir); + else + { + pg_log_debug("creating symbolic link from \"%s\" to \"%s\"", + linkpath, ts->new_dir); + if (symlink(ts->new_dir, linkpath) != 0) + pg_fatal("could not create symbolic link from \"%s\" to \"%s\": %m", + linkpath, ts->new_dir); + } + } + else + { + if (opt.dry_run) + pg_log_debug("would create directory \"%s\"", ts->new_dir); + else + { + pg_log_debug("creating directory \"%s\"", ts->new_dir); + if (pg_mkdir_p(ts->new_dir, pg_dir_create_mode) == -1) + pg_fatal("could not create directory \"%s\": %m", + ts->new_dir); + } + } + + /* OK, now handle the directory contents. */ + process_directory_recursively(ts->oid, ts->old_dir, ts->new_dir, + NULL, n_prior_backups, prior_backup_dirs, + manifests, mwriter, &opt); + } + + /* Finalize the backup_manifest, if we're generating one. */ + if (mwriter != NULL) + finalize_manifest(mwriter, + manifests[n_prior_backups]->first_wal_range); + + /* fsync that output directory unless we've been told not to do so */ + if (!opt.no_sync) + { + if (opt.dry_run) + pg_log_debug("would recursively fsync \"%s\"", opt.output); + else + { + pg_log_debug("recursively fsyncing \"%s\"", opt.output); + sync_pgdata(opt.output, version * 10000, opt.sync_method); + } + } + + /* It's a success, so don't remove the output directories. */ + reset_directory_cleanup_list(); + exit(0); +} + +/* + * Process the option argument for the -T, --tablespace-mapping switch. + */ +static void +add_tablespace_mapping(cb_options *opt, char *arg) +{ + cb_tablespace_mapping *tsmap = pg_malloc0(sizeof(cb_tablespace_mapping)); + char *dst; + char *dst_ptr; + char *arg_ptr; + + /* + * Basically, we just want to copy everything before the equals sign to + * tsmap->old_dir and everything afterwards to tsmap->new_dir, but if + * there's more or less than one equals sign, that's an error, and if + * there's an equals sign preceded by a backslash, don't treat it as a + * field separator but instead copy a literal equals sign. + */ + dst_ptr = dst = tsmap->old_dir; + for (arg_ptr = arg; *arg_ptr != '\0'; arg_ptr++) + { + if (dst_ptr - dst >= MAXPGPATH) + pg_fatal("directory name too long"); + + if (*arg_ptr == '\\' && *(arg_ptr + 1) == '=') + ; /* skip backslash escaping = */ + else if (*arg_ptr == '=' && (arg_ptr == arg || *(arg_ptr - 1) != '\\')) + { + if (tsmap->new_dir[0] != '\0') + pg_fatal("multiple \"=\" signs in tablespace mapping"); + else + dst = dst_ptr = tsmap->new_dir; + } + else + *dst_ptr++ = *arg_ptr; + } + if (!tsmap->old_dir[0] || !tsmap->new_dir[0]) + pg_fatal("invalid tablespace mapping format \"%s\", must be \"OLDDIR=NEWDIR\"", arg); + + /* + * All tablespaces are created with absolute directories, so specifying a + * non-absolute path here would never match, possibly confusing users. + * + * In contrast to pg_basebackup, both the old and new directories are on + * the local machine, so the local machine's definition of an absolute + * path is the only relevant one. + */ + if (!is_absolute_path(tsmap->old_dir)) + pg_fatal("old directory is not an absolute path in tablespace mapping: %s", + tsmap->old_dir); + + if (!is_absolute_path(tsmap->new_dir)) + pg_fatal("old directory is not an absolute path in tablespace mapping: %s", + tsmap->new_dir); + + /* Canonicalize paths to avoid spurious failures when comparing. */ + canonicalize_path(tsmap->old_dir); + canonicalize_path(tsmap->new_dir); + + /* Add it to the list. */ + tsmap->next = opt->tsmappings; + opt->tsmappings = tsmap; +} + +/* + * Check that the backup_label files form a coherent backup chain, and return + * the contents of the backup_label file from the latest backup. + */ +static StringInfo +check_backup_label_files(int n_backups, char **backup_dirs) +{ + StringInfo buf = makeStringInfo(); + StringInfo lastbuf = buf; + int i; + TimeLineID check_tli = 0; + XLogRecPtr check_lsn = InvalidXLogRecPtr; + + /* Try to read each backup_label file in turn, last to first. */ + for (i = n_backups - 1; i >= 0; --i) + { + char pathbuf[MAXPGPATH]; + int fd; + TimeLineID start_tli; + TimeLineID previous_tli; + XLogRecPtr start_lsn; + XLogRecPtr previous_lsn; + + /* Open the backup_label file. */ + snprintf(pathbuf, MAXPGPATH, "%s/backup_label", backup_dirs[i]); + pg_log_debug("reading \"%s\"", pathbuf); + if ((fd = open(pathbuf, O_RDONLY, 0)) < 0) + pg_fatal("could not open file \"%s\": %m", pathbuf); + + /* + * Slurp the whole file into memory. + * + * The exact size limit that we impose here doesn't really matter -- + * most of what's supposed to be in the file is fixed size and quite + * short. However, the length of the backup_label is limited (at least + * by some parts of the code) to MAXGPATH, so include that value in + * the maximum length that we tolerate. + */ + slurp_file(fd, pathbuf, buf, 10000 + MAXPGPATH); + + /* Close the file. */ + if (close(fd) != 0) + pg_fatal("could not close \"%s\": %m", pathbuf); + + /* Parse the file contents. */ + parse_backup_label(pathbuf, buf, &start_tli, &start_lsn, + &previous_tli, &previous_lsn); + + /* + * Sanity checks. + * + * XXX. It's actually not required that start_lsn == check_lsn. It + * would be OK if start_lsn > check_lsn provided that start_lsn is + * less than or equal to the relevant switchpoint. But at the moment + * we don't have that information. + */ + if (i > 0 && previous_tli == 0) + pg_fatal("backup at \"%s\" is a full backup, but only the first backup should be a full backup", + backup_dirs[i]); + if (i == 0 && previous_tli != 0) + pg_fatal("backup at \"%s\" is an incremental backup, but the first backup should be a full backup", + backup_dirs[i]); + if (i < n_backups - 1 && start_tli != check_tli) + pg_fatal("backup at \"%s\" starts on timeline %u, but expected %u", + backup_dirs[i], start_tli, check_tli); + if (i < n_backups - 1 && start_lsn != check_lsn) + pg_fatal("backup at \"%s\" starts at LSN %X/%X, but expected %X/%X", + backup_dirs[i], + LSN_FORMAT_ARGS(start_lsn), + LSN_FORMAT_ARGS(check_lsn)); + check_tli = previous_tli; + check_lsn = previous_lsn; + + /* + * The last backup label in the chain needs to be saved for later use, + * while the others are only needed within this loop. + */ + if (lastbuf == buf) + buf = makeStringInfo(); + else + resetStringInfo(buf); + } + + /* Free memory that we don't need any more. */ + if (lastbuf != buf) + { + pfree(buf->data); + pfree(buf); + } + + /* + * Return the data from the first backup_info that we read (which is the + * backup_label from the last directory specified on the command line). + */ + return lastbuf; +} + +/* + * Sanity check control files. + */ +static void +check_control_files(int n_backups, char **backup_dirs) +{ + int i; + uint64 system_identifier; + + /* Try to read each control file in turn, last to first. */ + for (i = n_backups - 1; i >= 0; --i) + { + ControlFileData *control_file; + bool crc_ok; + + pg_log_debug("reading \"%s/global/pg_control\"", backup_dirs[i]); + control_file = get_controlfile(backup_dirs[i], &crc_ok); + + /* Control file contents not meaningful if CRC is bad. */ + if (!crc_ok) + pg_fatal("%s/global/pg_control: crc is incorrect", backup_dirs[i]); + + /* Can't interpret control file if not current version. */ + if (control_file->pg_control_version != PG_CONTROL_VERSION) + pg_fatal("%s/global/pg_control: unexpected control file version", + backup_dirs[i]); + + /* System identifiers should all match. */ + if (i == n_backups - 1) + system_identifier = control_file->system_identifier; + else if (system_identifier != control_file->system_identifier) + pg_fatal("%s/global/pg_control: expected system identifier %llu, but found %llu", + backup_dirs[i], (unsigned long long) system_identifier, + (unsigned long long) control_file->system_identifier); + + /* Release memory. */ + pfree(control_file); + } + + /* + * If debug output is enabled, make a note of the system identifier that + * we found in all of the relevant control files. + */ + pg_log_debug("system identifier is %llu", + (unsigned long long) system_identifier); +} + +/* + * Set default permissions for new files and directories based on the + * permissions of the given directory. The intent here is that the output + * directory should use the same permissions scheme as the final input + * directory. + */ +static void +check_input_dir_permissions(char *dir) +{ + struct stat st; + + if (stat(dir, &st) != 0) + pg_fatal("could not stat \"%s\": %m", dir); + + SetDataDirectoryCreatePerm(st.st_mode); +} + +/* + * Clean up output directories before exiting. + */ +static void +cleanup_directories_atexit(void) +{ + while (cleanup_dir_list != NULL) + { + cb_cleanup_dir *dir = cleanup_dir_list; + + if (dir->rmtopdir) + { + pg_log_info("removing output directory \"%s\"", dir->target_path); + if (!rmtree(dir->target_path, dir->rmtopdir)) + pg_log_error("failed to remove output directory"); + } + else + { + pg_log_info("removing contents of output directory \"%s\"", + dir->target_path); + if (!rmtree(dir->target_path, dir->rmtopdir)) + pg_log_error("failed to remove contents of output directory"); + } + + cleanup_dir_list = cleanup_dir_list->next; + pfree(dir); + } +} + +/* + * Create the named output directory, unless it already exists or we're in + * dry-run mode. If it already exists but is not empty, that's a fatal error. + * + * Adds the created directory to the list of directories to be cleaned up + * at process exit. + */ +static void +create_output_directory(char *dirname, cb_options *opt) +{ + switch (pg_check_dir(dirname)) + { + case 0: + if (opt->dry_run) + { + pg_log_debug("would create directory \"%s\"", dirname); + return; + } + pg_log_debug("creating directory \"%s\"", dirname); + if (pg_mkdir_p(dirname, pg_dir_create_mode) == -1) + pg_fatal("could not create directory \"%s\": %m", dirname); + remember_to_cleanup_directory(dirname, true); + break; + + case 1: + pg_log_debug("using existing directory \"%s\"", dirname); + remember_to_cleanup_directory(dirname, false); + break; + + case 2: + case 3: + case 4: + pg_fatal("directory \"%s\" exists but is not empty", dirname); + + case -1: + pg_fatal("could not access directory \"%s\": %m", dirname); + } +} + +/* + * help + * + * Prints help page for the program + * + * progname: the name of the executed program, such as "pg_combinebackup" + */ +static void +help(const char *progname) +{ + printf(_("%s combines incremental backups.\n\n"), progname); + printf(_("Usage:\n")); + printf(_(" %s [OPTION]... DIRECTORY...\n"), progname); + printf(_("\nOptions:\n")); + printf(_(" -d, --debug generate lots of debugging output\n")); + printf(_(" -n, --dry-run don't actually do anything\n")); + printf(_(" -N, --no-sync do not wait for changes to be written safely to disk\n")); + printf(_(" -o, --output output directory\n")); + printf(_(" -T, --tablespace-mapping=OLDDIR=NEWDIR\n")); + printf(_(" relocate tablespace in OLDDIR to NEWDIR\n")); + printf(_(" --manifest-checksums=SHA{224,256,384,512}|CRC32C|NONE\n" + " use algorithm for manifest checksums\n")); + printf(_(" --no-manifest suppress generation of backup manifest\n")); + printf(_(" --sync-method=METHOD set method for syncing files to disk\n")); + printf(_(" -?, --help show this help, then exit\n")); + + printf(_("\nReport bugs to <%s>.\n"), PACKAGE_BUGREPORT); + printf(_("%s home page: <%s>\n"), PACKAGE_NAME, PACKAGE_URL); +} + +/* + * Try to parse a string as a non-zero OID without leading zeroes. + * + * If it works, return true and set *result to the answer, else return false. + */ +static bool +parse_oid(char *s, Oid *result) +{ + Oid oid; + char *ep; + + errno = 0; + oid = strtoul(s, &ep, 10); + if (errno != 0 || *ep != '\0' || oid < 1 || oid > PG_UINT32_MAX) + return false; + + *result = oid; + return true; +} + +/* + * Copy files from the input directory to the output directory, reconstructing + * full files from incremental files as required. + * + * If processing is a user-defined tablespace, the tsoid should be the OID + * of that tablespace and input_directory and output_directory should be the + * toplevel input and output directories for that tablespace. Otherwise, + * tsoid should be InvalidOid and input_directory and output_directory should + * be the main input and output directories. + * + * relative_path is the path beneath the given input and output directories + * that we are currently processing. If NULL, it indicates that we're + * processing the input and output directories themselves. + * + * n_prior_backups is the number of prior backups that we have available. + * This doesn't count the very last backup, which is referenced by + * output_directory, just the older ones. prior_backup_dirs is an array of + * the locations of those previous backups. + */ +static void +process_directory_recursively(Oid tsoid, + char *input_directory, + char *output_directory, + char *relative_path, + int n_prior_backups, + char **prior_backup_dirs, + manifest_data **manifests, + manifest_writer *mwriter, + cb_options *opt) +{ + char ifulldir[MAXPGPATH]; + char ofulldir[MAXPGPATH]; + char manifest_prefix[MAXPGPATH]; + DIR *dir; + struct dirent *de; + bool is_pg_tblspc; + bool is_pg_wal; + manifest_data *latest_manifest = manifests[n_prior_backups]; + pg_checksum_type checksum_type; + + StaticAssertStmt(strlen(INCREMENTAL_PREFIX) == INCREMENTAL_PREFIX_LENGTH, + "INCREMENTAL_PREFIX_LENGTH is incorrect"); + + /* + * pg_tblspc and pg_wal are special cases, so detect those here. + * + * pg_tblspc is only special at the top level, but subdirectories of + * pg_wal are just as special as the top level directory. + * + * Since incremental backup does not exist in pre-v10 versions, we don't + * have to worry about the old pg_xlog naming. + */ + is_pg_tblspc = !OidIsValid(tsoid) && relative_path != NULL && + strcmp(relative_path, "pg_tblspc") == 0; + is_pg_wal = !OidIsValid(tsoid) && relative_path != NULL && + (strcmp(relative_path, "pg_wal") == 0 || + strncmp(relative_path, "pg_wal/", 7) == 0); + + /* + * If we're under pg_wal, then we don't need checksums, because these + * files aren't included in the backup manifest. Otherwise use whatever + * type of checksum is configured. + */ + if (!is_pg_wal) + checksum_type = opt->manifest_checksums; + else + checksum_type = CHECKSUM_TYPE_NONE; + + /* + * Append the relative path to the input and output directories, and + * figure out the appropriate prefix to add to files in this directory + * when looking them up in a backup manifest. + */ + if (relative_path == NULL) + { + strncpy(ifulldir, input_directory, MAXPGPATH); + strncpy(ofulldir, output_directory, MAXPGPATH); + if (OidIsValid(tsoid)) + snprintf(manifest_prefix, MAXPGPATH, "pg_tblspc/%u/", tsoid); + else + manifest_prefix[0] = '\0'; + } + else + { + snprintf(ifulldir, MAXPGPATH, "%s/%s", input_directory, + relative_path); + snprintf(ofulldir, MAXPGPATH, "%s/%s", output_directory, + relative_path); + if (OidIsValid(tsoid)) + snprintf(manifest_prefix, MAXPGPATH, "pg_tblspc/%u/%s/", + tsoid, relative_path); + else + snprintf(manifest_prefix, MAXPGPATH, "%s/", relative_path); + } + + /* + * Toplevel output directories have already been created by the time this + * function is called, but any subdirectories are our responsibility. + */ + if (relative_path != NULL) + { + if (opt->dry_run) + pg_log_debug("would create directory \"%s\"", ofulldir); + else + { + pg_log_debug("creating directory \"%s\"", ofulldir); + if (mkdir(ofulldir, pg_dir_create_mode) == -1) + pg_fatal("could not create directory \"%s\": %m", ofulldir); + } + } + + /* It's time to scan the directory. */ + if ((dir = opendir(ifulldir)) == NULL) + pg_fatal("could not open directory \"%s\": %m", ifulldir); + while (errno = 0, (de = readdir(dir)) != NULL) + { + PGFileType type; + char ifullpath[MAXPGPATH]; + char ofullpath[MAXPGPATH]; + char manifest_path[MAXPGPATH]; + Oid oid = InvalidOid; + int checksum_length = 0; + uint8 *checksum_payload = NULL; + pg_checksum_context checksum_ctx; + + /* Ignore "." and ".." entries. */ + if (strcmp(de->d_name, ".") == 0 || + strcmp(de->d_name, "..") == 0) + continue; + + /* Construct input path. */ + snprintf(ifullpath, MAXPGPATH, "%s/%s", ifulldir, de->d_name); + + /* Figure out what kind of directory entry this is. */ + type = get_dirent_type(ifullpath, de, false, PG_LOG_ERROR); + if (type == PGFILETYPE_ERROR) + exit(1); + + /* + * If we're processing pg_tblspc, then check whether the filename + * looks like it could be a tablespace OID. If so, and if the + * directory entry is a symbolic link or a directory, skip it. + * + * Our goal here is to ignore anything that would have been considered + * by scan_for_existing_tablespaces to be a tablespace. + */ + if (is_pg_tblspc && parse_oid(de->d_name, &oid) && + (type == PGFILETYPE_LNK || type == PGFILETYPE_DIR)) + continue; + + /* If it's a directory, recurse. */ + if (type == PGFILETYPE_DIR) + { + char new_relative_path[MAXPGPATH]; + + /* Append new pathname component to relative path. */ + if (relative_path == NULL) + strncpy(new_relative_path, de->d_name, MAXPGPATH); + else + snprintf(new_relative_path, MAXPGPATH, "%s/%s", relative_path, + de->d_name); + + /* And recurse. */ + process_directory_recursively(tsoid, + input_directory, output_directory, + new_relative_path, + n_prior_backups, prior_backup_dirs, + manifests, mwriter, opt); + continue; + } + + /* Skip anything that's not a regular file. */ + if (type != PGFILETYPE_REG) + { + if (type == PGFILETYPE_LNK) + pg_log_warning("skipping symbolic link \"%s\"", ifullpath); + else + pg_log_warning("skipping special file \"%s\"", ifullpath); + continue; + } + + /* + * Skip the backup_label and backup_manifest files; they require + * special handling and are handled elsewhere. + */ + if (relative_path == NULL && + (strcmp(de->d_name, "backup_label") == 0 || + strcmp(de->d_name, "backup_manifest") == 0)) + continue; + + /* + * If it's an incremental file, hand it off to the reconstruction + * code, which will figure out what to do. + */ + if (strncmp(de->d_name, INCREMENTAL_PREFIX, + INCREMENTAL_PREFIX_LENGTH) == 0) + { + /* Output path should not include "INCREMENTAL." prefix. */ + snprintf(ofullpath, MAXPGPATH, "%s/%s", ofulldir, + de->d_name + INCREMENTAL_PREFIX_LENGTH); + + + /* Manifest path likewise omits incremental prefix. */ + snprintf(manifest_path, MAXPGPATH, "%s%s", manifest_prefix, + de->d_name + INCREMENTAL_PREFIX_LENGTH); + + /* Reconstruction logic will do the rest. */ + reconstruct_from_incremental_file(ifullpath, ofullpath, + relative_path, + de->d_name + INCREMENTAL_PREFIX_LENGTH, + n_prior_backups, + prior_backup_dirs, + manifests, + manifest_path, + checksum_type, + &checksum_length, + &checksum_payload, + opt->dry_run); + } + else + { + /* Construct the path that the backup_manifest will use. */ + snprintf(manifest_path, MAXPGPATH, "%s%s", manifest_prefix, + de->d_name); + + /* + * It's not an incremental file, so we need to copy the entire + * file to the output directory. + * + * If a checksum of the required type already exists in the + * backup_manifest for the final input directory, we can save some + * work by reusing that checksum instead of computing a new one. + */ + if (checksum_type != CHECKSUM_TYPE_NONE && + latest_manifest != NULL) + { + manifest_file *mfile; + + mfile = manifest_files_lookup(latest_manifest->files, + manifest_path); + if (mfile == NULL) + { + /* + * The directory is out of sync with the backup_manifest, + * so emit a warning. + */ + pg_log_warning("\"%s/backup_manifest\" contains no entry for \"%s\"", + input_directory, manifest_path); + } + else if (mfile->checksum_type == checksum_type) + { + checksum_length = mfile->checksum_length; + checksum_payload = mfile->checksum_payload; + } + } + + /* + * If we're reusing a checksum, then we don't need copy_file() to + * compute one for us, but otherwise, it needs to compute whatever + * type of checksum we need. + */ + if (checksum_length != 0) + pg_checksum_init(&checksum_ctx, CHECKSUM_TYPE_NONE); + else + pg_checksum_init(&checksum_ctx, checksum_type); + + /* Actually copy the file. */ + snprintf(ofullpath, MAXPGPATH, "%s/%s", ofulldir, de->d_name); + copy_file(ifullpath, ofullpath, &checksum_ctx, opt->dry_run); + + /* + * If copy_file() performed a checksum calculation for us, then + * save the results (except in dry-run mode, when there's no + * point). + */ + if (checksum_ctx.type != CHECKSUM_TYPE_NONE && !opt->dry_run) + { + checksum_payload = pg_malloc(PG_CHECKSUM_MAX_LENGTH); + checksum_length = pg_checksum_final(&checksum_ctx, + checksum_payload); + } + } + + /* Generate manifest entry, if needed. */ + if (mwriter != NULL) + { + struct stat sb; + + /* + * In order to generate a manifest entry, we need the file size + * and mtime. We have no way to know the correct mtime except to + * stat() the file, so just do that and get the size as well. + * + * If we didn't need the mtime here, we could try to obtain the + * file size from the reconstruction or file copy process above, + * although that is actually not convenient in all cases. If we + * write the file ourselves then clearly we can keep a count of + * bytes, but if we use something like CopyFile() then it's + * trickier. Since we have to stat() anyway to get the mtime, + * there's no point in worrying about it. + */ + if (stat(ofullpath, &sb) < 0) + pg_fatal("could not stat file \"%s\": %m", ofullpath); + + /* OK, now do the work. */ + add_file_to_manifest(mwriter, manifest_path, + sb.st_size, sb.st_mtime, + checksum_type, checksum_length, + checksum_payload); + } + + /* Avoid leaking memory. */ + if (checksum_payload != NULL) + pfree(checksum_payload); + } + + closedir(dir); +} + +/* + * Read the version number from PG_VERSION and convert it to the usual server + * version number format. (e.g. If PG_VERSION contains "14\n" this function + * will return 140000) + */ +static int +read_pg_version_file(char *directory) +{ + char filename[MAXPGPATH]; + StringInfoData buf; + int fd; + int version; + char *ep; + + /* Construct pathname. */ + snprintf(filename, MAXPGPATH, "%s/PG_VERSION", directory); + + /* Open file. */ + if ((fd = open(filename, O_RDONLY, 0)) < 0) + pg_fatal("could not open file \"%s\": %m", filename); + + /* Read into memory. Length limit of 128 should be more than generous. */ + initStringInfo(&buf); + slurp_file(fd, filename, &buf, 128); + + /* Close the file. */ + if (close(fd) != 0) + pg_fatal("could not close \"%s\": %m", filename); + + /* Convert to integer. */ + errno = 0; + version = strtoul(buf.data, &ep, 10); + if (errno != 0 || *ep != '\n') + { + /* + * Incremental backup is not relevant to very old server versions that + * used multi-part version number (e.g. 9.6, or 8.4). So if we see + * what looks like the beginning of such a version number, just bail + * out. + */ + if (version < 10 && *ep == '.') + pg_fatal("%s: server version too old\n", filename); + pg_fatal("%s: could not parse version number\n", filename); + } + + /* Debugging output. */ + pg_log_debug("read server version %d from \"%s\"", version, filename); + + /* Release memory and return result. */ + pfree(buf.data); + return version * 10000; +} + +/* + * Add a directory to the list of output directories to clean up. + */ +static void +remember_to_cleanup_directory(char *target_path, bool rmtopdir) +{ + cb_cleanup_dir *dir = pg_malloc(sizeof(cb_cleanup_dir)); + + dir->target_path = target_path; + dir->rmtopdir = rmtopdir; + dir->next = cleanup_dir_list; + cleanup_dir_list = dir; +} + +/* + * Empty out the list of directories scheduled for cleanup a exit. + * + * We want to remove the output directories only on a failure, so call this + * function when we know that the operation has succeeded. + * + * Since we only expect this to be called when we're about to exit, we could + * just set cleanup_dir_list to NULL and be done with it, but we free the + * memory to be tidy. + */ +static void +reset_directory_cleanup_list(void) +{ + while (cleanup_dir_list != NULL) + { + cb_cleanup_dir *dir = cleanup_dir_list; + + cleanup_dir_list = cleanup_dir_list->next; + pfree(dir); + } +} + +/* + * Scan the pg_tblspc directory of the final input backup to get a canonical + * list of what tablespaces are part of the backup. + * + * 'pathname' should be the path to the toplevel backup directory for the + * final backup in the backup chain. + */ +static cb_tablespace * +scan_for_existing_tablespaces(char *pathname, cb_options *opt) +{ + char pg_tblspc[MAXPGPATH]; + DIR *dir; + struct dirent *de; + cb_tablespace *tslist = NULL; + + snprintf(pg_tblspc, MAXPGPATH, "%s/pg_tblspc", pathname); + pg_log_debug("scanning \"%s\"", pg_tblspc); + + if ((dir = opendir(pg_tblspc)) == NULL) + pg_fatal("could not open directory \"%s\": %m", pathname); + + while (errno = 0, (de = readdir(dir)) != NULL) + { + Oid oid; + char tblspcdir[MAXPGPATH]; + char link_target[MAXPGPATH]; + int link_length; + cb_tablespace *ts; + cb_tablespace *otherts; + PGFileType type; + + /* Silently ignore "." and ".." entries. */ + if (strcmp(de->d_name, ".") == 0 || strcmp(de->d_name, "..") == 0) + continue; + + /* Construct full pathname. */ + snprintf(tblspcdir, MAXPGPATH, "%s/%s", pg_tblspc, de->d_name); + + /* Ignore any file name that doesn't look like a proper OID. */ + if (!parse_oid(de->d_name, &oid)) + { + pg_log_debug("skipping \"%s\" because the filename is not a legal tablespace OID", + tblspcdir); + continue; + } + + /* Only symbolic links and directories are tablespaces. */ + type = get_dirent_type(tblspcdir, de, false, PG_LOG_ERROR); + if (type == PGFILETYPE_ERROR) + exit(1); + if (type != PGFILETYPE_LNK && type != PGFILETYPE_DIR) + { + pg_log_debug("skipping \"%s\" because it is neither a symbolic link nor a directory", + tblspcdir); + continue; + } + + /* Create a new tablespace object. */ + ts = pg_malloc0(sizeof(cb_tablespace)); + ts->oid = oid; + + /* + * If it's a link, it's not an in-place tablespace. Otherwise, it must + * be a directory, and thus an in-place tablespace. + */ + if (type == PGFILETYPE_LNK) + { + cb_tablespace_mapping *tsmap; + + /* Read the link target. */ + link_length = readlink(tblspcdir, link_target, sizeof(link_target)); + if (link_length < 0) + pg_fatal("could not read symbolic link \"%s\": %m", + tblspcdir); + if (link_length >= sizeof(link_target)) + pg_fatal("symbolic link \"%s\" is too long", tblspcdir); + link_target[link_length] = '\0'; + if (!is_absolute_path(link_target)) + pg_fatal("symbolic link \"%s\" is relative", tblspcdir); + + /* Caonicalize the link target. */ + canonicalize_path(link_target); + + /* + * Find the corresponding tablespace mapping and copy the relevant + * details into the new tablespace entry. + */ + for (tsmap = opt->tsmappings; tsmap != NULL; tsmap = tsmap->next) + { + if (strcmp(tsmap->old_dir, link_target) == 0) + { + strncpy(ts->old_dir, tsmap->old_dir, MAXPGPATH); + strncpy(ts->new_dir, tsmap->new_dir, MAXPGPATH); + ts->in_place = false; + break; + } + } + + /* Every non-in-place tablespace must be mapped. */ + if (tsmap == NULL) + pg_fatal("tablespace at \"%s\" has no tablespace mapping", + link_target); + } + else + { + /* + * For an in-place tablespace, there's no separate directory, so + * we just record the paths within the data directories. + */ + snprintf(ts->old_dir, MAXPGPATH, "%s/%s", pg_tblspc, de->d_name); + snprintf(ts->new_dir, MAXPGPATH, "%s/pg_tblpc/%s", opt->output, + de->d_name); + ts->in_place = true; + } + + /* Tablespaces should not share a directory. */ + for (otherts = tslist; otherts != NULL; otherts = otherts->next) + if (strcmp(ts->new_dir, otherts->new_dir) == 0) + pg_fatal("tablespaces with OIDs %u and %u both point at \"%s\"", + otherts->oid, oid, ts->new_dir); + + /* Add this tablespace to the list. */ + ts->next = tslist; + tslist = ts; + } + + return tslist; +} + +/* + * Read a file into a StringInfo. + * + * fd is used for the actual file I/O, filename for error reporting purposes. + * A file longer than maxlen is a fatal error. + */ +static void +slurp_file(int fd, char *filename, StringInfo buf, int maxlen) +{ + struct stat st; + ssize_t rb; + + /* Check file size, and complain if it's too large. */ + if (fstat(fd, &st) != 0) + pg_fatal("could not stat \"%s\": %m", filename); + if (st.st_size > maxlen) + pg_fatal("file \"%s\" is too large", filename); + + /* Make sure we have enough space. */ + enlargeStringInfo(buf, st.st_size); + + /* Read the data. */ + rb = read(fd, &buf->data[buf->len], st.st_size); + + /* + * We don't expect any concurrent changes, so we should read exactly the + * expected number of bytes. + */ + if (rb != st.st_size) + { + if (rb < 0) + pg_fatal("could not read file \"%s\": %m", filename); + else + pg_fatal("could not read file \"%s\": read only %d of %d bytes", + filename, (int) rb, (int) st.st_size); + } + + /* Adjust buffer length for new data and restore trailing-\0 invariant */ + buf->len += rb; + buf->data[buf->len] = '\0'; +} diff --git a/src/bin/pg_combinebackup/reconstruct.c b/src/bin/pg_combinebackup/reconstruct.c new file mode 100644 index 0000000000..c774bf1842 --- /dev/null +++ b/src/bin/pg_combinebackup/reconstruct.c @@ -0,0 +1,618 @@ +/*------------------------------------------------------------------------- + * + * reconstruct.c + * Reconstruct full file from incremental file and backup chain. + * + * Copyright (c) 2017-2023, PostgreSQL Global Development Group + * + * IDENTIFICATION + * src/bin/pg_combinebackup/reconstruct.c + * + *------------------------------------------------------------------------- + */ +#include "postgres_fe.h" + +#include <unistd.h> + +#include "backup/basebackup_incremental.h" +#include "common/logging.h" +#include "common/file_perm.h" +#include "copy_file.h" +#include "reconstruct.h" +#include "storage/block.h" + +/* + * An rfile stores the data that we need in order to be able to use some file + * on disk for reconstruction. For any given output file, we create one rfile + * per backup that we need to consult when we constructing that output file. + * + * If we find a full version of the file in the backup chain, then only + * filename and fd are initialized; the remaining fields are 0 or NULL. + * For an incremental file, header_length, num_blocks, relative_block_numbers, + * and truncation_block_length are also set. + * + * num_blocks_read and highest_offset_read always start out as 0. + */ +typedef struct rfile +{ + char *filename; + int fd; + size_t header_length; + unsigned num_blocks; + BlockNumber *relative_block_numbers; + unsigned truncation_block_length; + unsigned num_blocks_read; + off_t highest_offset_read; +} rfile; + +static void debug_reconstruction(int n_source, + rfile **sources, + bool dry_run); +static unsigned find_reconstructed_block_length(rfile *s); +static rfile *make_incremental_rfile(char *filename); +static rfile *make_rfile(char *filename, bool missing_ok); +static void write_reconstructed_file(char *input_filename, + char *output_filename, + unsigned block_length, + rfile **sourcemap, + off_t *offsetmap, + pg_checksum_context *checksum_ctx, + bool dry_run); +static void read_bytes(rfile *rf, void *buffer, unsigned length); + +/* + * Reconstruct a full file from an incremental file and a chain of prior + * backups. + * + * input_filename should be the path to the incremental file, and + * output_filename should be the path where the reconstructed file is to be + * written. + * + * relative_path should be the relative path to the directory containing this + * file. bare_file_name should be the name of the file within that directory, + * without "INCREMENTAL.". + * + * n_prior_backups is the number of prior backups, and prior_backup_dirs is + * an array of pathnames where those backups can be found. + */ +void +reconstruct_from_incremental_file(char *input_filename, + char *output_filename, + char *relative_path, + char *bare_file_name, + int n_prior_backups, + char **prior_backup_dirs, + manifest_data **manifests, + char *manifest_path, + pg_checksum_type checksum_type, + int *checksum_length, + uint8 **checksum_payload, + bool dry_run) +{ + rfile **source; + rfile *latest_source = NULL; + rfile **sourcemap; + off_t *offsetmap; + unsigned block_length; + unsigned num_missing_blocks; + unsigned i; + unsigned sidx = n_prior_backups; + bool full_copy_possible = true; + int copy_source_index = -1; + rfile *copy_source = NULL; + pg_checksum_context checksum_ctx; + + /* + * Every block must come either from the latest version of the file or + * from one of the prior backups. + */ + source = pg_malloc0(sizeof(rfile *) * (1 + n_prior_backups)); + + /* + * Use the information from the latest incremental file to figure out how + * long the reconstructed file should be. + */ + latest_source = make_incremental_rfile(input_filename); + source[n_prior_backups] = latest_source; + block_length = find_reconstructed_block_length(latest_source); + + /* + * For each block in the output file, we need to know from which file we + * need to obtain it and at what offset in that file it's stored. + * sourcemap gives us the first of these things, and offsetmap the latter. + */ + sourcemap = pg_malloc0(sizeof(rfile *) * block_length); + offsetmap = pg_malloc0(sizeof(off_t) * block_length); + + /* + * Blocks prior to the truncation_block_length threshold must be obtained + * from some prior backup, while those after that threshold are left as + * zeroes if not present in the newest incremental file. + * num_missing_blocks counts the number of blocks that we must be found + * somewhere in the backup chain, and is thus initially equal to + * truncation_block_length. + */ + num_missing_blocks = latest_source->truncation_block_length; + + /* + * Every block that is present in the newest incremental file should be + * sourced from that file. If it precedes the truncation_block_length, + * it's a block that we would otherwise have had to find in an older + * backup and thus reduces the number of blocks remaining to be found by + * one; otherwise, it's an extra block that needs to be included in the + * output but would not have needed to be found in an older backup if it + * had not been present. + */ + for (i = 0; i < latest_source->num_blocks; ++i) + { + BlockNumber b = latest_source->relative_block_numbers[i]; + + Assert(b < block_length); + sourcemap[b] = latest_source; + offsetmap[b] = latest_source->header_length + (i * BLCKSZ); + if (b < latest_source->truncation_block_length) + num_missing_blocks--; + + /* + * A full copy of a file from an earlier backup is only possible if no + * blocks are needed from any later incremental file. + */ + full_copy_possible = false; + } + + while (num_missing_blocks > 0) + { + char source_filename[MAXPGPATH]; + rfile *s; + + /* + * Move to the next backup in the chain. If there are no more, then + * something has gone wrong and reconstruction has failed. + */ + if (sidx == 0) + pg_fatal("reconstruction for file \"%s\" failed to find %u required blocks", + output_filename, num_missing_blocks); + --sidx; + + /* + * Look for the full file in the previous backup. If not found, then + * look for an incremental file instead. + */ + snprintf(source_filename, MAXPGPATH, "%s/%s/%s", + prior_backup_dirs[sidx], relative_path, bare_file_name); + if ((s = make_rfile(source_filename, true)) == NULL) + { + snprintf(source_filename, MAXPGPATH, "%s/%s/INCREMENTAL.%s", + prior_backup_dirs[sidx], relative_path, bare_file_name); + s = make_incremental_rfile(source_filename); + } + source[sidx] = s; + + /* + * If s->header_length == 0, then this is a full file; otherwise, it's + * an incremental file. + */ + if (s->header_length != 0) + { + /* + * Since we found another incremental file, source all blocks from + * it that we need but don't yet have. + */ + for (i = 0; i < s->num_blocks; ++i) + { + BlockNumber b = s->relative_block_numbers[i]; + + if (b < latest_source->truncation_block_length && + sourcemap[b] == NULL) + { + sourcemap[b] = s; + offsetmap[b] = s->header_length + (i * BLCKSZ); + + Assert(num_missing_blocks > 0); + --num_missing_blocks; + + /* + * A full copy of a file from an earlier backup is only + * possible if no blocks are needed from any later + * incremental file. + */ + full_copy_possible = false; + } + } + } + else + { + BlockNumber b; + + /* + * Since we found a full file, source all remaining required + * blocks from it. + */ + for (b = 0; b < latest_source->truncation_block_length; ++b) + { + if (sourcemap[b] == NULL) + { + sourcemap[b] = s; + offsetmap[b] = b * BLCKSZ; + + Assert(num_missing_blocks > 0); + --num_missing_blocks; + } + } + Assert(num_missing_blocks == 0); + + /* + * If a full copy looks possible, check whether the resulting file + * should be exactly as long as the source file is. If so, a full + * copy is acceptable, otherwise not. + */ + if (full_copy_possible) + { + struct stat sb; + uint64 expected_length; + + if (fstat(s->fd, &sb) < 0) + pg_fatal("could not stat \"%s\": %m", s->filename); + expected_length = + (uint64) latest_source->truncation_block_length; + expected_length *= BLCKSZ; + if (expected_length == sb.st_size) + { + copy_source = s; + copy_source_index = sidx; + } + } + } + } + + /* + * If a checksum of the required type already exists in the + * backup_manifest for the relevant input directory, we can save some work + * by reusing that checksum instead of computing a new one. + */ + if (copy_source_index >= 0 && manifests[copy_source_index] != NULL && + checksum_type != CHECKSUM_TYPE_NONE) + { + manifest_file *mfile; + + mfile = manifest_files_lookup(manifests[copy_source_index]->files, + manifest_path); + if (mfile == NULL) + { + /* + * The directory is out of sync with the backup_manifest, so emit + * a warning. + */ + pg_log_warning("\"%s/backup_manifest\" contains no entry for \"%s\"", + prior_backup_dirs[copy_source_index], + manifest_path); + } + else if (mfile->checksum_type == checksum_type) + { + *checksum_length = mfile->checksum_length; + *checksum_payload = pg_malloc(*checksum_length); + memcpy(*checksum_payload, mfile->checksum_payload, + *checksum_length); + checksum_type = CHECKSUM_TYPE_NONE; + } + } + + /* Prepare for checksum calculation, if required. */ + pg_checksum_init(&checksum_ctx, checksum_type); + + /* + * If the full file can be created by copying a file from an older backup + * in the chain without needing to overwrite any blocks or truncate the + * result, then forget about performing reconstruction and just copy that + * file in its entirety. + * + * Otherwise, reconstruct. + */ + if (copy_source != NULL) + copy_file(copy_source->filename, output_filename, + &checksum_ctx, dry_run); + else + { + write_reconstructed_file(input_filename, output_filename, + block_length, sourcemap, offsetmap, + &checksum_ctx, dry_run); + debug_reconstruction(n_prior_backups + 1, source, dry_run); + } + + /* Save results of checksum calculation. */ + if (checksum_type != CHECKSUM_TYPE_NONE) + { + *checksum_payload = pg_malloc(PG_CHECKSUM_MAX_LENGTH); + *checksum_length = pg_checksum_final(&checksum_ctx, + *checksum_payload); + } + + /* + * Close files and release memory. + */ + for (i = 0; i <= n_prior_backups; ++i) + { + rfile *s = source[i]; + + if (s == NULL) + continue; + if (close(s->fd) != 0) + pg_fatal("could not close \"%s\": %m", s->filename); + if (s->relative_block_numbers != NULL) + pfree(s->relative_block_numbers); + pg_free(s->filename); + } + pfree(sourcemap); + pfree(offsetmap); + pfree(source); +} + +/* + * Perform post-reconstruction logging and sanity checks. + */ +static void +debug_reconstruction(int n_source, rfile **sources, bool dry_run) +{ + unsigned i; + + for (i = 0; i < n_source; ++i) + { + rfile *s = sources[i]; + + /* Ignore source if not used. */ + if (s == NULL) + continue; + + /* If no data is needed from this file, we can ignore it. */ + if (s->num_blocks_read == 0) + continue; + + /* Debug logging. */ + if (dry_run) + pg_log_debug("would have read %u blocks from \"%s\"", + s->num_blocks_read, s->filename); + else + pg_log_debug("read %u blocks from \"%s\"", + s->num_blocks_read, s->filename); + + /* + * In dry-run mode, we don't actually try to read data from the file, + * but we do try to verify that the file is long enough that we could + * have read the data if we'd tried. + * + * If this fails, then it means that a non-dry-run attempt would fail, + * complaining of not being able to read the required bytes from the + * file. + */ + if (dry_run) + { + struct stat sb; + + if (fstat(s->fd, &sb) < 0) + pg_fatal("could not stat \"%s\": %m", s->filename); + if (sb.st_size < s->highest_offset_read) + pg_fatal("file \"%s\" is too short: expected %llu, found %llu", + s->filename, + (unsigned long long) s->highest_offset_read, + (unsigned long long) sb.st_size); + } + } +} + +/* + * When we perform reconstruction using an incremental file, the output file + * should be at least as long as the truncation_block_length. Any blocks + * present in the incremental file increase the output length as far as is + * necessary to include those blocks. + */ +static unsigned +find_reconstructed_block_length(rfile *s) +{ + unsigned block_length = s->truncation_block_length; + unsigned i; + + for (i = 0; i < s->num_blocks; ++i) + if (s->relative_block_numbers[i] >= block_length) + block_length = s->relative_block_numbers[i] + 1; + + return block_length; +} + +/* + * Initialize an incremental rfile, reading the header so that we know which + * blocks it contains. + */ +static rfile * +make_incremental_rfile(char *filename) +{ + rfile *rf; + unsigned magic; + + rf = make_rfile(filename, false); + + /* Read and validate magic number. */ + read_bytes(rf, &magic, sizeof(magic)); + if (magic != INCREMENTAL_MAGIC) + pg_fatal("file \"%s\" has bad incremental magic number (0x%x not 0x%x)", + filename, magic, INCREMENTAL_MAGIC); + + /* Read block count. */ + read_bytes(rf, &rf->num_blocks, sizeof(rf->num_blocks)); + if (rf->num_blocks > RELSEG_SIZE) + pg_fatal("file \"%s\" has block count %u in excess of segment size %u", + filename, rf->num_blocks, RELSEG_SIZE); + + /* Read truncation block length. */ + read_bytes(rf, &rf->truncation_block_length, + sizeof(rf->truncation_block_length)); + if (rf->truncation_block_length > RELSEG_SIZE) + pg_fatal("file \"%s\" has truncation block length %u in excess of segment size %u", + filename, rf->truncation_block_length, RELSEG_SIZE); + + /* Read block numbers if there are any. */ + if (rf->num_blocks > 0) + { + rf->relative_block_numbers = + pg_malloc0(sizeof(BlockNumber) * rf->num_blocks); + read_bytes(rf, rf->relative_block_numbers, + sizeof(BlockNumber) * rf->num_blocks); + } + + /* Remember length of header. */ + rf->header_length = sizeof(magic) + sizeof(rf->num_blocks) + + sizeof(rf->truncation_block_length) + + sizeof(BlockNumber) * rf->num_blocks; + + return rf; +} + +/* + * Allocate and perform basic initialization of an rfile. + */ +static rfile * +make_rfile(char *filename, bool missing_ok) +{ + rfile *rf; + + rf = pg_malloc0(sizeof(rfile)); + rf->filename = pstrdup(filename); + if ((rf->fd = open(filename, O_RDONLY | PG_BINARY, 0)) < 0) + { + if (missing_ok && errno == ENOENT) + { + pg_free(rf); + return NULL; + } + pg_fatal("could not open file \"%s\": %m", filename); + } + + return rf; +} + +/* + * Read the indicated number of bytes from an rfile into the buffer. + */ +static void +read_bytes(rfile *rf, void *buffer, unsigned length) +{ + unsigned rb = read(rf->fd, buffer, length); + + if (rb != length) + { + if (rb < 0) + pg_fatal("could not read file \"%s\": %m", rf->filename); + else + pg_fatal("could not read file \"%s\": read only %d of %d bytes", + rf->filename, (int) rb, length); + } +} + +/* + * Write out a reconstructed file. + */ +static void +write_reconstructed_file(char *input_filename, + char *output_filename, + unsigned block_length, + rfile **sourcemap, + off_t *offsetmap, + pg_checksum_context *checksum_ctx, + bool dry_run) +{ + int wfd = -1; + unsigned i; + unsigned zero_blocks = 0; + + /* Debugging output. */ + if (dry_run) + pg_log_debug("would reconstruct \"%s\" (%u blocks, checksum %s)", + output_filename, block_length, + pg_checksum_type_name(checksum_ctx->type)); + else + pg_log_debug("reconstructing \"%s\" (%u blocks, checksum %s)", + output_filename, block_length, + pg_checksum_type_name(checksum_ctx->type)); + + /* Open the output file, except in dry_run mode. */ + if (!dry_run && + (wfd = open(output_filename, + O_RDWR | PG_BINARY | O_CREAT | O_EXCL, + pg_file_create_mode)) < 0) + pg_fatal("could not open file \"%s\": %m", output_filename); + + /* Read and write the blocks as required. */ + for (i = 0; i < block_length; ++i) + { + uint8 buffer[BLCKSZ]; + rfile *s = sourcemap[i]; + unsigned wb; + + /* Update accounting information. */ + if (s == NULL) + ++zero_blocks; + else + { + s->num_blocks_read++; + s->highest_offset_read = Max(s->highest_offset_read, + offsetmap[i] + BLCKSZ); + } + + /* Skip the rest of this in dry-run mode. */ + if (dry_run) + continue; + + /* Read or zero-fill the block as appropriate. */ + if (s == NULL) + { + /* + * New block not mentioned in the WAL summary. Should have been an + * uninitialized block, so just zero-fill it. + */ + memset(buffer, 0, BLCKSZ); + } + else + { + unsigned rb; + + /* Read the block from the correct source, except if dry-run. */ + rb = pg_pread(s->fd, buffer, BLCKSZ, offsetmap[i]); + if (rb != BLCKSZ) + { + if (rb < 0) + pg_fatal("could not read file \"%s\": %m", s->filename); + else + pg_fatal("could not read file \"%s\": read only %d of %d bytes at offset %u", + s->filename, (int) rb, BLCKSZ, + (unsigned) offsetmap[i]); + } + } + + /* Write out the block. */ + if ((wb = write(wfd, buffer, BLCKSZ)) != BLCKSZ) + { + if (wb < 0) + pg_fatal("could not write file \"%s\": %m", output_filename); + else + pg_fatal("could not write file \"%s\": wrote only %d of %d bytes", + output_filename, (int) wb, BLCKSZ); + } + + /* Update the checksum computation. */ + if (pg_checksum_update(checksum_ctx, buffer, BLCKSZ) < 0) + pg_fatal("could not update checksum of file \"%s\"", + output_filename); + } + + /* Debugging output. */ + if (zero_blocks > 0) + { + if (dry_run) + pg_log_debug("would have zero-filled %u blocks", zero_blocks); + else + pg_log_debug("zero-filled %u blocks", zero_blocks); + } + + /* Close the output file. */ + if (wfd >= 0 && close(wfd) != 0) + pg_fatal("could not close \"%s\": %m", output_filename); +} diff --git a/src/bin/pg_combinebackup/reconstruct.h b/src/bin/pg_combinebackup/reconstruct.h new file mode 100644 index 0000000000..c599a70d42 --- /dev/null +++ b/src/bin/pg_combinebackup/reconstruct.h @@ -0,0 +1,32 @@ +/*------------------------------------------------------------------------- + * + * reconstruct.h + * Reconstruct full file from incremental file and backup chain. + * + * Copyright (c) 2017-2023, PostgreSQL Global Development Group + * + * IDENTIFICATION + * src/bin/pg_combinebackup/reconstruct.h + * + *------------------------------------------------------------------------- + */ +#ifndef RECONSTRUCT_H +#define RECONSTRUCT_H + +#include "common/checksum_helper.h" +#include "load_manifest.h" + +extern void reconstruct_from_incremental_file(char *input_filename, + char *output_filename, + char *relative_path, + char *bare_file_name, + int n_prior_backups, + char **prior_backup_dirs, + manifest_data **manifests, + char *manifest_path, + pg_checksum_type checksum_type, + int *checksum_length, + uint8 **checksum_payload, + bool dry_run); + +#endif diff --git a/src/bin/pg_combinebackup/t/001_basic.pl b/src/bin/pg_combinebackup/t/001_basic.pl new file mode 100644 index 0000000000..fb66075d1a --- /dev/null +++ b/src/bin/pg_combinebackup/t/001_basic.pl @@ -0,0 +1,23 @@ +# Copyright (c) 2021-2023, PostgreSQL Global Development Group + +use strict; +use warnings; +use PostgreSQL::Test::Utils; +use Test::More; + +my $tempdir = PostgreSQL::Test::Utils::tempdir; + +program_help_ok('pg_combinebackup'); +program_version_ok('pg_combinebackup'); +program_options_handling_ok('pg_combinebackup'); + +command_fails_like( + ['pg_combinebackup'], + qr/no input directories specified/, + 'input directories must be specified'); +command_fails_like( + [ 'pg_combinebackup', $tempdir ], + qr/no output directory specified/, + 'output directory must be specified'); + +done_testing(); diff --git a/src/bin/pg_combinebackup/t/002_compare_backups.pl b/src/bin/pg_combinebackup/t/002_compare_backups.pl new file mode 100644 index 0000000000..3d9238f366 --- /dev/null +++ b/src/bin/pg_combinebackup/t/002_compare_backups.pl @@ -0,0 +1,154 @@ +# Copyright (c) 2021-2023, PostgreSQL Global Development Group + +use strict; +use warnings; +use File::Compare; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Set up a new database instance. +my $primary = PostgreSQL::Test::Cluster->new('primary'); +$primary->init(has_archiving => 1, allows_streaming => 1); +$primary->append_conf('postgresql.conf', 'autovacuum = off'); +$primary->start; + +# Create some test tables, each containing one row of data, plus a whole +# extra database. +$primary->safe_psql('postgres', <<EOM); +CREATE TABLE will_change (a int, b text); +INSERT INTO will_change VALUES (1, 'initial test row'); +CREATE TABLE will_grow (a int, b text); +INSERT INTO will_grow VALUES (1, 'initial test row'); +CREATE TABLE will_shrink (a int, b text); +INSERT INTO will_shrink VALUES (1, 'initial test row'); +CREATE TABLE will_get_vacuumed (a int, b text); +INSERT INTO will_get_vacuumed VALUES (1, 'initial test row'); +CREATE TABLE will_get_dropped (a int, b text); +INSERT INTO will_get_dropped VALUES (1, 'initial test row'); +CREATE TABLE will_get_rewritten (a int, b text); +INSERT INTO will_get_rewritten VALUES (1, 'initial test row'); +CREATE DATABASE db_will_get_dropped; +EOM + +# Take a full backup. +my $backup1path = $primary->backup_dir . '/backup1'; +$primary->command_ok( + [ 'pg_basebackup', '-D', $backup1path, '--no-sync', '-cfast' ], + "full backup"); + +# Now make some database changes. +$primary->safe_psql('postgres', <<EOM); +UPDATE will_change SET b = 'modified value' WHERE a = 1; +INSERT INTO will_grow + SELECT g, 'additional row' FROM generate_series(2, 5000) g; +TRUNCATE will_shrink; +VACUUM will_get_vacuumed; +DROP TABLE will_get_dropped; +CREATE TABLE newly_created (a int, b text); +INSERT INTO newly_created VALUES (1, 'row for new table'); +VACUUM FULL will_get_rewritten; +DROP DATABASE db_will_get_dropped; +CREATE DATABASE db_newly_created; +EOM + +# Take an incremental backup. +my $backup2path = $primary->backup_dir . '/backup2'; +$primary->command_ok( + [ 'pg_basebackup', '-D', $backup2path, '--no-sync', '-cfast', + '--incremental', $backup1path . '/backup_manifest' ], + "incremental backup"); + +# Find an LSN to which either backup can be recovered. +my $lsn = $primary->safe_psql('postgres', "SELECT pg_current_wal_lsn();"); + +# Make sure that the WAL segment containing that LSN has been archived. +# PostgreSQL won't issue two consecutive XLOG_SWITCH records, and the backup +# just issued one, so call txid_current() to generate some WAL activity +# before calling pg_switch_wal(). +$primary->safe_psql('postgres', 'SELECT txid_current();'); +$primary->safe_psql('postgres', 'SELECT pg_switch_wal()'); + +# Now wait for the LSN we chose above to be archived. +my $archive_wait_query = + "SELECT pg_walfile_name('$lsn') <= last_archived_wal FROM pg_stat_archiver;"; +$primary->poll_query_until('postgres', $archive_wait_query) + or die "Timed out while waiting for WAL segment to be archived"; + +# Perform PITR from the full backup. Disable archive_mode so that the archive +# doesn't find out about the new timeline; that way, the later PITR below will +# choose the same timeline. +my $pitr1 = PostgreSQL::Test::Cluster->new('pitr1'); +$pitr1->init_from_backup($primary, 'backup1', + standby => 1, has_restoring => 1); +$pitr1->append_conf('postgresql.conf', qq{ +recovery_target_lsn = '$lsn' +recovery_target_action = 'promote' +archive_mode = 'off' +}); +$pitr1->start(); + +# Perform PITR to the same LSN from the incremental backup. Use the same +# basic configuration as before. +my $pitr2 = PostgreSQL::Test::Cluster->new('pitr2'); +$pitr2->init_from_backup($primary, 'backup2', + standby => 1, has_restoring => 1, + combine_with_prior => [ 'backup1' ]); +$pitr2->append_conf('postgresql.conf', qq{ +recovery_target_lsn = '$lsn' +recovery_target_action = 'promote' +archive_mode = 'off' +}); +$pitr2->start(); + +# Wait until both servers exit recovery. +$pitr1->poll_query_until('postgres', + "SELECT NOT pg_is_in_recovery();") + or die "Timed out while waiting apply to reach LSN $lsn"; +$pitr2->poll_query_until('postgres', + "SELECT NOT pg_is_in_recovery();") + or die "Timed out while waiting apply to reach LSN $lsn"; + +# Perform a logical dump of each server, and check that they match. +# It would be much nicer if we could physically compare the data files, but +# that doesn't really work. The contents of the page hole aren't guaranteed to +# be identical, and there can be other discrepancies as well. To make this work +# we'd need the equivalent of each AM's rm_mask functon written or at least +# callable from Perl, and that doesn't seem practical. +# +# NB: We're just using the primary's backup directory for scratch space here. +# This could equally well be any other directory we wanted to pick. +my $backupdir = $primary->backup_dir; +my $dump1 = $backupdir . '/pitr1.dump'; +my $dump2 = $backupdir . '/pitr2.dump'; +$pitr1->command_ok([ + 'pg_dumpall', '-f', $dump1, '--no-sync', '--no-unlogged-table-data', + '-d', $pitr1->connstr('postgres'), + ], + 'dump from PITR 1'); +$pitr1->command_ok([ + 'pg_dumpall', '-f', $dump2, '--no-sync', '--no-unlogged-table-data', + '-d', $pitr1->connstr('postgres'), + ], + 'dump from PITR 2'); + +# Compare the two dumps, there should be no differences. +my $compare_res = compare($dump1, $dump2); +note($dump1); +note($dump2); +is($compare_res, 0, "dumps are identical"); + +# Provide more context if the dumps do not match. +if ($compare_res != 0) +{ + my ($stdout, $stderr) = + run_command([ 'diff', '-u', $dump1, $dump2 ]); + print "=== diff of $dump1 and $dump2\n"; + print "=== stdout ===\n"; + print $stdout; + print "=== stderr ===\n"; + print $stderr; + print "=== EOF ===\n"; +} + +done_testing(); diff --git a/src/bin/pg_combinebackup/write_manifest.c b/src/bin/pg_combinebackup/write_manifest.c new file mode 100644 index 0000000000..82160134d8 --- /dev/null +++ b/src/bin/pg_combinebackup/write_manifest.c @@ -0,0 +1,293 @@ +/*------------------------------------------------------------------------- + * + * Write a new backup manifest. + * + * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group + * Portions Copyright (c) 1994, Regents of the University of California + * + * src/bin/pg_combinebackup/write_manifest.c + * + *------------------------------------------------------------------------- + */ + +#include "postgres_fe.h" + +#include <fcntl.h> +#include <time.h> +#include <unistd.h> + +#include "common/checksum_helper.h" +#include "common/file_perm.h" +#include "common/logging.h" +#include "lib/stringinfo.h" +#include "load_manifest.h" +#include "mb/pg_wchar.h" +#include "write_manifest.h" + +struct manifest_writer +{ + char pathname[MAXPGPATH]; + int fd; + StringInfoData buf; + bool first_file; + bool still_checksumming; + pg_checksum_context manifest_ctx; +}; + +static void escape_json(StringInfo buf, const char *str); +static void flush_manifest(manifest_writer *mwriter); +static size_t hex_encode(const uint8 *src, size_t len, char *dst); + +/* + * Create a new backup manifest writer. + * + * The backup manifest will be written into a file named backup_manifest + * in the specified directory. + */ +manifest_writer * +create_manifest_writer(char *directory) +{ + manifest_writer *mwriter = pg_malloc(sizeof(manifest_writer)); + + snprintf(mwriter->pathname, MAXPGPATH, "%s/backup_manifest", directory); + mwriter->fd = -1; + initStringInfo(&mwriter->buf); + mwriter->first_file = true; + mwriter->still_checksumming = true; + pg_checksum_init(&mwriter->manifest_ctx, CHECKSUM_TYPE_SHA256); + + appendStringInfo(&mwriter->buf, + "{ \"PostgreSQL-Backup-Manifest-Version\": 1,\n" + "\"Files\": ["); + + return mwriter; +} + +/* + * Add an entry for a file to a backup manifest. + * + * This is very similar to the backend's AddFileToBackupManifest, but + * various adjustments are required due to frontend/backend differences + * and other details. + */ +void +add_file_to_manifest(manifest_writer *mwriter, const char *manifest_path, + size_t size, pg_time_t mtime, + pg_checksum_type checksum_type, + int checksum_length, + uint8 *checksum_payload) +{ + int pathlen = strlen(manifest_path); + + if (mwriter->first_file) + { + appendStringInfoChar(&mwriter->buf, '\n'); + mwriter->first_file = false; + } + else + appendStringInfoString(&mwriter->buf, ",\n"); + + if (pg_encoding_verifymbstr(PG_UTF8, manifest_path, pathlen) == pathlen) + { + appendStringInfoString(&mwriter->buf, "{ \"Path\": "); + escape_json(&mwriter->buf, manifest_path); + appendStringInfoString(&mwriter->buf, ", "); + } + else + { + appendStringInfoString(&mwriter->buf, "{ \"Encoded-Path\": \""); + enlargeStringInfo(&mwriter->buf, 2 * pathlen); + mwriter->buf.len += hex_encode((const uint8 *) manifest_path, pathlen, + &mwriter->buf.data[mwriter->buf.len]); + appendStringInfoString(&mwriter->buf, "\", "); + } + + appendStringInfo(&mwriter->buf, "\"Size\": %zu, ", size); + + appendStringInfoString(&mwriter->buf, "\"Last-Modified\": \""); + enlargeStringInfo(&mwriter->buf, 128); + mwriter->buf.len += strftime(&mwriter->buf.data[mwriter->buf.len], 128, + "%Y-%m-%d %H:%M:%S %Z", + gmtime(&mtime)); + appendStringInfoChar(&mwriter->buf, '"'); + + if (mwriter->buf.len > 128 * 1024) + flush_manifest(mwriter); + + if (checksum_length > 0) + { + appendStringInfo(&mwriter->buf, + ", \"Checksum-Algorithm\": \"%s\", \"Checksum\": \"", + pg_checksum_type_name(checksum_type)); + + enlargeStringInfo(&mwriter->buf, 2 * checksum_length); + mwriter->buf.len += hex_encode(checksum_payload, checksum_length, + &mwriter->buf.data[mwriter->buf.len]); + + appendStringInfoChar(&mwriter->buf, '"'); + } + + appendStringInfoString(&mwriter->buf, " }"); + + if (mwriter->buf.len > 128 * 1024) + flush_manifest(mwriter); +} + +/* + * Finalize the backup_manifest. + */ +void +finalize_manifest(manifest_writer *mwriter, + manifest_wal_range *first_wal_range) +{ + uint8 checksumbuf[PG_SHA256_DIGEST_LENGTH]; + int len; + manifest_wal_range *wal_range; + + /* Terminate the list of files. */ + appendStringInfoString(&mwriter->buf, "\n],\n"); + + /* Start a list of LSN ranges. */ + appendStringInfoString(&mwriter->buf, "\"WAL-Ranges\": [\n"); + + for (wal_range = first_wal_range; wal_range != NULL; + wal_range = wal_range->next) + appendStringInfo(&mwriter->buf, + "%s{ \"Timeline\": %u, \"Start-LSN\": \"%X/%X\", \"End-LSN\": \"%X/%X\" }", + wal_range == first_wal_range ? "" : ",\n", + wal_range->tli, + LSN_FORMAT_ARGS(wal_range->start_lsn), + LSN_FORMAT_ARGS(wal_range->end_lsn)); + + /* Terminate the list of WAL ranges. */ + appendStringInfoString(&mwriter->buf, "\n],\n"); + + /* Flush accumulated data and update checksum calculation. */ + flush_manifest(mwriter); + + /* Checksum only includes data up to this point. */ + mwriter->still_checksumming = false; + + /* Compute and insert manifest checksum. */ + appendStringInfoString(&mwriter->buf, "\"Manifest-Checksum\": \""); + enlargeStringInfo(&mwriter->buf, 2 * PG_SHA256_DIGEST_STRING_LENGTH); + len = pg_checksum_final(&mwriter->manifest_ctx, checksumbuf); + Assert(len == PG_SHA256_DIGEST_LENGTH); + mwriter->buf.len += + hex_encode(checksumbuf, len, &mwriter->buf.data[mwriter->buf.len]); + appendStringInfoString(&mwriter->buf, "\"}\n"); + + /* Flush the last manifest checksum itself. */ + flush_manifest(mwriter); + + /* Close the file. */ + if (close(mwriter->fd) != 0) + pg_fatal("could not close \"%s\": %m", mwriter->pathname); + mwriter->fd = -1; +} + +/* + * Produce a JSON string literal, properly escaping characters in the text. + */ +static void +escape_json(StringInfo buf, const char *str) +{ + const char *p; + + appendStringInfoCharMacro(buf, '"'); + for (p = str; *p; p++) + { + switch (*p) + { + case '\b': + appendStringInfoString(buf, "\\b"); + break; + case '\f': + appendStringInfoString(buf, "\\f"); + break; + case '\n': + appendStringInfoString(buf, "\\n"); + break; + case '\r': + appendStringInfoString(buf, "\\r"); + break; + case '\t': + appendStringInfoString(buf, "\\t"); + break; + case '"': + appendStringInfoString(buf, "\\\""); + break; + case '\\': + appendStringInfoString(buf, "\\\\"); + break; + default: + if ((unsigned char) *p < ' ') + appendStringInfo(buf, "\\u%04x", (int) *p); + else + appendStringInfoCharMacro(buf, *p); + break; + } + } + appendStringInfoCharMacro(buf, '"'); +} + +/* + * Flush whatever portion of the backup manifest we have generated and + * buffered in memory out to a file on disk. + * + * The first call to this function will create the file. After that, we + * keep it open and just append more data. + */ +static void +flush_manifest(manifest_writer *mwriter) +{ + char pathname[MAXPGPATH]; + + if (mwriter->fd == -1 && + (mwriter->fd = open(mwriter->pathname, + O_WRONLY | O_CREAT | O_EXCL | PG_BINARY, + pg_file_create_mode)) < 0) + pg_fatal("could not open file \"%s\": %m", mwriter->pathname); + + if (mwriter->buf.len > 0) + { + ssize_t wb; + + wb = write(mwriter->fd, mwriter->buf.data, mwriter->buf.len); + if (wb != mwriter->buf.len) + { + if (wb < 0) + pg_fatal("could not write \"%s\": %m", mwriter->pathname); + else + pg_fatal("could not write file \"%s\": wrote only %d of %d bytes", + pathname, (int) wb, mwriter->buf.len); + } + + if (mwriter->still_checksumming) + pg_checksum_update(&mwriter->manifest_ctx, + (uint8 *) mwriter->buf.data, + mwriter->buf.len); + resetStringInfo(&mwriter->buf); + } +} + +/* + * Encode bytes using two hexademical digits for each one. + */ +static size_t +hex_encode(const uint8 *src, size_t len, char *dst) +{ + const uint8 *end = src + len; + + while (src < end) + { + unsigned n1 = (*src >> 4) & 0xF; + unsigned n2 = *src & 0xF; + + *dst++ = n1 < 10 ? '0' + n1 : 'a' + n1 - 10; + *dst++ = n2 < 10 ? '0' + n2 : 'a' + n2 - 10; + ++src; + } + + return len * 2; +} diff --git a/src/bin/pg_combinebackup/write_manifest.h b/src/bin/pg_combinebackup/write_manifest.h new file mode 100644 index 0000000000..8fd7fe02c8 --- /dev/null +++ b/src/bin/pg_combinebackup/write_manifest.h @@ -0,0 +1,33 @@ +/*------------------------------------------------------------------------- + * + * Write a new backup manifest. + * + * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group + * Portions Copyright (c) 1994, Regents of the University of California + * + * src/bin/pg_combinebackup/write_manifest.h + * + *------------------------------------------------------------------------- + */ +#ifndef WRITE_MANIFEST_H +#define WRITE_MANIFEST_H + +#include "common/checksum_helper.h" +#include "pgtime.h" + +struct manifest_wal_range; + +struct manifest_writer; +typedef struct manifest_writer manifest_writer; + +extern manifest_writer *create_manifest_writer(char *directory); +extern void add_file_to_manifest(manifest_writer *mwriter, + const char *manifest_path, + size_t size, pg_time_t mtime, + pg_checksum_type checksum_type, + int checksum_length, + uint8 *checksum_payload); +extern void finalize_manifest(manifest_writer *mwriter, + struct manifest_wal_range *first_wal_range); + +#endif /* WRITE_MANIFEST_H */ diff --git a/src/bin/pg_resetwal/pg_resetwal.c b/src/bin/pg_resetwal/pg_resetwal.c index 04567f349d..c3b9e07841 100644 --- a/src/bin/pg_resetwal/pg_resetwal.c +++ b/src/bin/pg_resetwal/pg_resetwal.c @@ -85,6 +85,7 @@ static void RewriteControlFile(void); static void FindEndOfXLOG(void); static void KillExistingXLOG(void); static void KillExistingArchiveStatus(void); +static void KillExistingWALSummaries(void); static void WriteEmptyXLOG(void); static void usage(void); @@ -493,6 +494,7 @@ main(int argc, char *argv[]) RewriteControlFile(); KillExistingXLOG(); KillExistingArchiveStatus(); + KillExistingWALSummaries(); WriteEmptyXLOG(); printf(_("Write-ahead log reset\n")); @@ -1034,6 +1036,40 @@ KillExistingArchiveStatus(void) pg_fatal("could not close directory \"%s\": %m", ARCHSTATDIR); } +/* + * Remove existing WAL summary files + */ +static void +KillExistingWALSummaries(void) +{ +#define WALSUMMARYDIR XLOGDIR "/summaries" +#define WALSUMMARY_NHEXCHARS 40 + + DIR *xldir; + struct dirent *xlde; + char path[MAXPGPATH + sizeof(WALSUMMARYDIR)]; + + xldir = opendir(WALSUMMARYDIR); + if (xldir == NULL) + pg_fatal("could not open directory \"%s\": %m", WALSUMMARYDIR); + + while (errno = 0, (xlde = readdir(xldir)) != NULL) + { + if (strspn(xlde->d_name, "0123456789ABCDEF") == WALSUMMARY_NHEXCHARS && + strcmp(xlde->d_name + WALSUMMARY_NHEXCHARS, ".summary") == 0) + { + snprintf(path, sizeof(path), "%s/%s", WALSUMMARYDIR, xlde->d_name); + if (unlink(path) < 0) + pg_fatal("could not delete file \"%s\": %m", path); + } + } + + if (errno) + pg_fatal("could not read directory \"%s\": %m", WALSUMMARYDIR); + + if (closedir(xldir)) + pg_fatal("could not close directory \"%s\": %m", ARCHSTATDIR); +} /* * Write an empty XLOG file, containing only the checkpoint record diff --git a/src/common/Makefile b/src/common/Makefile index ff60666f5c..ebff20b1d3 100644 --- a/src/common/Makefile +++ b/src/common/Makefile @@ -49,6 +49,7 @@ OBJS_COMMON = \ archive.o \ base64.o \ binaryheap.o \ + blkreftable.o \ checksum_helper.o \ compression.o \ config_info.o \ diff --git a/src/common/blkreftable.c b/src/common/blkreftable.c new file mode 100644 index 0000000000..012a443584 --- /dev/null +++ b/src/common/blkreftable.c @@ -0,0 +1,1309 @@ +/*------------------------------------------------------------------------- + * + * blkreftable.c + * Block reference tables. + * + * A block reference table is used to keep track of which blocks have + * been modified by WAL records within a certain LSN range. + * + * For each relation fork, we keep track of all blocks that have appeared + * in block reference in the WAL. We also keep track of the "limit block", + * which is the smallest relation length in blocks known to have occurred + * during that range of WAL records. This should be set to 0 if the relation + * fork is created or destroyed, and to the post-truncation length if + * truncated. + * + * Whenever we set the limit block, we also forget about any modified blocks + * beyond that point. Those blocks don't exist any more. Such blocks can + * later be marked as modified again; if that happens, it means the relation + * was re-extended. + * + * Portions Copyright (c) 2010-2022, PostgreSQL Global Development Group + * + * src/common/blkreftable.c + * + *------------------------------------------------------------------------- + */ + + +#ifndef FRONTEND +#include "postgres.h" +#else +#include "postgres_fe.h" +#endif + +#ifdef FRONTEND +#include "common/logging.h" +#endif + +#include "common/blkreftable.h" +#include "common/hashfn.h" +#include "port/pg_crc32c.h" + +/* + * A block reference table keeps track of the status of each relation + * fork individually. + */ +typedef struct BlockRefTableKey +{ + RelFileLocator rlocator; + ForkNumber forknum; +} BlockRefTableKey; + +/* + * We could need to store data either for a relation in which only a + * tiny fraction of the blocks have been modified or for a relation in + * which nearly every block has been modified, and we want a + * space-efficient representation in both cases. To accomplish this, + * we divide the relation into chunks of 2^16 blocks and choose between + * an array representation and a bitmap representation for each chunk. + * + * When the number of modified blocks in a given chunk is small, we + * essentially store an array of block numbers, but we need not store the + * entire block number: instead, we store each block number as a 2-byte + * offset from the start of the chunk. + * + * When the number of modified blocks in a given chunk is large, we switch + * to a bitmap representation. + * + * These same basic representational choices are used both when a block + * reference table is stored in memory and when it is serialized to disk. + * + * In the in-memory representation, we initially allocate each chunk with + * space for a number of entries given by INITIAL_ENTRIES_PER_CHUNK and + * increase that as necessary until we reach MAX_ENTRIES_PER_CHUNK. + * Any chunk whose allocated size reaches MAX_ENTRIES_PER_CHUNK is converted + * to a bitmap, and thus never needs to grow further. + */ +#define BLOCKS_PER_CHUNK (1 << 16) +#define BLOCKS_PER_ENTRY (BITS_PER_BYTE * sizeof(uint16)) +#define MAX_ENTRIES_PER_CHUNK (BLOCKS_PER_CHUNK / BLOCKS_PER_ENTRY) +#define INITIAL_ENTRIES_PER_CHUNK 16 +typedef uint16 *BlockRefTableChunk; + +/* + * State for one relation fork. + * + * 'rlocator' and 'forknum' identify the relation fork to which this entry + * pertains. + * + * 'limit_block' is the shortest known length of the relation in blocks + * within the LSN range covered by a particular block reference table. + * It should be set to 0 if the relation fork is created or dropped. If the + * relation fork is truncated, it should be set to the number of blocks that + * remain after truncation. + * + * 'nchunks' is the allocated length of each of the three arrays that follow. + * We can only represent the status of block numbers less than nchunks * + * BLOCKS_PER_CHUNK. + * + * 'chunk_size' is an array storing the allocated size of each chunk. + * + * 'chunk_usage' is an array storing the number of elements used in each + * chunk. If that value is less than MAX_ENTRIES_PER_CHUNK, the corresonding + * chunk is used as an array; else the corresponding chunk is used as a bitmap. + * When used as a bitmap, the least significant bit of the first array element + * is the status of the lowest-numbered block covered by this chunk. + * + * 'chunk_data' is the array of chunks. + */ +struct BlockRefTableEntry +{ + BlockRefTableKey key; + BlockNumber limit_block; + char status; + uint32 nchunks; + uint16 *chunk_size; + uint16 *chunk_usage; + BlockRefTableChunk *chunk_data; +}; + +/* Declare and define a hash table over type BlockRefTableEntry. */ +#define SH_PREFIX blockreftable +#define SH_ELEMENT_TYPE BlockRefTableEntry +#define SH_KEY_TYPE BlockRefTableKey +#define SH_KEY key +#define SH_HASH_KEY(tb, key) \ + hash_bytes((const unsigned char *) &key, sizeof(BlockRefTableKey)) +#define SH_EQUAL(tb, a, b) memcmp(&a, &b, sizeof(BlockRefTableKey)) == 0 +#define SH_SCOPE static inline +#ifdef FRONTEND +#define SH_RAW_ALLOCATOR pg_malloc0 +#endif +#define SH_DEFINE +#define SH_DECLARE +#include "lib/simplehash.h" + +/* + * A block reference table is basically just the hash table, but we don't + * want to expose that to outside callers. + * + * We keep track of the memory context in use explicitly too, so that it's + * easy to place all of our allocations in the same context. + */ +struct BlockRefTable +{ + blockreftable_hash *hash; +#ifndef FRONTEND + MemoryContext mcxt; +#endif +}; + +/* + * On-disk serialization format for block reference table entries. + */ +typedef struct BlockRefTableSerializedEntry +{ + RelFileLocator rlocator; + ForkNumber forknum; + BlockNumber limit_block; + uint32 nchunks; +} BlockRefTableSerializedEntry; + +/* + * Buffer size, so that we avoid doing many small I/Os. + */ +#define BUFSIZE 65536 + +/* + * Ad-hoc buffer for file I/O. + */ +typedef struct BlockRefTableBuffer +{ + io_callback_fn io_callback; + void *io_callback_arg; + char data[BUFSIZE]; + int used; + int cursor; + pg_crc32c crc; +} BlockRefTableBuffer; + +/* + * State for keeping track of progress while incrementally reading a block + * table reference file from disk. + * + * total_chunks means the number of chunks for the RelFileLocator/ForkNumber + * combination that is curently being read, and consumed_chunks is the number + * of those that have been read. (We always read all the information for + * a single chunk at one time, so we don't need to be able to represent the + * state where a chunk has been partially read.) + * + * chunk_size is the array of chunk sizes. The length is given by total_chunks. + * + * chunk_data holds the current chunk. + * + * chunk_position helps us figure out how much progress we've made in returning + * the block numbers for the current chunk to the caller. If the chunk is a + * bitmap, it's the number of bits we've scanned; otherwise, it's the number + * of chunk entries we've scanned. + */ +struct BlockRefTableReader +{ + BlockRefTableBuffer buffer; + char *error_filename; + report_error_fn error_callback; + void *error_callback_arg; + uint32 total_chunks; + uint32 consumed_chunks; + uint16 *chunk_size; + uint16 chunk_data[MAX_ENTRIES_PER_CHUNK]; + uint32 chunk_position; +}; + +/* + * State for keeping track of progress while incrementally writing a block + * reference table file to disk. + */ +struct BlockRefTableWriter +{ + BlockRefTableBuffer buffer; +}; + +/* Function prototypes. */ +static int BlockRefTableComparator(const void *a, const void *b); +static void BlockRefTableFlush(BlockRefTableBuffer *buffer); +static void BlockRefTableRead(BlockRefTableReader *reader, void *data, + int length); +static void BlockRefTableWrite(BlockRefTableBuffer *buffer, void *data, + int length); +static void BlockRefTableFileTerminate(BlockRefTableBuffer *buffer); + +/* + * Create an empty block reference table. + */ +BlockRefTable * +CreateEmptyBlockRefTable(void) +{ + BlockRefTable *brtab = palloc(sizeof(BlockRefTable)); + + /* + * Even completely empty database has a few hundred relation forks, so it + * seems best to size the hash on the assumption that we're going to have + * at least a few thousand entries. + */ +#ifdef FRONTEND + brtab->hash = blockreftable_create(4096, NULL); +#else + brtab->mcxt = CurrentMemoryContext; + brtab->hash = blockreftable_create(brtab->mcxt, 4096, NULL); +#endif + + return brtab; +} + +/* + * Set the "limit block" for a relation fork and forget any modified blocks + * with equal or higher block numbers. + * + * The "limit block" is the shortest known length of the relation within the + * range of WAL records covered by this block reference table. + */ +void +BlockRefTableSetLimitBlock(BlockRefTable *brtab, + const RelFileLocator *rlocator, + ForkNumber forknum, + BlockNumber limit_block) +{ + BlockRefTableEntry *brtentry; + BlockRefTableKey key; + bool found; + + memcpy(&key.rlocator, rlocator, sizeof(RelFileLocator)); + key.forknum = forknum; + brtentry = blockreftable_insert(brtab->hash, key, &found); + + if (!found) + { + /* + * We have no existing data about this relation fork, so just record + * the limit_block value supplied by the caller, and make sure other + * parts of the entry are properly initialized. + */ + brtentry->limit_block = limit_block; + brtentry->nchunks = 0; + brtentry->chunk_size = NULL; + brtentry->chunk_usage = NULL; + brtentry->chunk_data = NULL; + return; + } + + BlockRefTableEntrySetLimitBlock(brtentry, limit_block); +} + +/* + * Mark a block in a given relation fork as known to have been modified. + */ +void +BlockRefTableMarkBlockModified(BlockRefTable *brtab, + const RelFileLocator *rlocator, + ForkNumber forknum, + BlockNumber blknum) +{ + BlockRefTableEntry *brtentry; + BlockRefTableKey key; + bool found; +#ifndef FRONTEND + MemoryContext oldcontext = MemoryContextSwitchTo(brtab->mcxt); +#endif + + memcpy(&key.rlocator, rlocator, sizeof(RelFileLocator)); + key.forknum = forknum; + brtentry = blockreftable_insert(brtab->hash, key, &found); + + if (!found) + { + /* + * We want to set the initial limit block value to something higher + * than any legal block number. InvalidBlockNumber fits the bill. + */ + brtentry->limit_block = InvalidBlockNumber; + brtentry->nchunks = 0; + brtentry->chunk_size = NULL; + brtentry->chunk_usage = NULL; + brtentry->chunk_data = NULL; + } + + BlockRefTableEntryMarkBlockModified(brtentry, forknum, blknum); + +#ifndef FRONTEND + MemoryContextSwitchTo(oldcontext); +#endif +} + +/* + * Get an entry from a block reference table. + * + * If the entry does not exist, this function returns NULL. Otherwise, it + * returns the entry and sets *limit_block to the value from the entry. + */ +BlockRefTableEntry * +BlockRefTableGetEntry(BlockRefTable *brtab, const RelFileLocator *rlocator, + ForkNumber forknum, BlockNumber *limit_block) +{ + BlockRefTableKey key; + BlockRefTableEntry *entry; + + Assert(limit_block != NULL); + + memcpy(&key.rlocator, rlocator, sizeof(RelFileLocator)); + key.forknum = forknum; + entry = blockreftable_lookup(brtab->hash, key); + + if (entry != NULL) + *limit_block = entry->limit_block; + + return entry; +} + +/* + * Get block numbers from a table entry. + * + * 'blocks' must point to enough space to hold at least 'nblocks' block + * numbers, and any block numbers we manage to get will be written there. + * The return value is the number of block numbers actually written. + * + * We do not return block numbers unless they are greater than or equal to + * start_blkno and strictly less than stop_blkno. + */ +int +BlockRefTableEntryGetBlocks(BlockRefTableEntry *entry, + BlockNumber start_blkno, + BlockNumber stop_blkno, + BlockNumber *blocks, + int nblocks) +{ + uint32 start_chunkno; + uint32 stop_chunkno; + uint32 chunkno; + int nresults = 0; + + Assert(entry != NULL); + + /* + * Figure out which chunks could potentially contain blocks of interest. + * + * We need to be careful about overflow here, because stop_blkno could be + * InvalidBlockNumber or something very close to it. + */ + start_chunkno = start_blkno / BLOCKS_PER_CHUNK; + stop_chunkno = stop_blkno / BLOCKS_PER_CHUNK; + if ((stop_blkno % BLOCKS_PER_CHUNK) != 0) + ++stop_chunkno; + if (stop_chunkno > entry->nchunks) + stop_chunkno = entry->nchunks; + + /* + * Loop over chunks. + */ + for (chunkno = start_chunkno; chunkno < stop_chunkno; ++chunkno) + { + uint16 chunk_usage = entry->chunk_usage[chunkno]; + BlockRefTableChunk chunk_data = entry->chunk_data[chunkno]; + unsigned start_offset = 0; + unsigned stop_offset = BLOCKS_PER_CHUNK; + + /* + * If the start and/or stop block number falls within this chunk, the + * whole chunk may not be of interest. Figure out which portion we + * care about, if it's not the whole thing. + */ + if (chunkno == start_chunkno) + start_offset = start_blkno % BLOCKS_PER_CHUNK; + if (chunkno == stop_chunkno) + stop_offset = stop_blkno % BLOCKS_PER_CHUNK; + + /* + * Handling differs depending on whether this is an array of offsets + * or a bitmap. + */ + if (chunk_usage == MAX_ENTRIES_PER_CHUNK) + { + unsigned i; + + /* It's a bitmap, so test every relevant bit. */ + for (i = start_offset; i < BLOCKS_PER_CHUNK; ++i) + { + uint16 w = chunk_data[i / BLOCKS_PER_ENTRY]; + + if ((w & (1 << (i % BLOCKS_PER_ENTRY))) != 0) + { + BlockNumber blkno = chunkno * BLOCKS_PER_CHUNK + i; + + blocks[nresults++] = blkno; + + /* Early exit if we run out of output space. */ + if (nresults == nblocks) + return nresults; + } + } + } + else + { + unsigned i; + + /* It's an array of offsets, so check each one. */ + for (i = 0; i < chunk_usage; ++i) + { + uint16 offset = chunk_data[i]; + + if (offset >= start_offset && offset < stop_offset) + { + BlockNumber blkno = chunkno * BLOCKS_PER_CHUNK + offset; + + blocks[nresults++] = blkno; + + /* Early exit if we run out of output space. */ + if (nresults == nblocks) + return nresults; + } + } + } + } + + return nresults; +} + +/* + * Serialize a block reference table to a file. + */ +void +WriteBlockRefTable(BlockRefTable *brtab, + io_callback_fn write_callback, + void *write_callback_arg) +{ + BlockRefTableSerializedEntry *sdata = NULL; + BlockRefTableBuffer buffer; + uint32 magic = BLOCKREFTABLE_MAGIC; + + /* Prepare buffer. */ + memset(&buffer, 0, sizeof(BlockRefTableBuffer)); + buffer.io_callback = write_callback; + buffer.io_callback_arg = write_callback_arg; + INIT_CRC32C(buffer.crc); + + /* Write magic number. */ + BlockRefTableWrite(&buffer, &magic, sizeof(uint32)); + + /* Write the entries, assuming there are some. */ + if (brtab->hash->members > 0) + { + unsigned i = 0; + blockreftable_iterator it; + BlockRefTableEntry *brtentry; + + /* Extract entries into serializable format and sort them. */ + sdata = + palloc(brtab->hash->members * sizeof(BlockRefTableSerializedEntry)); + blockreftable_start_iterate(brtab->hash, &it); + while ((brtentry = blockreftable_iterate(brtab->hash, &it)) != NULL) + { + BlockRefTableSerializedEntry *sentry = &sdata[i++]; + + sentry->rlocator = brtentry->key.rlocator; + sentry->forknum = brtentry->key.forknum; + sentry->limit_block = brtentry->limit_block; + sentry->nchunks = brtentry->nchunks; + + /* trim trailing zero entries */ + while (sentry->nchunks > 0 && + brtentry->chunk_usage[sentry->nchunks - 1] == 0) + sentry->nchunks--; + } + Assert(i == brtab->hash->members); + qsort(sdata, i, sizeof(BlockRefTableSerializedEntry), + BlockRefTableComparator); + + /* Loop over entries in sorted order and serialize each one. */ + for (i = 0; i < brtab->hash->members; ++i) + { + BlockRefTableSerializedEntry *sentry = &sdata[i]; + BlockRefTableEntry *brtentry; + BlockRefTableKey key; + unsigned j; + + /* Write the serialized entry itself. */ + BlockRefTableWrite(&buffer, sentry, + sizeof(BlockRefTableSerializedEntry)); + + /* Look up the original entry so we can access the chunks. */ + memcpy(&key.rlocator, &sentry->rlocator, sizeof(RelFileLocator)); + key.forknum = sentry->forknum; + brtentry = blockreftable_lookup(brtab->hash, key); + Assert(brtentry != NULL); + + /* Write the untruncated portion of the chunk length array. */ + if (sentry->nchunks != 0) + BlockRefTableWrite(&buffer, brtentry->chunk_usage, + sentry->nchunks * sizeof(uint16)); + + /* Write the contents of each chunk. */ + for (j = 0; j < brtentry->nchunks; ++j) + { + if (brtentry->chunk_usage[j] == 0) + continue; + BlockRefTableWrite(&buffer, brtentry->chunk_data[j], + brtentry->chunk_usage[j] * sizeof(uint16)); + } + } + } + + /* Write out appropriate terminator and CRC and flush buffer. */ + BlockRefTableFileTerminate(&buffer); +} + +/* + * Prepare to incrementally read a block reference table file. + * + * 'read_callback' is a function that can be called to read data from the + * underlying file (or other data source) into our internal buffer. + * + * 'read_callback_arg' is an opaque argument to be passed to read_callback. + * + * 'error_filename' is the filename that should be included in error messages + * if the file is found to be malformed. The value is not copied, so the + * caller should ensure that it remains valid until done with this + * BlockRefTableReader. + * + * 'error_callback' is a function to be called if the file is found to be + * malformed. This is not used for I/O errors, which must be handled internally + * by read_callback. + * + * 'error_callback_arg' is an opaque arguent to be passed to error_callback. + */ +BlockRefTableReader * +CreateBlockRefTableReader(io_callback_fn read_callback, + void *read_callback_arg, + char *error_filename, + report_error_fn error_callback, + void *error_callback_arg) +{ + BlockRefTableReader *reader; + uint32 magic; + + /* Initialize data structure. */ + reader = palloc0(sizeof(BlockRefTableReader)); + reader->buffer.io_callback = read_callback; + reader->buffer.io_callback_arg = read_callback_arg; + reader->error_filename = error_filename; + reader->error_callback = error_callback; + reader->error_callback_arg = error_callback_arg; + INIT_CRC32C(reader->buffer.crc); + + /* Verify magic number. */ + BlockRefTableRead(reader, &magic, sizeof(uint32)); + if (magic != BLOCKREFTABLE_MAGIC) + error_callback(error_callback_arg, + "file \"%s\" has wrong magic number: expected %u, found %u", + error_filename, + BLOCKREFTABLE_MAGIC, magic); + + return reader; +} + +/* + * Read next relation fork covered by this block reference table file. + * + * After calling this function, you must call BlockRefTableReaderGetBlocks + * until it returns 0 before calling it again. + */ +bool +BlockRefTableReaderNextRelation(BlockRefTableReader *reader, + RelFileLocator *rlocator, + ForkNumber *forknum, + BlockNumber *limit_block) +{ + BlockRefTableSerializedEntry sentry; + BlockRefTableSerializedEntry zentry = {0}; + + /* + * Sanity check: caller must read all blocks from all chunks before moving + * on to the next relation. + */ + Assert(reader->total_chunks == reader->consumed_chunks); + + /* Read serialized entry. */ + BlockRefTableRead(reader, &sentry, + sizeof(BlockRefTableSerializedEntry)); + + /* + * If we just read the sentinel entry indicating that we've reached the + * end, read and check the CRC. + */ + if (memcmp(&sentry, &zentry, sizeof(BlockRefTableSerializedEntry)) == 0) + { + pg_crc32c expected_crc; + pg_crc32c actual_crc; + + /* + * We want to know the CRC of the file excluding the 4-byte CRC + * itself, so copy the current value of the CRC accumulator before + * reading those bytes, and use the copy to finalize the calculation. + */ + expected_crc = reader->buffer.crc; + FIN_CRC32C(expected_crc); + + /* Now we can read the actual value. */ + BlockRefTableRead(reader, &actual_crc, sizeof(pg_crc32c)); + + /* Throw an error if there is a mismatch. */ + if (!EQ_CRC32C(expected_crc, actual_crc)) + reader->error_callback(reader->error_callback_arg, + "file \"%s\" has wrong checksum: expected %08X, found %08X", + reader->error_filename, expected_crc, actual_crc); + + return false; + } + + /* Read chunk size array. */ + if (reader->chunk_size != NULL) + pfree(reader->chunk_size); + reader->chunk_size = palloc(sentry.nchunks * sizeof(uint16)); + BlockRefTableRead(reader, reader->chunk_size, + sentry.nchunks * sizeof(uint16)); + + /* Set up for chunk scan. */ + reader->total_chunks = sentry.nchunks; + reader->consumed_chunks = 0; + + /* Return data to caller. */ + memcpy(rlocator, &sentry.rlocator, sizeof(RelFileLocator)); + *forknum = sentry.forknum; + *limit_block = sentry.limit_block; + return true; +} + +/* + * Get modified blocks associated with the relation fork returned by + * the most recent call to BlockRefTableReaderNextRelation. + * + * On return, block numbers will be written into the 'blocks' array, whose + * length should be passed via 'nblocks'. The return value is the number of + * entries actually written into the 'blocks' array, which may be less than + * 'nblocks' if we run out of modified blocks in the relation fork before + * we run out of room in the array. + */ +unsigned +BlockRefTableReaderGetBlocks(BlockRefTableReader *reader, + BlockNumber *blocks, + int nblocks) +{ + unsigned blocks_found = 0; + + /* Must provide space for at least one block number to be returned. */ + Assert(nblocks > 0); + + /* Loop collecting blocks to return to caller. */ + for (;;) + { + uint16 next_chunk_size; + + /* + * If we've read at least one chunk, maybe it contains some block + * numbers that could satisfy caller's request. + */ + if (reader->consumed_chunks > 0) + { + uint32 chunkno = reader->consumed_chunks - 1; + uint16 chunk_size = reader->chunk_size[chunkno]; + + if (chunk_size == MAX_ENTRIES_PER_CHUNK) + { + /* Bitmap format, so search for bits that are set. */ + while (reader->chunk_position < BLOCKS_PER_CHUNK && + blocks_found < nblocks) + { + uint16 chunkoffset = reader->chunk_position; + uint16 w; + + w = reader->chunk_data[chunkoffset / BLOCKS_PER_ENTRY]; + if ((w & (1u << (chunkoffset % BLOCKS_PER_ENTRY))) != 0) + blocks[blocks_found++] = + chunkno * BLOCKS_PER_CHUNK + chunkoffset; + ++reader->chunk_position; + } + } + else + { + /* Not in bitmap format, so each entry is a 2-byte offset. */ + while (reader->chunk_position < chunk_size && + blocks_found < nblocks) + { + blocks[blocks_found++] = chunkno * BLOCKS_PER_CHUNK + + reader->chunk_data[reader->chunk_position]; + ++reader->chunk_position; + } + } + } + + /* We found enough blocks, so we're done. */ + if (blocks_found >= nblocks) + break; + + /* + * We didn't find enough blocks, so we must need the next chunk. If + * there are none left, though, then we're done anyway. + */ + if (reader->consumed_chunks == reader->total_chunks) + break; + + /* + * Read data for next chunk and reset scan position to beginning of + * chunk. Note that the next chunk might be empty, in which case we + * consume the chunk without actually consuming any bytes from the + * underlying file. + */ + next_chunk_size = reader->chunk_size[reader->consumed_chunks]; + if (next_chunk_size > 0) + BlockRefTableRead(reader, reader->chunk_data, + next_chunk_size * sizeof(uint16)); + ++reader->consumed_chunks; + reader->chunk_position = 0; + } + + return blocks_found; +} + +/* + * Release memory used while reading a block reference table from a file. + */ +void +DestroyBlockRefTableReader(BlockRefTableReader *reader) +{ + if (reader->chunk_size != NULL) + { + pfree(reader->chunk_size); + reader->chunk_size = NULL; + } + pfree(reader); +} + +/* + * Prepare to write a block reference table file incrementally. + * + * Caller must be able to supply BlockRefTableEntry objects sorted in the + * appropriate order. + */ +BlockRefTableWriter * +CreateBlockRefTableWriter(io_callback_fn write_callback, + void *write_callback_arg) +{ + BlockRefTableWriter *writer; + uint32 magic = BLOCKREFTABLE_MAGIC; + + /* Prepare buffer and CRC check and save callbacks. */ + writer = palloc0(sizeof(BlockRefTableWriter)); + writer->buffer.io_callback = write_callback; + writer->buffer.io_callback_arg = write_callback_arg; + INIT_CRC32C(writer->buffer.crc); + + /* Write magic number. */ + BlockRefTableWrite(&writer->buffer, &magic, sizeof(uint32)); + + return writer; +} + +/* + * Append one entry to a block reference table file. + * + * Note that entries must be written in the proper order, that is, sorted by + * tablespace, then database, then relfilenumber, then fork number. Caller + * is responsible for supplying data in the correct order. If that seems hard, + * use an in-memory BlockRefTable instead. + */ +void +BlockRefTableWriteEntry(BlockRefTableWriter *writer, BlockRefTableEntry *entry) +{ + BlockRefTableSerializedEntry sentry; + unsigned j; + + /* Convert to serialized entry format. */ + sentry.rlocator = entry->key.rlocator; + sentry.forknum = entry->key.forknum; + sentry.limit_block = entry->limit_block; + sentry.nchunks = entry->nchunks; + + /* Trim trailing zero entries. */ + while (sentry.nchunks > 0 && entry->chunk_usage[sentry.nchunks - 1] == 0) + sentry.nchunks--; + + /* Write the serialized entry itself. */ + BlockRefTableWrite(&writer->buffer, &sentry, + sizeof(BlockRefTableSerializedEntry)); + + /* Write the untruncated portion of the chunk length array. */ + if (sentry.nchunks != 0) + BlockRefTableWrite(&writer->buffer, entry->chunk_usage, + sentry.nchunks * sizeof(uint16)); + + /* Write the contents of each chunk. */ + for (j = 0; j < entry->nchunks; ++j) + { + if (entry->chunk_usage[j] == 0) + continue; + BlockRefTableWrite(&writer->buffer, entry->chunk_data[j], + entry->chunk_usage[j] * sizeof(uint16)); + } +} + +/* + * Finalize an incremental write of a block reference table file. + */ +void +DestroyBlockRefTableWriter(BlockRefTableWriter *writer) +{ + BlockRefTableFileTerminate(&writer->buffer); + pfree(writer); +} + +/* + * Allocate a standalone BlockRefTableEntry. + * + * When we're manipulating a full in-memory BlockRefTable, the entries are + * part of the hash table and are allocated by simplehash. This routine is + * used by callers that want to write out a BlockRefTable to a file without + * needing to store the whole thing in memory at once. + * + * Entries allocated by this function can be manipulated using the functions + * BlockRefTableEntrySetLimitBlock and BlockRefTableEntryMarkBlockModified + * and then written using BlockRefTableWriteEntry and freed using + * BlockRefTableFreeEntry. + */ +BlockRefTableEntry * +CreateBlockRefTableEntry(RelFileLocator rlocator, ForkNumber forknum) +{ + BlockRefTableEntry *entry = palloc0(sizeof(BlockRefTableEntry)); + + memcpy(&entry->key.rlocator, &rlocator, sizeof(RelFileLocator)); + entry->key.forknum = forknum; + entry->limit_block = InvalidBlockNumber; + + return entry; +} + +/* + * Update a BlockRefTableEntry with a new value for the "limit block" and + * forget any equal-or-higher-numbered modified blocks. + * + * The "limit block" is the shortest known length of the relation within the + * range of WAL records covered by this block reference table. + */ +void +BlockRefTableEntrySetLimitBlock(BlockRefTableEntry *entry, + BlockNumber limit_block) +{ + unsigned chunkno; + unsigned limit_chunkno; + unsigned limit_chunkoffset; + BlockRefTableChunk limit_chunk; + + /* If we already have an equal or lower limit block, do nothing. */ + if (limit_block >= entry->limit_block) + return; + + /* Record the new limit block value. */ + entry->limit_block = limit_block; + + /* + * Figure out which chunk would store the state of the new limit block, + * and which offset within that chunk. + */ + limit_chunkno = limit_block / BLOCKS_PER_CHUNK; + limit_chunkoffset = limit_block % BLOCKS_PER_CHUNK; + + /* + * If the number of chunks is not large enough for any blocks with equal + * or higher block numbers to exist, then there is nothing further to do. + */ + if (limit_chunkno >= entry->nchunks) + return; + + /* Discard entire contents of any higher-numbered chunks. */ + for (chunkno = limit_chunkno + 1; chunkno < entry->nchunks; ++chunkno) + entry->chunk_usage[chunkno] = 0; + + /* + * Next, we need to discard any offsets within the chunk that would + * contain the limit_block. We must handle this differenly depending on + * whether the chunk that would contain limit_block is a bitmap or an + * array of offsets. + */ + limit_chunk = entry->chunk_data[limit_chunkno]; + if (entry->chunk_usage[limit_chunkno] == MAX_ENTRIES_PER_CHUNK) + { + unsigned chunkoffset; + + /* It's a bitmap. Unset bits. */ + for (chunkoffset = limit_chunkoffset; chunkoffset < BLOCKS_PER_CHUNK; + ++chunkoffset) + limit_chunk[chunkoffset / BLOCKS_PER_ENTRY] &= + ~(1 << (chunkoffset % BLOCKS_PER_ENTRY)); + } + else + { + unsigned i, + j = 0; + + /* It's an offset array. Filter out large offsets. */ + for (i = 0; i < entry->chunk_usage[limit_chunkno]; ++i) + { + Assert(j <= i); + if (limit_chunk[i] < limit_chunkoffset) + limit_chunk[j++] = limit_chunk[i]; + } + Assert(j <= entry->chunk_usage[limit_chunkno]); + entry->chunk_usage[limit_chunkno] = j; + } +} + +/* + * Mark a block in a given BlkRefTableEntry as known to have been modified. + */ +void +BlockRefTableEntryMarkBlockModified(BlockRefTableEntry *entry, + ForkNumber forknum, + BlockNumber blknum) +{ + unsigned chunkno; + unsigned chunkoffset; + unsigned i; + + /* + * Which chunk should store the state of this block? And what is the + * offset of this block relative to the start of that chunk? + */ + chunkno = blknum / BLOCKS_PER_CHUNK; + chunkoffset = blknum % BLOCKS_PER_CHUNK; + + /* + * If 'nchunks' isn't big enough for us to be able to represent the state + * of this block, we need to enlarge our arrays. + */ + if (chunkno >= entry->nchunks) + { + unsigned max_chunks; + unsigned extra_chunks; + + /* + * New array size is a power of 2, at least 16, big enough so that + * chunkno will be a valid array index. + */ + max_chunks = Max(16, entry->nchunks); + while (max_chunks < chunkno + 1) + chunkno *= 2; + Assert(max_chunks > chunkno); + extra_chunks = max_chunks - entry->nchunks; + + if (entry->nchunks == 0) + { + entry->chunk_size = palloc0(sizeof(uint16) * max_chunks); + entry->chunk_usage = palloc0(sizeof(uint16) * max_chunks); + entry->chunk_data = + palloc0(sizeof(BlockRefTableChunk) * max_chunks); + } + else + { + entry->chunk_size = repalloc(entry->chunk_size, + sizeof(uint16) * max_chunks); + memset(&entry->chunk_size[entry->nchunks], 0, + extra_chunks * sizeof(uint16)); + entry->chunk_usage = repalloc(entry->chunk_usage, + sizeof(uint16) * max_chunks); + memset(&entry->chunk_usage[entry->nchunks], 0, + extra_chunks * sizeof(uint16)); + entry->chunk_data = repalloc(entry->chunk_data, + sizeof(BlockRefTableChunk) * max_chunks); + memset(&entry->chunk_data[entry->nchunks], 0, + extra_chunks * sizeof(BlockRefTableChunk)); + } + entry->nchunks = max_chunks; + } + + /* + * If the chunk that covers this block number doesn't exist yet, create it + * as an array and add the appropriate offset to it. We make it pretty + * small initially, because there might only be 1 or a few block + * references in this chunk and we don't want to use up too much memory. + */ + if (entry->chunk_size[chunkno] == 0) + { + entry->chunk_data[chunkno] = + palloc(sizeof(uint16) * INITIAL_ENTRIES_PER_CHUNK); + entry->chunk_size[chunkno] = INITIAL_ENTRIES_PER_CHUNK; + entry->chunk_data[chunkno][0] = chunkoffset; + entry->chunk_usage[chunkno] = 1; + return; + } + + /* + * If the number of entries in this chunk is already maximum, it must be a + * bitmap. Just set the appropriate bit. + */ + if (entry->chunk_usage[chunkno] == MAX_ENTRIES_PER_CHUNK) + { + BlockRefTableChunk chunk = entry->chunk_data[chunkno]; + + chunk[chunkoffset / BLOCKS_PER_ENTRY] |= + 1 << (chunkoffset % BLOCKS_PER_ENTRY); + return; + } + + /* + * There is an existing chunk and it's in array format. Let's find out + * whether it already has an entry for this block. If so, we do not need + * to do anything. + */ + for (i = 0; i < entry->chunk_usage[chunkno]; ++i) + { + if (entry->chunk_data[chunkno][i] == chunkoffset) + return; + } + + /* + * If the number of entries currently used is one less than the maximum, + * it's time to convert to bitmap format. + */ + if (entry->chunk_usage[chunkno] == MAX_ENTRIES_PER_CHUNK - 1) + { + BlockRefTableChunk newchunk; + unsigned j; + + /* Allocate a new chunk. */ + newchunk = palloc0(MAX_ENTRIES_PER_CHUNK * sizeof(uint16)); + + /* Set the bit for each existing entry. */ + for (j = 0; j < entry->chunk_usage[chunkno]; ++j) + { + unsigned coff = entry->chunk_data[chunkno][j]; + + newchunk[coff / BLOCKS_PER_ENTRY] |= + 1 << (coff % BLOCKS_PER_ENTRY); + } + + /* Set the bit for the new entry. */ + newchunk[chunkoffset / BLOCKS_PER_ENTRY] |= + 1 << (chunkoffset % BLOCKS_PER_ENTRY); + + /* Swap the new chunk into place and update metadata. */ + pfree(entry->chunk_data[chunkno]); + entry->chunk_data[chunkno] = newchunk; + entry->chunk_size[chunkno] = MAX_ENTRIES_PER_CHUNK; + entry->chunk_usage[chunkno] = MAX_ENTRIES_PER_CHUNK; + return; + } + + /* + * OK, we currently have an array, and we don't need to convert to a + * bitmap, but we do need to add a new element. If there's not enough + * room, we'll have to expand the array. + */ + if (entry->chunk_usage[chunkno] == entry->chunk_size[chunkno]) + { + unsigned newsize = entry->chunk_size[chunkno] * 2; + + Assert(newsize <= MAX_ENTRIES_PER_CHUNK); + entry->chunk_data[chunkno] = repalloc(entry->chunk_data[chunkno], + newsize * sizeof(uint16)); + entry->chunk_size[chunkno] = newsize; + } + + /* Now we can add the new entry. */ + entry->chunk_data[chunkno][entry->chunk_usage[chunkno]] = + chunkoffset; + entry->chunk_usage[chunkno]++; +} + +/* + * Release memory for a BlockRefTablEntry that was created by + * CreateBlockRefTableEntry. + */ +void +BlockRefTableFreeEntry(BlockRefTableEntry *entry) +{ + if (entry->chunk_size != NULL) + { + pfree(entry->chunk_size); + entry->chunk_size = NULL; + } + + if (entry->chunk_usage != NULL) + { + pfree(entry->chunk_usage); + entry->chunk_usage = NULL; + } + + if (entry->chunk_data != NULL) + { + pfree(entry->chunk_data); + entry->chunk_data = NULL; + } + + pfree(entry); +} + +/* + * Comparator for BlockRefTableSerializedEntry objects. + * + * We make the tablespace OID the first column of the sort key to match + * the on-disk tree structure. + */ +static int +BlockRefTableComparator(const void *a, const void *b) +{ + const BlockRefTableSerializedEntry *sa = a; + const BlockRefTableSerializedEntry *sb = b; + + if (sa->rlocator.spcOid > sb->rlocator.spcOid) + return 1; + if (sa->rlocator.spcOid < sb->rlocator.spcOid) + return -1; + + if (sa->rlocator.dbOid > sb->rlocator.dbOid) + return 1; + if (sa->rlocator.dbOid < sb->rlocator.dbOid) + return -1; + + if (sa->rlocator.relNumber > sb->rlocator.relNumber) + return 1; + if (sa->rlocator.relNumber < sb->rlocator.relNumber) + return -1; + + if (sa->forknum > sb->forknum) + return 1; + if (sa->forknum < sb->forknum) + return -1; + + return 0; +} + +/* + * Flush any buffered data out of a BlockRefTableBuffer. + */ +static void +BlockRefTableFlush(BlockRefTableBuffer *buffer) +{ + buffer->io_callback(buffer->io_callback_arg, buffer->data, buffer->used); + buffer->used = 0; +} + +/* + * Read data from a BlockRefTableBuffer, and update the running CRC + * calculation for the returned data (but not any data that we may have + * buffered but not yet actually returned). + */ +static void +BlockRefTableRead(BlockRefTableReader *reader, void *data, int length) +{ + BlockRefTableBuffer *buffer = &reader->buffer; + + /* Loop until read is fully satisfied. */ + while (length > 0) + { + if (buffer->cursor < buffer->used) + { + /* + * If any buffered data is available, use that to satisfy as much + * of the request as possible. + */ + int bytes_to_copy = Min(length, buffer->used - buffer->cursor); + + memcpy(data, &buffer->data[buffer->cursor], bytes_to_copy); + COMP_CRC32C(buffer->crc, &buffer->data[buffer->cursor], + bytes_to_copy); + buffer->cursor += bytes_to_copy; + data = ((char *) data) + bytes_to_copy; + length -= bytes_to_copy; + } + else if (length >= BUFSIZE) + { + /* + * If the request length is long, read directly into caller's + * buffer. + */ + int bytes_read; + + bytes_read = buffer->io_callback(buffer->io_callback_arg, + data, length); + COMP_CRC32C(buffer->crc, data, bytes_read); + data = ((char *) data) + bytes_read; + length -= bytes_read; + + /* If we didn't get anything, that's bad. */ + if (bytes_read == 0) + reader->error_callback(reader->error_callback_arg, + "file \"%s\" ends unexpectedly", + reader->error_filename); + } + else + { + /* + * Refill our buffer. + */ + buffer->used = buffer->io_callback(buffer->io_callback_arg, + buffer->data, BUFSIZE); + buffer->cursor = 0; + + /* If we didn't get anything, that's bad. */ + if (buffer->used == 0) + reader->error_callback(reader->error_callback_arg, + "file \"%s\" ends unexpectedly", + reader->error_filename); + } + } +} + +/* + * Supply data to a BlockRefTableBuffer for write to the underlying File, + * and update the running CRC calculation for that data. + */ +static void +BlockRefTableWrite(BlockRefTableBuffer *buffer, void *data, int length) +{ + /* Update running CRC calculation. */ + COMP_CRC32C(buffer->crc, data, length); + + /* If the new data can't fit into the buffer, flush the buffer. */ + if (buffer->used + length > BUFSIZE) + { + buffer->io_callback(buffer->io_callback_arg, buffer->data, + buffer->used); + buffer->used = 0; + } + + /* If the new data would fill the buffer, or more, write it directly. */ + if (length >= BUFSIZE) + { + buffer->io_callback(buffer->io_callback_arg, data, length); + return; + } + + /* Otherwise, copy the new data into the buffer. */ + memcpy(&buffer->data[buffer->used], data, length); + buffer->used += length; + Assert(buffer->used <= BUFSIZE); +} + +/* + * Generate the sentinel and CRC required at the end of a block reference + * table file and flush them out of our internal buffer. + */ +static void +BlockRefTableFileTerminate(BlockRefTableBuffer *buffer) +{ + BlockRefTableSerializedEntry zentry = {0}; + pg_crc32c crc; + + /* Write a sentinel indicating that there are no more entries. */ + BlockRefTableWrite(buffer, &zentry, + sizeof(BlockRefTableSerializedEntry)); + + /* + * Writing the checksum will perturb the ongoing checksum calculation, so + * copy the state first and finalize the computation using the copy. + */ + crc = buffer->crc; + FIN_CRC32C(crc); + BlockRefTableWrite(buffer, &crc, sizeof(pg_crc32c)); + + /* Flush any leftover data out of our buffer. */ + BlockRefTableFlush(buffer); +} diff --git a/src/common/meson.build b/src/common/meson.build index fcc0c4fe8d..6e51257b1c 100644 --- a/src/common/meson.build +++ b/src/common/meson.build @@ -4,6 +4,7 @@ common_sources = files( 'archive.c', 'base64.c', 'binaryheap.c', + 'blkreftable.c', 'checksum_helper.c', 'compression.c', 'controldata_utils.c', diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 48ca852381..fed5d790cc 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -206,6 +206,7 @@ extern int XLogFileOpen(XLogSegNo segno, TimeLineID tli); extern void CheckXLogRemoved(XLogSegNo segno, TimeLineID tli); extern XLogSegNo XLogGetLastRemovedSegno(void); +extern XLogSegNo XLogGetOldestSegno(TimeLineID tli); extern void XLogSetAsyncXactLSN(XLogRecPtr asyncXactLSN); extern void XLogSetReplicationSlotMinimumLSN(XLogRecPtr lsn); diff --git a/src/include/access/xlogbackup.h b/src/include/access/xlogbackup.h index 1611358137..90e04cad56 100644 --- a/src/include/access/xlogbackup.h +++ b/src/include/access/xlogbackup.h @@ -28,6 +28,8 @@ typedef struct BackupState XLogRecPtr checkpointloc; /* last checkpoint location */ pg_time_t starttime; /* backup start time */ bool started_in_recovery; /* backup started in recovery? */ + XLogRecPtr istartpoint; /* incremental based on backup at this LSN */ + TimeLineID istarttli; /* incremental based on backup on this TLI */ /* Fields saved at the end of backup */ XLogRecPtr stoppoint; /* backup stop WAL location */ diff --git a/src/include/backup/basebackup.h b/src/include/backup/basebackup.h index 1432d9c206..345bd22534 100644 --- a/src/include/backup/basebackup.h +++ b/src/include/backup/basebackup.h @@ -34,6 +34,9 @@ typedef struct int64 size; /* total size as sent; -1 if not known */ } tablespaceinfo; -extern void SendBaseBackup(BaseBackupCmd *cmd); +struct IncrementalBackupInfo; + +extern void SendBaseBackup(BaseBackupCmd *cmd, + struct IncrementalBackupInfo *ib); #endif /* _BASEBACKUP_H */ diff --git a/src/include/backup/basebackup_incremental.h b/src/include/backup/basebackup_incremental.h new file mode 100644 index 0000000000..c300235a2f --- /dev/null +++ b/src/include/backup/basebackup_incremental.h @@ -0,0 +1,56 @@ +/*------------------------------------------------------------------------- + * + * basebackup_incremental.h + * API for incremental backup support + * + * Portions Copyright (c) 2010-2022, PostgreSQL Global Development Group + * + * src/include/backup/basebackup_incremental.h + * + *------------------------------------------------------------------------- + */ +#ifndef BASEBACKUP_INCREMENTAL_H +#define BASEBACKUP_INCREMENTAL_H + +#include "access/xlogbackup.h" +#include "common/relpath.h" +#include "storage/block.h" +#include "utils/palloc.h" + +#define INCREMENTAL_MAGIC 0xd3ae1f0d + +typedef enum +{ + BACK_UP_FILE_FULLY, + BACK_UP_FILE_INCREMENTALLY, + DO_NOT_BACK_UP_FILE +} FileBackupMethod; + +struct IncrementalBackupInfo; +typedef struct IncrementalBackupInfo IncrementalBackupInfo; + +extern IncrementalBackupInfo *CreateIncrementalBackupInfo(MemoryContext); + +extern void AppendIncrementalManifestData(IncrementalBackupInfo *ib, + const char *data, + int len); +extern void FinalizeIncrementalManifest(IncrementalBackupInfo *ib); + +extern void PrepareForIncrementalBackup(IncrementalBackupInfo *ib, + BackupState *backup_state); + +extern char *GetIncrementalFilePath(Oid dboid, Oid spcoid, + RelFileNumber relfilenumber, + ForkNumber forknum, unsigned segno); +extern FileBackupMethod GetFileBackupMethod(IncrementalBackupInfo *ib, + char *path, + Oid dboid, Oid spcoid, + RelFileNumber relfilenumber, + ForkNumber forknum, + unsigned segno, size_t size, + unsigned *num_blocks_required, + BlockNumber *relative_block_numbers, + unsigned *truncation_block_length); +extern size_t GetIncrementalFileSize(unsigned num_blocks_required); + +#endif diff --git a/src/include/backup/walsummary.h b/src/include/backup/walsummary.h new file mode 100644 index 0000000000..d086e64019 --- /dev/null +++ b/src/include/backup/walsummary.h @@ -0,0 +1,49 @@ +/*------------------------------------------------------------------------- + * + * walsummary.h + * WAL summary management + * + * Portions Copyright (c) 2010-2022, PostgreSQL Global Development Group + * + * src/include/backup/walsummary.h + * + *------------------------------------------------------------------------- + */ +#ifndef WALSUMMARY_H +#define WALSUMMARY_H + +#include <time.h> + +#include "access/xlogdefs.h" +#include "nodes/pg_list.h" +#include "storage/fd.h" + +typedef struct WalSummaryIO +{ + File file; + off_t filepos; +} WalSummaryIO; + +typedef struct WalSummaryFile +{ + XLogRecPtr start_lsn; + XLogRecPtr end_lsn; + TimeLineID tli; +} WalSummaryFile; + +extern List *GetWalSummaries(TimeLineID tli, XLogRecPtr start_lsn, + XLogRecPtr end_lsn); +extern List *FilterWalSummaries(List *wslist, TimeLineID tli, + XLogRecPtr start_lsn, XLogRecPtr end_lsn); +extern bool WalSummariesAreComplete(List *wslist, + XLogRecPtr start_lsn, XLogRecPtr end_lsn, + XLogRecPtr *missing_lsn); +extern File OpenWalSummaryFile(WalSummaryFile *ws, bool missing_ok); +extern void RemoveWalSummaryIfOlderThan(WalSummaryFile *ws, + time_t cutoff_time); + +extern int ReadWalSummary(void *wal_summary_io, void *data, int length); +extern int WriteWalSummary(void *wal_summary_io, void *data, int length); +extern void ReportWalSummaryError(void *callback_arg, char *fmt,...); + +#endif /* WALSUMMARY_H */ diff --git a/src/include/catalog/pg_proc.dat b/src/include/catalog/pg_proc.dat index f0b7b9cbd8..f68e6d4987 100644 --- a/src/include/catalog/pg_proc.dat +++ b/src/include/catalog/pg_proc.dat @@ -12062,4 +12062,23 @@ proname => 'any_value_transfn', prorettype => 'anyelement', proargtypes => 'anyelement anyelement', prosrc => 'any_value_transfn' }, +{ oid => '8436', + descr => 'list of available WAL summary files', + proname => 'pg_available_wal_summaries', prorows => '100', + proretset => 't', provolatile => 'v', proparallel => 's', + prorettype => 'record', proargtypes => '', + proallargtypes => '{int8,pg_lsn,pg_lsn}', + proargmodes => '{o,o,o}', + proargnames => '{tli,start_lsn,end_lsn}', + prosrc => 'pg_available_wal_summaries' }, +{ oid => '8437', + descr => 'contents of a WAL sumamry file', + proname => 'pg_wal_summary_contents', prorows => '100', + proretset => 't', provolatile => 'v', proparallel => 's', + prorettype => 'record', proargtypes => 'int8 pg_lsn pg_lsn', + proallargtypes => '{int8,pg_lsn,pg_lsn,oid,oid,oid,int2,int8,bool}', + proargmodes => '{i,i,i,o,o,o,o,o,o}', + proargnames => '{tli,start_lsn,end_lsn,relfilenode,reltablespace,reldatabase,relforknumber,relblocknumber,is_limit_block}', + prosrc => 'pg_wal_summary_contents' }, + ] diff --git a/src/include/common/blkreftable.h b/src/include/common/blkreftable.h new file mode 100644 index 0000000000..22d9883dc5 --- /dev/null +++ b/src/include/common/blkreftable.h @@ -0,0 +1,120 @@ +/*------------------------------------------------------------------------- + * + * blkreftable.h + * Block reference tables. + * + * A block reference table is used to keep track of which blocks have + * been modified by WAL records within a certain LSN range. + * + * For each relation fork, there is a "limit block number". All existing + * blocks greater than or equal to the limit block number must be + * considered modified; for those less than the limit block number, + * we maintain a bitmap. When a relation fork is created or dropped, + * the limit block number should be set to 0. When it's truncated, + * the limit block number should be set to the length in blocks to + * which it was truncated. + * + * Portions Copyright (c) 2010-2022, PostgreSQL Global Development Group + * + * src/include/common/blkreftable.h + * + *------------------------------------------------------------------------- + */ +#ifndef BLKREFTABLE_H +#define BLKREFTABLE_H + +#include "storage/block.h" +#include "storage/relfilelocator.h" + +/* Magic number for serialization file format. */ +#define BLOCKREFTABLE_MAGIC 0x652b137b + +struct BlockRefTable; +struct BlockRefTableEntry; +struct BlockRefTableReader; +struct BlockRefTableWriter; +typedef struct BlockRefTable BlockRefTable; +typedef struct BlockRefTableEntry BlockRefTableEntry; +typedef struct BlockRefTableReader BlockRefTableReader; +typedef struct BlockRefTableWriter BlockRefTableWriter; + +/* + * The return value of io_callback_fn should be the number of bytes read + * or written. If an error occurs, the functions should report it and + * not return. When used as a write callback, short writes should be retried + * or treated as errors, so that if the callback returns, the return value + * is always the request length. + * + * report_error_fn should not return. + */ +typedef int (*io_callback_fn) (void *callback_arg, void *data, int length); +typedef void (*report_error_fn) (void *calblack_arg, char *msg,...); + + +/* + * Functions for manipulating an entire in-memory block reference table. + */ +extern BlockRefTable *CreateEmptyBlockRefTable(void); +extern void BlockRefTableSetLimitBlock(BlockRefTable *brtab, + const RelFileLocator *rlocator, + ForkNumber forknum, + BlockNumber limit_block); +extern void BlockRefTableMarkBlockModified(BlockRefTable *brtab, + const RelFileLocator *rlocator, + ForkNumber forknum, + BlockNumber blknum); +extern void WriteBlockRefTable(BlockRefTable *brtab, + io_callback_fn write_callback, + void *write_callback_arg); + +extern BlockRefTableEntry *BlockRefTableGetEntry(BlockRefTable *brtab, + const RelFileLocator *rlocator, + ForkNumber forknum, + BlockNumber *limit_block); +extern int BlockRefTableEntryGetBlocks(BlockRefTableEntry *entry, + BlockNumber start_blkno, + BlockNumber stop_blkno, + BlockNumber *blocks, + int nblocks); + +/* + * Functions for reading a block reference table incrementally from disk. + */ +extern BlockRefTableReader *CreateBlockRefTableReader(io_callback_fn read_callback, + void *read_callback_arg, + char *error_filename, + report_error_fn error_callback, + void *error_callback_arg); +extern bool BlockRefTableReaderNextRelation(BlockRefTableReader *reader, + RelFileLocator *rlocator, + ForkNumber *forknum, + BlockNumber *limit_block); +extern unsigned BlockRefTableReaderGetBlocks(BlockRefTableReader *reader, + BlockNumber *blocks, + int nblocks); +extern void DestroyBlockRefTableReader(BlockRefTableReader *reader); + +/* + * Functions for writing a block reference table incrementally to disk. + * + * Note that entries must be written in the proper order, that is, sorted by + * database, then tablespace, then relfilenumber, then fork number. Caller + * is responsible for supplying data in the correct order. If that seems hard, + * use an in-memory BlockRefTable instead. + */ +extern BlockRefTableWriter *CreateBlockRefTableWriter(io_callback_fn write_callback, + void *write_callback_arg); +extern void BlockRefTableWriteEntry(BlockRefTableWriter *writer, + BlockRefTableEntry *entry); +extern void DestroyBlockRefTableWriter(BlockRefTableWriter *writer); + +extern BlockRefTableEntry *CreateBlockRefTableEntry(RelFileLocator rlocator, + ForkNumber forknum); +extern void BlockRefTableEntrySetLimitBlock(BlockRefTableEntry *entry, + BlockNumber limit_block); +extern void BlockRefTableEntryMarkBlockModified(BlockRefTableEntry *entry, + ForkNumber forknum, + BlockNumber blknum); +extern void BlockRefTableFreeEntry(BlockRefTableEntry *entry); + +#endif /* BLKREFTABLE_H */ diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 14bd574fc2..898adccb25 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -338,6 +338,7 @@ typedef enum BackendType B_STARTUP, B_WAL_RECEIVER, B_WAL_SENDER, + B_WAL_SUMMARIZER, B_WAL_WRITER, } BackendType; @@ -443,6 +444,7 @@ typedef enum CheckpointerProcess, WalWriterProcess, WalReceiverProcess, + WalSummarizerProcess, NUM_AUXPROCTYPES /* Must be last! */ } AuxProcType; @@ -455,6 +457,7 @@ extern PGDLLIMPORT AuxProcType MyAuxProcType; #define AmCheckpointerProcess() (MyAuxProcType == CheckpointerProcess) #define AmWalWriterProcess() (MyAuxProcType == WalWriterProcess) #define AmWalReceiverProcess() (MyAuxProcType == WalReceiverProcess) +#define AmWalSummarizerProcess() (MyAuxProcType == WalSummarizerProcess) /***************************************************************************** diff --git a/src/include/nodes/replnodes.h b/src/include/nodes/replnodes.h index 4321ba8f86..856491eecd 100644 --- a/src/include/nodes/replnodes.h +++ b/src/include/nodes/replnodes.h @@ -108,4 +108,13 @@ typedef struct TimeLineHistoryCmd TimeLineID timeline; } TimeLineHistoryCmd; +/* ---------------------- + * UPLOAD_MANIFEST command + * ---------------------- + */ +typedef struct UploadManifestCmd +{ + NodeTag type; +} UploadManifestCmd; + #endif /* REPLNODES_H */ diff --git a/src/include/postmaster/walsummarizer.h b/src/include/postmaster/walsummarizer.h new file mode 100644 index 0000000000..7584cb69a7 --- /dev/null +++ b/src/include/postmaster/walsummarizer.h @@ -0,0 +1,31 @@ +/*------------------------------------------------------------------------- + * + * walsummarizer.h + * + * Header file for background WAL summarization process. + * + * Portions Copyright (c) 1996-2022, PostgreSQL Global Development Group + * + * IDENTIFICATION + * src/include/postmaster/walsummarizer.h + * + *------------------------------------------------------------------------- + */ +#ifndef WALSUMMARIZER_H +#define WALSUMMARIZER_H + +#include "access/xlogdefs.h" + +extern int wal_summarize_mb; +extern int wal_summarize_keep_time; + +extern Size WalSummarizerShmemSize(void); +extern void WalSummarizerShmemInit(void); +extern void WalSummarizerMain(void) pg_attribute_noreturn(); + +extern XLogRecPtr GetOldestUnsummarizedLSN(TimeLineID *tli, + bool *lsn_is_exact); +extern void SetWalSummarizerLatch(void); +extern XLogRecPtr WaitForWalSummarization(XLogRecPtr lsn, long timeout); + +#endif diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index ef74f32693..ee55008082 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -417,11 +417,12 @@ extern PGDLLIMPORT PGPROC *PreparedXactProcs; * We set aside some extra PGPROC structures for auxiliary processes, * ie things that aren't full-fledged backends but need shmem access. * - * Background writer, checkpointer, WAL writer and archiver run during normal - * operation. Startup process and WAL receiver also consume 2 slots, but WAL - * writer is launched only after startup has exited, so we only need 5 slots. + * Background writer, checkpointer, WAL writer, WAL summarizer, and archiver + * run during normal operation. Startup process and WAL receiver also consume + * 2 slots, but WAL writer is launched only after startup has exited, so we + * only need 6 slots. */ -#define NUM_AUXILIARY_PROCS 5 +#define NUM_AUXILIARY_PROCS 6 /* configurable options */ extern PGDLLIMPORT int DeadlockTimeout; diff --git a/src/include/utils/guc_tables.h b/src/include/utils/guc_tables.h index d5a0880678..7d3bc0f671 100644 --- a/src/include/utils/guc_tables.h +++ b/src/include/utils/guc_tables.h @@ -72,6 +72,7 @@ enum config_group WAL_RECOVERY, WAL_ARCHIVE_RECOVERY, WAL_RECOVERY_TARGET, + WAL_SUMMARIZATION, REPLICATION_SENDING, REPLICATION_PRIMARY, REPLICATION_STANDBY, diff --git a/src/test/perl/PostgreSQL/Test/Cluster.pm b/src/test/perl/PostgreSQL/Test/Cluster.pm index c3d46c7c70..b711d60fc4 100644 --- a/src/test/perl/PostgreSQL/Test/Cluster.pm +++ b/src/test/perl/PostgreSQL/Test/Cluster.pm @@ -779,6 +779,10 @@ a tar-format backup, pass the name of the tar program to use in the keyword parameter tar_program. Note that tablespace tar files aren't handled here. +To restore from an incremental backup, pass the parameter combine_with_prior +as a reference to an array of prior backup names with which this backup +is to be combined using pg_combinebackup. + Streaming replication can be enabled on this node by passing the keyword parameter has_streaming => 1. This is disabled by default. @@ -816,7 +820,22 @@ sub init_from_backup mkdir $self->archive_dir; my $data_path = $self->data_dir; - if (defined $params{tar_program}) + if (defined $params{combine_with_prior}) + { + my @prior_backups = @{$params{combine_with_prior}}; + my @prior_backup_path; + + for my $prior_backup_name (@prior_backups) + { + push @prior_backup_path, + $root_node->backup_dir . '/' . $prior_backup_name; + } + + local %ENV = $self->_get_env(); + PostgreSQL::Test::Utils::system_or_bail('pg_combinebackup', + @prior_backup_path, $backup_path, '-o', $data_path); + } + elsif (defined $params{tar_program}) { mkdir($data_path); PostgreSQL::Test::Utils::system_or_bail($params{tar_program}, 'xf', diff --git a/src/test/recovery/t/001_stream_rep.pl b/src/test/recovery/t/001_stream_rep.pl index 0c72ba0944..353db33a9f 100644 --- a/src/test/recovery/t/001_stream_rep.pl +++ b/src/test/recovery/t/001_stream_rep.pl @@ -15,6 +15,8 @@ my $node_primary = PostgreSQL::Test::Cluster->new('primary'); $node_primary->init( allows_streaming => 1, auth_extra => [ '--create-role', 'repl_role' ]); +# WAL summarization can postpone WAL recycling, leading to test failures +$node_primary->append_conf('postgresql.conf', "wal_summarize_mb = 0"); $node_primary->start; my $backup_name = 'my_backup'; diff --git a/src/test/recovery/t/019_replslot_limit.pl b/src/test/recovery/t/019_replslot_limit.pl index 7d94f15778..4f52ddbe79 100644 --- a/src/test/recovery/t/019_replslot_limit.pl +++ b/src/test/recovery/t/019_replslot_limit.pl @@ -22,6 +22,7 @@ $node_primary->append_conf( min_wal_size = 2MB max_wal_size = 4MB log_checkpoints = yes +wal_summarize_mb = 0 )); $node_primary->start; $node_primary->safe_psql('postgres', @@ -256,6 +257,7 @@ $node_primary2->append_conf( min_wal_size = 32MB max_wal_size = 32MB log_checkpoints = yes +wal_summarize_mb = 0 )); $node_primary2->start; $node_primary2->safe_psql('postgres', @@ -310,6 +312,7 @@ $node_primary3->append_conf( max_wal_size = 2MB log_checkpoints = yes max_slot_wal_keep_size = 1MB + wal_summarize_mb = 0 )); $node_primary3->start; $node_primary3->safe_psql('postgres', diff --git a/src/test/recovery/t/035_standby_logical_decoding.pl b/src/test/recovery/t/035_standby_logical_decoding.pl index 480e6d6caa..a91437dfa7 100644 --- a/src/test/recovery/t/035_standby_logical_decoding.pl +++ b/src/test/recovery/t/035_standby_logical_decoding.pl @@ -250,6 +250,7 @@ $node_primary->append_conf( wal_level = 'logical' max_replication_slots = 4 max_wal_senders = 4 +wal_summarize_mb = 0 }); $node_primary->dump_info; $node_primary->start; diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 8de90c4958..ff3cff8c28 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -3991,3 +3991,26 @@ yyscan_t z_stream z_streamp zic_t +BlockRefTable +BlockRefTableBuffer +BlockRefTableEntry +BlockRefTableKey +BlockRefTableReader +BlockRefTableSerializedEntry +BlockRefTableWriter +FileBackupMethod +IncrementalBackupInfo +SummarizerReadLocalXLogPrivate +UploadManifestCmd +WalSummarizerData +WalSummaryFile +WalSummaryIO +backup_file_entry +backup_wal_range +cb_cleanup_dir +cb_options +cb_tablespace +cb_tablespace_mapping +manifest_data +manifest_writer +rfile -- 2.37.1 (Apple Git-137.1) [application/octet-stream] v4-0002-Change-struct-tablespaceinfo-s-oid-member-from-ch.patch (11.7K, ../../CA+TgmoYw=8CnKEPQX9qk2estSTCaB7OGcnVM8oa-hAePG2O4Yw@mail.gmail.com/7-v4-0002-Change-struct-tablespaceinfo-s-oid-member-from-ch.patch) download | inline diff: From 94280254510de2f50cee7a597158df7bf1fca16c Mon Sep 17 00:00:00 2001 From: Robert Haas <[email protected]> Date: Tue, 3 Oct 2023 13:30:16 -0400 Subject: [PATCH v4 2/6] Change struct tablespaceinfo's oid member from 'char *' to 'Oid' This shouldn't change behavior except in the unusual case where there are file in the tablespace directory that have entirely numeric names but are nevertheless not possible names for a tablespace directory, either because their names have leading zeroes that shouldn't be there, or the value is actually zero, or because the value is too large to represent as an OID. In those cases, the directory would previously have made it into the list of tablespaceinfo objects and no longer will. Thus, base backups will now ignore such directories, instead of treating them as legitimate tablespace directories. Similarly, if entries for such tablespaces occur in a tablespace_map file, they will now be rejected as erroneous, instead of being honored. This is infrastructure for future work that wants to be able to know the tablespace of each relation that is part of a backup *as an OID*. By strengthening the up-front validation, we don't have to worry about weird cases later, and can more easily avoid repeated string->integer conversions. --- src/backend/access/transam/xlog.c | 19 ++++++++++-- src/backend/access/transam/xlogrecovery.c | 12 ++++++-- src/backend/backup/backup_manifest.c | 6 ++-- src/backend/backup/basebackup.c | 35 ++++++++++++----------- src/backend/backup/basebackup_copy.c | 2 +- src/include/backup/backup_manifest.h | 2 +- src/include/backup/basebackup.h | 2 +- 7 files changed, 49 insertions(+), 29 deletions(-) diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index fcbde10529..677a5bf51b 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -8483,9 +8483,22 @@ do_pg_backup_start(const char *backupidstr, bool fast, List **tablespaces, char *relpath = NULL; char *s; PGFileType de_type; + char *badp; + Oid tsoid; - /* Skip anything that doesn't look like a tablespace */ - if (strspn(de->d_name, "0123456789") != strlen(de->d_name)) + /* + * Try to parse the directory name as an unsigned integer. + * + * Tablespace directories should be positive integers that can be + * represented in 32 bits, with no leading zeroes or trailing + * garbage. If we come across a name that doesn't meet those + * criteria, skip it. + */ + if (de->d_name[0] < '1' || de->d_name[1] > '9') + continue; + errno = 0; + tsoid = strtoul(de->d_name, &badp, 10); + if (*badp != '\0' || errno == EINVAL || errno == ERANGE) continue; snprintf(fullpath, sizeof(fullpath), "pg_tblspc/%s", de->d_name); @@ -8560,7 +8573,7 @@ do_pg_backup_start(const char *backupidstr, bool fast, List **tablespaces, } ti = palloc(sizeof(tablespaceinfo)); - ti->oid = pstrdup(de->d_name); + ti->oid = tsoid; ti->path = pstrdup(linkpath); ti->rpath = relpath; ti->size = -1; diff --git a/src/backend/access/transam/xlogrecovery.c b/src/backend/access/transam/xlogrecovery.c index becc2bda62..5549e1afc5 100644 --- a/src/backend/access/transam/xlogrecovery.c +++ b/src/backend/access/transam/xlogrecovery.c @@ -678,7 +678,7 @@ InitWalRecovery(ControlFileData *ControlFile, bool *wasShutdown_ptr, tablespaceinfo *ti = lfirst(lc); char *linkloc; - linkloc = psprintf("pg_tblspc/%s", ti->oid); + linkloc = psprintf("pg_tblspc/%u", ti->oid); /* * Remove the existing symlink if any and Create the symlink @@ -692,7 +692,6 @@ InitWalRecovery(ControlFileData *ControlFile, bool *wasShutdown_ptr, errmsg("could not create symbolic link \"%s\": %m", linkloc))); - pfree(ti->oid); pfree(ti->path); pfree(ti); } @@ -1341,6 +1340,8 @@ read_tablespace_map(List **tablespaces) { if (!was_backslash && (ch == '\n' || ch == '\r')) { + char *endp; + if (i == 0) continue; /* \r immediately followed by \n */ @@ -1360,7 +1361,12 @@ read_tablespace_map(List **tablespaces) str[n++] = '\0'; ti = palloc0(sizeof(tablespaceinfo)); - ti->oid = pstrdup(str); + errno = 0; + ti->oid = strtoul(str, &endp, 10); + if (*endp != '\0' || errno == EINVAL || errno == ERANGE) + ereport(FATAL, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("invalid data in file \"%s\"", TABLESPACE_MAP))); ti->path = pstrdup(str + n); *tablespaces = lappend(*tablespaces, ti); diff --git a/src/backend/backup/backup_manifest.c b/src/backend/backup/backup_manifest.c index cee6216524..aeed362a9a 100644 --- a/src/backend/backup/backup_manifest.c +++ b/src/backend/backup/backup_manifest.c @@ -97,7 +97,7 @@ FreeBackupManifest(backup_manifest_info *manifest) * Add an entry to the backup manifest for a file. */ void -AddFileToBackupManifest(backup_manifest_info *manifest, const char *spcoid, +AddFileToBackupManifest(backup_manifest_info *manifest, Oid spcoid, const char *pathname, size_t size, pg_time_t mtime, pg_checksum_context *checksum_ctx) { @@ -114,9 +114,9 @@ AddFileToBackupManifest(backup_manifest_info *manifest, const char *spcoid, * pathname relative to the data directory (ignoring the intermediate * symlink traversal). */ - if (spcoid != NULL) + if (OidIsValid(spcoid)) { - snprintf(pathbuf, sizeof(pathbuf), "pg_tblspc/%s/%s", spcoid, + snprintf(pathbuf, sizeof(pathbuf), "pg_tblspc/%u/%s", spcoid, pathname); pathname = pathbuf; } diff --git a/src/backend/backup/basebackup.c b/src/backend/backup/basebackup.c index b126d9c890..b537f46219 100644 --- a/src/backend/backup/basebackup.c +++ b/src/backend/backup/basebackup.c @@ -75,14 +75,15 @@ typedef struct pg_checksum_type manifest_checksum_type; } basebackup_options; -static int64 sendTablespace(bbsink *sink, char *path, char *spcoid, bool sizeonly, +static int64 sendTablespace(bbsink *sink, char *path, Oid spcoid, bool sizeonly, struct backup_manifest_info *manifest); static int64 sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, List *tablespaces, bool sendtblspclinks, - backup_manifest_info *manifest, const char *spcoid); + backup_manifest_info *manifest, Oid spcoid); static bool sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, - struct stat *statbuf, bool missing_ok, Oid dboid, - backup_manifest_info *manifest, const char *spcoid); + struct stat *statbuf, bool missing_ok, + Oid dboid, Oid spcoid, + backup_manifest_info *manifest); static off_t read_file_data_into_buffer(bbsink *sink, const char *readfilename, int fd, off_t offset, size_t length, @@ -305,7 +306,7 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) if (tmp->path == NULL) tmp->size = sendDir(sink, ".", 1, true, state.tablespaces, - true, NULL, NULL); + true, NULL, InvalidOid); else tmp->size = sendTablespace(sink, tmp->path, tmp->oid, true, NULL); @@ -346,7 +347,7 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) /* Then the bulk of the files... */ sendDir(sink, ".", 1, false, state.tablespaces, - sendtblspclinks, &manifest, NULL); + sendtblspclinks, &manifest, InvalidOid); /* ... and pg_control after everything else. */ if (lstat(XLOG_CONTROL_FILE, &statbuf) != 0) @@ -355,11 +356,11 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) errmsg("could not stat file \"%s\": %m", XLOG_CONTROL_FILE))); sendFile(sink, XLOG_CONTROL_FILE, XLOG_CONTROL_FILE, &statbuf, - false, InvalidOid, &manifest, NULL); + false, InvalidOid, InvalidOid, &manifest); } else { - char *archive_name = psprintf("%s.tar", ti->oid); + char *archive_name = psprintf("%u.tar", ti->oid); bbsink_begin_archive(sink, archive_name); @@ -623,8 +624,8 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) (errcode_for_file_access(), errmsg("could not stat file \"%s\": %m", pathbuf))); - sendFile(sink, pathbuf, pathbuf, &statbuf, false, InvalidOid, - &manifest, NULL); + sendFile(sink, pathbuf, pathbuf, &statbuf, false, + InvalidOid, InvalidOid, &manifest); /* unconditionally mark file as archived */ StatusFilePath(pathbuf, fname, ".done"); @@ -1087,7 +1088,7 @@ sendFileWithContent(bbsink *sink, const char *filename, const char *content, _tarWritePadding(sink, len); - AddFileToBackupManifest(manifest, NULL, filename, len, + AddFileToBackupManifest(manifest, InvalidOid, filename, len, (pg_time_t) statbuf.st_mtime, &checksum_ctx); } @@ -1099,7 +1100,7 @@ sendFileWithContent(bbsink *sink, const char *filename, const char *content, * Only used to send auxiliary tablespaces, not PGDATA. */ static int64 -sendTablespace(bbsink *sink, char *path, char *spcoid, bool sizeonly, +sendTablespace(bbsink *sink, char *path, Oid spcoid, bool sizeonly, backup_manifest_info *manifest) { int64 size; @@ -1154,7 +1155,7 @@ sendTablespace(bbsink *sink, char *path, char *spcoid, bool sizeonly, static int64 sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, List *tablespaces, bool sendtblspclinks, backup_manifest_info *manifest, - const char *spcoid) + Oid spcoid) { DIR *dir; struct dirent *de; @@ -1416,8 +1417,8 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, if (!sizeonly) sent = sendFile(sink, pathbuf, pathbuf + basepathlen + 1, &statbuf, - true, isDbDir ? atooid(lastDir + 1) : InvalidOid, - manifest, spcoid); + true, isDbDir ? atooid(lastDir + 1) : InvalidOid, spcoid, + manifest); if (sent || sizeonly) { @@ -1486,8 +1487,8 @@ is_checksummed_file(const char *fullpath, const char *filename) */ static bool sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, - struct stat *statbuf, bool missing_ok, Oid dboid, - backup_manifest_info *manifest, const char *spcoid) + struct stat *statbuf, bool missing_ok, Oid dboid, Oid spcoid, + backup_manifest_info *manifest) { int fd; BlockNumber blkno = 0; diff --git a/src/backend/backup/basebackup_copy.c b/src/backend/backup/basebackup_copy.c index fee30c21e1..3bdbe1f989 100644 --- a/src/backend/backup/basebackup_copy.c +++ b/src/backend/backup/basebackup_copy.c @@ -407,7 +407,7 @@ SendTablespaceList(List *tablespaces) } else { - values[0] = ObjectIdGetDatum(strtoul(ti->oid, NULL, 10)); + values[0] = ObjectIdGetDatum(ti->oid); values[1] = CStringGetTextDatum(ti->path); } if (ti->size >= 0) diff --git a/src/include/backup/backup_manifest.h b/src/include/backup/backup_manifest.h index d41b439980..5a481dbcf5 100644 --- a/src/include/backup/backup_manifest.h +++ b/src/include/backup/backup_manifest.h @@ -39,7 +39,7 @@ extern void InitializeBackupManifest(backup_manifest_info *manifest, backup_manifest_option want_manifest, pg_checksum_type manifest_checksum_type); extern void AddFileToBackupManifest(backup_manifest_info *manifest, - const char *spcoid, + Oid spcoid, const char *pathname, size_t size, pg_time_t mtime, pg_checksum_context *checksum_ctx); diff --git a/src/include/backup/basebackup.h b/src/include/backup/basebackup.h index 3e68abc2bb..1432d9c206 100644 --- a/src/include/backup/basebackup.h +++ b/src/include/backup/basebackup.h @@ -27,7 +27,7 @@ */ typedef struct { - char *oid; /* tablespace's OID, as a decimal string */ + Oid oid; /* tablespace's OID */ char *path; /* full path to tablespace's directory */ char *rpath; /* relative path if it's within PGDATA, else * NULL */ -- 2.37.1 (Apple Git-137.1) ^ permalink raw reply [nested|flat] 22+ messages in thread
* Re: trying again to get incremental backup @ 2023-10-19 16:05 Robert Haas <[email protected]> parent: Robert Haas <[email protected]> 2 siblings, 1 reply; 22+ messages in thread From: Robert Haas @ 2023-10-19 16:05 UTC (permalink / raw) To: Dilip Kumar <[email protected]>; +Cc: Andres Freund <[email protected]>; pgsql-hackers On Wed, Oct 4, 2023 at 4:08 PM Robert Haas <[email protected]> wrote: > Clearly there's a good amount of stuff to sort out here, but we've > still got quite a bit of time left before feature freeze so I'd like > to have a go at it. Please let me know your thoughts, if you have any. Apparently, nobody has any thoughts, but here's an updated patch set anyway. The main change, other than rebasing, is that I did a bunch more documentation work on the main patch (0005). I'm much happier with it now, although I expect it may need more adjustments here and there as outstanding design questions get settled. After some thought, I think that it should be fine to commit 0001 and 0002 as independent refactoring patches, and I plan to go ahead and do that pretty soon unless somebody objects. Thanks, -- Robert Haas EDB: http://www.enterprisedb.com Attachments: [application/octet-stream] v5-0004-Move-src-bin-pg_verifybackup-parse_manifest.c-int.patch (4.3K, ../../CA+TgmoZNVeBzoqDL8xvr-nkaepq815jtDR4nJzPew7=3iEuM1g@mail.gmail.com/2-v5-0004-Move-src-bin-pg_verifybackup-parse_manifest.c-int.patch) download | inline diff: From b9355ab0f9c9f10736a13e7730294b82fc374963 Mon Sep 17 00:00:00 2001 From: Robert Haas <[email protected]> Date: Tue, 3 Oct 2023 13:32:45 -0400 Subject: [PATCH v5 4/6] Move src/bin/pg_verifybackup/parse_manifest.c into src/common. This makes it possible for the code to be easily reused by other client-side tools, and/or by the server. --- src/bin/pg_verifybackup/Makefile | 1 - src/bin/pg_verifybackup/meson.build | 1 - src/bin/pg_verifybackup/pg_verifybackup.c | 2 +- src/common/Makefile | 1 + src/common/meson.build | 1 + src/{bin/pg_verifybackup => common}/parse_manifest.c | 4 ++-- src/{bin/pg_verifybackup => include/common}/parse_manifest.h | 2 +- 7 files changed, 6 insertions(+), 6 deletions(-) rename src/{bin/pg_verifybackup => common}/parse_manifest.c (99%) rename src/{bin/pg_verifybackup => include/common}/parse_manifest.h (97%) diff --git a/src/bin/pg_verifybackup/Makefile b/src/bin/pg_verifybackup/Makefile index 596df15118..8f04fa662c 100644 --- a/src/bin/pg_verifybackup/Makefile +++ b/src/bin/pg_verifybackup/Makefile @@ -21,7 +21,6 @@ LDFLAGS_INTERNAL += -L$(top_builddir)/src/fe_utils -lpgfeutils $(libpq_pgport) OBJS = \ $(WIN32RES) \ - parse_manifest.o \ pg_verifybackup.o all: pg_verifybackup diff --git a/src/bin/pg_verifybackup/meson.build b/src/bin/pg_verifybackup/meson.build index 9369da1bc6..58f780d1a6 100644 --- a/src/bin/pg_verifybackup/meson.build +++ b/src/bin/pg_verifybackup/meson.build @@ -1,7 +1,6 @@ # Copyright (c) 2022-2023, PostgreSQL Global Development Group pg_verifybackup_sources = files( - 'parse_manifest.c', 'pg_verifybackup.c' ) diff --git a/src/bin/pg_verifybackup/pg_verifybackup.c b/src/bin/pg_verifybackup/pg_verifybackup.c index 059836f0e6..ce423a03d4 100644 --- a/src/bin/pg_verifybackup/pg_verifybackup.c +++ b/src/bin/pg_verifybackup/pg_verifybackup.c @@ -20,9 +20,9 @@ #include "common/hashfn.h" #include "common/logging.h" +#include "common/parse_manifest.h" #include "fe_utils/simple_list.h" #include "getopt_long.h" -#include "parse_manifest.h" #include "pgtime.h" /* diff --git a/src/common/Makefile b/src/common/Makefile index 70884be00c..3c8effc533 100644 --- a/src/common/Makefile +++ b/src/common/Makefile @@ -66,6 +66,7 @@ OBJS_COMMON = \ kwlookup.o \ link-canary.o \ md5_common.o \ + parse_manifest.o \ percentrepl.o \ pg_get_line.o \ pg_lzcompress.o \ diff --git a/src/common/meson.build b/src/common/meson.build index ae05ac63cf..aa646f96a3 100644 --- a/src/common/meson.build +++ b/src/common/meson.build @@ -18,6 +18,7 @@ common_sources = files( 'kwlookup.c', 'link-canary.c', 'md5_common.c', + 'parse_manifest.c', 'percentrepl.c', 'pg_get_line.c', 'pg_lzcompress.c', diff --git a/src/bin/pg_verifybackup/parse_manifest.c b/src/common/parse_manifest.c similarity index 99% rename from src/bin/pg_verifybackup/parse_manifest.c rename to src/common/parse_manifest.c index f0acd9f1e7..9895f2f17d 100644 --- a/src/bin/pg_verifybackup/parse_manifest.c +++ b/src/common/parse_manifest.c @@ -6,15 +6,15 @@ * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group * Portions Copyright (c) 1994, Regents of the University of California * - * src/bin/pg_verifybackup/parse_manifest.c + * src/common/parse_manifest.c * *------------------------------------------------------------------------- */ #include "postgres_fe.h" -#include "parse_manifest.h" #include "common/jsonapi.h" +#include "common/parse_manifest.h" /* * Semantic states for JSON manifest parsing. diff --git a/src/bin/pg_verifybackup/parse_manifest.h b/src/include/common/parse_manifest.h similarity index 97% rename from src/bin/pg_verifybackup/parse_manifest.h rename to src/include/common/parse_manifest.h index 7387a917a2..7b24c5d785 100644 --- a/src/bin/pg_verifybackup/parse_manifest.h +++ b/src/include/common/parse_manifest.h @@ -6,7 +6,7 @@ * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group * Portions Copyright (c) 1994, Regents of the University of California * - * src/bin/pg_verifybackup/parse_manifest.h + * src/include/common/parse_manifest.h * *------------------------------------------------------------------------- */ -- 2.37.1 (Apple Git-137.1) [application/octet-stream] v5-0006-Add-new-pg_walsummary-tool.patch (11.4K, ../../CA+TgmoZNVeBzoqDL8xvr-nkaepq815jtDR4nJzPew7=3iEuM1g@mail.gmail.com/3-v5-0006-Add-new-pg_walsummary-tool.patch) download | inline diff: From 5a3e4b4d41faa184f03cddf45f546de764eac6de Mon Sep 17 00:00:00 2001 From: Robert Haas <[email protected]> Date: Wed, 14 Jun 2023 11:31:39 -0400 Subject: [PATCH v5 6/6] Add new pg_walsummary tool. This can dump the contents of WAL summary files, either those in pg_wal/summaries, or the INCREMENTAL_BACKUP files that are part of an incremental backup proper. XXX. Needs documentation and tests. --- src/bin/Makefile | 1 + src/bin/meson.build | 1 + src/bin/pg_walsummary/.gitignore | 1 + src/bin/pg_walsummary/Makefile | 42 ++++ src/bin/pg_walsummary/meson.build | 24 +++ src/bin/pg_walsummary/pg_walsummary.c | 278 ++++++++++++++++++++++++++ 6 files changed, 347 insertions(+) create mode 100644 src/bin/pg_walsummary/.gitignore create mode 100644 src/bin/pg_walsummary/Makefile create mode 100644 src/bin/pg_walsummary/meson.build create mode 100644 src/bin/pg_walsummary/pg_walsummary.c diff --git a/src/bin/Makefile b/src/bin/Makefile index aa2210925e..f98f58d39e 100644 --- a/src/bin/Makefile +++ b/src/bin/Makefile @@ -31,6 +31,7 @@ SUBDIRS = \ pg_upgrade \ pg_verifybackup \ pg_waldump \ + pg_walsummary \ pgbench \ psql \ scripts diff --git a/src/bin/meson.build b/src/bin/meson.build index 4cb6fd59bb..d1e9ef4409 100644 --- a/src/bin/meson.build +++ b/src/bin/meson.build @@ -17,6 +17,7 @@ subdir('pg_test_timing') subdir('pg_upgrade') subdir('pg_verifybackup') subdir('pg_waldump') +subdir('pg_walsummary') subdir('pgbench') subdir('pgevent') subdir('psql') diff --git a/src/bin/pg_walsummary/.gitignore b/src/bin/pg_walsummary/.gitignore new file mode 100644 index 0000000000..d71ec192fa --- /dev/null +++ b/src/bin/pg_walsummary/.gitignore @@ -0,0 +1 @@ +pg_walsummary diff --git a/src/bin/pg_walsummary/Makefile b/src/bin/pg_walsummary/Makefile new file mode 100644 index 0000000000..852f7208f6 --- /dev/null +++ b/src/bin/pg_walsummary/Makefile @@ -0,0 +1,42 @@ +#------------------------------------------------------------------------- +# +# Makefile for src/bin/pg_walsummary +# +# Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group +# Portions Copyright (c) 1994, Regents of the University of California +# +# src/bin/pg_walsummary/Makefile +# +#------------------------------------------------------------------------- + +PGFILEDESC = "pg_walsummary - print contents of WAL summary files" +PGAPPICON=win32 + +subdir = src/bin/pg_walsummary +top_builddir = ../../.. +include $(top_builddir)/src/Makefile.global + +override CPPFLAGS := -I$(libpq_srcdir) $(CPPFLAGS) +LDFLAGS_INTERNAL += -L$(top_builddir)/src/fe_utils -lpgfeutils + +OBJS = \ + $(WIN32RES) \ + pg_walsummary.o + +all: pg_walsummary + +pg_walsummary: $(OBJS) | submake-libpgport submake-libpgfeutils + $(CC) $(CFLAGS) $^ $(LDFLAGS) $(LDFLAGS_EX) $(LIBS) -o $@$(X) + + +install: all installdirs + $(INSTALL_PROGRAM) pg_walsummary$(X) '$(DESTDIR)$(bindir)/pg_walsummary$(X)' + +installdirs: + $(MKDIR_P) '$(DESTDIR)$(bindir)' + +uninstall: + rm -f '$(DESTDIR)$(bindir)/pg_walsummary$(X)' + +clean distclean maintainer-clean: + rm -f pg_walsummary$(X) $(OBJS) diff --git a/src/bin/pg_walsummary/meson.build b/src/bin/pg_walsummary/meson.build new file mode 100644 index 0000000000..c2092960c6 --- /dev/null +++ b/src/bin/pg_walsummary/meson.build @@ -0,0 +1,24 @@ +# Copyright (c) 2022-2023, PostgreSQL Global Development Group + +pg_walsummary_sources = files( + 'pg_walsummary.c', +) + +if host_system == 'windows' + pg_walsummary_sources += rc_bin_gen.process(win32ver_rc, extra_args: [ + '--NAME', 'pg_walsummary', + '--FILEDESC', 'pg_walsummary - print contents of WAL summary files',]) +endif + +pg_walsummary = executable('pg_walsummary', + pg_walsummary_sources, + dependencies: [frontend_code], + kwargs: default_bin_args, +) +bin_targets += pg_walsummary + +tests += { + 'name': 'pg_walsummary', + 'sd': meson.current_source_dir(), + 'bd': meson.current_build_dir() +} diff --git a/src/bin/pg_walsummary/pg_walsummary.c b/src/bin/pg_walsummary/pg_walsummary.c new file mode 100644 index 0000000000..0304a42026 --- /dev/null +++ b/src/bin/pg_walsummary/pg_walsummary.c @@ -0,0 +1,278 @@ +/*------------------------------------------------------------------------- + * + * pg_walsummary.c + * Prints the contents of WAL summary files. + * + * Copyright (c) 2017-2023, PostgreSQL Global Development Group + * + * IDENTIFICATION + * src/bin/pg_walsummary/pg_walsummary.c + * + *------------------------------------------------------------------------- + */ +#include "postgres_fe.h" + +#include <fcntl.h> +#include <limits.h> + +#include "common/blkreftable.h" +#include "common/logging.h" +#include "fe_utils/option_utils.h" +#include "lib/stringinfo.h" +#include "getopt_long.h" + +typedef struct ws_options +{ + bool individual; + bool quiet; +} ws_options; + +typedef struct ws_file_info +{ + int fd; + char *filename; +} ws_file_info; + +static BlockNumber *block_buffer = NULL; +static unsigned block_buffer_size = 512; /* Initial size. */ + +static void dump_one_relation(ws_options *opt, RelFileLocator *rlocator, + ForkNumber forknum, BlockNumber limit_block, + BlockRefTableReader *reader); +static void help(const char *progname); +static int compare_block_numbers(const void *a, const void *b); +static int walsummary_read_callback(void *callback_arg, void *data, + int length); +static void walsummary_error_callback(void *callback_arg, char *fmt,...); + +/* + * Main program. + */ +int +main(int argc, char *argv[]) +{ + static struct option long_options[] = { + {"individual", no_argument, NULL, 'i'}, + {"quiet", no_argument, NULL, 'q'}, + {NULL, 0, NULL, 0} + }; + + const char *progname; + int optindex; + int c; + ws_options opt; + + pg_logging_init(argv[0]); + progname = get_progname(argv[0]); + handle_help_version_opts(argc, argv, progname, help); + + /* process command-line options */ + while ((c = getopt_long(argc, argv, "f:iqw:", + long_options, &optindex)) != -1) + { + switch (c) + { + case 'i': + opt.individual = true; + break; + case 'q': + opt.quiet = true; + break; + default: + /* getopt_long already emitted a complaint */ + pg_log_error_hint("Try \"%s --help\" for more information.", progname); + exit(1); + } + } + + if (optind >= argc) + { + pg_log_error("%s: no input files specified", progname); + pg_log_error_hint("Try \"%s --help\" for more information.", progname); + exit(1); + } + + while (optind < argc) + { + ws_file_info ws; + BlockRefTableReader *reader; + RelFileLocator rlocator; + ForkNumber forknum; + BlockNumber limit_block; + + ws.filename = argv[optind++]; + if ((ws.fd = open(ws.filename, O_RDONLY | PG_BINARY, 0)) < 0) + pg_fatal("could not open file \"%s\": %m", ws.filename); + + reader = CreateBlockRefTableReader(walsummary_read_callback, &ws, + ws.filename, + walsummary_error_callback, NULL); + while (BlockRefTableReaderNextRelation(reader, &rlocator, &forknum, + &limit_block)) + dump_one_relation(&opt, &rlocator, forknum, limit_block, reader); + + DestroyBlockRefTableReader(reader); + close(ws.fd); + } + + exit(0); +} + +/* + * Dump details for one relation. + */ +static void +dump_one_relation(ws_options *opt, RelFileLocator *rlocator, + ForkNumber forknum, BlockNumber limit_block, + BlockRefTableReader *reader) +{ + unsigned i = 0; + unsigned nblocks; + BlockNumber startblock = InvalidBlockNumber; + BlockNumber endblock = InvalidBlockNumber; + + /* Dump limit block, if any. */ + if (limit_block != InvalidBlockNumber) + printf("TS %u, DB %u, REL %u, FORK %s: limit %u\n", + rlocator->spcOid, rlocator->dbOid, rlocator->relNumber, + forkNames[forknum], limit_block); + + /* If we haven't allocated a block buffer yet, do that now. */ + if (block_buffer == NULL) + block_buffer = palloc_array(BlockNumber, block_buffer_size); + + /* Try to fill the block buffer. */ + nblocks = BlockRefTableReaderGetBlocks(reader, + block_buffer, + block_buffer_size); + + /* If we filled the block buffer completely, we must enlarge it. */ + while (nblocks >= block_buffer_size) + { + unsigned new_size; + + /* Double the size, being careful about overflow. */ + new_size = block_buffer_size * 2; + if (new_size < block_buffer_size) + new_size = PG_UINT32_MAX; + block_buffer = repalloc_array(block_buffer, BlockNumber, new_size); + + /* Try to fill the newly-allocated space. */ + nblocks += + BlockRefTableReaderGetBlocks(reader, + block_buffer + block_buffer_size, + new_size - block_buffer_size); + + /* Save the new size for later calls. */ + block_buffer_size = new_size; + } + + /* If we don't need to produce any output, skip the rest of this. */ + if (opt->quiet) + return; + + /* + * Sort the returned block numbers. If the block reference table was using + * the bitmap representation for a given chunk, the block numbers in that + * chunk will already be sorted, but when the array-of-offsets + * representation is used, we can receive block numbers here out of order. + */ + qsort(block_buffer, nblocks, sizeof(BlockNumber), compare_block_numbers); + + /* Dump block references. */ + while (i < nblocks) + { + /* + * Find the next range of blocks to print, but if --individual was + * specified, then consider each block a separate range. + */ + startblock = endblock = block_buffer[i++]; + if (!opt->individual) + { + while (i < nblocks && block_buffer[i] == endblock + 1) + { + endblock++; + i++; + } + } + + /* Format this range of block numbers as a string. */ + if (startblock == endblock) + printf("TS %u, DB %u, REL %u, FORK %s: block %u\n", + rlocator->spcOid, rlocator->dbOid, rlocator->relNumber, + forkNames[forknum], startblock); + else + printf("TS %u, DB %u, REL %u, FORK %s: blocks %u..%u\n", + rlocator->spcOid, rlocator->dbOid, rlocator->relNumber, + forkNames[forknum], startblock, endblock); + } +} + +/* + * Quicksort comparator for block numbers. + */ +static int +compare_block_numbers(const void *a, const void *b) +{ + BlockNumber aa = *(BlockNumber *) a; + BlockNumber bb = *(BlockNumber *) b; + + if (aa > bb) + return 1; + else if (aa == bb) + return 0; + else + return -1; +} + +/* + * Error callback. + */ +void +walsummary_error_callback(void *callback_arg, char *fmt,...) +{ + va_list ap; + + va_start(ap, fmt); + pg_log_generic_v(PG_LOG_ERROR, PG_LOG_PRIMARY, fmt, ap); + va_end(ap); + + exit(1); +} + +/* + * Read callback. + */ +int +walsummary_read_callback(void *callback_arg, void *data, int length) +{ + ws_file_info *ws = callback_arg; + int rc; + + if ((rc = read(ws->fd, data, length)) < 0) + pg_fatal("could not read file \"%s\": %m", ws->filename); + + return rc; +} + +/* + * help + * + * Prints help page for the program + * + * progname: the name of the executed program, such as "pg_walsummary" + */ +static void +help(const char *progname) +{ + printf(_("%s prints the contents of a WAL summary file.\n\n"), progname); + printf(_("Usage:\n")); + printf(_(" %s [OPTION]... FILE...\n"), progname); + printf(_("\nOptions:\n")); + printf(_(" -i, --individual list block numbers individually, not as ranges\n")); + printf(_(" -q, --quiet don't print anything, just parse the files\n")); + printf(_(" -?, --help show this help, then exit\n")); + + printf(_("\nReport bugs to <%s>.\n"), PACKAGE_BUGREPORT); + printf(_("%s home page: <%s>\n"), PACKAGE_NAME, PACKAGE_URL); +} -- 2.37.1 (Apple Git-137.1) [application/octet-stream] v5-0003-Change-how-a-base-backup-decides-which-files-have.patch (11.4K, ../../CA+TgmoZNVeBzoqDL8xvr-nkaepq815jtDR4nJzPew7=3iEuM1g@mail.gmail.com/4-v5-0003-Change-how-a-base-backup-decides-which-files-have.patch) download | inline diff: From ea3ed36f4767cc9d5bb3edf992f6cf291de59bbc Mon Sep 17 00:00:00 2001 From: Robert Haas <[email protected]> Date: Tue, 3 Oct 2023 13:32:28 -0400 Subject: [PATCH v5 3/6] Change how a base backup decides which files have checksums. Previously, it thought that any plain file located under global, base, or a tablespace directory had checksums unless it was in a short list of excluded files. Now, it thinks that files in those directories have checksums if parse_filename_for_nontemp_relation says that they are relation files. (Temporary relation files don't matter because they're excluded from the backup anyway.) This changes the behavior if you have stray files not managed by PostgreSQL in the relevant directories. Previously, you'd get some kind of checksum-related complaint if such files existed, assuming that the cluster had checksums enabled and that the base backup wasn't run with NOVERIFY_CHECKSUMS. Now, you won't get those complaints any more. That seems like an improvement to me, because those files were presumably not created by PostgreSQL and so there is no reason to think that they would be checksummed like a PostgreSQL relation file. (If we want to complain about such files, we should complain about them existing at all, not just about their checksums.) The point of this change is to make the code more consistent. sendDir() was already calling parse_filename_for_nontemp_relation() as part of an effort to determine which files to include in the backup. So, it already had the information about whether a certain file was a relation file. sendFile() then used a separate method, embodied in is_checksummed_file(), to make what is essentially the same determination. It's better not to make the same decision using two different methods, especially in closely-related code. --- src/backend/backup/basebackup.c | 172 ++++++++++---------------------- 1 file changed, 55 insertions(+), 117 deletions(-) diff --git a/src/backend/backup/basebackup.c b/src/backend/backup/basebackup.c index b537f46219..4ba63ad8a6 100644 --- a/src/backend/backup/basebackup.c +++ b/src/backend/backup/basebackup.c @@ -82,7 +82,8 @@ static int64 sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeo backup_manifest_info *manifest, Oid spcoid); static bool sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, struct stat *statbuf, bool missing_ok, - Oid dboid, Oid spcoid, + Oid dboid, Oid spcoid, RelFileNumber relfilenumber, + unsigned segno, backup_manifest_info *manifest); static off_t read_file_data_into_buffer(bbsink *sink, const char *readfilename, int fd, @@ -104,7 +105,6 @@ static void convert_link_to_directory(const char *pathbuf, struct stat *statbuf) static void perform_base_backup(basebackup_options *opt, bbsink *sink); static void parse_basebackup_options(List *options, basebackup_options *opt); static int compareWalFileNames(const ListCell *a, const ListCell *b); -static bool is_checksummed_file(const char *fullpath, const char *filename); static int basebackup_read_file(int fd, char *buf, size_t nbytes, off_t offset, const char *filename, bool partial_read_ok); @@ -213,23 +213,6 @@ static const struct exclude_list_item excludeFiles[] = {NULL, false} }; -/* - * List of files excluded from checksum validation. - * - * Note: this list should be kept in sync with what pg_checksums.c - * includes. - */ -static const struct exclude_list_item noChecksumFiles[] = { - {"pg_control", false}, - {"pg_filenode.map", false}, - {"pg_internal.init", true}, - {"PG_VERSION", false}, -#ifdef EXEC_BACKEND - {"config_exec_params", true}, -#endif - {NULL, false} -}; - /* * Actually do a base backup for the specified tablespaces. * @@ -356,7 +339,8 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) errmsg("could not stat file \"%s\": %m", XLOG_CONTROL_FILE))); sendFile(sink, XLOG_CONTROL_FILE, XLOG_CONTROL_FILE, &statbuf, - false, InvalidOid, InvalidOid, &manifest); + false, InvalidOid, InvalidOid, + InvalidRelFileNumber, 0, &manifest); } else { @@ -625,7 +609,8 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) errmsg("could not stat file \"%s\": %m", pathbuf))); sendFile(sink, pathbuf, pathbuf, &statbuf, false, - InvalidOid, InvalidOid, &manifest); + InvalidOid, InvalidOid, InvalidRelFileNumber, 0, + &manifest); /* unconditionally mark file as archived */ StatusFilePath(pathbuf, fname, ".done"); @@ -1163,7 +1148,8 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, struct stat statbuf; int64 size = 0; const char *lastDir; /* Split last dir from parent path. */ - bool isDbDir = false; /* Does this directory contain relations? */ + bool isRelationDir = false; /* Does directory contain relations? */ + Oid dboid = InvalidOid; /* * Determine if the current path is a database directory that can contain @@ -1190,17 +1176,23 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, strncmp(lastDir - (sizeof(TABLESPACE_VERSION_DIRECTORY) - 1), TABLESPACE_VERSION_DIRECTORY, sizeof(TABLESPACE_VERSION_DIRECTORY) - 1) == 0)) - isDbDir = true; + { + isRelationDir = true; + dboid = atooid(lastDir + 1); + } } + else if (strcmp(path, "./global") == 0) + isRelationDir = true; dir = AllocateDir(path); while ((de = ReadDir(dir, path)) != NULL) { int excludeIdx; bool excludeFound; - RelFileNumber relNumber; - ForkNumber relForkNum; - unsigned segno; + RelFileNumber relfilenumber = InvalidRelFileNumber; + ForkNumber relForkNum = InvalidForkNumber; + unsigned segno = 0; + bool isRelationFile = false; /* Skip special stuff */ if (strcmp(de->d_name, ".") == 0 || strcmp(de->d_name, "..") == 0) @@ -1248,37 +1240,40 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, if (excludeFound) continue; + /* + * If there could be non-temporary relation files in this directory, + * try to parse the filename. + */ + if (isRelationDir) + isRelationFile = + parse_filename_for_nontemp_relation(de->d_name, + &relfilenumber, + &relForkNum, &segno); + /* Exclude all forks for unlogged tables except the init fork */ - if (isDbDir && - parse_filename_for_nontemp_relation(de->d_name, &relNumber, - &relForkNum, &segno)) + if (isRelationFile && relForkNum != INIT_FORKNUM) { - /* Never exclude init forks */ - if (relForkNum != INIT_FORKNUM) - { - char initForkFile[MAXPGPATH]; + char initForkFile[MAXPGPATH]; - /* - * If any other type of fork, check if there is an init fork - * with the same RelFileNumber. If so, the file can be - * excluded. - */ - snprintf(initForkFile, sizeof(initForkFile), "%s/%u_init", - path, relNumber); + /* + * If any other type of fork, check if there is an init fork with + * the same RelFileNumber. If so, the file can be excluded. + */ + snprintf(initForkFile, sizeof(initForkFile), "%s/%u_init", + path, relfilenumber); - if (lstat(initForkFile, &statbuf) == 0) - { - elog(DEBUG2, - "unlogged relation file \"%s\" excluded from backup", - de->d_name); + if (lstat(initForkFile, &statbuf) == 0) + { + elog(DEBUG2, + "unlogged relation file \"%s\" excluded from backup", + de->d_name); - continue; - } + continue; } } /* Exclude temporary relations */ - if (isDbDir && looks_like_temp_rel_name(de->d_name)) + if (OidIsValid(dboid) && looks_like_temp_rel_name(de->d_name)) { elog(DEBUG2, "temporary relation file \"%s\" excluded from backup", @@ -1417,8 +1412,8 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, if (!sizeonly) sent = sendFile(sink, pathbuf, pathbuf + basepathlen + 1, &statbuf, - true, isDbDir ? atooid(lastDir + 1) : InvalidOid, spcoid, - manifest); + true, dboid, spcoid, + relfilenumber, segno, manifest); if (sent || sizeonly) { @@ -1440,40 +1435,6 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, return size; } -/* - * Check if a file should have its checksum validated. - * We validate checksums on files in regular tablespaces - * (including global and default) only, and in those there - * are some files that are explicitly excluded. - */ -static bool -is_checksummed_file(const char *fullpath, const char *filename) -{ - /* Check that the file is in a tablespace */ - if (strncmp(fullpath, "./global/", 9) == 0 || - strncmp(fullpath, "./base/", 7) == 0 || - strncmp(fullpath, "/", 1) == 0) - { - int excludeIdx; - - /* Compare file against noChecksumFiles skip list */ - for (excludeIdx = 0; noChecksumFiles[excludeIdx].name != NULL; excludeIdx++) - { - int cmplen = strlen(noChecksumFiles[excludeIdx].name); - - if (!noChecksumFiles[excludeIdx].match_prefix) - cmplen++; - if (strncmp(filename, noChecksumFiles[excludeIdx].name, - cmplen) == 0) - return false; - } - - return true; - } - else - return false; -} - /* * Given the member, write the TAR header & send the file. * @@ -1488,6 +1449,7 @@ is_checksummed_file(const char *fullpath, const char *filename) static bool sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, struct stat *statbuf, bool missing_ok, Oid dboid, Oid spcoid, + RelFileNumber relfilenumber, unsigned segno, backup_manifest_info *manifest) { int fd; @@ -1495,8 +1457,6 @@ sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, int checksum_failures = 0; off_t cnt; pgoff_t bytes_done = 0; - int segmentno = 0; - char *segmentpath; bool verify_checksum = false; pg_checksum_context checksum_ctx; @@ -1522,36 +1482,14 @@ sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, */ Assert((sink->bbs_buffer_length % BLCKSZ) == 0); - if (!noverify_checksums && DataChecksumsEnabled()) - { - char *filename; - - /* - * Get the filename (excluding path). As last_dir_separator() - * includes the last directory separator, we chop that off by - * incrementing the pointer. - */ - filename = last_dir_separator(readfilename) + 1; - - if (is_checksummed_file(readfilename, filename)) - { - verify_checksum = true; - - /* - * Cut off at the segment boundary (".") to get the segment number - * in order to mix it into the checksum. - */ - segmentpath = strstr(filename, "."); - if (segmentpath != NULL) - { - segmentno = atoi(segmentpath + 1); - if (segmentno == 0) - ereport(ERROR, - (errmsg("invalid segment number %d in file \"%s\"", - segmentno, filename))); - } - } - } + /* + * If we weren't told not to verify checksums, and if checksums are + * enabled for this cluster, and if this is a relation file, then verify + * the checksum. + */ + if (!noverify_checksums && DataChecksumsEnabled() && + RelFileNumberIsValid(relfilenumber)) + verify_checksum = true; /* * Loop until we read the amount of data the caller told us to expect. The @@ -1566,7 +1504,7 @@ sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, /* Try to read some more data. */ cnt = read_file_data_into_buffer(sink, readfilename, fd, bytes_done, remaining, - blkno + segmentno * RELSEG_SIZE, + blkno + segno * RELSEG_SIZE, verify_checksum, &checksum_failures); -- 2.37.1 (Apple Git-137.1) [application/octet-stream] v5-0002-Change-struct-tablespaceinfo-s-oid-member-from-ch.patch (11.7K, ../../CA+TgmoZNVeBzoqDL8xvr-nkaepq815jtDR4nJzPew7=3iEuM1g@mail.gmail.com/5-v5-0002-Change-struct-tablespaceinfo-s-oid-member-from-ch.patch) download | inline diff: From 8ecb8c9b5fd628b1e1df876518cc35973ce53518 Mon Sep 17 00:00:00 2001 From: Robert Haas <[email protected]> Date: Tue, 3 Oct 2023 13:30:16 -0400 Subject: [PATCH v5 2/6] Change struct tablespaceinfo's oid member from 'char *' to 'Oid' This shouldn't change behavior except in the unusual case where there are file in the tablespace directory that have entirely numeric names but are nevertheless not possible names for a tablespace directory, either because their names have leading zeroes that shouldn't be there, or the value is actually zero, or because the value is too large to represent as an OID. In those cases, the directory would previously have made it into the list of tablespaceinfo objects and no longer will. Thus, base backups will now ignore such directories, instead of treating them as legitimate tablespace directories. Similarly, if entries for such tablespaces occur in a tablespace_map file, they will now be rejected as erroneous, instead of being honored. This is infrastructure for future work that wants to be able to know the tablespace of each relation that is part of a backup *as an OID*. By strengthening the up-front validation, we don't have to worry about weird cases later, and can more easily avoid repeated string->integer conversions. --- src/backend/access/transam/xlog.c | 19 ++++++++++-- src/backend/access/transam/xlogrecovery.c | 12 ++++++-- src/backend/backup/backup_manifest.c | 6 ++-- src/backend/backup/basebackup.c | 35 ++++++++++++----------- src/backend/backup/basebackup_copy.c | 2 +- src/include/backup/backup_manifest.h | 2 +- src/include/backup/basebackup.h | 2 +- 7 files changed, 49 insertions(+), 29 deletions(-) diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index c0e4ca5089..6c724745b5 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -8502,9 +8502,22 @@ do_pg_backup_start(const char *backupidstr, bool fast, List **tablespaces, char *relpath = NULL; char *s; PGFileType de_type; + char *badp; + Oid tsoid; - /* Skip anything that doesn't look like a tablespace */ - if (strspn(de->d_name, "0123456789") != strlen(de->d_name)) + /* + * Try to parse the directory name as an unsigned integer. + * + * Tablespace directories should be positive integers that can be + * represented in 32 bits, with no leading zeroes or trailing + * garbage. If we come across a name that doesn't meet those + * criteria, skip it. + */ + if (de->d_name[0] < '1' || de->d_name[1] > '9') + continue; + errno = 0; + tsoid = strtoul(de->d_name, &badp, 10); + if (*badp != '\0' || errno == EINVAL || errno == ERANGE) continue; snprintf(fullpath, sizeof(fullpath), "pg_tblspc/%s", de->d_name); @@ -8579,7 +8592,7 @@ do_pg_backup_start(const char *backupidstr, bool fast, List **tablespaces, } ti = palloc(sizeof(tablespaceinfo)); - ti->oid = pstrdup(de->d_name); + ti->oid = tsoid; ti->path = pstrdup(linkpath); ti->rpath = relpath; ti->size = -1; diff --git a/src/backend/access/transam/xlogrecovery.c b/src/backend/access/transam/xlogrecovery.c index becc2bda62..5549e1afc5 100644 --- a/src/backend/access/transam/xlogrecovery.c +++ b/src/backend/access/transam/xlogrecovery.c @@ -678,7 +678,7 @@ InitWalRecovery(ControlFileData *ControlFile, bool *wasShutdown_ptr, tablespaceinfo *ti = lfirst(lc); char *linkloc; - linkloc = psprintf("pg_tblspc/%s", ti->oid); + linkloc = psprintf("pg_tblspc/%u", ti->oid); /* * Remove the existing symlink if any and Create the symlink @@ -692,7 +692,6 @@ InitWalRecovery(ControlFileData *ControlFile, bool *wasShutdown_ptr, errmsg("could not create symbolic link \"%s\": %m", linkloc))); - pfree(ti->oid); pfree(ti->path); pfree(ti); } @@ -1341,6 +1340,8 @@ read_tablespace_map(List **tablespaces) { if (!was_backslash && (ch == '\n' || ch == '\r')) { + char *endp; + if (i == 0) continue; /* \r immediately followed by \n */ @@ -1360,7 +1361,12 @@ read_tablespace_map(List **tablespaces) str[n++] = '\0'; ti = palloc0(sizeof(tablespaceinfo)); - ti->oid = pstrdup(str); + errno = 0; + ti->oid = strtoul(str, &endp, 10); + if (*endp != '\0' || errno == EINVAL || errno == ERANGE) + ereport(FATAL, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("invalid data in file \"%s\"", TABLESPACE_MAP))); ti->path = pstrdup(str + n); *tablespaces = lappend(*tablespaces, ti); diff --git a/src/backend/backup/backup_manifest.c b/src/backend/backup/backup_manifest.c index cee6216524..aeed362a9a 100644 --- a/src/backend/backup/backup_manifest.c +++ b/src/backend/backup/backup_manifest.c @@ -97,7 +97,7 @@ FreeBackupManifest(backup_manifest_info *manifest) * Add an entry to the backup manifest for a file. */ void -AddFileToBackupManifest(backup_manifest_info *manifest, const char *spcoid, +AddFileToBackupManifest(backup_manifest_info *manifest, Oid spcoid, const char *pathname, size_t size, pg_time_t mtime, pg_checksum_context *checksum_ctx) { @@ -114,9 +114,9 @@ AddFileToBackupManifest(backup_manifest_info *manifest, const char *spcoid, * pathname relative to the data directory (ignoring the intermediate * symlink traversal). */ - if (spcoid != NULL) + if (OidIsValid(spcoid)) { - snprintf(pathbuf, sizeof(pathbuf), "pg_tblspc/%s/%s", spcoid, + snprintf(pathbuf, sizeof(pathbuf), "pg_tblspc/%u/%s", spcoid, pathname); pathname = pathbuf; } diff --git a/src/backend/backup/basebackup.c b/src/backend/backup/basebackup.c index b126d9c890..b537f46219 100644 --- a/src/backend/backup/basebackup.c +++ b/src/backend/backup/basebackup.c @@ -75,14 +75,15 @@ typedef struct pg_checksum_type manifest_checksum_type; } basebackup_options; -static int64 sendTablespace(bbsink *sink, char *path, char *spcoid, bool sizeonly, +static int64 sendTablespace(bbsink *sink, char *path, Oid spcoid, bool sizeonly, struct backup_manifest_info *manifest); static int64 sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, List *tablespaces, bool sendtblspclinks, - backup_manifest_info *manifest, const char *spcoid); + backup_manifest_info *manifest, Oid spcoid); static bool sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, - struct stat *statbuf, bool missing_ok, Oid dboid, - backup_manifest_info *manifest, const char *spcoid); + struct stat *statbuf, bool missing_ok, + Oid dboid, Oid spcoid, + backup_manifest_info *manifest); static off_t read_file_data_into_buffer(bbsink *sink, const char *readfilename, int fd, off_t offset, size_t length, @@ -305,7 +306,7 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) if (tmp->path == NULL) tmp->size = sendDir(sink, ".", 1, true, state.tablespaces, - true, NULL, NULL); + true, NULL, InvalidOid); else tmp->size = sendTablespace(sink, tmp->path, tmp->oid, true, NULL); @@ -346,7 +347,7 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) /* Then the bulk of the files... */ sendDir(sink, ".", 1, false, state.tablespaces, - sendtblspclinks, &manifest, NULL); + sendtblspclinks, &manifest, InvalidOid); /* ... and pg_control after everything else. */ if (lstat(XLOG_CONTROL_FILE, &statbuf) != 0) @@ -355,11 +356,11 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) errmsg("could not stat file \"%s\": %m", XLOG_CONTROL_FILE))); sendFile(sink, XLOG_CONTROL_FILE, XLOG_CONTROL_FILE, &statbuf, - false, InvalidOid, &manifest, NULL); + false, InvalidOid, InvalidOid, &manifest); } else { - char *archive_name = psprintf("%s.tar", ti->oid); + char *archive_name = psprintf("%u.tar", ti->oid); bbsink_begin_archive(sink, archive_name); @@ -623,8 +624,8 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) (errcode_for_file_access(), errmsg("could not stat file \"%s\": %m", pathbuf))); - sendFile(sink, pathbuf, pathbuf, &statbuf, false, InvalidOid, - &manifest, NULL); + sendFile(sink, pathbuf, pathbuf, &statbuf, false, + InvalidOid, InvalidOid, &manifest); /* unconditionally mark file as archived */ StatusFilePath(pathbuf, fname, ".done"); @@ -1087,7 +1088,7 @@ sendFileWithContent(bbsink *sink, const char *filename, const char *content, _tarWritePadding(sink, len); - AddFileToBackupManifest(manifest, NULL, filename, len, + AddFileToBackupManifest(manifest, InvalidOid, filename, len, (pg_time_t) statbuf.st_mtime, &checksum_ctx); } @@ -1099,7 +1100,7 @@ sendFileWithContent(bbsink *sink, const char *filename, const char *content, * Only used to send auxiliary tablespaces, not PGDATA. */ static int64 -sendTablespace(bbsink *sink, char *path, char *spcoid, bool sizeonly, +sendTablespace(bbsink *sink, char *path, Oid spcoid, bool sizeonly, backup_manifest_info *manifest) { int64 size; @@ -1154,7 +1155,7 @@ sendTablespace(bbsink *sink, char *path, char *spcoid, bool sizeonly, static int64 sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, List *tablespaces, bool sendtblspclinks, backup_manifest_info *manifest, - const char *spcoid) + Oid spcoid) { DIR *dir; struct dirent *de; @@ -1416,8 +1417,8 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, if (!sizeonly) sent = sendFile(sink, pathbuf, pathbuf + basepathlen + 1, &statbuf, - true, isDbDir ? atooid(lastDir + 1) : InvalidOid, - manifest, spcoid); + true, isDbDir ? atooid(lastDir + 1) : InvalidOid, spcoid, + manifest); if (sent || sizeonly) { @@ -1486,8 +1487,8 @@ is_checksummed_file(const char *fullpath, const char *filename) */ static bool sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, - struct stat *statbuf, bool missing_ok, Oid dboid, - backup_manifest_info *manifest, const char *spcoid) + struct stat *statbuf, bool missing_ok, Oid dboid, Oid spcoid, + backup_manifest_info *manifest) { int fd; BlockNumber blkno = 0; diff --git a/src/backend/backup/basebackup_copy.c b/src/backend/backup/basebackup_copy.c index fee30c21e1..3bdbe1f989 100644 --- a/src/backend/backup/basebackup_copy.c +++ b/src/backend/backup/basebackup_copy.c @@ -407,7 +407,7 @@ SendTablespaceList(List *tablespaces) } else { - values[0] = ObjectIdGetDatum(strtoul(ti->oid, NULL, 10)); + values[0] = ObjectIdGetDatum(ti->oid); values[1] = CStringGetTextDatum(ti->path); } if (ti->size >= 0) diff --git a/src/include/backup/backup_manifest.h b/src/include/backup/backup_manifest.h index d41b439980..5a481dbcf5 100644 --- a/src/include/backup/backup_manifest.h +++ b/src/include/backup/backup_manifest.h @@ -39,7 +39,7 @@ extern void InitializeBackupManifest(backup_manifest_info *manifest, backup_manifest_option want_manifest, pg_checksum_type manifest_checksum_type); extern void AddFileToBackupManifest(backup_manifest_info *manifest, - const char *spcoid, + Oid spcoid, const char *pathname, size_t size, pg_time_t mtime, pg_checksum_context *checksum_ctx); diff --git a/src/include/backup/basebackup.h b/src/include/backup/basebackup.h index 3e68abc2bb..1432d9c206 100644 --- a/src/include/backup/basebackup.h +++ b/src/include/backup/basebackup.h @@ -27,7 +27,7 @@ */ typedef struct { - char *oid; /* tablespace's OID, as a decimal string */ + Oid oid; /* tablespace's OID */ char *path; /* full path to tablespace's directory */ char *rpath; /* relative path if it's within PGDATA, else * NULL */ -- 2.37.1 (Apple Git-137.1) [application/octet-stream] v5-0005-Prototype-patch-for-incremental-backup.patch (335.4K, ../../CA+TgmoZNVeBzoqDL8xvr-nkaepq815jtDR4nJzPew7=3iEuM1g@mail.gmail.com/6-v5-0005-Prototype-patch-for-incremental-backup.patch) download | inline diff: From 7b150a8eb4ec49792535768ac1b5bafe42840c35 Mon Sep 17 00:00:00 2001 From: Robert Haas <[email protected]> Date: Wed, 14 Jun 2023 11:31:29 -0400 Subject: [PATCH v5 5/6] Prototype patch for incremental backup. We don't differentiate between incremental and differential backups; an incremental backup can be based either on a full backup or on a previous incremental backup. This adds a new background process, the WAL summarizer, whose behavor is governed by new GUCs wal_summarize_mb and wal_summarize_keep_time. This writes out WAL summary files to $PGDATA/pg_wal/summaries. Each summary file contains information for a certain range of LSNs on a certain TLI. For each relation, it stores a "limit block" which is 0 if a relation is created or destroyed within a certain range of WAL records, or otherwise the shortest length to which the relation was truncated during that range of WAL records, or otherwise InvalidBlockNumber. In addition, it stores any blocks which have been modified during that range of WAL records, but excluding blocks which were removed by truncation after they were modified and which were never modified thereafter. In other words, it tells us which blocks need to copied in case of an incremental backup covering that range of WAL records. To take an incremental backup, you use the new replication command UPLOAD_MANIFEST to upload the manifest for the prior backup. This prior backup could either be a full backup or another incremental backup. You then use BASE_BACKUP with the INCREMENTAL option to take the backup. pg_basebackup now has an --incremental=PATH_TO_MANIFEST option to trigger this behavior. An incremental backup is like a regular full backup except that some relation files are replaced with files with names like INCREMENTAL.${ORIGINAL_NAME}, and the backup_label file contains additional lines identifying it as an incremental backup. The new pg_combinebackup tool can be used to reconstruct a data directory from a full backup and a series of incremental backups. Open issues: - Needs some rework once XLOG_CHECKPOINT_REDO patch is committed. - How should we control generation and retention of summary files? What should be the defaults? - Needs to be tested on a standby. - Should we send the whole backup manifest to the server or, say, just an LSN? - Should the timeout when waiting for WAL summaries be configurable? - It would be nice (but not essential) to do something about incremental JSON parsing. - Might need more tests. Patch by me. Thanks to Dilip Kumar and Andres Freund for some helpful design discussions. Reviewed by Dilip Kumar and Jakub Wartak. --- doc/src/sgml/backup.sgml | 89 +- doc/src/sgml/ref/allfiles.sgml | 1 + doc/src/sgml/ref/pg_basebackup.sgml | 37 +- doc/src/sgml/ref/pg_combinebackup.sgml | 228 +++ doc/src/sgml/reference.sgml | 1 + src/backend/access/transam/xlog.c | 93 +- src/backend/access/transam/xlogbackup.c | 10 + src/backend/access/transam/xlogrecovery.c | 6 + src/backend/backup/Makefile | 5 +- src/backend/backup/basebackup.c | 334 +++- src/backend/backup/basebackup_incremental.c | 873 ++++++++++ src/backend/backup/meson.build | 3 + src/backend/backup/walsummary.c | 356 +++++ src/backend/backup/walsummaryfuncs.c | 169 ++ src/backend/postmaster/Makefile | 1 + src/backend/postmaster/auxprocess.c | 8 + src/backend/postmaster/meson.build | 1 + src/backend/postmaster/postmaster.c | 53 + src/backend/postmaster/walsummarizer.c | 1414 +++++++++++++++++ src/backend/replication/repl_gram.y | 14 +- src/backend/replication/repl_scanner.l | 2 + src/backend/replication/walsender.c | 162 +- src/backend/storage/ipc/ipci.c | 3 + src/backend/storage/lmgr/lwlocknames.txt | 1 + src/backend/utils/activity/pgstat_io.c | 4 +- .../utils/activity/wait_event_names.txt | 5 + src/backend/utils/init/miscinit.c | 3 + src/backend/utils/misc/guc_tables.c | 29 + src/backend/utils/misc/postgresql.conf.sample | 5 + src/bin/Makefile | 1 + src/bin/initdb/initdb.c | 1 + src/bin/meson.build | 1 + src/bin/pg_basebackup/bbstreamer_file.c | 1 + src/bin/pg_basebackup/pg_basebackup.c | 110 +- src/bin/pg_basebackup/t/010_pg_basebackup.pl | 4 +- src/bin/pg_combinebackup/.gitignore | 1 + src/bin/pg_combinebackup/Makefile | 52 + src/bin/pg_combinebackup/backup_label.c | 281 ++++ src/bin/pg_combinebackup/backup_label.h | 29 + src/bin/pg_combinebackup/copy_file.c | 169 ++ src/bin/pg_combinebackup/copy_file.h | 19 + src/bin/pg_combinebackup/load_manifest.c | 245 +++ src/bin/pg_combinebackup/load_manifest.h | 67 + src/bin/pg_combinebackup/meson.build | 35 + src/bin/pg_combinebackup/pg_combinebackup.c | 1270 +++++++++++++++ src/bin/pg_combinebackup/reconstruct.c | 618 +++++++ src/bin/pg_combinebackup/reconstruct.h | 32 + src/bin/pg_combinebackup/t/001_basic.pl | 23 + .../pg_combinebackup/t/002_compare_backups.pl | 154 ++ src/bin/pg_combinebackup/write_manifest.c | 293 ++++ src/bin/pg_combinebackup/write_manifest.h | 33 + src/bin/pg_resetwal/pg_resetwal.c | 36 + src/common/Makefile | 1 + src/common/blkreftable.c | 1309 +++++++++++++++ src/common/meson.build | 1 + src/include/access/xlog.h | 1 + src/include/access/xlogbackup.h | 2 + src/include/backup/basebackup.h | 5 +- src/include/backup/basebackup_incremental.h | 56 + src/include/backup/walsummary.h | 49 + src/include/catalog/pg_proc.dat | 19 + src/include/common/blkreftable.h | 120 ++ src/include/miscadmin.h | 3 + src/include/nodes/replnodes.h | 9 + src/include/postmaster/walsummarizer.h | 31 + src/include/storage/proc.h | 9 +- src/include/utils/guc_tables.h | 1 + src/test/perl/PostgreSQL/Test/Cluster.pm | 21 +- src/test/recovery/t/001_stream_rep.pl | 2 + src/test/recovery/t/019_replslot_limit.pl | 3 + .../t/035_standby_logical_decoding.pl | 1 + src/tools/pgindent/typedefs.list | 23 + 72 files changed, 8981 insertions(+), 70 deletions(-) create mode 100644 doc/src/sgml/ref/pg_combinebackup.sgml create mode 100644 src/backend/backup/basebackup_incremental.c create mode 100644 src/backend/backup/walsummary.c create mode 100644 src/backend/backup/walsummaryfuncs.c create mode 100644 src/backend/postmaster/walsummarizer.c create mode 100644 src/bin/pg_combinebackup/.gitignore create mode 100644 src/bin/pg_combinebackup/Makefile create mode 100644 src/bin/pg_combinebackup/backup_label.c create mode 100644 src/bin/pg_combinebackup/backup_label.h create mode 100644 src/bin/pg_combinebackup/copy_file.c create mode 100644 src/bin/pg_combinebackup/copy_file.h create mode 100644 src/bin/pg_combinebackup/load_manifest.c create mode 100644 src/bin/pg_combinebackup/load_manifest.h create mode 100644 src/bin/pg_combinebackup/meson.build create mode 100644 src/bin/pg_combinebackup/pg_combinebackup.c create mode 100644 src/bin/pg_combinebackup/reconstruct.c create mode 100644 src/bin/pg_combinebackup/reconstruct.h create mode 100644 src/bin/pg_combinebackup/t/001_basic.pl create mode 100644 src/bin/pg_combinebackup/t/002_compare_backups.pl create mode 100644 src/bin/pg_combinebackup/write_manifest.c create mode 100644 src/bin/pg_combinebackup/write_manifest.h create mode 100644 src/common/blkreftable.c create mode 100644 src/include/backup/basebackup_incremental.h create mode 100644 src/include/backup/walsummary.h create mode 100644 src/include/common/blkreftable.h create mode 100644 src/include/postmaster/walsummarizer.h diff --git a/doc/src/sgml/backup.sgml b/doc/src/sgml/backup.sgml index 8cb24d6ae5..b3468eea3c 100644 --- a/doc/src/sgml/backup.sgml +++ b/doc/src/sgml/backup.sgml @@ -857,12 +857,79 @@ test ! -f /mnt/server/archivedir/00000001000000A900000065 && cp pg_wal/0 </para> </sect2> + <sect2 id="backup-incremental-backup"> + <title>Making an Incremental Backup</title> + + <para> + You can use <xref linkend="app-pgbasebackup"/> to take an incremental + backup by specifying the <literal>--incremental</literal> option. You must + supply, as an argument to <literal>--incremental</literal>, the backup + manifest to an earlier backup from the same server. In the resulting + backup, non-relation files will be included in their entirety, but some + relation files may be replaced by smaller incremental files which contain + only the blocks which have been changed since the earlier backup and enough + metadata to reconstruct the current version of the file. + </para> + + <para> + To figure out which blocks need to be backed up, the server uses WAL + summaries, which are stored in the data directory, inside the directory + <literal>pg_wal/summaries</literal>. If the required summary files are not + present, an attempt to take an incremental backup will fail. The summaries + present in this directory must cover all LSNs from the start LSN of the + prior backup to the start LSN of the current backup. Since the server looks + for WAL summaries just after establishing the start LSN of the current + backup, the necessary summary files probably won't be instantly present + on disk, but the server will wait for any missing files to show up. + This also helps if the WAL summarization process has fallen behind. + However, if the necessary files have already been removed, or if the WAL + summarizer doesn't catch up quickly enough, the incremental backup will + fail. + </para> + + <para> + When restoring an incremental backup, it will be necessary to have not + only the incremental backup itself but also all earlier backups that + are required to supply the blocks omitted from the incremental backup. + See <xref linkend="app-pgcombinebackup"/> for further information about + this requirement. + </para> + + <para> + Note that all of the requirements for making use of a full backup also + apply to an incremental backup. For instance, you still need all of the + WAL segment files generated during and after the file system backup, and + any relevant WAL history files. And you still need to create a + <literal>recovery.signal</literal> (or <literal>standby.signal</literal>) + and perform recovery, as described in + <xref linkend="backup-pitr-recovery" />. The requirement to have earlier + backups available at restore time and to use + <literal>pg_combinebackup</literal> is an additional requirement on top of + everything else. Keep in mind that <application>PostgreSQL</application> + has no built-in mechanism to figure out which backups are still needed as + a basis for restoring later incremental backups. You must keep track of + the relationships between your full and incremental backups on your own, + and be certain not to remove earlier backups if they might be needed when + restoring later incremental backups. + </para> + + <para> + Incremental backups typically only make sense for relatively large + databases where a significant portion of the data does not change, or only + changes slowly. For a small database, it's simpler to ignore the existence + of incremental backups and simply take full backups, which are simpler + to manage. For a large database all of which is heavily modified, + incremental backups won't be much smaller than full backups. + </para> + </sect2> + <sect2 id="backup-lowlevel-base-backup"> <title>Making a Base Backup Using the Low Level API</title> <para> - The procedure for making a base backup using the low level - APIs contains a few more steps than - the <xref linkend="app-pgbasebackup"/> method, but is relatively + Instead of taking a full or incremental base backup using + <xref linkend="app-pgbasebackup"/>, you can take a base backup using the + low-level API. This procedure contains a few more steps than + the <application>pg_basebackup</application> method, but is relatively simple. It is very important that these steps are executed in sequence, and that the success of a step is verified before proceeding to the next step. @@ -1118,7 +1185,8 @@ SELECT * FROM pg_backup_stop(wait_for_archive => true); </listitem> <listitem> <para> - Restore the database files from your file system backup. Be sure that they + If you're restoring a full backup, you can restore the database files + directly into the target directories. Be sure that they are restored with the right ownership (the database system user, not <literal>root</literal>!) and with the right permissions. If you are using tablespaces, @@ -1126,6 +1194,19 @@ SELECT * FROM pg_backup_stop(wait_for_archive => true); were correctly restored. </para> </listitem> + <listitem> + <para> + If you're restoring an incremental backup, you'll need to restore the + incremental backup and all earlier backups upon which it directly or + indirectly depends to the machine where you are performing the restore. + These backups will need to be placed in separate directories, not the + target directories where you want the running server to end up. + Once this is done, use <xref linkend="app-pgcombinebackup"/> to pull + data from the full backup and all of the subsequent incremental backups + and write out a synthetic full backup to the target directories. As above, + verify that permissions and tablespace links are correct. + </para> + </listitem> <listitem> <para> Remove any files present in <filename>pg_wal/</filename>; these came from the diff --git a/doc/src/sgml/ref/allfiles.sgml b/doc/src/sgml/ref/allfiles.sgml index 54b5f22d6e..fda4690eab 100644 --- a/doc/src/sgml/ref/allfiles.sgml +++ b/doc/src/sgml/ref/allfiles.sgml @@ -202,6 +202,7 @@ Complete list of usable sgml source files in this directory. <!ENTITY pgBasebackup SYSTEM "pg_basebackup.sgml"> <!ENTITY pgbench SYSTEM "pgbench.sgml"> <!ENTITY pgChecksums SYSTEM "pg_checksums.sgml"> +<!ENTITY pgCombinebackup SYSTEM "pg_combinebackup.sgml"> <!ENTITY pgConfig SYSTEM "pg_config-ref.sgml"> <!ENTITY pgControldata SYSTEM "pg_controldata.sgml"> <!ENTITY pgCtl SYSTEM "pg_ctl-ref.sgml"> diff --git a/doc/src/sgml/ref/pg_basebackup.sgml b/doc/src/sgml/ref/pg_basebackup.sgml index 712568a62d..50536d0521 100644 --- a/doc/src/sgml/ref/pg_basebackup.sgml +++ b/doc/src/sgml/ref/pg_basebackup.sgml @@ -38,11 +38,25 @@ PostgreSQL documentation </para> <para> - <application>pg_basebackup</application> makes an exact copy of the database - cluster's files, while making sure the server is put into and - out of backup mode automatically. Backups are always taken of the entire - database cluster; it is not possible to back up individual databases or - database objects. For selective backups, another tool such as + <application>pg_basebackup</application> can take a full or incremental + base backup of the database. When used to take a full backup, it makes an + exact copy of the database cluster's files. When used to take an incremental + backup, some files that would have been part of a full backup may be + replaced with incremental versions of the same files, containing only those + blocks that have been modified since the reference backup. An incremental + backup cannot be used directly; instead, + <xref linkend="app-pgcombinebackup"/> must first + be used to combine it with the previous backups upon which it depends. + See <xref linkend="backup-incremental-backup" /> for more information + about incremental backups, and <xref linkend="backup-pitr-recovery" /> + for steps to recover from a backup. + </para> + + <para> + In any mode, <application>pg_basebackup</application> makes sure the server + is put into and out of backup mode automatically. Backups are always taken of + the entire database cluster; it is not possible to back up individual + databases or database objects. For selective backups, another tool such as <xref linkend="app-pgdump"/> must be used. </para> @@ -197,6 +211,19 @@ PostgreSQL documentation </listitem> </varlistentry> + <varlistentry> + <term><option>-i <replaceable class="parameter">old_manifest_file</replaceable></option></term> + <term><option>--incremental=<replaceable class="parameter">old_meanifest_file</replaceable></option></term> + <listitem> + <para> + Performs an <link linkend="backup-incremental-backup">incremental + backup</link>. The backup manifest for the reference + backup must be provided, and will be uploaded to the server, which will + respond by sending the requested incremental backup. + </para> + </listitem> + </varlistentry> + <varlistentry> <term><option>-R</option></term> <term><option>--write-recovery-conf</option></term> diff --git a/doc/src/sgml/ref/pg_combinebackup.sgml b/doc/src/sgml/ref/pg_combinebackup.sgml new file mode 100644 index 0000000000..6cac73573f --- /dev/null +++ b/doc/src/sgml/ref/pg_combinebackup.sgml @@ -0,0 +1,228 @@ +<!-- +doc/src/sgml/ref/pg_combinebackup.sgml +PostgreSQL documentation +--> + +<refentry id="app-pgcombinebackup"> + <indexterm zone="app-pgcombinebackup"> + <primary>pg_combinebackup</primary> + </indexterm> + + <refmeta> + <refentrytitle><application>pg_combinebackup</application></refentrytitle> + <manvolnum>1</manvolnum> + <refmiscinfo>Application</refmiscinfo> + </refmeta> + + <refnamediv> + <refname>pg_combinebackup</refname> + <refpurpose>reconstruct a full backup from an incremental backup and dependent backups</refpurpose> + </refnamediv> + + <refsynopsisdiv> + <cmdsynopsis> + <command>pg_combinebackup</command> + <arg rep="repeat"><replaceable>option</replaceable></arg> + <arg rep="repeat"><replaceable>backup_directory</replaceable></arg> + </cmdsynopsis> + </refsynopsisdiv> + + <refsect1> + <title>Description</title> + <para> + <application>pg_combinebackup</application> is used to reconstruct a + synthetic full backup from an + <link linkend="backup-incremental-backup">incremental backup</link> and the + earlier backups upon which it depends. + </para> + + <para> + Specify all of the required backups on the command line from oldest to newest. + That is, the first backup directory should be the path to the full backup, and + the last should be the path to the final incremental backup + that you wish to restore. The reconstructed backup will be written to the + output directory specified by the <option>-o</option> option. + </para> + + <para> + Although <application>pg_combinebackup</application> will attempt to verify + that the backups you specify form a legal backup chain from which a correct + full backup can be reconstructed, it is not designed to help you keep track + of which backups depend on which other backups. If you remove the one or + more of the previous backups upon which your incremental + backup relies, you will not be able to restore it. + </para> + + <para> + Since the output of <application>pg_combinebackup</application> is a + synthetic full backup, it can be used as an input to a future invocation of + <application>pg_combinebackup</application>. The synthetic full backup would + be specified on the command line in lieu of the chain of backups from which + it was reconstructed. + </para> + </refsect1> + + <refsect1> + <title>Options</title> + + <para> + <variablelist> + <varlistentry> + <term><option>-d</option></term> + <term><option>--debug</option></term> + <listitem> + <para> + Print lots of debug logging output on <filename>stderr</filename>. + </para> + </listitem> + </varlistentry> + + <varlistentry> + <term><option>-T <replaceable class="parameter">olddir</replaceable>=<replaceable class="parameter">newdir</replaceable></option></term> + <term><option>--tablespace-mapping=<replaceable class="parameter">olddir</replaceable>=<replaceable class="parameter">newdir</replaceable></option></term> + <listitem> + <para> + Relocates the tablespace in directory <replaceable>olddir</replaceable> + to <replaceable>newdir</replaceable> during the backup. + <replaceable>olddir</replaceable> is the absolute path of the tablespace + as it exists in the first backup specified on the command line, + and <replaceable>newdir</replaceable> is the absolute path to use for the + tablespace in the reconstructed backup. If either path needs to contain + an equal sign (<literal>=</literal>), precede that with a backslash. + This option can be specified multiple times for multiple tablespaces. + </para> + </listitem> + </varlistentry> + + <varlistentry> + <term><option>-N</option></term> + <term><option>--no-sync</option></term> + <listitem> + <para> + By default, <command>pg_combinebackup</command> will wait for all files + to be written safely to disk. This option causes + <command>pg_combinebackup</command> to return without waiting, which is + faster, but means that a subsequent operating system crash can leave + the output backup corrupt. Generally, this option is useful for testing + but should not be used when creating a production installation. + </para> + </listitem> + </varlistentry> + + <varlistentry> + <term><option>-o <replaceable class="parameter">outputdir</replaceable></option></term> + <term><option>--output=<replaceable class="parameter">outputdir</replaceable></option></term> + <listitem> + <para> + Specifies the output directory to which the synthetic full backup + should be written. Currently, this argument is required. + </para> + </listitem> + </varlistentry> + + <varlistentry> + <term><option>--sync-method</option></term> + <listitem> + <para> + When set to <literal>fsync</literal>, which is the default, + <command>pg_combinebackup</command> will recursively open and synchronize + all files in the backup directory. When the plain format is used, the + search for files will follow symbolic links for the WAL directory and + each configured tablespace. + </para> + <para> + On Linux, <literal>syncfs</literal> may be used instead to ask the + operating system to synchronize the whole file system that contains the + backup directory. When the plain format is used, + <command>pg_combinebackup</command> will also synchronize the file systems + that contain the WAL files and each tablespace. See + <xref linkend="syncfs"/> for more information about using + <function>syncfs()</function>. + </para> + <para> + This option has no effect when <option>--no-sync</option> is used. + </para> + </listitem> + </varlistentry> + + <varlistentry> + <term><option>--manifest-checksums=<replaceable class="parameter">algorithm</replaceable></option></term> + <listitem> + <para> + Like <xref linkend="app-pgbasebackup"/>, + <application>pg_combinebackup</application> writes a backup manifest + in the output directory. This option specifies the checksum algorithm + that should be applied to each file included in the backup manifest. + Currently, the available algorithms are <literal>NONE</literal>, + <literal>CRC32C</literal>, <literal>SHA224</literal>, + <literal>SHA256</literal>, <literal>SHA384</literal>, + and <literal>SHA512</literal>. The default is <literal>CRC32C</literal>. + </para> + </listitem> + </varlistentry> + + <varlistentry> + <term><option>--no-manifest</option></term> + <listitem> + <para> + Disables generation of a backup manifest. If this option is not + specified, a backup manifest for the reconstructed backup will be + written to the output directory. + </para> + </listitem> + </varlistentry> + </variablelist> + + <variablelist> + <varlistentry> + <term><option>-V</option></term> + <term><option>--version</option></term> + <listitem> + <para> + Prints the <application>pg_combinebackup</application> version and exits. + </para> + </listitem> + </varlistentry> + + <varlistentry> + <term><option>-?</option></term> + <term><option>--help</option></term> + <listitem> + <para> + Shows help about <application>pg_combinebackup</application> command line + arguments, and exits. + </para> + </listitem> + </varlistentry> + + </variablelist> + </para> + + </refsect1> + + <refsect1> + <title>Environment</title> + + <para> + This utility, like most other <productname>PostgreSQL</productname> utilities, + uses the environment variables supported by <application>libpq</application> + (see <xref linkend="libpq-envars"/>). + </para> + + <para> + The environment variable <envar>PG_COLOR</envar> specifies whether to use + color in diagnostic messages. Possible values are + <literal>always</literal>, <literal>auto</literal> and + <literal>never</literal>. + </para> + </refsect1> + + <refsect1> + <title>See Also</title> + + <simplelist type="inline"> + <member><xref linkend="app-pgbasebackup"/></member> + </simplelist> + </refsect1> + +</refentry> diff --git a/doc/src/sgml/reference.sgml b/doc/src/sgml/reference.sgml index e11b4b6130..a07d2b5e01 100644 --- a/doc/src/sgml/reference.sgml +++ b/doc/src/sgml/reference.sgml @@ -250,6 +250,7 @@ &pgamcheck; &pgBasebackup; &pgbench; + &pgCombinebackup; &pgConfig; &pgDump; &pgDumpall; diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 6c724745b5..d6f3cddfa0 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -77,6 +77,7 @@ #include "port/pg_iovec.h" #include "postmaster/bgwriter.h" #include "postmaster/startup.h" +#include "postmaster/walsummarizer.h" #include "postmaster/walwriter.h" #include "replication/logical.h" #include "replication/origin.h" @@ -3514,6 +3515,43 @@ XLogGetLastRemovedSegno(void) return lastRemovedSegNo; } +/* + * Return the oldest WAL segment on the given TLI that still exists in + * XLOGDIR, or 0 if none. + */ +XLogSegNo +XLogGetOldestSegno(TimeLineID tli) +{ + DIR *xldir; + struct dirent *xlde; + XLogSegNo oldest_segno = 0; + + xldir = AllocateDir(XLOGDIR); + while ((xlde = ReadDir(xldir, XLOGDIR)) != NULL) + { + TimeLineID file_tli; + XLogSegNo file_segno; + + /* Ignore files that are not XLOG segments */ + if (!IsXLogFileName(xlde->d_name)) + continue; + + /* Parse filename to get TLI and segno. */ + XLogFromFileName(xlde->d_name, &file_tli, &file_segno, + wal_segment_size); + + /* Ignore anything that's not from the TLI of interest. */ + if (tli != file_tli) + continue; + + /* If it's the oldest so far, update oldest_segno. */ + if (oldest_segno == 0 || file_segno < oldest_segno) + oldest_segno = file_segno; + } + + FreeDir(xldir); + return oldest_segno; +} /* * Update the last removed segno pointer in shared memory, to reflect that the @@ -3793,8 +3831,8 @@ RemoveXlogFile(const struct dirent *segment_de, } /* - * Verify whether pg_wal and pg_wal/archive_status exist. - * If the latter does not exist, recreate it. + * Verify whether pg_wal, pg_wal/archive_status, and pg_wal/summaries exist. + * If the latter do not exist, recreate them. * * It is not the goal of this function to verify the contents of these * directories, but to help in cases where someone has performed a cluster @@ -3837,6 +3875,26 @@ ValidateXLOGDirectoryStructure(void) (errmsg("could not create missing directory \"%s\": %m", path))); } + + /* Check for summaries */ + snprintf(path, MAXPGPATH, XLOGDIR "/summaries"); + if (stat(path, &stat_buf) == 0) + { + /* Check for weird cases where it exists but isn't a directory */ + if (!S_ISDIR(stat_buf.st_mode)) + ereport(FATAL, + (errmsg("required WAL directory \"%s\" does not exist", + path))); + } + else + { + ereport(LOG, + (errmsg("creating missing WAL directory \"%s\"", path))); + if (MakePGDirectory(path) < 0) + ereport(FATAL, + (errmsg("could not create missing directory \"%s\": %m", + path))); + } } /* @@ -5161,9 +5219,9 @@ StartupXLOG(void) #endif /* - * Verify that pg_wal and pg_wal/archive_status exist. In cases where - * someone has performed a copy for PITR, these directories may have been - * excluded and need to be re-created. + * Verify that pg_wal, pg_wal/archive_status, and pg_wal/summaries exist. + * In cases where someone has performed a copy for PITR, these directories + * may have been excluded and need to be re-created. */ ValidateXLOGDirectoryStructure(); @@ -6848,6 +6906,17 @@ CreateCheckPoint(int flags) */ END_CRIT_SECTION(); + /* + * If there hasn't been much system activity in a while, the WAL + * summarizer may be sleeping for relatively long periods, which could + * delay an incremental backup that has started concurrently. In the hopes + * of avoiding that, poke the WAL summarizer here. + * + * Possibly this should instead be done at some earlier point in this + * function, but it's not clear that it matters much. + */ + SetWalSummarizerLatch(); + /* * Let smgr do post-checkpoint cleanup (eg, deleting old files). */ @@ -7522,6 +7591,20 @@ KeepLogSeg(XLogRecPtr recptr, XLogSegNo *logSegNo) } } + /* + * If WAL summarization is in use, don't remove WAL that has yet to be + * summarized. + */ + keep = GetOldestUnsummarizedLSN(NULL, NULL); + if (keep != InvalidXLogRecPtr) + { + XLogSegNo unsummarized_segno; + + XLByteToSeg(keep, unsummarized_segno, wal_segment_size); + if (unsummarized_segno < segno) + segno = unsummarized_segno; + } + /* but, keep at least wal_keep_size if that's set */ if (wal_keep_size_mb > 0) { diff --git a/src/backend/access/transam/xlogbackup.c b/src/backend/access/transam/xlogbackup.c index 21d68133ae..f51d4282bb 100644 --- a/src/backend/access/transam/xlogbackup.c +++ b/src/backend/access/transam/xlogbackup.c @@ -77,6 +77,16 @@ build_backup_content(BackupState *state, bool ishistoryfile) appendStringInfo(result, "STOP TIMELINE: %u\n", state->stoptli); } + /* either both istartpoint and istarttli should be set, or neither */ + Assert(XLogRecPtrIsInvalid(state->istartpoint) == (state->istarttli == 0)); + if (!XLogRecPtrIsInvalid(state->istartpoint)) + { + appendStringInfo(result, "INCREMENTAL FROM LSN: %X/%X\n", + LSN_FORMAT_ARGS(state->istartpoint)); + appendStringInfo(result, "INCREMENTAL FROM TLI: %u\n", + state->istarttli); + } + data = result->data; pfree(result); diff --git a/src/backend/access/transam/xlogrecovery.c b/src/backend/access/transam/xlogrecovery.c index 5549e1afc5..89ddec5bf9 100644 --- a/src/backend/access/transam/xlogrecovery.c +++ b/src/backend/access/transam/xlogrecovery.c @@ -1284,6 +1284,12 @@ read_backup_label(XLogRecPtr *checkPointLoc, TimeLineID *backupLabelTLI, tli_from_file, BACKUP_LABEL_FILE))); } + if (fscanf(lfp, "INCREMENTAL FROM LSN: %X/%X\n", &hi, &lo) > 0) + ereport(FATAL, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("this is an incremental backup, not a data directory"), + errhint("Use pg_combinebackup to reconstruct a valid data directory."))); + if (ferror(lfp) || FreeFile(lfp)) ereport(FATAL, (errcode_for_file_access(), diff --git a/src/backend/backup/Makefile b/src/backend/backup/Makefile index b21bd8ff43..751e6d3d5e 100644 --- a/src/backend/backup/Makefile +++ b/src/backend/backup/Makefile @@ -19,12 +19,15 @@ OBJS = \ basebackup.o \ basebackup_copy.o \ basebackup_gzip.o \ + basebackup_incremental.o \ basebackup_lz4.o \ basebackup_zstd.o \ basebackup_progress.o \ basebackup_server.o \ basebackup_sink.o \ basebackup_target.o \ - basebackup_throttle.o + basebackup_throttle.o \ + walsummary.o \ + walsummaryfuncs.o include $(top_srcdir)/src/backend/common.mk diff --git a/src/backend/backup/basebackup.c b/src/backend/backup/basebackup.c index 4ba63ad8a6..8a70a9ae41 100644 --- a/src/backend/backup/basebackup.c +++ b/src/backend/backup/basebackup.c @@ -20,8 +20,10 @@ #include "access/xlogbackup.h" #include "backup/backup_manifest.h" #include "backup/basebackup.h" +#include "backup/basebackup_incremental.h" #include "backup/basebackup_sink.h" #include "backup/basebackup_target.h" +#include "catalog/pg_tablespace_d.h" #include "commands/defrem.h" #include "common/compression.h" #include "common/file_perm.h" @@ -64,6 +66,7 @@ typedef struct bool fastcheckpoint; bool nowait; bool includewal; + bool incremental; uint32 maxrate; bool sendtblspcmapfile; bool send_to_client; @@ -76,21 +79,28 @@ typedef struct } basebackup_options; static int64 sendTablespace(bbsink *sink, char *path, Oid spcoid, bool sizeonly, - struct backup_manifest_info *manifest); + struct backup_manifest_info *manifest, + IncrementalBackupInfo *ib); static int64 sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, List *tablespaces, bool sendtblspclinks, - backup_manifest_info *manifest, Oid spcoid); + backup_manifest_info *manifest, Oid spcoid, + IncrementalBackupInfo *ib); static bool sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, struct stat *statbuf, bool missing_ok, Oid dboid, Oid spcoid, RelFileNumber relfilenumber, unsigned segno, - backup_manifest_info *manifest); + backup_manifest_info *manifest, + unsigned num_incremental_blocks, + BlockNumber *incremental_blocks, + unsigned truncation_block_length); static off_t read_file_data_into_buffer(bbsink *sink, const char *readfilename, int fd, off_t offset, size_t length, BlockNumber blkno, bool verify_checksum, int *checksum_failures); +static void push_to_sink(bbsink *sink, pg_checksum_context *checksum_ctx, + size_t *bytes_done, void *data, size_t length); static bool verify_page_checksum(Page page, XLogRecPtr start_lsn, BlockNumber blkno, uint16 *expected_checksum); @@ -102,7 +112,8 @@ static int64 _tarWriteHeader(bbsink *sink, const char *filename, bool sizeonly); static void _tarWritePadding(bbsink *sink, int len); static void convert_link_to_directory(const char *pathbuf, struct stat *statbuf); -static void perform_base_backup(basebackup_options *opt, bbsink *sink); +static void perform_base_backup(basebackup_options *opt, bbsink *sink, + IncrementalBackupInfo *ib); static void parse_basebackup_options(List *options, basebackup_options *opt); static int compareWalFileNames(const ListCell *a, const ListCell *b); static int basebackup_read_file(int fd, char *buf, size_t nbytes, off_t offset, @@ -220,7 +231,8 @@ static const struct exclude_list_item excludeFiles[] = * clobbered by longjmp" from stupider versions of gcc. */ static void -perform_base_backup(basebackup_options *opt, bbsink *sink) +perform_base_backup(basebackup_options *opt, bbsink *sink, + IncrementalBackupInfo *ib) { bbsink_state state; XLogRecPtr endptr; @@ -270,6 +282,10 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) ListCell *lc; tablespaceinfo *newti; + /* If this is an incremental backup, execute preparatory steps. */ + if (ib != NULL) + PrepareForIncrementalBackup(ib, backup_state); + /* Add a node for the base directory at the end */ newti = palloc0(sizeof(tablespaceinfo)); newti->size = -1; @@ -289,10 +305,10 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) if (tmp->path == NULL) tmp->size = sendDir(sink, ".", 1, true, state.tablespaces, - true, NULL, InvalidOid); + true, NULL, InvalidOid, NULL); else tmp->size = sendTablespace(sink, tmp->path, tmp->oid, true, - NULL); + NULL, NULL); state.bytes_total += tmp->size; } state.bytes_total_is_valid = true; @@ -330,7 +346,7 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) /* Then the bulk of the files... */ sendDir(sink, ".", 1, false, state.tablespaces, - sendtblspclinks, &manifest, InvalidOid); + sendtblspclinks, &manifest, InvalidOid, ib); /* ... and pg_control after everything else. */ if (lstat(XLOG_CONTROL_FILE, &statbuf) != 0) @@ -340,7 +356,7 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) XLOG_CONTROL_FILE))); sendFile(sink, XLOG_CONTROL_FILE, XLOG_CONTROL_FILE, &statbuf, false, InvalidOid, InvalidOid, - InvalidRelFileNumber, 0, &manifest); + InvalidRelFileNumber, 0, &manifest, 0, NULL, 0); } else { @@ -348,7 +364,7 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) bbsink_begin_archive(sink, archive_name); - sendTablespace(sink, ti->path, ti->oid, false, &manifest); + sendTablespace(sink, ti->path, ti->oid, false, &manifest, ib); } /* @@ -610,7 +626,7 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) sendFile(sink, pathbuf, pathbuf, &statbuf, false, InvalidOid, InvalidOid, InvalidRelFileNumber, 0, - &manifest); + &manifest, 0, NULL, 0); /* unconditionally mark file as archived */ StatusFilePath(pathbuf, fname, ".done"); @@ -686,6 +702,7 @@ parse_basebackup_options(List *options, basebackup_options *opt) bool o_checkpoint = false; bool o_nowait = false; bool o_wal = false; + bool o_incremental = false; bool o_maxrate = false; bool o_tablespace_map = false; bool o_noverify_checksums = false; @@ -764,6 +781,15 @@ parse_basebackup_options(List *options, basebackup_options *opt) opt->includewal = defGetBoolean(defel); o_wal = true; } + else if (strcmp(defel->defname, "incremental") == 0) + { + if (o_incremental) + ereport(ERROR, + (errcode(ERRCODE_SYNTAX_ERROR), + errmsg("duplicate option \"%s\"", defel->defname))); + opt->incremental = defGetBoolean(defel); + o_incremental = true; + } else if (strcmp(defel->defname, "max_rate") == 0) { int64 maxrate; @@ -956,7 +982,7 @@ parse_basebackup_options(List *options, basebackup_options *opt) * the filesystem, bypassing the buffer cache. */ void -SendBaseBackup(BaseBackupCmd *cmd) +SendBaseBackup(BaseBackupCmd *cmd, IncrementalBackupInfo *ib) { basebackup_options opt; bbsink *sink; @@ -980,6 +1006,20 @@ SendBaseBackup(BaseBackupCmd *cmd) set_ps_display(activitymsg); } + /* + * If we're asked to perform an incremental backup and the user has not + * supplied a manifest, that's an ERROR. + * + * If we're asked to perform a full backup and the user did supply a + * manifest, just ignore it. + */ + if (!opt.incremental) + ib = NULL; + else if (ib == NULL) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("must UPLOAD_MANIFEST before performing an incremental BASE_BACKUP"))); + /* * If the target is specifically 'client' then set up to stream the backup * to the client; otherwise, it's being sent someplace else and should not @@ -1011,7 +1051,7 @@ SendBaseBackup(BaseBackupCmd *cmd) */ PG_TRY(); { - perform_base_backup(&opt, sink); + perform_base_backup(&opt, sink, ib); } PG_FINALLY(); { @@ -1086,7 +1126,7 @@ sendFileWithContent(bbsink *sink, const char *filename, const char *content, */ static int64 sendTablespace(bbsink *sink, char *path, Oid spcoid, bool sizeonly, - backup_manifest_info *manifest) + backup_manifest_info *manifest, IncrementalBackupInfo *ib) { int64 size; char pathbuf[MAXPGPATH]; @@ -1120,7 +1160,7 @@ sendTablespace(bbsink *sink, char *path, Oid spcoid, bool sizeonly, /* Send all the files in the tablespace version directory */ size += sendDir(sink, pathbuf, strlen(path), sizeonly, NIL, true, manifest, - spcoid); + spcoid, ib); return size; } @@ -1140,7 +1180,7 @@ sendTablespace(bbsink *sink, char *path, Oid spcoid, bool sizeonly, static int64 sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, List *tablespaces, bool sendtblspclinks, backup_manifest_info *manifest, - Oid spcoid) + Oid spcoid, IncrementalBackupInfo *ib) { DIR *dir; struct dirent *de; @@ -1149,7 +1189,16 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, int64 size = 0; const char *lastDir; /* Split last dir from parent path. */ bool isRelationDir = false; /* Does directory contain relations? */ + bool isGlobalDir = false; Oid dboid = InvalidOid; + BlockNumber *relative_block_numbers = NULL; + + /* + * Since this array is relatively large, avoid putting it on the stack. + * But we don't need it at all if this is not an incremental backup. + */ + if (ib != NULL) + relative_block_numbers = palloc(sizeof(BlockNumber) * RELSEG_SIZE); /* * Determine if the current path is a database directory that can contain @@ -1182,7 +1231,10 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, } } else if (strcmp(path, "./global") == 0) + { isRelationDir = true; + isGlobalDir = true; + } dir = AllocateDir(path); while ((de = ReadDir(dir, path)) != NULL) @@ -1331,11 +1383,13 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, &statbuf, sizeonly); /* - * Also send archive_status directory (by hackishly reusing - * statbuf from above ...). + * Also send archive_status and summaries directories (by + * hackishly reusing statbuf from above ...). */ size += _tarWriteHeader(sink, "./pg_wal/archive_status", NULL, &statbuf, sizeonly); + size += _tarWriteHeader(sink, "./pg_wal/summaries", NULL, + &statbuf, sizeonly); continue; /* don't recurse into pg_wal */ } @@ -1404,33 +1458,88 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, if (!skip_this_dir) size += sendDir(sink, pathbuf, basepathlen, sizeonly, tablespaces, - sendtblspclinks, manifest, spcoid); + sendtblspclinks, manifest, spcoid, ib); } else if (S_ISREG(statbuf.st_mode)) { bool sent = false; + unsigned num_blocks_required = 0; + unsigned truncation_block_length = 0; + char tarfilenamebuf[MAXPGPATH * 2]; + char *tarfilename = pathbuf + basepathlen + 1; + FileBackupMethod method = BACK_UP_FILE_FULLY; - if (!sizeonly) - sent = sendFile(sink, pathbuf, pathbuf + basepathlen + 1, &statbuf, - true, dboid, spcoid, - relfilenumber, segno, manifest); + if (ib != NULL && isRelationFile) + { + Oid relspcoid; + char *lookup_path; - if (sent || sizeonly) + if (OidIsValid(spcoid)) + { + relspcoid = spcoid; + lookup_path = psprintf("pg_tblspc/%u/%s", spcoid, + pathbuf + basepathlen + 1); + } + else + { + if (isGlobalDir) + relspcoid = GLOBALTABLESPACE_OID; + else + relspcoid = DEFAULTTABLESPACE_OID; + lookup_path = pstrdup(pathbuf + basepathlen + 1); + } + + method = GetFileBackupMethod(ib, lookup_path, dboid, relspcoid, + relfilenumber, relForkNum, + segno, statbuf.st_size, + &num_blocks_required, + relative_block_numbers, + &truncation_block_length); + if (method == BACK_UP_FILE_INCREMENTALLY) + { + statbuf.st_size = + GetIncrementalFileSize(num_blocks_required); + snprintf(tarfilenamebuf, sizeof(tarfilenamebuf), + "%s/INCREMENTAL.%s", + path + basepathlen + 1, + de->d_name); + tarfilename = tarfilenamebuf; + } + + pfree(lookup_path); + } + + if (method != DO_NOT_BACK_UP_FILE) { - /* Add size. */ - size += statbuf.st_size; + if (!sizeonly) + sent = sendFile(sink, pathbuf, tarfilename, &statbuf, + true, dboid, spcoid, + relfilenumber, segno, manifest, + num_blocks_required, + method == BACK_UP_FILE_INCREMENTALLY ? relative_block_numbers : NULL, + truncation_block_length); + + if (sent || sizeonly) + { + /* Add size. */ + size += statbuf.st_size; - /* Pad to a multiple of the tar block size. */ - size += tarPaddingBytesRequired(statbuf.st_size); + /* Pad to a multiple of the tar block size. */ + size += tarPaddingBytesRequired(statbuf.st_size); - /* Size of the header for the file. */ - size += TAR_BLOCK_SIZE; + /* Size of the header for the file. */ + size += TAR_BLOCK_SIZE; + } } } else ereport(WARNING, (errmsg("skipping special file \"%s\"", pathbuf))); } + + if (relative_block_numbers != NULL) + pfree(relative_block_numbers); + FreeDir(dir); return size; } @@ -1443,6 +1552,12 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, * If dboid is anything other than InvalidOid then any checksum failures * detected will get reported to the cumulative stats system. * + * If the file is to be sent incrementally, then num_incremental_blocks + * should be the number of blocks to be sent, and incremental_blocks + * an array of block numbers relative to the start of the current segment. + * If the whole file is to be sent, then incremental_blocks should be NULL, + * and num_incremental_blocks can have any value, as it will be ignored. + * * Returns true if the file was successfully sent, false if 'missing_ok', * and the file did not exist. */ @@ -1450,7 +1565,8 @@ static bool sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, struct stat *statbuf, bool missing_ok, Oid dboid, Oid spcoid, RelFileNumber relfilenumber, unsigned segno, - backup_manifest_info *manifest) + backup_manifest_info *manifest, unsigned num_incremental_blocks, + BlockNumber *incremental_blocks, unsigned truncation_block_length) { int fd; BlockNumber blkno = 0; @@ -1459,6 +1575,7 @@ sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, pgoff_t bytes_done = 0; bool verify_checksum = false; pg_checksum_context checksum_ctx; + int ibindex = 0; if (pg_checksum_init(&checksum_ctx, manifest->checksum_type) < 0) elog(ERROR, "could not initialize checksum of file \"%s\"", @@ -1491,22 +1608,111 @@ sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, RelFileNumberIsValid(relfilenumber)) verify_checksum = true; + /* + * If we're sending an incremental file, write the file header. + */ + if (incremental_blocks != NULL) + { + unsigned magic = INCREMENTAL_MAGIC; + size_t header_bytes_done = 0; + + /* Emit header data. */ + push_to_sink(sink, &checksum_ctx, &header_bytes_done, + &magic, sizeof(magic)); + push_to_sink(sink, &checksum_ctx, &header_bytes_done, + &num_incremental_blocks, sizeof(num_incremental_blocks)); + push_to_sink(sink, &checksum_ctx, &header_bytes_done, + &truncation_block_length, sizeof(truncation_block_length)); + push_to_sink(sink, &checksum_ctx, &header_bytes_done, + incremental_blocks, + sizeof(BlockNumber) * num_incremental_blocks); + + /* Flush out any data still in the buffer so it's again empty. */ + if (header_bytes_done > 0) + { + bbsink_archive_contents(sink, header_bytes_done); + if (pg_checksum_update(&checksum_ctx, + (uint8 *) sink->bbs_buffer, + header_bytes_done) < 0) + elog(ERROR, "could not update checksum of base backup"); + } + + /* Update our notion of file position. */ + bytes_done += sizeof(magic); + bytes_done += sizeof(num_incremental_blocks); + bytes_done += sizeof(truncation_block_length); + bytes_done += sizeof(BlockNumber) * num_incremental_blocks; + } + /* * Loop until we read the amount of data the caller told us to expect. The * file could be longer, if it was extended while we were sending it, but * for a base backup we can ignore such extended data. It will be restored * from WAL. */ - while (bytes_done < statbuf->st_size) + while (1) { - size_t remaining = statbuf->st_size - bytes_done; + /* + * Determine whether we've read all the data that we need, and if not, + * read some more. + */ + if (incremental_blocks == NULL) + { + size_t remaining = statbuf->st_size - bytes_done; + + /* + * If we've read the required number of bytes, then it's time to + * stop. + */ + if (bytes_done >= statbuf->st_size) + break; + + /* + * Read as many bytes as will fit in the buffer, or however many + * are left to read, whichever is less. + */ + cnt = read_file_data_into_buffer(sink, readfilename, fd, + bytes_done, remaining, + blkno + segno * RELSEG_SIZE, + verify_checksum, + &checksum_failures); + } + else + { + BlockNumber relative_blkno; - /* Try to read some more data. */ - cnt = read_file_data_into_buffer(sink, readfilename, fd, bytes_done, - remaining, - blkno + segno * RELSEG_SIZE, - verify_checksum, - &checksum_failures); + /* + * If we've read all the blocks, then it's time to stop. + */ + if (ibindex >= num_incremental_blocks) + break; + + /* + * Read just one block, whichever one is the next that we're + * supposed to include. + */ + relative_blkno = incremental_blocks[ibindex++]; + cnt = read_file_data_into_buffer(sink, readfilename, fd, + relative_blkno * BLCKSZ, + BLCKSZ, + relative_blkno + segno * RELSEG_SIZE, + verify_checksum, + &checksum_failures); + + /* + * If we get a partial read, that must mean that the relation is + * being truncated. Ultimately, it should be truncated to a + * multiple of BLCKSZ, since this path should only be reached for + * relation files, but we might transiently observe an + * intermediate value. + * + * It should be fine to treat this just as if the entire block had + * been truncated away - i.e. fill this and all later blocks with + * zeroes. WAL replay will fix things up. + */ + if (cnt < BLCKSZ) + break; + } /* * If the amount of data we were able to read was not a multiple of @@ -1689,6 +1895,56 @@ read_file_data_into_buffer(bbsink *sink, const char *readfilename, int fd, return cnt; } +/* + * Push data into a bbsink. + * + * It's better, when possible, to read data directly into the bbsink's buffer, + * rather than using this function to copy it into the buffer; this function is + * for cases where that approach is not practical. + * + * bytes_done should point to a count of the number of bytes that are + * currently used in the bbsink's buffer. Upon return, the bytes identified by + * data and length will have been copied into the bbsink's buffer, flushing + * as required, and *bytes_done will have been updated accordingly. If the + * buffer was flushed, the previous contents will also have been fed to + * checksum_ctx. + * + * Note that after one or more calls to this function it is the caller's + * responsibility to perform any required final flush. + */ +static void +push_to_sink(bbsink *sink, pg_checksum_context *checksum_ctx, + size_t *bytes_done, void *data, size_t length) +{ + while (length > 0) + { + size_t bytes_to_copy; + + /* + * We use < here rather than <= so that if the data exactly fills the + * remaining buffer space, we trigger a flush now. + */ + if (length < sink->bbs_buffer_length - *bytes_done) + { + /* Append remaining data to buffer. */ + memcpy(sink->bbs_buffer + *bytes_done, data, length); + *bytes_done += length; + return; + } + + /* Copy until buffer is full and flush it. */ + bytes_to_copy = sink->bbs_buffer_length - *bytes_done; + memcpy(sink->bbs_buffer + *bytes_done, data, bytes_to_copy); + data = ((char *) data) + bytes_to_copy; + length -= bytes_to_copy; + bbsink_archive_contents(sink, sink->bbs_buffer_length); + if (pg_checksum_update(checksum_ctx, (uint8 *) sink->bbs_buffer, + sink->bbs_buffer_length) < 0) + elog(ERROR, "could not update checksum"); + *bytes_done = 0; + } +} + /* * Try to verify the checksum for the provided page, if it seems appropriate * to do so. diff --git a/src/backend/backup/basebackup_incremental.c b/src/backend/backup/basebackup_incremental.c new file mode 100644 index 0000000000..20cc00bded --- /dev/null +++ b/src/backend/backup/basebackup_incremental.c @@ -0,0 +1,873 @@ +/*------------------------------------------------------------------------- + * + * basebackup_incremental.c + * code for incremental backup support + * + * This code isn't actually in charge of taking an incremental backup; + * the actual construction of the incremental backup happens in + * basebackup.c. Here, we're concerned with providing the necessary + * supports for that operation. In particular, we need to parse the + * backup manifest supplied by the user taking the incremental backup + * and extract the required information from it. + * + * Portions Copyright (c) 2010-2022, PostgreSQL Global Development Group + * + * IDENTIFICATION + * src/backend/backup/basebackup_incremental.c + * + *------------------------------------------------------------------------- + */ +#include "postgres.h" + +#include "access/timeline.h" +#include "access/xlog.h" +#include "access/xlogrecovery.h" +#include "backup/basebackup_incremental.h" +#include "backup/walsummary.h" +#include "common/blkreftable.h" +#include "common/parse_manifest.h" +#include "common/hashfn.h" +#include "postmaster/walsummarizer.h" + +#define BLOCKS_PER_READ 512 + +/* + * Details extracted from the WAL ranges present in the supplied backup manifest. + */ +typedef struct +{ + TimeLineID tli; + XLogRecPtr start_lsn; + XLogRecPtr end_lsn; +} backup_wal_range; + +/* + * Details extracted from the file list present in the supplied backup manifest. + */ +typedef struct +{ + uint32 status; + const char *path; + size_t size; +} backup_file_entry; + +static uint32 hash_string_pointer(const char *s); +#define SH_PREFIX backup_file +#define SH_ELEMENT_TYPE backup_file_entry +#define SH_KEY_TYPE const char * +#define SH_KEY path +#define SH_HASH_KEY(tb, key) hash_string_pointer(key) +#define SH_EQUAL(tb, a, b) (strcmp(a, b) == 0) +#define SH_SCOPE static inline +#define SH_DECLARE +#define SH_DEFINE +#include "lib/simplehash.h" + +struct IncrementalBackupInfo +{ + /* Memory context for this object and its subsidiary objects. */ + MemoryContext mcxt; + + /* Temporary buffer for storing the manifest while parsing it. */ + StringInfoData buf; + + /* WAL ranges extracted from the backup manifest. */ + List *manifest_wal_ranges; + + /* + * Files extracted from the backup manifest. + * + * We don't really need this information, because we use WAL summaries to + * figure what's changed. It would be unsafe to just rely on the list of + * files that existed before, because it's possible for a file to be + * removed and a new one created with the same name and different + * contents. In such cases, the whole file must still be sent. We can tell + * from the WAL summaries whether that happened, but not from the file + * list. + * + * Nonetheless, this data is useful for sanity checking. If a file that we + * think we shouldn't need to send is not present in the manifest for the + * prior backup, something has gone terribly wrong. We retain the file + * names and sizes, but not the checksums or last modified times, for + * which we have no use. + * + * One significant downside of storing this data is that it consumes + * memory. If that turns out to be a problem, we might have to decide not + * to retain this information, or to make it optional. + */ + backup_file_hash *manifest_files; + + /* + * Block-reference table for the incremental backup. + * + * It's possible that storing the entire block-reference table in memory + * will be a problem for some users. The in-memory format that we're using + * here is pretty efficient, converging to little more than 1 bit per + * block for relation forks with large numbers of modified blocks. It's + * possible, however, that if you try to perform an incremental backup of + * a database with a sufficiently large number of relations on a + * sufficiently small machine, you could run out of memory here. If that + * turns out to be a problem in practice, we'll need to be more clever. + */ + BlockRefTable *brtab; +}; + +static void manifest_process_file(JsonManifestParseContext *, + char *pathname, + size_t size, + pg_checksum_type checksum_type, + int checksum_length, + uint8 *checksum_payload); +static void manifest_process_wal_range(JsonManifestParseContext *, + TimeLineID tli, + XLogRecPtr start_lsn, + XLogRecPtr end_lsn); +static void manifest_report_error(JsonManifestParseContext *ib, + const char *fmt,...) + pg_attribute_printf(2, 3) pg_attribute_noreturn(); + +/* + * Create a new object for storing information extracted from the manifest + * supplied when creating an incremental backup. + */ +IncrementalBackupInfo * +CreateIncrementalBackupInfo(MemoryContext mcxt) +{ + IncrementalBackupInfo *ib; + MemoryContext oldcontext; + + oldcontext = MemoryContextSwitchTo(mcxt); + + ib = palloc0(sizeof(IncrementalBackupInfo)); + ib->mcxt = mcxt; + initStringInfo(&ib->buf); + + /* + * It's hard to guess how many files a "typical" installation will have in + * the data directory, but a fresh initdb creates almost 1000 files as of + * this writing, so it seems to make sense for our estimate to + * substantially higher. + */ + ib->manifest_files = backup_file_create(mcxt, 10000, NULL); + + MemoryContextSwitchTo(oldcontext); + + return ib; +} + +/* + * Before taking an incremental backup, the caller must supply the backup + * manifest from a prior backup. Each chunk of manifest data recieved + * from the client should be passed to this function. + */ +void +AppendIncrementalManifestData(IncrementalBackupInfo *ib, const char *data, + int len) +{ + MemoryContext oldcontext; + + /* Switch to our memory context. */ + oldcontext = MemoryContextSwitchTo(ib->mcxt); + + /* + * XXX. Our json parser is at present incapable of parsing json blobs + * incrementally, so we have to accumulate the entire backup manifest + * before we can do anything with it. This should really be fixed, since + * some users might have very large numbers of files in the data + * directory. + */ + appendBinaryStringInfo(&ib->buf, data, len); + + /* Switch back to previous memory context. */ + MemoryContextSwitchTo(oldcontext); +} + +/* + * Finalize an IncrementalBackupInfo object after all manifest data has + * been supplied via calls to AppendIncrementalManifestData. + */ +void +FinalizeIncrementalManifest(IncrementalBackupInfo *ib) +{ + JsonManifestParseContext context; + MemoryContext oldcontext; + + /* Switch to our memory context. */ + oldcontext = MemoryContextSwitchTo(ib->mcxt); + + /* Parse the manifest. */ + context.private_data = ib; + context.perfile_cb = manifest_process_file; + context.perwalrange_cb = manifest_process_wal_range; + context.error_cb = manifest_report_error; + json_parse_manifest(&context, ib->buf.data, ib->buf.len); + + /* Done with the buffer, so release memory. */ + pfree(ib->buf.data); + ib->buf.data = NULL; + + /* Switch back to previous memory context. */ + MemoryContextSwitchTo(oldcontext); +} + +/* + * Prepare to take an incremental backup. + * + * Before this function is called, AppendIncrementalManifestData and + * FinalizeIncrementalManifest should have already been called to pass all + * the manifest data to this object. + * + * This function performs sanity checks on the data extracted from the + * manifest and figures out for which WAL ranges we need summaries, and + * whether those summaries are available. Then, it reads and combines the + * data from those summary files. It also updates the backup_state with the + * reference TLI and LSN for the prior backup. + */ +void +PrepareForIncrementalBackup(IncrementalBackupInfo *ib, + BackupState *backup_state) +{ + MemoryContext oldcontext; + List *expectedTLEs; + List *all_wslist, + *required_wslist = NIL; + ListCell *lc; + TimeLineHistoryEntry **tlep; + int num_wal_ranges; + int i; + bool found_backup_start_tli = false; + TimeLineID earliest_wal_range_tli = 0; + XLogRecPtr earliest_wal_range_start_lsn; + TimeLineID latest_wal_range_tli = 0; + XLogRecPtr summarized_lsn; + + Assert(ib->buf.data == NULL); + + /* Switch to our memory context. */ + oldcontext = MemoryContextSwitchTo(ib->mcxt); + + /* + * Match up the TLIs that appear in the WAL ranges of the backup manifest + * with those that appear in this server's timeline history. We expect + * every backup_wal_range to match to a TimeLineHistoryEntry; if it does + * not, that's an error. + * + * This loop also decides which of the WAL ranges is the manifest is most + * ancient and which one is the newest, according to the timeline history + * of this server, and stores TLIs of those WAL ranges into + * earliest_wal_range_tli and latest_wal_range_tli. It also updates + * earliest_wal_range_start_lsn to the start LSN of the WAL range for + * earliest_wal_range_tli. + * + * Note that the return value of readTimeLineHistory puts the latest + * timeline at the beginning of the list, not the end. Hence, the earliest + * TLI is the one that occurs nearest the end of the list returned by + * readTimeLineHistory, and the latest TLI is the one that occurs closest + * to the beginning. + */ + expectedTLEs = readTimeLineHistory(backup_state->starttli); + num_wal_ranges = list_length(ib->manifest_wal_ranges); + tlep = palloc0(num_wal_ranges * sizeof(TimeLineHistoryEntry *)); + for (i = 0; i < num_wal_ranges; ++i) + { + backup_wal_range *range = list_nth(ib->manifest_wal_ranges, i); + bool saw_earliest_wal_range_tli = false; + bool saw_latest_wal_range_tli = false; + + /* Search this server's history for this WAL range's TLI. */ + foreach(lc, expectedTLEs) + { + TimeLineHistoryEntry *tle = lfirst(lc); + + if (tle->tli == range->tli) + { + tlep[i] = tle; + break; + } + + if (tle->tli == earliest_wal_range_tli) + saw_earliest_wal_range_tli = true; + if (tle->tli == latest_wal_range_tli) + saw_latest_wal_range_tli = true; + } + + /* + * An incremental backup can only be taken relative to a backup that + * represents a previous state of this server. If the backup requires + * WAL from a timeline that's not in our history, that definitely + * isn't the case. + */ + if (tlep[i] == NULL) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("timeline %u found in manifest, but not in this server's history", + range->tli))); + + /* + * If we found this TLI in the server's history before encountering + * the latest TLI seen so far in the server's history, then this TLI + * is the latest one seen so far. + * + * If on the other hand we saw the earliest TLI seen so far before + * finding this TLI, this TLI is earlier than the earliest one seen so + * far. And if this is the first TLI for which we've searched, it's + * also the earliest one seen so far. + * + * On the first loop iteration, both things should necessarily be + * true. + */ + if (!saw_latest_wal_range_tli) + latest_wal_range_tli = range->tli; + if (earliest_wal_range_tli == 0 || saw_earliest_wal_range_tli) + { + earliest_wal_range_tli = range->tli; + earliest_wal_range_start_lsn = range->start_lsn; + } + } + + /* + * Propagate information about the prior backup into the backup_label that + * will be generated for this backup. + */ + backup_state->istartpoint = earliest_wal_range_start_lsn; + backup_state->istarttli = earliest_wal_range_tli; + + /* + * Sanity check start and end LSNs for the WAL ranges in the manifest. + * + * Commonly, there won't be any timeline switches during the prior backup + * at all, but if there are, they should happen at the same LSNs that this + * server switched timelines. + * + * Whether there are any timeline switches during the prior backup or not, + * the prior backup shouldn't require any WAL from a timeline prior to the + * start of that timeline. It also shouldn't require any WAL from later + * than the start of this backup. + * + * If any of these sanity checks fail, one possible explanation is that + * the user has generated WAL on the same timeline with the same LSNs more + * than once. For instance, if two standbys running on timeline 1 were + * both promoted and (due to a broken archiving setup) both selected new + * timeline ID 2, then it's possible that one of these checks might trip. + * + * Note that there are lots of ways for the user to do something very bad + * without tripping any of these checks, and they are not intended to be + * comprehensive. It's pretty hard to see how we could be certain of + * anything here. However, if there's a problem staring us right in the + * face, it's best to report it, so we do. + */ + for (i = 0; i < num_wal_ranges; ++i) + { + backup_wal_range *range = list_nth(ib->manifest_wal_ranges, i); + + if (range->tli == earliest_wal_range_tli) + { + if (range->start_lsn < tlep[i]->begin) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("manifest requires WAL from initial timeline %u starting at %X/%X, but that timeline begins at %X/%X", + range->tli, + LSN_FORMAT_ARGS(range->start_lsn), + LSN_FORMAT_ARGS(tlep[i]->begin)))); + } + else + { + if (range->start_lsn != tlep[i]->begin) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("manifest requires WAL from continuation timeline %u starting at %X/%X, but that timeline begins at %X/%X", + range->tli, + LSN_FORMAT_ARGS(range->start_lsn), + LSN_FORMAT_ARGS(tlep[i]->begin)))); + } + + if (range->tli == latest_wal_range_tli) + { + if (range->end_lsn > backup_state->startpoint) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("manifest requires WAL from final timeline %u ending at %X/%X, but this backup starts at %X/%X", + range->tli, + LSN_FORMAT_ARGS(range->end_lsn), + LSN_FORMAT_ARGS(backup_state->startpoint)))); + } + else + { + if (range->end_lsn != tlep[i]->end) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("manifest requires WAL from non-final timeline %u ending at %X/%X, but this server switched timelines at %X/%X", + range->tli, + LSN_FORMAT_ARGS(range->end_lsn), + LSN_FORMAT_ARGS(tlep[i]->end)))); + } + + } + + /* + * Wait for WAL summarization to catch up to the backup start LSN (but + * time out if it doesn't do so quickly enough). + */ + /* XXX make timeout configurable */ + summarized_lsn = WaitForWalSummarization(backup_state->startpoint, 60000); + if (summarized_lsn < backup_state->startpoint) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("timeout waiting for WAL summarization"), + errdetail("This backup requires WAL to be summarized up to %X/%X, but summarizer has only reached %X/%X.", + LSN_FORMAT_ARGS(backup_state->startpoint), + LSN_FORMAT_ARGS(summarized_lsn)))); + + /* + * Retrieve a list of all WAL summaries on any timeline that overlap with + * the LSN range of interest. We could instead call GetWalSummaries() once + * per timeline in the loop that follows, but that would involve reading + * the directory multiple times. It should be mildly faster - and perhaps + * a bit safer - to do it just once. + */ + all_wslist = GetWalSummaries(0, earliest_wal_range_start_lsn, + backup_state->startpoint); + + /* + * We need WAL summaries for everything that happened during the prior + * backup and everything that happened afterward up until the point where + * the current backup started. + */ + foreach(lc, expectedTLEs) + { + TimeLineHistoryEntry *tle = lfirst(lc); + XLogRecPtr tli_start_lsn = tle->begin; + XLogRecPtr tli_end_lsn = tle->end; + XLogRecPtr tli_missing_lsn = InvalidXLogRecPtr; + List *tli_wslist; + + /* + * Working through the history of this server from the current + * timeline backwards, we skip everything until we find the timeline + * where this backup started. Most of the time, this means we won't + * skip anything at all, as it's unlikely that the timeline has + * changed since the beginning of the backup moments ago. + */ + if (tle->tli == backup_state->starttli) + { + found_backup_start_tli = true; + tli_end_lsn = backup_state->startpoint; + } + else if (!found_backup_start_tli) + continue; + + /* + * Find the summaries that overlap the LSN range of interest for this + * timeline. If this is the earliest timeline involved, the range of + * interest begins with the start LSN of the prior backup; otherwise, + * it begins at the LSN at which this timeline came into existence. If + * this is the latest TLI involved, the range of interest ends at the + * start LSN of the current backup; otherwise, it ends at the point + * where we switched from this timeline to the next one. + */ + if (tle->tli == earliest_wal_range_tli) + tli_start_lsn = earliest_wal_range_start_lsn; + tli_wslist = FilterWalSummaries(all_wslist, tle->tli, + tli_start_lsn, tli_end_lsn); + + /* + * There is no guarantee that the WAL summaries we found cover the + * entire range of LSNs for which summaries are required, or indeed + * that we found any WAL summaries at all. Check whether we have a + * problem of that sort. + */ + if (!WalSummariesAreComplete(tli_wslist, tli_start_lsn, tli_end_lsn, + &tli_missing_lsn)) + { + if (XLogRecPtrIsInvalid(tli_missing_lsn)) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("WAL summaries are required on timeline %u from %X/%X to %X/%X, but no summaries for that timeline and LSN range exist", + tle->tli, + LSN_FORMAT_ARGS(tli_start_lsn), + LSN_FORMAT_ARGS(tli_end_lsn)))); + else + ereport(ERROR, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("WAL summaries are required on timeline %u from %X/%X to %X/%X, but the summaries for that timeline and LSN range are incomplete", + tle->tli, + LSN_FORMAT_ARGS(tli_start_lsn), + LSN_FORMAT_ARGS(tli_end_lsn)), + errdetail("The first unsummarized LSN is this range is %X/%X.", + LSN_FORMAT_ARGS(tli_missing_lsn)))); + } + + /* + * Remember that we need to read these summaries. + * + * Technically, it's possible that this could read more files than + * required, since tli_wslist in theory could contain redundant + * summaries. For instance, if we have a summary from 0/10000000 to + * 0/20000000 and also one from 0/00000000 to 0/30000000, then the + * latter subsumes the former and the former could be ignored. + * + * We ignore this possibility because the WAL summarizer only tries to + * generate summaries that do not overlap. If somehow they exist, + * we'll do a bit of extra work but the results should still be + * correct. + */ + required_wslist = list_concat(required_wslist, tli_wslist); + + /* + * Timelines earlier than the one in which the prior backup began are + * not relevant. + */ + if (tle->tli == earliest_wal_range_tli) + break; + } + + /* + * Read all of the required block reference table files and merge all of + * the data into a single in-memory block reference table. + * + * See the comments for struct IncrementalBackupInfo for some thoughts on + * memory usage. + */ + ib->brtab = CreateEmptyBlockRefTable(); + foreach(lc, required_wslist) + { + WalSummaryFile *ws = lfirst(lc); + WalSummaryIO wsio; + BlockRefTableReader *reader; + RelFileLocator rlocator; + ForkNumber forknum; + BlockNumber limit_block; + BlockNumber blocks[BLOCKS_PER_READ]; + + wsio.file = OpenWalSummaryFile(ws, false); + wsio.filepos = 0; + ereport(DEBUG1, + (errmsg_internal("reading WAL summary file \"%s\"", + FilePathName(wsio.file)))); + reader = CreateBlockRefTableReader(ReadWalSummary, &wsio, + FilePathName(wsio.file), + ReportWalSummaryError, NULL); + while (BlockRefTableReaderNextRelation(reader, &rlocator, &forknum, + &limit_block)) + { + BlockRefTableSetLimitBlock(ib->brtab, &rlocator, + forknum, limit_block); + + while (1) + { + unsigned nblocks; + unsigned i; + + nblocks = BlockRefTableReaderGetBlocks(reader, blocks, + BLOCKS_PER_READ); + if (nblocks == 0) + break; + + for (i = 0; i < nblocks; ++i) + BlockRefTableMarkBlockModified(ib->brtab, &rlocator, + forknum, blocks[i]); + } + } + DestroyBlockRefTableReader(reader); + FileClose(wsio.file); + } + + /* Switch back to previous memory context. */ + MemoryContextSwitchTo(oldcontext); +} + +/* + * Get the pathname that should be used when a file is sent incrementally. + * + * The result is a palloc'd string. + */ +char * +GetIncrementalFilePath(Oid dboid, Oid spcoid, RelFileNumber relfilenumber, + ForkNumber forknum, unsigned segno) +{ + char *path; + char *lastslash; + char *ipath; + + path = GetRelationPath(dboid, spcoid, relfilenumber, InvalidBackendId, + forknum); + + lastslash = strrchr(path, '/'); + Assert(lastslash != NULL); + *lastslash = '\0'; + + if (segno > 0) + ipath = psprintf("%s/INCREMENTAL.%s.%u", path, lastslash + 1, segno); + else + ipath = psprintf("%s/INCREMENTAL.%s", path, lastslash + 1); + + pfree(path); + + return ipath; +} + +/* + * How should we back up a particular file as part of an incremental backup? + * + * If the return value is BACK_UP_FILE_FULLY, caller should back up the whole + * file just as if this were not an incremental backup. + * + * If the return value is BACK_UP_FILE_INCREMENTALLY, caller should include + * an incremental file in the backup instead of the entire file. On return, + * *num_blocks_required will be set to the number of blocks that need to be + * sent, and the actual block numbers will have been stored in + * relative_block_numbers, which should be an array of at least RELSEG_SIZE. + * In addition, *truncation_block_length will be set to the value that should + * be included in the incremental file. + * + * If the return value is DO_NOT_BACK_UP_FILE, the caller should not include + * the file in the backup at all. + */ +FileBackupMethod +GetFileBackupMethod(IncrementalBackupInfo *ib, char *path, + Oid dboid, Oid spcoid, + RelFileNumber relfilenumber, ForkNumber forknum, + unsigned segno, size_t size, + unsigned *num_blocks_required, + BlockNumber *relative_block_numbers, + unsigned *truncation_block_length) +{ + BlockNumber absolute_block_numbers[RELSEG_SIZE]; + BlockNumber limit_block; + BlockNumber start_blkno; + BlockNumber stop_blkno; + RelFileLocator rlocator; + BlockRefTableEntry *brtentry; + unsigned i; + unsigned nblocks; + + /* Should only be called after PrepareForIncrementalBackup. */ + Assert(ib->buf.data == NULL); + + /* + * dboid could be InvalidOid if shared rel, but spcoid and relfilenumber + * should have legal values. + */ + Assert(OidIsValid(spcoid)); + Assert(RelFileNumberIsValid(relfilenumber)); + + /* + * If the file size is too large or not a multiple of BLCKSZ, then + * something weird is happening, so give up and send the whole file. + */ + if ((size % BLCKSZ) != 0 || size / BLCKSZ > RELSEG_SIZE) + return BACK_UP_FILE_FULLY; + + /* + * The free-space map fork is not properly WAL-logged, so we need to + * backup the entire file every time. + */ + if (forknum == FSM_FORKNUM) + return BACK_UP_FILE_FULLY; + + /* + * Check whether this file is part of the prior backup. If it isn't, back + * up the whole file. + */ + if (backup_file_lookup(ib->manifest_files, path) == NULL) + { + char *ipath; + + ipath = GetIncrementalFilePath(dboid, spcoid, relfilenumber, + forknum, segno); + if (backup_file_lookup(ib->manifest_files, ipath) == NULL) + return BACK_UP_FILE_FULLY; + } + + /* Look up the block reference table entry. */ + rlocator.spcOid = spcoid; + rlocator.dbOid = dboid; + rlocator.relNumber = relfilenumber; + brtentry = BlockRefTableGetEntry(ib->brtab, &rlocator, forknum, + &limit_block); + + /* + * If there is no entry, then there have been no WAL-logged changes to the + * relation since the predecessor backup was taken, so we can back it up + * incrementally and need not include any modified blocks. + * + * However, if the file is zero-length, we should do a full backup, + * because an incremental file is always more than zero length, and it's + * silly to take an incremental backup when a full backup would be + * smaller. + */ + if (brtentry == NULL) + { + *num_blocks_required = 0; + *truncation_block_length = size / BLCKSZ; + if (size == 0) + return BACK_UP_FILE_FULLY; + return BACK_UP_FILE_INCREMENTALLY; + } + + /* + * If the limit_block is less than or equal to the point where this + * segment starts, send the whole file. + */ + if (limit_block <= segno * RELSEG_SIZE) + return BACK_UP_FILE_FULLY; + + /* + * Get relevant entries from the block reference table entry. + * + * We shouldn't overflow computing the start or stop block numbers, but if + * it manages to happen somehow, detect it and throw an error. + */ + start_blkno = segno * RELSEG_SIZE; + stop_blkno = start_blkno + (size / BLCKSZ); + if (start_blkno / RELSEG_SIZE != segno || stop_blkno < start_blkno) + ereport(ERROR, + errcode(ERRCODE_INTERNAL_ERROR), + errmsg_internal("overflow computing block number bounds for segment %u with size %lu", + segno, size)); + nblocks = BlockRefTableEntryGetBlocks(brtentry, start_blkno, stop_blkno, + absolute_block_numbers, RELSEG_SIZE); + Assert(nblocks <= RELSEG_SIZE); + + /* + * If we're going to have to send nearly all of the blocks, then just send + * the whole file, because that won't require much extra storage or + * transfer and will speed up and simplify backup restoration. It's not + * clear what threshold is most appropriate here and perhaps it ought to + * be configurable, but for now we're just going to say that if we'd need + * to send 90% of the blocks anyway, give up and send the whole file. + * + * NB: If you change the threshold here, at least make sure to back up the + * file fully when every single block must be sent, because there's + * nothing good about sending an incremental file in that case. + */ + if (nblocks * BLCKSZ > size * 0.9) + return BACK_UP_FILE_FULLY; + + /* + * Looks like we can send an incremental file. + * + * Return the relevant details to the caller, transposing absolute block + * numbers to relative block numbers. + * + * The truncation block length is the minimum length of the reconstructed + * file. Any block numbers below this threshold that are not present in + * the backup need to be fetched from the prior backup. At or above this + * threshold, blocks should only be included in the result if they are + * present in the backup. (This may require inserting zero blocks if the + * blocks included in the backup are non-consecutive.) + */ + for (i = 0; i < nblocks; ++i) + relative_block_numbers[i] = absolute_block_numbers[i] - start_blkno; + *num_blocks_required = nblocks; + *truncation_block_length = + Min(size / BLCKSZ, limit_block - segno * RELSEG_SIZE); + return BACK_UP_FILE_INCREMENTALLY; +} + +/* + * Compute the size for an incremental file containing a given number of blocks. + */ +extern size_t +GetIncrementalFileSize(unsigned num_blocks_required) +{ + size_t result; + + /* Make sure we're not going to overflow. */ + Assert(num_blocks_required <= RELSEG_SIZE); + + /* + * Three four byte quantities (magic number, truncation block length, + * block count) followed by block numbers followed by block contents. + */ + result = 3 * sizeof(uint32); + result += (BLCKSZ + sizeof(BlockNumber)) * num_blocks_required; + + return result; +} + +/* + * Helper function for filemap hash table. + */ +static uint32 +hash_string_pointer(const char *s) +{ + unsigned char *ss = (unsigned char *) s; + + return hash_bytes(ss, strlen(s)); +} + +/* + * This callback is invoked for each file mentioned in the backup manifest. + * + * We store the path to each file and the size of each file for sanity-checking + * purposes. For further details, see comments for IncrementalBackupInfo. + */ +static void +manifest_process_file(JsonManifestParseContext *context, + char *pathname, size_t size, + pg_checksum_type checksum_type, + int checksum_length, + uint8 *checksum_payload) +{ + IncrementalBackupInfo *ib = context->private_data; + backup_file_entry *entry; + bool found; + + entry = backup_file_insert(ib->manifest_files, pathname, &found); + if (!found) + { + entry->path = MemoryContextStrdup(ib->manifest_files->ctx, + pathname); + entry->size = size; + } +} + +/* + * This callback is invoked for each WAL range mentioned in the backup + * manifest. + * + * We're just interested in learning the oldest LSN and the corresponding TLI + * that appear in any WAL range. + */ +static void +manifest_process_wal_range(JsonManifestParseContext *context, + TimeLineID tli, XLogRecPtr start_lsn, + XLogRecPtr end_lsn) +{ + IncrementalBackupInfo *ib = context->private_data; + backup_wal_range *range = palloc(sizeof(backup_wal_range)); + + range->tli = tli; + range->start_lsn = start_lsn; + range->end_lsn = end_lsn; + ib->manifest_wal_ranges = lappend(ib->manifest_wal_ranges, range); +} + +/* + * This callback is invoked if an error occurs while parsing the backup + * manifest. + */ +static void +manifest_report_error(JsonManifestParseContext *context, const char *fmt,...) +{ + StringInfoData errbuf; + + initStringInfo(&errbuf); + + for (;;) + { + va_list ap; + int needed; + + va_start(ap, fmt); + needed = appendStringInfoVA(&errbuf, fmt, ap); + va_end(ap); + if (needed == 0) + break; + enlargeStringInfo(&errbuf, needed); + } + + ereport(ERROR, + errmsg_internal("%s", errbuf.data)); +} diff --git a/src/backend/backup/meson.build b/src/backend/backup/meson.build index 11a79bbf80..19c355ceca 100644 --- a/src/backend/backup/meson.build +++ b/src/backend/backup/meson.build @@ -5,6 +5,7 @@ backend_sources += files( 'basebackup.c', 'basebackup_copy.c', 'basebackup_gzip.c', + 'basebackup_incremental.c', 'basebackup_lz4.c', 'basebackup_progress.c', 'basebackup_server.c', @@ -12,4 +13,6 @@ backend_sources += files( 'basebackup_target.c', 'basebackup_throttle.c', 'basebackup_zstd.c', + 'walsummary.c', + 'walsummaryfuncs.c' ) diff --git a/src/backend/backup/walsummary.c b/src/backend/backup/walsummary.c new file mode 100644 index 0000000000..ebf4ea038d --- /dev/null +++ b/src/backend/backup/walsummary.c @@ -0,0 +1,356 @@ +/*------------------------------------------------------------------------- + * + * walsummary.c + * Functions for accessing and managing WAL summary data. + * + * Portions Copyright (c) 2010-2022, PostgreSQL Global Development Group + * + * src/backend/backup/walsummary.c + * + *------------------------------------------------------------------------- + */ + +#include "postgres.h" + +#include <sys/stat.h> +#include <unistd.h> + +#include "access/xlog_internal.h" +#include "backup/walsummary.h" +#include "utils/wait_event.h" + +static bool IsWalSummaryFilename(char *filename); +static int ListComparatorForWalSummaryFiles(const ListCell *a, + const ListCell *b); + +/* + * Get a list of WAL summaries. + * + * If tli != 0, only WAL summaries with the indicated TLI will be included. + * + * If start_lsn != InvalidXLogRecPtr, only summaries that end before the + * indicated LSN will be included. + * + * If end_lsn != InvalidXLogRecPtr, only summaries that start before the + * indicated LSN will be included. + * + * The intent is that you can call GetWalSummaries(tli, start_lsn, end_lsn) + * to get all WAL summaries on the indicated timeline that overlap the + * specified LSN range. + */ +List * +GetWalSummaries(TimeLineID tli, XLogRecPtr start_lsn, XLogRecPtr end_lsn) +{ + DIR *sdir; + struct dirent *dent; + List *result = NIL; + + sdir = AllocateDir(XLOGDIR "/summaries"); + while ((dent = ReadDir(sdir, XLOGDIR "/summaries")) != NULL) + { + WalSummaryFile *ws; + uint32 tmp[5]; + TimeLineID file_tli; + XLogRecPtr file_start_lsn; + XLogRecPtr file_end_lsn; + + /* Decode filename, or skip if it's not in the expected format. */ + if (!IsWalSummaryFilename(dent->d_name)) + continue; + sscanf(dent->d_name, "%08X%08X%08X%08X%08X", + &tmp[0], &tmp[1], &tmp[2], &tmp[3], &tmp[4]); + file_tli = tmp[0]; + file_start_lsn = ((uint64) tmp[1]) << 32 | tmp[2]; + file_end_lsn = ((uint64) tmp[3]) << 32 | tmp[4]; + + /* Skip if it doesn't match the filter criteria. */ + if (tli != 0 && tli != file_tli) + continue; + if (!XLogRecPtrIsInvalid(start_lsn) && start_lsn > file_end_lsn) + continue; + if (!XLogRecPtrIsInvalid(end_lsn) && end_lsn < file_start_lsn) + continue; + + /* Add it to the list. */ + ws = palloc(sizeof(WalSummaryFile)); + ws->tli = file_tli; + ws->start_lsn = file_start_lsn; + ws->end_lsn = file_end_lsn; + result = lappend(result, ws); + } + FreeDir(sdir); + + return result; +} + +/* + * Build a new list of WAL summaries based on an existing list, but filtering + * out summaries that don't match the search parameters. + * + * If tli != 0, only WAL summaries with the indicated TLI will be included. + * + * If start_lsn != InvalidXLogRecPtr, only summaries that end before the + * indicated LSN will be included. + * + * If end_lsn != InvalidXLogRecPtr, only summaries that start before the + * indicated LSN will be included. + */ +List * +FilterWalSummaries(List *wslist, TimeLineID tli, + XLogRecPtr start_lsn, XLogRecPtr end_lsn) +{ + List *result = NIL; + ListCell *lc; + + /* Loop over input. */ + foreach(lc, wslist) + { + WalSummaryFile *ws = lfirst(lc); + + /* Skip if it doesn't match the filter criteria. */ + if (tli != 0 && tli != ws->tli) + continue; + if (!XLogRecPtrIsInvalid(start_lsn) && start_lsn > ws->end_lsn) + continue; + if (!XLogRecPtrIsInvalid(end_lsn) && end_lsn < ws->start_lsn) + continue; + + /* Add it to the result list. */ + result = lappend(result, ws); + } + + return result; +} + +/* + * Check whether the supplied list of WalSummaryFile objects covers the + * whole range of LSNs from start_lsn to end_lsn. This function ignores + * timelines, so the caller should probably filter using the appropriate + * timeline before calling this. + * + * If the whole range of LSNs is covered, returns true, otherwise false. + * If false is returned, *missing_lsn is set either to InvalidXLogRecPtr + * if there are no WAL summary files in the input list, or to the first LSN + * in the range that is not covered by a WAL summary file in the input list. + */ +bool +WalSummariesAreComplete(List *wslist, XLogRecPtr start_lsn, + XLogRecPtr end_lsn, XLogRecPtr *missing_lsn) +{ + XLogRecPtr current_lsn = start_lsn; + ListCell *lc; + + /* Special case for empty list. */ + if (wslist == NIL) + { + *missing_lsn = InvalidXLogRecPtr; + return false; + } + + /* Make a private copy of the list and sort it by start LSN. */ + wslist = list_copy(wslist); + list_sort(wslist, ListComparatorForWalSummaryFiles); + + /* + * Consider summary files in order of increasing start_lsn, advancing the + * known-summarized range from start_lsn toward end_lsn. + * + * Normally, the summary files should cover non-overlapping WAL ranges, + * but this algorithm is intended to be correct even in case of overlap. + */ + foreach(lc, wslist) + { + WalSummaryFile *ws = lfirst(lc); + + if (ws->start_lsn > current_lsn) + { + /* We found a gap. */ + break; + } + if (ws->end_lsn > current_lsn) + { + /* + * Next summary extends beyond end of previous summary, so extend + * the end of the range known to be summarized. + */ + current_lsn = ws->end_lsn; + + /* + * If the range we know to be summarized has reached the required + * end LSN, we have proved completeness. + */ + if (current_lsn >= end_lsn) + return true; + } + } + + /* + * We either ran out of summary files without reaching the end LSN, or we + * hit a gap in the sequence that resulted in us bailing out of the loop + * above. + */ + *missing_lsn = current_lsn; + return false; +} + +/* + * Open a WAL summary file. + * + * This will throw an error in case of trouble. As an exception, if + * missing_ok = true and the trouble is specifically that the file does + * not exist, it will not throw an error and will return a value less than 0. + */ +File +OpenWalSummaryFile(WalSummaryFile *ws, bool missing_ok) +{ + char path[MAXPGPATH]; + File file; + + snprintf(path, MAXPGPATH, + XLOGDIR "/summaries/%08X%08X%08X%08X%08X.summary", + ws->tli, + LSN_FORMAT_ARGS(ws->start_lsn), + LSN_FORMAT_ARGS(ws->end_lsn)); + + file = PathNameOpenFile(path, O_RDONLY); + if (file < 0 && (errno != EEXIST || !missing_ok)) + ereport(ERROR, + (errcode_for_file_access(), + errmsg("could not open file \"%s\": %m", path))); + + return file; +} + +/* + * Remove a WAL summary file if the last modification time precedes the + * cutoff time. + */ +void +RemoveWalSummaryIfOlderThan(WalSummaryFile *ws, time_t cutoff_time) +{ + char path[MAXPGPATH]; + struct stat statbuf; + + snprintf(path, MAXPGPATH, + XLOGDIR "/summaries/%08X%08X%08X%08X%08X.summary", + ws->tli, + LSN_FORMAT_ARGS(ws->start_lsn), + LSN_FORMAT_ARGS(ws->end_lsn)); + + if (lstat(path, &statbuf) != 0) + { + if (errno == ENOENT) + return; + ereport(ERROR, + (errcode_for_file_access(), + errmsg("could not stat file \"%s\": %m", path))); + } + if (statbuf.st_mtime >= cutoff_time) + return; + if (unlink(path) != 0) + ereport(ERROR, + (errcode_for_file_access(), + errmsg("could not stat file \"%s\": %m", path))); + ereport(DEBUG2, + (errmsg_internal("removing file \"%s\"", path))); +} + +/* + * Test whether a filename looks like a WAL summary file. + */ +static bool +IsWalSummaryFilename(char *filename) +{ + return strspn(filename, "0123456789ABCDEF") == 40 && + strcmp(filename + 40, ".summary") == 0; +} + +/* + * Data read callback for use with CreateBlockRefTableReader. + */ +int +ReadWalSummary(void *wal_summary_io, void *data, int length) +{ + WalSummaryIO *io = wal_summary_io; + int nbytes; + + nbytes = FileRead(io->file, data, length, io->filepos, + WAIT_EVENT_WAL_SUMMARY_READ); + if (nbytes < 0) + ereport(ERROR, + (errcode_for_file_access(), + errmsg("could not write file \"%s\": %m", + FilePathName(io->file)))); + + io->filepos += nbytes; + return nbytes; +} + +/* + * Data write callback for use with WriteBlockRefTable. + */ +int +WriteWalSummary(void *wal_summary_io, void *data, int length) +{ + WalSummaryIO *io = wal_summary_io; + int nbytes; + + nbytes = FileWrite(io->file, data, length, io->filepos, + WAIT_EVENT_WAL_SUMMARY_WRITE); + if (nbytes < 0) + ereport(ERROR, + (errcode_for_file_access(), + errmsg("could not write file \"%s\": %m", + FilePathName(io->file)))); + if (nbytes != length) + ereport(ERROR, + (errcode_for_file_access(), + errmsg("could not write file \"%s\": wrote only %d of %d bytes at offset %u", + FilePathName(io->file), nbytes, + length, (unsigned) io->filepos), + errhint("Check free disk space."))); + + io->filepos += nbytes; + return nbytes; +} + +/* + * Error-reporting callback for use with CreateBlockRefTableReader. + */ +void +ReportWalSummaryError(void *callback_arg, char *fmt,...) +{ + StringInfoData buf; + va_list ap; + int needed; + + initStringInfo(&buf); + for (;;) + { + va_start(ap, fmt); + needed = appendStringInfoVA(&buf, fmt, ap); + va_end(ap); + if (needed == 0) + break; + enlargeStringInfo(&buf, needed); + } + ereport(ERROR, + errcode(ERRCODE_DATA_CORRUPTED), + errmsg_internal("%s", buf.data)); +} + +/* + * Comparator to sort a List of WalSummaryFile objects by start_lsn. + */ +static int +ListComparatorForWalSummaryFiles(const ListCell *a, const ListCell *b) +{ + WalSummaryFile *ws1 = lfirst(a); + WalSummaryFile *ws2 = lfirst(b); + + if (ws1->start_lsn < ws2->start_lsn) + return -1; + if (ws1->start_lsn > ws2->start_lsn) + return 1; + return 0; +} diff --git a/src/backend/backup/walsummaryfuncs.c b/src/backend/backup/walsummaryfuncs.c new file mode 100644 index 0000000000..2e77d38b4a --- /dev/null +++ b/src/backend/backup/walsummaryfuncs.c @@ -0,0 +1,169 @@ +/*------------------------------------------------------------------------- + * + * walsummaryfuncs.c + * SQL-callable functions for accessing WAL summary data. + * + * Portions Copyright (c) 2010-2022, PostgreSQL Global Development Group + * + * src/backend/backup/walsummaryfuncs.c + * + *------------------------------------------------------------------------- + */ + +#include "postgres.h" + +#include "backup/walsummary.h" +#include "common/blkreftable.h" +#include "funcapi.h" +#include "miscadmin.h" +#include "utils/fmgrprotos.h" +#include "utils/pg_lsn.h" + +#define NUM_WS_ATTS 3 +#define NUM_SUMMARY_ATTS 6 +#define MAX_BLOCKS_PER_CALL 256 + +/* + * List the WAL summary files available in pg_wal/summaries. + */ +Datum +pg_available_wal_summaries(PG_FUNCTION_ARGS) +{ + ReturnSetInfo *rsi; + List *wslist; + ListCell *lc; + Datum values[NUM_WS_ATTS]; + bool nulls[NUM_WS_ATTS]; + + InitMaterializedSRF(fcinfo, 0); + rsi = (ReturnSetInfo *) fcinfo->resultinfo; + + memset(nulls, 0, sizeof(nulls)); + + wslist = GetWalSummaries(0, InvalidXLogRecPtr, InvalidXLogRecPtr); + foreach(lc, wslist) + { + WalSummaryFile *ws = (WalSummaryFile *) lfirst(lc); + HeapTuple tuple; + + CHECK_FOR_INTERRUPTS(); + + values[0] = Int64GetDatum((int64) ws->tli); + values[1] = LSNGetDatum(ws->start_lsn); + values[2] = LSNGetDatum(ws->end_lsn); + + tuple = heap_form_tuple(rsi->setDesc, values, nulls); + tuplestore_puttuple(rsi->setResult, tuple); + } + + return (Datum) 0; +} + +/* + * List the contents of a WAL summary file identified by TLI, start LSN, + * and end LSN. + */ +Datum +pg_wal_summary_contents(PG_FUNCTION_ARGS) +{ + ReturnSetInfo *rsi; + Datum values[NUM_SUMMARY_ATTS]; + bool nulls[NUM_SUMMARY_ATTS]; + WalSummaryFile ws; + WalSummaryIO io; + BlockRefTableReader *reader; + int64 raw_tli; + RelFileLocator rlocator; + ForkNumber forknum; + BlockNumber limit_block; + + InitMaterializedSRF(fcinfo, 0); + rsi = (ReturnSetInfo *) fcinfo->resultinfo; + memset(nulls, 0, sizeof(nulls)); + + /* + * Since the timeline could at least in theory be more than 2^31, and + * since we don't have unsigned types at the SQL level, it is passed as a + * 64-bit integer. Test whether it's out of range. + */ + raw_tli = PG_GETARG_INT64(0); + if (raw_tli < 1 || raw_tli > PG_INT32_MAX) + ereport(ERROR, + errcode(ERRCODE_INVALID_PARAMETER_VALUE), + errmsg("invalid timeline %lld", (long long) raw_tli)); + + /* Prepare to read the specified WAL summry file. */ + ws.tli = (TimeLineID) raw_tli; + ws.start_lsn = PG_GETARG_LSN(1); + ws.end_lsn = PG_GETARG_LSN(2); + io.filepos = 0; + io.file = OpenWalSummaryFile(&ws, false); + reader = CreateBlockRefTableReader(ReadWalSummary, &io, + FilePathName(io.file), + ReportWalSummaryError, NULL); + + /* Loop over relation forks. */ + while (BlockRefTableReaderNextRelation(reader, &rlocator, &forknum, + &limit_block)) + { + BlockNumber blocks[MAX_BLOCKS_PER_CALL]; + HeapTuple tuple; + + CHECK_FOR_INTERRUPTS(); + + values[0] = ObjectIdGetDatum(rlocator.relNumber); + values[1] = ObjectIdGetDatum(rlocator.spcOid); + values[2] = ObjectIdGetDatum(rlocator.dbOid); + values[3] = Int16GetDatum((int16) forknum); + + /* Loop over blocks within the current relation fork. */ + while (true) + { + unsigned nblocks; + unsigned i; + + CHECK_FOR_INTERRUPTS(); + + nblocks = BlockRefTableReaderGetBlocks(reader, blocks, + MAX_BLOCKS_PER_CALL); + if (nblocks == 0) + break; + + /* + * For each block that we specifically know to have been modified, + * emit a row with that block number and limit_block = false. + */ + values[5] = BoolGetDatum(false); + for (i = 0; i < nblocks; ++i) + { + values[4] = Int64GetDatum((int64) blocks[i]); + + tuple = heap_form_tuple(rsi->setDesc, values, nulls); + tuplestore_puttuple(rsi->setResult, tuple); + } + + /* + * If the limit block is not InvalidBlockNumber, emit an exta row + * with that block number and limit_block = true. + * + * There is no point in doing this when the limit_block is + * InvalidBlockNumber, because no block with that number or any + * higher number can ever exist. + */ + if (BlockNumberIsValid(limit_block)) + { + values[4] = Int64GetDatum((int64) limit_block); + values[5] = BoolGetDatum(true); + + tuple = heap_form_tuple(rsi->setDesc, values, nulls); + tuplestore_puttuple(rsi->setResult, tuple); + } + } + } + + /* Cleanup */ + DestroyBlockRefTableReader(reader); + FileClose(io.file); + + return (Datum) 0; +} diff --git a/src/backend/postmaster/Makefile b/src/backend/postmaster/Makefile index 047448b34e..367a46c617 100644 --- a/src/backend/postmaster/Makefile +++ b/src/backend/postmaster/Makefile @@ -24,6 +24,7 @@ OBJS = \ postmaster.o \ startup.o \ syslogger.o \ + walsummarizer.o \ walwriter.o include $(top_srcdir)/src/backend/common.mk diff --git a/src/backend/postmaster/auxprocess.c b/src/backend/postmaster/auxprocess.c index cae6feb356..0c15c1777d 100644 --- a/src/backend/postmaster/auxprocess.c +++ b/src/backend/postmaster/auxprocess.c @@ -21,6 +21,7 @@ #include "postmaster/auxprocess.h" #include "postmaster/bgwriter.h" #include "postmaster/startup.h" +#include "postmaster/walsummarizer.h" #include "postmaster/walwriter.h" #include "replication/walreceiver.h" #include "storage/bufmgr.h" @@ -80,6 +81,9 @@ AuxiliaryProcessMain(AuxProcType auxtype) case WalReceiverProcess: MyBackendType = B_WAL_RECEIVER; break; + case WalSummarizerProcess: + MyBackendType = B_WAL_SUMMARIZER; + break; default: elog(PANIC, "unrecognized process type: %d", (int) MyAuxProcType); MyBackendType = B_INVALID; @@ -161,6 +165,10 @@ AuxiliaryProcessMain(AuxProcType auxtype) WalReceiverMain(); proc_exit(1); + case WalSummarizerProcess: + WalSummarizerMain(); + proc_exit(1); + default: elog(PANIC, "unrecognized process type: %d", (int) MyAuxProcType); proc_exit(1); diff --git a/src/backend/postmaster/meson.build b/src/backend/postmaster/meson.build index cda921fd10..a30eb6692f 100644 --- a/src/backend/postmaster/meson.build +++ b/src/backend/postmaster/meson.build @@ -12,5 +12,6 @@ backend_sources += files( 'postmaster.c', 'startup.c', 'syslogger.c', + 'walsummarizer.c', 'walwriter.c', ) diff --git a/src/backend/postmaster/postmaster.c b/src/backend/postmaster/postmaster.c index 9cb624eab8..86f6cf2feb 100644 --- a/src/backend/postmaster/postmaster.c +++ b/src/backend/postmaster/postmaster.c @@ -115,6 +115,7 @@ #include "postmaster/pgarch.h" #include "postmaster/postmaster.h" #include "postmaster/syslogger.h" +#include "postmaster/walsummarizer.h" #include "replication/logicallauncher.h" #include "replication/walsender.h" #include "storage/fd.h" @@ -252,6 +253,7 @@ static pid_t StartupPID = 0, CheckpointerPID = 0, WalWriterPID = 0, WalReceiverPID = 0, + WalSummarizerPID = 0, AutoVacPID = 0, PgArchPID = 0, SysLoggerPID = 0; @@ -443,6 +445,7 @@ static bool CreateOptsFile(int argc, char *argv[], char *fullprogname); static pid_t StartChildProcess(AuxProcType type); static void StartAutovacuumWorker(void); static void MaybeStartWalReceiver(void); +static void MaybeStartWalSummarizer(void); static void InitPostmasterDeathWatchHandle(void); /* @@ -562,6 +565,7 @@ static void ShmemBackendArrayRemove(Backend *bn); #define StartCheckpointer() StartChildProcess(CheckpointerProcess) #define StartWalWriter() StartChildProcess(WalWriterProcess) #define StartWalReceiver() StartChildProcess(WalReceiverProcess) +#define StartWalSummarizer() StartChildProcess(WalSummarizerProcess) /* Macros to check exit status of a child process */ #define EXIT_STATUS_0(st) ((st) == 0) @@ -1833,6 +1837,9 @@ ServerLoop(void) if (WalReceiverRequested) MaybeStartWalReceiver(); + /* If we need to start a WAL summarizer, try to do that now */ + MaybeStartWalSummarizer(); + /* Get other worker processes running, if needed */ if (StartWorkerNeeded || HaveCrashedWorker) maybe_start_bgworkers(); @@ -2657,6 +2664,8 @@ process_pm_reload_request(void) signal_child(WalWriterPID, SIGHUP); if (WalReceiverPID != 0) signal_child(WalReceiverPID, SIGHUP); + if (WalSummarizerPID != 0) + signal_child(WalSummarizerPID, SIGHUP); if (AutoVacPID != 0) signal_child(AutoVacPID, SIGHUP); if (PgArchPID != 0) @@ -3010,6 +3019,7 @@ process_pm_child_exit(void) BgWriterPID = StartBackgroundWriter(); if (WalWriterPID == 0) WalWriterPID = StartWalWriter(); + MaybeStartWalSummarizer(); /* * Likewise, start other special children as needed. In a restart @@ -3128,6 +3138,20 @@ process_pm_child_exit(void) continue; } + /* + * Was it the wal summarizer? Normal exit can be ignored; we'll start + * a new one at the next iteration of the postmaster's main loop, if + * necessary. Any other exit condition is treated as a crash. + */ + if (pid == WalSummarizerPID) + { + WalSummarizerPID = 0; + if (!EXIT_STATUS_0(exitstatus)) + HandleChildCrash(pid, exitstatus, + _("WAL summarizer process")); + continue; + } + /* * Was it the autovacuum launcher? Normal exit can be ignored; we'll * start a new one at the next iteration of the postmaster's main @@ -3523,6 +3547,12 @@ HandleChildCrash(int pid, int exitstatus, const char *procname) else if (WalReceiverPID != 0 && take_action) sigquit_child(WalReceiverPID); + /* Take care of the walsummarizer too */ + if (pid == WalSummarizerPID) + WalSummarizerPID = 0; + else if (WalSummarizerPID != 0 && take_action) + sigquit_child(WalSummarizerPID); + /* Take care of the autovacuum launcher too */ if (pid == AutoVacPID) AutoVacPID = 0; @@ -3673,6 +3703,8 @@ PostmasterStateMachine(void) signal_child(StartupPID, SIGTERM); if (WalReceiverPID != 0) signal_child(WalReceiverPID, SIGTERM); + if (WalSummarizerPID != 0) + signal_child(WalSummarizerPID, SIGTERM); /* checkpointer, archiver, stats, and syslogger may continue for now */ /* Now transition to PM_WAIT_BACKENDS state to wait for them to die */ @@ -3699,6 +3731,7 @@ PostmasterStateMachine(void) if (CountChildren(BACKEND_TYPE_ALL - BACKEND_TYPE_WALSND) == 0 && StartupPID == 0 && WalReceiverPID == 0 && + WalSummarizerPID == 0 && BgWriterPID == 0 && (CheckpointerPID == 0 || (!FatalError && Shutdown < ImmediateShutdown)) && @@ -3796,6 +3829,7 @@ PostmasterStateMachine(void) /* These other guys should be dead already */ Assert(StartupPID == 0); Assert(WalReceiverPID == 0); + Assert(WalSummarizerPID == 0); Assert(BgWriterPID == 0); Assert(CheckpointerPID == 0); Assert(WalWriterPID == 0); @@ -4017,6 +4051,8 @@ TerminateChildren(int signal) signal_child(WalWriterPID, signal); if (WalReceiverPID != 0) signal_child(WalReceiverPID, signal); + if (WalSummarizerPID != 0) + signal_child(WalSummarizerPID, signal); if (AutoVacPID != 0) signal_child(AutoVacPID, signal); if (PgArchPID != 0) @@ -5364,6 +5400,10 @@ StartChildProcess(AuxProcType type) ereport(LOG, (errmsg("could not fork WAL receiver process: %m"))); break; + case WalSummarizerProcess: + ereport(LOG, + (errmsg("could not fork WAL summarizer process: %m"))); + break; default: ereport(LOG, (errmsg("could not fork process: %m"))); @@ -5500,6 +5540,19 @@ MaybeStartWalReceiver(void) } } +/* + * MaybeStartWalSummarizer + * Start the WAL summarizer process, if not running and our state allows. + */ +static void +MaybeStartWalSummarizer(void) +{ + if (wal_summarize_mb != 0 && WalSummarizerPID == 0 && + (pmState == PM_RUN || pmState == PM_HOT_STANDBY) && + Shutdown <= SmartShutdown) + WalSummarizerPID = StartWalSummarizer(); +} + /* * Create the opts file diff --git a/src/backend/postmaster/walsummarizer.c b/src/backend/postmaster/walsummarizer.c new file mode 100644 index 0000000000..34bd254183 --- /dev/null +++ b/src/backend/postmaster/walsummarizer.c @@ -0,0 +1,1414 @@ +/*------------------------------------------------------------------------- + * + * walsummarizer.c + * + * Background process to perform WAL summarization, if it is enabled. + * It continuously scans the write-ahead log and periodically emits a + * summary file which indicates which blocks in which relation forks + * were modified by WAL records in the LSN range covered by the summary + * file. See walsummary.c and blkreftable.c for more details on the + * naming and contents of WAL summary files. + * + * If configured to do, this background process will also remove WAL + * summary files when the file timestamp is older than a configurable + * threshold (but only if the WAL has been removed first). + * + * Portions Copyright (c) 1996-2022, PostgreSQL Global Development Group + * + * IDENTIFICATION + * src/backend/postmaster/walsummarizer.c + * + *------------------------------------------------------------------------- + */ +#include "postgres.h" + +#include "access/timeline.h" +#include "access/xlog.h" +#include "access/xlog_internal.h" +#include "access/xlogrecovery.h" +#include "access/xlogutils.h" +#include "backup/walsummary.h" +#include "catalog/storage_xlog.h" +#include "common/blkreftable.h" +#include "libpq/pqsignal.h" +#include "miscadmin.h" +#include "postmaster/bgwriter.h" +#include "postmaster/interrupt.h" +#include "postmaster/walsummarizer.h" +#include "replication/walreceiver.h" +#include "storage/fd.h" +#include "storage/ipc.h" +#include "storage/lwlock.h" +#include "storage/latch.h" +#include "storage/proc.h" +#include "storage/procsignal.h" +#include "storage/shmem.h" +#include "storage/spin.h" +#include "utils/guc.h" +#include "utils/memutils.h" +#include "utils/wait_event.h" + +/* + * Data in shared memory related to WAL summarization. + */ +typedef struct +{ + /* + * These fields are protected by WALSummarizerLock. + * + * Until we've discovered what summary files already exist on disk and + * stored that information in shared memory, initialized is false and the + * other fields here contain no meaningful information. After that has + * been done, initialized is true. + * + * summarized_tli and summarized_lsn indicate the last LSN and TLI at + * which the next summary file will start. Normally, these are the LSN and + * TLI at which the last file ended; in such case, lsn_is_exact is true. + * If, however, the LSN is just an approximation, then lsn_is_exact is + * false. This can happen if, for example, there are no existing WAL + * summary files at startup. In that case, we have to derive the position + * at which to start summarizing from the WAL files that exist on disk, + * and so the LSN might point to the start of the next file even though + * that might happen to be in the middle of a WAL record. + * + * summarizer_pgprocno is the pgprocno value for the summarizer process, + * if one is running, or else INVALID_PGPROCNO. + * + * pending_lsn is used by the summarizer to advertise the ending LSN of a + * record it has recently read. It shouldn't ever be less than + * summarized_lsn, but might be greater, because the summarizer buffers + * data for a range of LSNs in memory before writing out a new file. + * + * switch_requested can be set to true to notify the summarizer that a new + * WAL summary file should be written as soon as possible, without trying + * to read more WAL first. + */ + bool initialized; + TimeLineID summarized_tli; + XLogRecPtr summarized_lsn; + bool lsn_is_exact; + int summarizer_pgprocno; + XLogRecPtr pending_lsn; + bool switch_requested; + + /* + * This field handles its own synchronizaton. + */ + ConditionVariable summary_file_cv; +} WalSummarizerData; + +/* + * Private data for our xlogreader's page read callback. + */ +typedef struct +{ + TimeLineID tli; + bool historic; + XLogRecPtr read_upto; + bool end_of_wal; + bool waited; + XLogRecPtr redo_pointer; + bool redo_pointer_reached; + XLogRecPtr redo_pointer_refresh_lsn; +} SummarizerReadLocalXLogPrivate; + +/* Pointer to shared memory state. */ +static WalSummarizerData *WalSummarizerCtl; + +/* + * When we reach end of WAL and need to read more, we sleep for a number of + * milliseconds that is a integer multiple of MS_PER_SLEEP_QUANTUM. This is + * the multiplier. It should vary between 1 and MAX_SLEEP_QUANTA, depending + * on system activity. See summarizer_wait_for_wal() for how we adjust this. + */ +static long sleep_quanta = 1; + +/* + * The sleep time will always be a multiple of 200ms and will not exceed + * one minute (300 * 200 = 60 * 1000). + */ +#define MAX_SLEEP_QUANTA 300 +#define MS_PER_SLEEP_QUANTUM 200 + +/* + * This is a count of the number of pages of WAL that we've read since the + * last time we waited for more WAL to appear. + */ +static long pages_read_since_last_sleep = 0; + +/* + * Most recent RedoRecPtr value observed by MaybeRemoveOldWalSummaries. + */ +static XLogRecPtr redo_pointer_at_last_summary_removal = InvalidXLogRecPtr; + +/* + * GUC parameters + */ +int wal_summarize_mb = 256; +int wal_summarize_keep_time = 7 * 24 * 60; + +static XLogRecPtr GetLatestLSN(TimeLineID *tli); +static void HandleWalSummarizerInterrupts(void); +static XLogRecPtr SummarizeWAL(TimeLineID tli, bool historic, + XLogRecPtr start_lsn, bool exact, + XLogRecPtr cutoff_lsn, XLogRecPtr maximum_lsn); +static void SummarizeSmgrRecord(XLogReaderState *xlogreader, + BlockRefTable *brtab); +static void SummarizeXactRecord(XLogReaderState *xlogreader, + BlockRefTable *brtab); +static int summarizer_read_local_xlog_page(XLogReaderState *state, + XLogRecPtr targetPagePtr, + int reqLen, + XLogRecPtr targetRecPtr, + char *cur_page); +static void summarizer_wait_for_wal(void); +static void MaybeRemoveOldWalSummaries(void); + +/* + * Amount of shared memory required for this module. + */ +Size +WalSummarizerShmemSize(void) +{ + return sizeof(WalSummarizerData); +} + +/* + * Create or attach to shared memory segment for this module. + */ +void +WalSummarizerShmemInit(void) +{ + bool found; + + WalSummarizerCtl = (WalSummarizerData *) + ShmemInitStruct("Wal Summarizer Ctl", WalSummarizerShmemSize(), + &found); + + if (!found) + { + /* + * First time through, so initialize. + * + * We're just filling in dummy values here -- the real initialization + * will happen when GetOldestUnsummarizedLSN() is called for the first + * time. + */ + WalSummarizerCtl->initialized = false; + WalSummarizerCtl->summarized_tli = 0; + WalSummarizerCtl->summarized_lsn = InvalidXLogRecPtr; + WalSummarizerCtl->lsn_is_exact = false; + WalSummarizerCtl->summarizer_pgprocno = INVALID_PGPROCNO; + WalSummarizerCtl->pending_lsn = InvalidXLogRecPtr; + WalSummarizerCtl->switch_requested = false; + ConditionVariableInit(&WalSummarizerCtl->summary_file_cv); + } +} + +/* + * Entry point for walsummarizer process. + */ +void +WalSummarizerMain(void) +{ + sigjmp_buf local_sigjmp_buf; + MemoryContext context; + + /* + * Within this function, 'current_lsn' and 'current_tli' refer to the + * point from which the next WAL summary file should start. 'exact' is + * true if 'current_lsn' is known to be the start of a WAL recod or WAL + * segment, and false if it might be in the middle of a record someplace. + * + * 'switch_lsn' and 'switch_tli', if set, are the LSN at which we need to + * switch to a new timeline and the timeline to which we need to switch. + * If not set, we either haven't figured out the answers yet or we're + * already on the latest timeline. + */ + XLogRecPtr current_lsn; + TimeLineID current_tli; + bool exact; + XLogRecPtr switch_lsn = InvalidXLogRecPtr; + TimeLineID switch_tli = 0; + + ereport(DEBUG1, + (errmsg_internal("WAL summarizer started"))); + + /* + * Properly accept or ignore signals the postmaster might send us + * + * We have no particular use for SIGINT at the moment, but seems + * reasonable to treat like SIGTERM. + */ + pqsignal(SIGHUP, SignalHandlerForConfigReload); + pqsignal(SIGINT, SignalHandlerForShutdownRequest); + pqsignal(SIGTERM, SignalHandlerForShutdownRequest); + /* SIGQUIT handler was already set up by InitPostmasterChild */ + pqsignal(SIGALRM, SIG_IGN); + pqsignal(SIGPIPE, SIG_IGN); + pqsignal(SIGUSR1, procsignal_sigusr1_handler); + pqsignal(SIGUSR2, SIG_IGN); /* not used */ + + /* Advertise ourselves. */ + LWLockAcquire(WALSummarizerLock, LW_EXCLUSIVE); + WalSummarizerCtl->summarizer_pgprocno = MyProc->pgprocno; + LWLockRelease(WALSummarizerLock); + + /* Create and switch to a memory context that we can reset on error. */ + context = AllocSetContextCreate(TopMemoryContext, + "Wal Summarizer", + ALLOCSET_DEFAULT_SIZES); + MemoryContextSwitchTo(context); + + /* + * Reset some signals that are accepted by postmaster but not here + */ + pqsignal(SIGCHLD, SIG_DFL); + + /* + * If an exception is encountered, processing resumes here. + */ + if (sigsetjmp(local_sigjmp_buf, 1) != 0) + { + /* Since not using PG_TRY, must reset error stack by hand */ + error_context_stack = NULL; + + /* Prevent interrupts while cleaning up */ + HOLD_INTERRUPTS(); + + /* Report the error to the server log */ + EmitErrorReport(); + + /* Release resources we might have acquired. */ + LWLockReleaseAll(); + ConditionVariableCancelSleep(); + pgstat_report_wait_end(); + ReleaseAuxProcessResources(false); + AtEOXact_Files(false); + AtEOXact_HashTables(false); + + /* + * Now return to normal top-level context and clear ErrorContext for + * next time. + */ + MemoryContextSwitchTo(context); + FlushErrorState(); + + /* Flush any leaked data in the top-level context */ + MemoryContextResetAndDeleteChildren(context); + + /* Now we can allow interrupts again */ + RESUME_INTERRUPTS(); + + /* + * Sleep for 10 seconds before attempting to resume operations in + * order to avoid excessing logging. + * + * Many of the likely error conditions are things that will repeat + * every time. For example, if the WAL can't be read or the summary + * can't be written, only administrator action will cure the problem. + * So a really fast retry time doesn't seem to be especially + * beneficial, and it will clutter the logs. + */ + (void) WaitLatch(MyLatch, + WL_TIMEOUT | WL_EXIT_ON_PM_DEATH, + 10000, + WAIT_EVENT_WAL_SUMMARIZER_ERROR); + } + + /* We can now handle ereport(ERROR) */ + PG_exception_stack = &local_sigjmp_buf; + + /* + * Unblock signals (they were blocked when the postmaster forked us) + */ + sigprocmask(SIG_SETMASK, &UnBlockSig, NULL); + + /* + * Fetch information about previous progress from shared memory. + * + * If we discover that WAL summarization is not enabled, just exit. + */ + current_lsn = GetOldestUnsummarizedLSN(¤t_tli, &exact); + if (XLogRecPtrIsInvalid(current_lsn)) + proc_exit(0); + + /* + * Loop forever + */ + for (;;) + { + XLogRecPtr latest_lsn; + TimeLineID latest_tli; + XLogRecPtr cutoff_lsn; + XLogRecPtr end_of_summary_lsn; + + /* Flush any leaked data in the top-level context */ + MemoryContextResetAndDeleteChildren(context); + + /* Process any signals received recently. */ + HandleWalSummarizerInterrupts(); + + /* If it's time to remove any old WAL summaries, do that now. */ + MaybeRemoveOldWalSummaries(); + + /* Find the LSN and TLI up to which we can safely summarize. */ + latest_lsn = GetLatestLSN(&latest_tli); + + /* + * If we're summarizing a historic timeline and we haven't yet + * computed the point at which to switch to the next timeline, do that + * now. + * + * Note that if this is a standby, what was previously the current + * timeline could become historic at any time. + * + * We could try to make this more efficient by caching the results of + * readTimeLineHistory when latest_tli has not changed, but since we + * only have to do this once per timeline switch, we probably wouldn't + * save any significant amount of work in practice. + */ + if (current_tli != latest_tli && XLogRecPtrIsInvalid(switch_lsn)) + { + List *tles = readTimeLineHistory(latest_tli); + + switch_lsn = tliSwitchPoint(current_tli, tles, &switch_tli); + elog(DEBUG2, + "switch point from TLI %u to TLI %u is at %X/%X", + current_tli, switch_tli, LSN_FORMAT_ARGS(switch_lsn)); + } + + /* + * wal_summarize_mb sets a soft limit on the amont of WAL covered by a + * single summary file. If we read a WAL record that ends after the + * cutoff LSN computed here, we'll stop the summary. In most cases, it + * will actually stop earlier than that, but this is here as a + * backstop. + */ + cutoff_lsn = current_lsn + wal_summarize_mb * 1024 * 1024; + if (!XLogRecPtrIsInvalid(switch_lsn) && cutoff_lsn > switch_lsn) + cutoff_lsn = switch_lsn; + elog(DEBUG2, + "WAL summarization cutoff is TLI %d @ %X/%X, flush position is %X/%X", + current_tli, LSN_FORMAT_ARGS(cutoff_lsn), LSN_FORMAT_ARGS(latest_lsn)); + + /* Summarize WAL. */ + end_of_summary_lsn = SummarizeWAL(current_tli, + current_tli != latest_tli, + current_lsn, exact, + cutoff_lsn, latest_lsn); + Assert(!XLogRecPtrIsInvalid(end_of_summary_lsn)); + Assert(end_of_summary_lsn >= current_lsn); + + /* + * Update state for next loop iteration. + * + * Next summary file should start from exactly where this one ended. + * Timeline remains unchanged unless a switch LSN was computed and we + * have reached it. + */ + current_lsn = end_of_summary_lsn; + exact = true; + if (!XLogRecPtrIsInvalid(switch_lsn) && cutoff_lsn >= switch_lsn) + { + current_tli = switch_tli; + switch_lsn = InvalidXLogRecPtr; + switch_tli = 0; + } + + /* Update state in shared memory. */ + LWLockAcquire(WALSummarizerLock, LW_EXCLUSIVE); + Assert(WalSummarizerCtl->pending_lsn <= end_of_summary_lsn); + WalSummarizerCtl->summarized_lsn = end_of_summary_lsn; + WalSummarizerCtl->summarized_tli = current_tli; + WalSummarizerCtl->lsn_is_exact = true; + WalSummarizerCtl->pending_lsn = end_of_summary_lsn; + WalSummarizerCtl->switch_requested = false; + LWLockRelease(WALSummarizerLock); + + /* Wake up anyone waiting for more summary files to be written. */ + ConditionVariableBroadcast(&WalSummarizerCtl->summary_file_cv); + } +} + +/* + * Get the oldest LSN in this server's timeline history that has not yet been + * summarized. + * + * If *tli != NULL, it will be set to the TLI for the LSN that is returned. + * + * If *lsn_is_exact != NULL, it will be set to true if the returned LSN is + * necessarily the start of a WAL record and false if it's just the beginning + * of a WAL segment. + */ +XLogRecPtr +GetOldestUnsummarizedLSN(TimeLineID *tli, bool *lsn_is_exact) +{ + TimeLineID latest_tli; + LWLockMode mode = LW_SHARED; + int n; + List *tles; + XLogRecPtr unsummarized_lsn; + TimeLineID unsummarized_tli = 0; + bool should_make_exact = false; + List *existing_summaries; + ListCell *lc; + + /* If not summarizing WAL, do nothing. */ + if (wal_summarize_mb == 0) + return InvalidXLogRecPtr; + + /* + * Initially, we acquire the lock in shared mode and try to fetch the + * required information. If the data structure hasn't been initialized, we + * reacquire the lock in shared mode so that we can initialize it. + * However, if someone else does that first before we get the lock, then + * we can just return the requested information after all. + */ + while (true) + { + LWLockAcquire(WALSummarizerLock, mode); + + if (WalSummarizerCtl->initialized) + { + unsummarized_lsn = WalSummarizerCtl->summarized_lsn; + if (tli != NULL) + *tli = WalSummarizerCtl->summarized_tli; + if (lsn_is_exact != NULL) + *lsn_is_exact = WalSummarizerCtl->lsn_is_exact; + LWLockRelease(WALSummarizerLock); + return unsummarized_lsn; + } + + if (mode == LW_EXCLUSIVE) + break; + + LWLockRelease(WALSummarizerLock); + mode = LW_EXCLUSIVE; + } + + /* + * The data structure needs to be initialized, and we are the first to + * obtain the lock in exclusive mode, so it's our job to do that + * initialization. + * + * So, find the oldest timeline on which WAL still exists, and the + * earliest segment for which it exists. + */ + (void) GetLatestLSN(&latest_tli); + tles = readTimeLineHistory(latest_tli); + for (n = list_length(tles) - 1; n >= 0; --n) + { + TimeLineHistoryEntry *tle = list_nth(tles, n); + XLogSegNo oldest_segno; + + oldest_segno = XLogGetOldestSegno(tle->tli); + if (oldest_segno != 0) + { + /* Compute oldest LSN that still exists on disk. */ + XLogSegNoOffsetToRecPtr(oldest_segno, 0, wal_segment_size, + unsummarized_lsn); + + unsummarized_tli = tle->tli; + break; + } + } + + /* It really should not be possible for us to find no WAL. */ + if (unsummarized_tli == 0) + ereport(ERROR, + errcode(ERRCODE_INTERNAL_ERROR), + errmsg_internal("no WAL found on timeline %d", latest_tli)); + + /* + * Don't try to summarize anything older than the end LSN of the newest + * summary file that exists for this timeline. + */ + existing_summaries = + GetWalSummaries(unsummarized_tli, + InvalidXLogRecPtr, InvalidXLogRecPtr); + foreach(lc, existing_summaries) + { + WalSummaryFile *ws = lfirst(lc); + + if (ws->end_lsn > unsummarized_lsn) + { + unsummarized_lsn = ws->end_lsn; + should_make_exact = true; + } + } + + /* Update shared memory with the discovered values. */ + WalSummarizerCtl->initialized = true; + WalSummarizerCtl->summarized_lsn = unsummarized_lsn; + WalSummarizerCtl->summarized_tli = unsummarized_tli; + WalSummarizerCtl->lsn_is_exact = should_make_exact; + WalSummarizerCtl->pending_lsn = unsummarized_lsn; + + /* Also return the to the caller as required. */ + if (tli != NULL) + *tli = WalSummarizerCtl->summarized_tli; + if (lsn_is_exact != NULL) + *lsn_is_exact = WalSummarizerCtl->lsn_is_exact; + LWLockRelease(WALSummarizerLock); + + return unsummarized_lsn; +} + +/* + * Attempt to set the WAL summarizer's latch. + * + * This might not work, because there's no guarantee that the WAL summarizer + * process was successfully started, and it also might have started but + * subsequently terminated. So, under normal circumstances, this will get the + * latch set, but there's no guarantee. + */ +void +SetWalSummarizerLatch(void) +{ + int pgprocno; + + LWLockAcquire(WALSummarizerLock, LW_EXCLUSIVE); + pgprocno = WalSummarizerCtl->summarizer_pgprocno; + LWLockRelease(WALSummarizerLock); + + if (pgprocno != INVALID_PGPROCNO) + SetLatch(&ProcGlobal->allProcs[pgprocno].procLatch); +} + +/* + * Wait until WAL summarization reaches the given LSN, but not longer than + * the given timeout. + * + * The return value is the first still-unsummarized LSN. If it's greater than + * or equal to the passed LSN, then that LSN was reached. If not, we timed out. + */ +XLogRecPtr +WaitForWalSummarization(XLogRecPtr lsn, long timeout) +{ + TimestampTz start_time = GetCurrentTimestamp(); + TimestampTz deadline = TimestampTzPlusMilliseconds(start_time, timeout); + XLogRecPtr summarized_lsn; + + Assert(!XLogRecPtrIsInvalid(lsn)); + Assert(timeout > 0); + + while (1) + { + TimestampTz now; + long remaining_timeout; + + /* + * If the LSN summarized on disk has reached the target value, stop. + * If it hasn't, but the in-memory value has reached the target value, + * request that a file be written as soon as possible. + */ + LWLockAcquire(WALSummarizerLock, LW_EXCLUSIVE); + summarized_lsn = WalSummarizerCtl->summarized_lsn; + if (summarized_lsn < lsn && + WalSummarizerCtl->pending_lsn >= lsn) + WalSummarizerCtl->switch_requested = true; + LWLockRelease(WALSummarizerLock); + if (summarized_lsn >= lsn) + break; + + /* Timeout reached? If yes, stop. */ + now = GetCurrentTimestamp(); + remaining_timeout = TimestampDifferenceMilliseconds(now, deadline); + if (remaining_timeout <= 0) + break; + + /* + * Limit the sleep to 1 second, because we may need to request a + * switch. + */ + if (remaining_timeout > 1000) + remaining_timeout = 1000; + + /* Wait and see. */ + ConditionVariableTimedSleep(&WalSummarizerCtl->summary_file_cv, + remaining_timeout, + WAIT_EVENT_WAL_SUMMARY_READY); + } + + return summarized_lsn; +} + +/* + * Get the latest LSN that is eligible to be summarized, and set *tli to the + * corresponding timeline. + */ +static XLogRecPtr +GetLatestLSN(TimeLineID *tli) +{ + if (!RecoveryInProgress()) + { + /* Don't summarize WAL before it's flushed. */ + return GetFlushRecPtr(tli); + } + else + { + XLogRecPtr flush_lsn; + TimeLineID flush_tli; + XLogRecPtr replay_lsn; + TimeLineID replay_tli; + + /* + * What we really want to know is how much WAL has been flushed to + * disk, but the only flush position available is the one provided by + * the walreceiver, which may not be running, because this could be + * crash recovery or recovery via restore_command. So use either the + * WAL receiver's flush position or the replay position, whichever is + * further ahead, on the theory that if the WAL has been replayed then + * it must also have been flushed to disk. + */ + flush_lsn = GetWalRcvFlushRecPtr(NULL, &flush_tli); + replay_lsn = GetXLogReplayRecPtr(&replay_tli); + if (flush_lsn > replay_lsn) + { + *tli = flush_tli; + return flush_lsn; + } + else + { + *tli = replay_tli; + return replay_lsn; + } + } +} + +/* + * Interrupt handler for main loop of WAL summarizer process. + */ +static void +HandleWalSummarizerInterrupts(void) +{ + if (ProcSignalBarrierPending) + ProcessProcSignalBarrier(); + + if (ConfigReloadPending) + { + ConfigReloadPending = false; + ProcessConfigFile(PGC_SIGHUP); + } + + if (ShutdownRequestPending || wal_summarize_mb == 0) + { + ereport(DEBUG1, + errmsg_internal("WAL summarizer shutting down")); + proc_exit(0); + } + + /* Perform logging of memory contexts of this process */ + if (LogMemoryContextPending) + ProcessLogMemoryContextInterrupt(); +} + +/* + * Summarize a range of WAL records on a single timeline. + * + * 'tli' is the timeline to be summarized. 'historic' should be false if the + * timeline in question is the latest one and true otherwise. + * + * 'start_lsn' is the point at which we should start summarizing. If this + * value comes from the end LSN of the previous record as returned by the + * xlograder machinery, 'exact' should be true; otherwise, 'exact' should + * be false, and this function will search forward for the start of a valid + * WAL record. + * + * 'cutoff_lsn' is the point at which we should stop summarizing. The first + * record that ends at or after cutoff_lsn will be the last one included + * in the summary. + * + * 'maximum_lsn' identifies the point beyond which we can't count on being + * able to read any more WAL. It should be the switch point when reading a + * historic timeline, or the most-recently-measured end of WAL when reading + * the current timeline. + * + * The return value is the LSN at which the WAL summary actually ends. Most + * often, a summary file ends because we notice that a checkpoint has + * occurred and reach the redo pointer of that checkpoint, but sometimes + * we stop for other reasons, such as a timeline switch, or reading a record + * that ends after the cutoff_lsn. + */ +static XLogRecPtr +SummarizeWAL(TimeLineID tli, bool historic, + XLogRecPtr start_lsn, bool exact, + XLogRecPtr cutoff_lsn, XLogRecPtr maximum_lsn) +{ + SummarizerReadLocalXLogPrivate *private_data; + XLogReaderState *xlogreader; + XLogRecPtr summary_start_lsn; + XLogRecPtr summary_end_lsn = cutoff_lsn; + char temp_path[MAXPGPATH]; + char final_path[MAXPGPATH]; + WalSummaryIO io; + BlockRefTable *brtab = CreateEmptyBlockRefTable(); + + /* Initialize private data for xlogreader. */ + private_data = (SummarizerReadLocalXLogPrivate *) + palloc0(sizeof(SummarizerReadLocalXLogPrivate)); + private_data->tli = tli; + private_data->historic = historic; + private_data->read_upto = maximum_lsn; + private_data->redo_pointer = GetRedoRecPtr(); + private_data->redo_pointer_refresh_lsn = start_lsn; + private_data->redo_pointer_reached = + (start_lsn >= private_data->redo_pointer); + + /* Create xlogreader. */ + xlogreader = XLogReaderAllocate(wal_segment_size, NULL, + XL_ROUTINE(.page_read = &summarizer_read_local_xlog_page, + .segment_open = &wal_segment_open, + .segment_close = &wal_segment_close), + private_data); + if (xlogreader == NULL) + ereport(ERROR, + (errcode(ERRCODE_OUT_OF_MEMORY), + errmsg("out of memory"), + errdetail("Failed while allocating a WAL reading processor."))); + + /* + * When exact = false, we're starting from an arbitrary point in the WAL + * and must search forward for the start of the next record. + * + * When exact = true, start_lsn should be either the LSN where a record + * begins, or the LSN of a page where the page header is immediately + * followed by the start of a new record. XLogBeginRead should tolerate + * either case. + * + * We need to allow for both cases because the behavior of xlogreader + * varies. When a record spans two or more xlog pages, the ending LSN + * reported by xlogreader will be the starting LSN of the following + * record, but when an xlog page boundary falls between two records, the + * end LSN for the first will be reported as the first byte of the + * following page. We can't know until we read that page how large the + * header will be, but we'll have to skip over it to find the next record. + */ + if (exact) + { + /* + * Even if start_lsn is the beginning of a page rather than the + * beginning of the first record on that page, we should still use it + * as the start LSN for the summary file. That's because we detect + * missing summary files by looking for cases where the end LSN of one + * file is less than the start LSN of the next file. When only a page + * header is skipped, nothing has been missed. + */ + XLogBeginRead(xlogreader, start_lsn); + summary_start_lsn = start_lsn; + } + else + { + summary_start_lsn = XLogFindNextRecord(xlogreader, start_lsn); + if (XLogRecPtrIsInvalid(summary_start_lsn)) + { + /* + * If we hit end-of-WAL while trying to find the next valid + * record, we must be on a historic timeline that has no valid + * records that begin after start_lsn and before end of WAL. + */ + if (private_data->end_of_wal) + { + ereport(DEBUG1, + errmsg_internal("could not read WAL from timeline %d at %X/%X: end of WAL at %X/%X", + tli, + LSN_FORMAT_ARGS(start_lsn), + LSN_FORMAT_ARGS(private_data->read_upto))); + + /* + * The timeline ends at or after start_lsn, without containing + * any records. Thus, we must make sure the main loop does not + * iterate. If start_lsn is the end of the timeline, then we + * won't actually emit an empty summary file, but otherwise, + * we must, to capture the fact that the LSN range in question + * contains no interesting WAL records. + */ + summary_start_lsn = start_lsn; + summary_end_lsn = private_data->read_upto; + cutoff_lsn = xlogreader->EndRecPtr; + } + else + ereport(ERROR, + (errmsg("could not find a valid record after %X/%X", + LSN_FORMAT_ARGS(start_lsn)))); + } + + /* We shouldn't go backward. */ + Assert(summary_start_lsn >= start_lsn); + } + + /* + * Main loop: read xlog records one by one. + */ + while (xlogreader->EndRecPtr < cutoff_lsn) + { + int block_id; + char *errormsg; + XLogRecord *record; + bool switch_requested; + + /* We shouldn't go backward. */ + Assert(summary_start_lsn <= xlogreader->EndRecPtr); + + /* + * This flag tracks whether the read of a particular record had to + * wait for more WAL to arrive, so reset it before reading the next + * record. + */ + private_data->waited = false; + + /* Now read the next record. */ + record = XLogReadRecord(xlogreader, &errormsg); + if (record == NULL) + { + SummarizerReadLocalXLogPrivate *private_data; + + private_data = (SummarizerReadLocalXLogPrivate *) + xlogreader->private_data; + if (private_data->end_of_wal) + { + /* + * This timeline must be historic and must end before we were + * able to read a complete record. + */ + ereport(DEBUG1, + errmsg_internal("could not read WAL from timeline %d at %X/%X: end of WAL at %X/%X", + tli, + LSN_FORMAT_ARGS(xlogreader->EndRecPtr), + LSN_FORMAT_ARGS(private_data->read_upto))); + /* Summary ends at end of WAL. */ + summary_end_lsn = private_data->read_upto; + break; + } + if (errormsg) + ereport(ERROR, + (errcode_for_file_access(), + errmsg("could not read WAL at %X/%X: %s", + LSN_FORMAT_ARGS(xlogreader->EndRecPtr), errormsg))); + else + ereport(ERROR, + (errcode_for_file_access(), + errmsg("could not read WAL at %X/%X", + LSN_FORMAT_ARGS(xlogreader->EndRecPtr)))); + } + + /* We shouldn't go backward. */ + Assert(summary_start_lsn <= xlogreader->EndRecPtr); + + if (xlogreader->ReadRecPtr >= cutoff_lsn) + { + /* + * Woops! We've read a record that *starts* after the cutoff LSN, + * contrary to our goal of reading only until we hit the first + * record that ends at or after the cutoff LSN. Pretend we didn't + * read it after all by bailing out of this loop right here, + * before we do anything with this record. + * + * This can happen because the last record before the cutoff LSN + * might be continued across multiple pages, and then we might + * come to a page with XLP_FIRST_IS_OVERWRITE_CONTRECORD set. In + * that case, the record that was continued across multiple pages + * is incomplete and will be disregarded, and the read will + * restart from the beginning of the page that is flagged + * XLP_FIRST_IS_OVERWRITE_CONTRECORD. + * + * If this case occurs, we can fairly say that the current summary + * file ends at the cutoff LSN exactly. The first record on the + * page marked XLP_FIRST_IS_OVERWRITE_CONTRECORD will be + * discovered when generating the next summary file. + */ + summary_end_lsn = cutoff_lsn; + break; + } + + /* + * We attempt, on a best effort basis only, to make WAL summary file + * boundaries line up with checkpoint cycles. So, if the last redo + * pointer we've seen was in the future, and this record starts at + * that redo pointer, stop before processing and let it be included in + * the next summary file. + * + * Note that in the case of a checkpoint triggered by a backup, the + * redo pointer is likely to be pointing to the first record on a + * page. Before reading the record, xlogreader->EndRecPtr will have + * pointed to the start of the page, which precedes the redo LSN. But + * after reading the next record, we'll advance over the page header + * and realize that the next record starts at the redo LSN exactly, + * making this the first point at which we can realize that it's time + * to stop. + */ + if (!private_data->redo_pointer_reached && + xlogreader->ReadRecPtr >= private_data->redo_pointer) + { + summary_end_lsn = xlogreader->ReadRecPtr; + break; + } + + /* Special handling for particular types of WAL records. */ + switch (XLogRecGetRmid(xlogreader)) + { + case RM_SMGR_ID: + SummarizeSmgrRecord(xlogreader, brtab); + break; + case RM_XACT_ID: + SummarizeXactRecord(xlogreader, brtab); + break; + default: + break; + } + + /* Feed block references from xlog record to block reference table. */ + for (block_id = 0; block_id <= XLogRecMaxBlockId(xlogreader); + block_id++) + { + RelFileLocator rlocator; + ForkNumber forknum; + BlockNumber blocknum; + + if (!XLogRecGetBlockTagExtended(xlogreader, block_id, &rlocator, + &forknum, &blocknum, NULL)) + continue; + + BlockRefTableMarkBlockModified(brtab, &rlocator, forknum, + blocknum); + } + + /* Update our notion of where this summary file ends. */ + summary_end_lsn = xlogreader->EndRecPtr; + + /* + * Also update shared memory, and handle any request for a WAL summary + * file switch. + */ + LWLockAcquire(WALSummarizerLock, LW_EXCLUSIVE); + Assert(summary_end_lsn >= WalSummarizerCtl->pending_lsn); + Assert(summary_end_lsn >= WalSummarizerCtl->summarized_lsn); + WalSummarizerCtl->pending_lsn = summary_end_lsn; + switch_requested = WalSummarizerCtl->switch_requested; + LWLockRelease(WALSummarizerLock); + if (switch_requested) + break; + + /* + * Periodically update our notion of the redo pointer, because it + * might be changing concurrently. There's no interlocking here: we + * might race past the new redo pointer before we learn about it. + * That's OK; we only use the redo pointer as a heuristic for where to + * stop summarizing. + * + * It would be nice if we could just fetch the updated redo pointer on + * every pass through this loop, but that seems a bit too expensive: + * GetRedoRecPtr acquires a heavily-contended spinlock. So, instead, + * just fetch the updated value if we've just had to sleep, or if + * we've read more than a segment's worth of WAL without sleeping. + */ + if (private_data->waited || xlogreader->EndRecPtr > + private_data->redo_pointer_refresh_lsn + wal_segment_size) + { + private_data->redo_pointer = GetRedoRecPtr(); + private_data->redo_pointer_refresh_lsn = xlogreader->EndRecPtr; + private_data->redo_pointer_reached = + (xlogreader->EndRecPtr >= private_data->redo_pointer); + } + + /* + * Recheck whether we've just caught up with the redo pointer, and if + * so, stop. This has the same purpose as the earlier check for the + * same condition above, but there we've just read a record and might + * decide against including it in the current summary file, whereas + * here we've already included it and might decide against reading the + * next one. Note that we may have just refreshed our notion of the + * redo pointer, so it's smart to check here before we do any more + * work. + */ + if (!private_data->redo_pointer_reached && + xlogreader->EndRecPtr >= private_data->redo_pointer) + break; + } + + /* Destroy xlogreader. */ + pfree(xlogreader->private_data); + XLogReaderFree(xlogreader); + + /* + * If a timeline switch occurs, we may fail to make any progress at all + * before exiting the loop above. If that happens, we don't write a WAL + * summary file at all. + */ + if (summary_end_lsn > summary_start_lsn) + { + /* Generate temporary and final path name. */ + snprintf(temp_path, MAXPGPATH, + XLOGDIR "/summaries/temp.summary"); + snprintf(final_path, MAXPGPATH, + XLOGDIR "/summaries/%08X%08X%08X%08X%08X.summary", + tli, + LSN_FORMAT_ARGS(summary_start_lsn), + LSN_FORMAT_ARGS(summary_end_lsn)); + + /* Open the temporary file for writing. */ + io.filepos = 0; + io.file = PathNameOpenFile(temp_path, O_WRONLY | O_CREAT | O_TRUNC); + if (io.file < 0) + ereport(ERROR, + (errcode_for_file_access(), + errmsg("could not create file \"%s\": %m", temp_path))); + + /* Write the data. */ + WriteBlockRefTable(brtab, WriteWalSummary, &io); + + /* Close temporary file and shut down xlogreader. */ + FileClose(io.file); + + /* Tell the user what we did. */ + ereport(LOG, + errmsg("summarized WAL on TLI %d from %X/%X to %X/%X", + tli, + LSN_FORMAT_ARGS(summary_start_lsn), + LSN_FORMAT_ARGS(summary_end_lsn))); + + /* Durably rename the new summary into place. */ + durable_rename(temp_path, final_path, ERROR); + } + + return summary_end_lsn; +} + +/* + * Special handling for WAL records with RM_SMGR_ID. + */ +static void +SummarizeSmgrRecord(XLogReaderState *xlogreader, BlockRefTable *brtab) +{ + uint8 info = XLogRecGetInfo(xlogreader) & ~XLR_INFO_MASK; + + if (info == XLOG_SMGR_CREATE) + { + xl_smgr_create *xlrec; + + /* + * If a new relation fork is created on disk, there is no point + * tracking anything about which blocks have been modified, because + * the whole thing will be new. Hence, set the limit block for this + * fork to 0. + * + * Ignore the FSM fork, which is not fully WAL-logged. + */ + xlrec = (xl_smgr_create *) XLogRecGetData(xlogreader); + + if (xlrec->forkNum != FSM_FORKNUM) + BlockRefTableSetLimitBlock(brtab, &xlrec->rlocator, + xlrec->forkNum, 0); + } + else if (info == XLOG_SMGR_TRUNCATE) + { + xl_smgr_truncate *xlrec; + + xlrec = (xl_smgr_truncate *) XLogRecGetData(xlogreader); + + /* + * If a relation fork is truncated on disk, there is in point in + * tracking anything about block modifications beyond the truncation + * point. + * + * We ignore SMGR_TRUNCATE_FSM here because the FSM isn't fully + * WAL-logged and thus we can't track modified blocks for it anyway. + */ + if ((xlrec->flags & SMGR_TRUNCATE_HEAP) != 0) + BlockRefTableSetLimitBlock(brtab, &xlrec->rlocator, + MAIN_FORKNUM, xlrec->blkno); + if ((xlrec->flags & SMGR_TRUNCATE_VM) != 0) + BlockRefTableSetLimitBlock(brtab, &xlrec->rlocator, + VISIBILITYMAP_FORKNUM, xlrec->blkno); + } +} + +/* + * Special handling for WAL recods with RM_XACT_ID. + */ +static void +SummarizeXactRecord(XLogReaderState *xlogreader, BlockRefTable *brtab) +{ + uint8 info = XLogRecGetInfo(xlogreader) & ~XLR_INFO_MASK; + uint8 xact_info = info & XLOG_XACT_OPMASK; + + if (xact_info == XLOG_XACT_COMMIT || + xact_info == XLOG_XACT_COMMIT_PREPARED) + { + xl_xact_commit *xlrec = (xl_xact_commit *) XLogRecGetData(xlogreader); + xl_xact_parsed_commit parsed; + int i; + + ParseCommitRecord(XLogRecGetInfo(xlogreader), xlrec, &parsed); + for (i = 0; i < parsed.nrels; ++i) + { + ForkNumber forknum; + + for (forknum = 0; forknum <= MAX_FORKNUM; ++forknum) + if (forknum != FSM_FORKNUM) + BlockRefTableSetLimitBlock(brtab, &parsed.xlocators[i], + forknum, 0); + } + } + else if (xact_info == XLOG_XACT_ABORT || + xact_info == XLOG_XACT_ABORT_PREPARED) + { + xl_xact_abort *xlrec = (xl_xact_abort *) XLogRecGetData(xlogreader); + xl_xact_parsed_abort parsed; + int i; + + ParseAbortRecord(XLogRecGetInfo(xlogreader), xlrec, &parsed); + for (i = 0; i < parsed.nrels; ++i) + { + ForkNumber forknum; + + for (forknum = 0; forknum <= MAX_FORKNUM; ++forknum) + if (forknum != FSM_FORKNUM) + BlockRefTableSetLimitBlock(brtab, &parsed.xlocators[i], + forknum, 0); + } + } +} + +/* + * Similar to read_local_xlog_page, but limited to read from one particular + * timeline. If the end of WAL is reached, it will wait for more if reading + * from the current timeline, or give up if reading from a historic timeline. + * In the latter case, it will also set private_data->end_of_wal = true. + * + * Caller must set private_data->tli to the TLI of interest, + * private_data->read_upto to the lowest LSN that is not known to be safe + * to read on that timeline, and private_data->historic to true if and only + * if the timeline is not the current timeline. This function will update + * private_data->read_upto and private_data->historic if more WAL appears + * on the current timeline or if the current timeline becomes historic. + */ +static int +summarizer_read_local_xlog_page(XLogReaderState *state, + XLogRecPtr targetPagePtr, int reqLen, + XLogRecPtr targetRecPtr, char *cur_page) +{ + int count; + WALReadError errinfo; + SummarizerReadLocalXLogPrivate *private_data; + + private_data = (SummarizerReadLocalXLogPrivate *) + state->private_data; + + while (true) + { + if (targetPagePtr + XLOG_BLCKSZ <= private_data->read_upto) + { + /* + * more than one block available; read only that block, have + * caller come back if they need more. + */ + count = XLOG_BLCKSZ; + break; + } + else if (targetPagePtr + reqLen > private_data->read_upto) + { + /* We don't seem to have enough data. */ + if (private_data->historic) + { + /* + * This is a historic timeline, so there will never be any + * more data than we have currently. + */ + private_data->end_of_wal = true; + return -1; + } + else + { + XLogRecPtr latest_lsn; + TimeLineID latest_tli; + + /* + * This is - or at least was up until very recently - the + * current timeline, so more data might show up. Delay here + * so we don't tight-loop. + */ + HandleWalSummarizerInterrupts(); + summarizer_wait_for_wal(); + private_data->waited = true; + + /* Recheck end-of-WAL. */ + latest_lsn = GetLatestLSN(&latest_tli); + if (private_data->tli == latest_tli) + { + /* Still the current timeline, update max LSN. */ + Assert(latest_lsn >= private_data->read_upto); + private_data->read_upto = latest_lsn; + } + else + { + List *tles = readTimeLineHistory(latest_tli); + XLogRecPtr switchpoint; + + /* + * The timeline we're scanning is no longer the latest + * one. Figure out when it ended and allow reads up to + * exactly that point. + */ + private_data->historic = true; + switchpoint = tliSwitchPoint(private_data->tli, tles, + NULL); + Assert(switchpoint >= private_data->read_upto); + private_data->read_upto = switchpoint; + } + + /* Go around and try again. */ + } + } + else + { + /* enough bytes available to satisfy the request */ + count = private_data->read_upto - targetPagePtr; + break; + } + } + + /* + * Even though we just determined how much of the page can be validly read + * as 'count', read the whole page anyway. It's guaranteed to be + * zero-padded up to the page boundary if it's incomplete. + */ + if (!WALRead(state, cur_page, targetPagePtr, XLOG_BLCKSZ, + private_data->tli, &errinfo)) + WALReadRaiseError(&errinfo); + + /* Track that we read a page, for sleep time calculation. */ + ++pages_read_since_last_sleep; + + /* number of valid bytes in the buffer */ + return count; +} + +/* + * Sleep for long enough that we believe it's likely that more WAL will + * be available afterwards. + */ +static void +summarizer_wait_for_wal(void) +{ + if (pages_read_since_last_sleep == 0) + { + /* + * No pages were read since the last sleep, so double the sleep time, + * but not beyond the maximum allowable value. + */ + sleep_quanta = Min(sleep_quanta * 2, MAX_SLEEP_QUANTA); + } + else if (pages_read_since_last_sleep > 1) + { + /* + * Multiple pages were read since the last sleep, so reduce the sleep + * time. + * + * A large burst of activity should be able to quickly reduce the + * sleep time to the minimum, but we don't want a handful of extra WAL + * records to provoke a strong reaction. We choose to reduce the sleep + * time by 1 quantum for each page read beyond the first, which is a + * fairly arbitrary way of trying to be reactive without + * overrreacting. + */ + if (pages_read_since_last_sleep > sleep_quanta - 1) + sleep_quanta = 1; + else + sleep_quanta -= pages_read_since_last_sleep; + } + + /* OK, now sleep. */ + (void) WaitLatch(MyLatch, + WL_LATCH_SET | WL_TIMEOUT | WL_EXIT_ON_PM_DEATH, + sleep_quanta * MS_PER_SLEEP_QUANTUM, + WAIT_EVENT_WAL_SUMMARIZER_WAL); + ResetLatch(MyLatch); + + /* Reset count of pages read. */ + pages_read_since_last_sleep = 0; +} + +/* + * Most recent RedoRecPtr value observed by RemoveOldWalSummaries. + */ +static void +MaybeRemoveOldWalSummaries(void) +{ + XLogRecPtr redo_pointer = GetRedoRecPtr(); + List *wslist; + time_t cutoff_time; + + /* If WAL summary removal is disabled, don't do anything. */ + if (wal_summarize_keep_time == 0) + return; + + /* + * If the redo pointer has not advanced, don't do anything. + * + * This has the effect that we only try to remove old WAL summary files + * once per checkpoint cycle. + */ + if (redo_pointer == redo_pointer_at_last_summary_removal) + return; + redo_pointer_at_last_summary_removal = redo_pointer; + + /* + * Files should only be removed if the last modification time precedes the + * cutoff time we compute here. + */ + cutoff_time = time(NULL) - 60 * wal_summarize_keep_time; + + /* Get all the summaries that currently exist. */ + wslist = GetWalSummaries(0, InvalidXLogRecPtr, InvalidXLogRecPtr); + + /* Loop until all summaries have been considered for removal. */ + while (wslist != NIL) + { + ListCell *lc; + XLogSegNo oldest_segno; + XLogRecPtr oldest_lsn = InvalidXLogRecPtr; + TimeLineID selected_tli; + + CHECK_FOR_INTERRUPTS(); + + /* + * Pick a timeline for which some summary files still exist on disk, + * and find the oldest LSN that still exists on disk for that + * timeline. + */ + selected_tli = ((WalSummaryFile *) linitial(wslist))->tli; + oldest_segno = XLogGetOldestSegno(selected_tli); + if (oldest_segno != 0) + XLogSegNoOffsetToRecPtr(oldest_segno, 0, wal_segment_size, + oldest_lsn); + + + /* Consider each WAL file on the selected timeline in turn. */ + foreach(lc, wslist) + { + WalSummaryFile *ws = lfirst(lc); + + CHECK_FOR_INTERRUPTS(); + + /* If it's not on this timeline, it's not time to consider it. */ + if (selected_tli != ws->tli) + continue; + + /* + * If the WAL doesn't exist any more, we can remove it if the file + * modification time is old enough. + */ + if (XLogRecPtrIsInvalid(oldest_lsn) || ws->end_lsn <= oldest_lsn) + RemoveWalSummaryIfOlderThan(ws, cutoff_time); + + /* + * Whether we we removed the file or not, we need not consider it + * again. + */ + wslist = foreach_delete_current(wslist, lc); + pfree(ws); + } + } +} diff --git a/src/backend/replication/repl_gram.y b/src/backend/replication/repl_gram.y index 0c874e33cf..a5d118ed68 100644 --- a/src/backend/replication/repl_gram.y +++ b/src/backend/replication/repl_gram.y @@ -76,11 +76,12 @@ Node *replication_parse_result; %token K_EXPORT_SNAPSHOT %token K_NOEXPORT_SNAPSHOT %token K_USE_SNAPSHOT +%token K_UPLOAD_MANIFEST %type <node> command %type <node> base_backup start_replication start_logical_replication create_replication_slot drop_replication_slot identify_system - read_replication_slot timeline_history show + read_replication_slot timeline_history show upload_manifest %type <list> generic_option_list %type <defelt> generic_option %type <uintval> opt_timeline @@ -114,6 +115,7 @@ command: | read_replication_slot | timeline_history | show + | upload_manifest ; /* @@ -307,6 +309,15 @@ timeline_history: } ; +/* UPLOAD_MANIFEST doesn't currently accept any arguments */ +upload_manifest: + K_UPLOAD_MANIFEST + { + UploadManifestCmd *cmd = makeNode(UploadManifestCmd); + + $$ = (Node *) cmd; + } + opt_physical: K_PHYSICAL | /* EMPTY */ @@ -411,6 +422,7 @@ ident_or_keyword: | K_EXPORT_SNAPSHOT { $$ = "export_snapshot"; } | K_NOEXPORT_SNAPSHOT { $$ = "noexport_snapshot"; } | K_USE_SNAPSHOT { $$ = "use_snapshot"; } + | K_UPLOAD_MANIFEST { $$ = "upload_manifest"; } ; %% diff --git a/src/backend/replication/repl_scanner.l b/src/backend/replication/repl_scanner.l index 1cc7fb858c..4805da08ee 100644 --- a/src/backend/replication/repl_scanner.l +++ b/src/backend/replication/repl_scanner.l @@ -136,6 +136,7 @@ EXPORT_SNAPSHOT { return K_EXPORT_SNAPSHOT; } NOEXPORT_SNAPSHOT { return K_NOEXPORT_SNAPSHOT; } USE_SNAPSHOT { return K_USE_SNAPSHOT; } WAIT { return K_WAIT; } +UPLOAD_MANIFEST { return K_UPLOAD_MANIFEST; } {space}+ { /* do nothing */ } @@ -303,6 +304,7 @@ replication_scanner_is_replication_command(void) case K_DROP_REPLICATION_SLOT: case K_READ_REPLICATION_SLOT: case K_TIMELINE_HISTORY: + case K_UPLOAD_MANIFEST: case K_SHOW: /* Yes; push back the first token so we can parse later. */ repl_pushed_back_token = first_token; diff --git a/src/backend/replication/walsender.c b/src/backend/replication/walsender.c index e250b0567e..b33b86671b 100644 --- a/src/backend/replication/walsender.c +++ b/src/backend/replication/walsender.c @@ -58,6 +58,7 @@ #include "access/xlogrecovery.h" #include "access/xlogutils.h" #include "backup/basebackup.h" +#include "backup/basebackup_incremental.h" #include "catalog/pg_authid.h" #include "catalog/pg_type.h" #include "commands/dbcommands.h" @@ -137,6 +138,17 @@ bool wake_wal_senders = false; */ static XLogReaderState *xlogreader = NULL; +/* + * If the UPLOAD_MANIFEST command is used to provide a backup manifest in + * preparation for an incremental backup, uploaded_manifest will be point + * to an object containing information about its contexts, and + * uploaded_manifest_mcxt will point to the memory context that contains + * that object and all of its subordinate data. Otherwise, both values will + * be NULL. + */ +static IncrementalBackupInfo *uploaded_manifest = NULL; +static MemoryContext uploaded_manifest_mcxt = NULL; + /* * These variables keep track of the state of the timeline we're currently * sending. sendTimeLine identifies the timeline. If sendTimeLineIsHistoric, @@ -233,6 +245,9 @@ static void XLogSendLogical(void); static void WalSndDone(WalSndSendDataCallback send_data); static XLogRecPtr GetStandbyFlushRecPtr(TimeLineID *tli); static void IdentifySystem(void); +static void UploadManifest(void); +static bool HandleUploadManifestPacket(StringInfo buf, off_t *offset, + IncrementalBackupInfo *ib); static void ReadReplicationSlot(ReadReplicationSlotCmd *cmd); static void CreateReplicationSlot(CreateReplicationSlotCmd *cmd); static void DropReplicationSlot(DropReplicationSlotCmd *cmd); @@ -660,6 +675,143 @@ SendTimeLineHistory(TimeLineHistoryCmd *cmd) pq_endmessage(&buf); } +/* + * Handle UPLOAD_MANIFEST command. + */ +static void +UploadManifest(void) +{ + MemoryContext mcxt; + IncrementalBackupInfo *ib; + off_t offset = 0; + StringInfoData buf; + + /* + * parsing the manifest will use the cryptohash stuff, which requires a + * resource owner + */ + Assert(CurrentResourceOwner == NULL); + CurrentResourceOwner = ResourceOwnerCreate(NULL, "base backup"); + + /* Prepare to read manifest data into a temporary context. */ + mcxt = AllocSetContextCreate(CurrentMemoryContext, + "incremental backup information", + ALLOCSET_DEFAULT_SIZES); + ib = CreateIncrementalBackupInfo(mcxt); + + /* Send a CopyInResponse message */ + pq_beginmessage(&buf, 'G'); + pq_sendbyte(&buf, 0); + pq_sendint16(&buf, 0); + pq_endmessage_reuse(&buf); + pq_flush(); + + /* Recieve packets from client until done. */ + while (HandleUploadManifestPacket(&buf, &offset, ib)) + ; + + /* Finish up manifest processing. */ + FinalizeIncrementalManifest(ib); + + /* + * Discard any old manifest information and arrange to preserve the new + * information we just got. + * + * We assume that MemoryContextDelete and MemoryContextSetParent won't + * fail, and thus we shouldn't end up bailing out of here in such a way as + * to leave dangling pointrs. + */ + if (uploaded_manifest_mcxt != NULL) + MemoryContextDelete(uploaded_manifest_mcxt); + MemoryContextSetParent(mcxt, CacheMemoryContext); + uploaded_manifest = ib; + uploaded_manifest_mcxt = mcxt; + + /* clean up the resource owner we created */ + WalSndResourceCleanup(true); +} + +/* + * Process one packet received during the handling of an UPLOAD_MANIFEST + * operation. + * + * 'buf' is scratch space. This function expects it to be initialized, doesn't + * care what the current contents are, and may override them with completely + * new contents. + * + * The return value is true if the caller should continue processing + * additional packets and false if the UPLOAD_MANIFEST operation is complete. + */ +static bool +HandleUploadManifestPacket(StringInfo buf, off_t *offset, + IncrementalBackupInfo *ib) +{ + int mtype; + int maxmsglen; + + HOLD_CANCEL_INTERRUPTS(); + + pq_startmsgread(); + mtype = pq_getbyte(); + if (mtype == EOF) + ereport(ERROR, + (errcode(ERRCODE_CONNECTION_FAILURE), + errmsg("unexpected EOF on client connection with an open transaction"))); + + switch (mtype) + { + case 'd': /* CopyData */ + maxmsglen = PQ_LARGE_MESSAGE_LIMIT; + break; + case 'c': /* CopyDone */ + case 'f': /* CopyFail */ + case 'H': /* Flush */ + case 'S': /* Sync */ + maxmsglen = PQ_SMALL_MESSAGE_LIMIT; + break; + default: + ereport(ERROR, + (errcode(ERRCODE_PROTOCOL_VIOLATION), + errmsg("unexpected message type 0x%02X during COPY from stdin", + mtype))); + maxmsglen = 0; /* keep compiler quiet */ + break; + } + + /* Now collect the message body */ + if (pq_getmessage(buf, maxmsglen)) + ereport(ERROR, + (errcode(ERRCODE_CONNECTION_FAILURE), + errmsg("unexpected EOF on client connection with an open transaction"))); + RESUME_CANCEL_INTERRUPTS(); + + /* Process the message */ + switch (mtype) + { + case 'd': /* CopyData */ + AppendIncrementalManifestData(ib, buf->data, buf->len); + return true; + + case 'c': /* CopyDone */ + return false; + + case 'H': /* Sync */ + case 'S': /* Flush */ + /* Ignore these while in CopyOut mode as we do elsewhere. */ + return true; + + case 'f': + ereport(ERROR, + (errcode(ERRCODE_QUERY_CANCELED), + errmsg("COPY from stdin failed: %s", + pq_getmsgstring(buf)))); + } + + /* Not reached. */ + Assert(false); + return false; +} + /* * Handle START_REPLICATION command. * @@ -1801,7 +1953,7 @@ exec_replication_command(const char *cmd_string) cmdtag = "BASE_BACKUP"; set_ps_display(cmdtag); PreventInTransactionBlock(true, cmdtag); - SendBaseBackup((BaseBackupCmd *) cmd_node); + SendBaseBackup((BaseBackupCmd *) cmd_node, uploaded_manifest); EndReplicationCommand(cmdtag); break; @@ -1863,6 +2015,14 @@ exec_replication_command(const char *cmd_string) } break; + case T_UploadManifestCmd: + cmdtag = "UPLOAD_MANIFEST"; + set_ps_display(cmdtag); + PreventInTransactionBlock(true, cmdtag); + UploadManifest(); + EndReplicationCommand(cmdtag); + break; + default: elog(ERROR, "unrecognized replication command node tag: %u", cmd_node->type); diff --git a/src/backend/storage/ipc/ipci.c b/src/backend/storage/ipc/ipci.c index a3d8eacb8d..3a6729003a 100644 --- a/src/backend/storage/ipc/ipci.c +++ b/src/backend/storage/ipc/ipci.c @@ -31,6 +31,7 @@ #include "postmaster/bgworker_internals.h" #include "postmaster/bgwriter.h" #include "postmaster/postmaster.h" +#include "postmaster/walsummarizer.h" #include "replication/logicallauncher.h" #include "replication/origin.h" #include "replication/slot.h" @@ -136,6 +137,7 @@ CalculateShmemSize(int *num_semaphores) size = add_size(size, ReplicationOriginShmemSize()); size = add_size(size, WalSndShmemSize()); size = add_size(size, WalRcvShmemSize()); + size = add_size(size, WalSummarizerShmemSize()); size = add_size(size, PgArchShmemSize()); size = add_size(size, ApplyLauncherShmemSize()); size = add_size(size, BTreeShmemSize()); @@ -291,6 +293,7 @@ CreateSharedMemoryAndSemaphores(void) ReplicationOriginShmemInit(); WalSndShmemInit(); WalRcvShmemInit(); + WalSummarizerShmemInit(); PgArchShmemInit(); ApplyLauncherShmemInit(); diff --git a/src/backend/storage/lmgr/lwlocknames.txt b/src/backend/storage/lmgr/lwlocknames.txt index f72f2906ce..d621f5507f 100644 --- a/src/backend/storage/lmgr/lwlocknames.txt +++ b/src/backend/storage/lmgr/lwlocknames.txt @@ -54,3 +54,4 @@ XactTruncationLock 44 WrapLimitsVacuumLock 46 NotifyQueueTailLock 47 WaitEventExtensionLock 48 +WALSummarizerLock 49 diff --git a/src/backend/utils/activity/pgstat_io.c b/src/backend/utils/activity/pgstat_io.c index 490d5a9ab7..8109aee6f0 100644 --- a/src/backend/utils/activity/pgstat_io.c +++ b/src/backend/utils/activity/pgstat_io.c @@ -296,7 +296,8 @@ pgstat_io_snapshot_cb(void) * - Syslogger because it is not connected to shared memory * - Archiver because most relevant archiving IO is delegated to a * specialized command or module -* - WAL Receiver and WAL Writer IO is not tracked in pg_stat_io for now +* - WAL Receiver, WAL Writer, and WAL Summarizer IO are not tracked in +* pg_stat_io for now * * Function returns true if BackendType participates in the cumulative stats * subsystem for IO and false if it does not. @@ -318,6 +319,7 @@ pgstat_tracks_io_bktype(BackendType bktype) case B_LOGGER: case B_WAL_RECEIVER: case B_WAL_WRITER: + case B_WAL_SUMMARIZER: return false; case B_AUTOVAC_LAUNCHER: diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index d7995931bd..7e79163466 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -56,6 +56,7 @@ RECOVERY_WAL_STREAM "Waiting in main loop of startup process for WAL to arrive, SYSLOGGER_MAIN "Waiting in main loop of syslogger process." WAL_RECEIVER_MAIN "Waiting in main loop of WAL receiver process." WAL_SENDER_MAIN "Waiting in main loop of WAL sender process." +WAL_SUMMARIZER_WAL "Waiting in WAL summarizer for more WAL to be generated." WAL_WRITER_MAIN "Waiting in main loop of WAL writer process." @@ -142,6 +143,7 @@ SAFE_SNAPSHOT "Waiting to obtain a valid snapshot for a <literal>READ ONLY DEFER SYNC_REP "Waiting for confirmation from a remote server during synchronous replication." WAL_RECEIVER_EXIT "Waiting for the WAL receiver to exit." WAL_RECEIVER_WAIT_START "Waiting for startup process to send initial data for streaming replication." +WAL_SUMMARY_READY "Waiting for a new WAL summary to be generated." XACT_GROUP_UPDATE "Waiting for the group leader to update transaction status at end of a parallel operation." @@ -162,6 +164,7 @@ REGISTER_SYNC_REQUEST "Waiting while sending synchronization requests to the che SPIN_DELAY "Waiting while acquiring a contended spinlock." VACUUM_DELAY "Waiting in a cost-based vacuum delay point." VACUUM_TRUNCATE "Waiting to acquire an exclusive lock to truncate off any empty pages at the end of a table vacuumed." +WAL_SUMMARIZER_ERROR "Waiting after a WAL summarizer error." # @@ -243,6 +246,8 @@ WAL_COPY_WRITE "Waiting for a write when creating a new WAL segment by copying a WAL_INIT_SYNC "Waiting for a newly initialized WAL file to reach durable storage." WAL_INIT_WRITE "Waiting for a write while initializing a new WAL file." WAL_READ "Waiting for a read from a WAL file." +WAL_SUMMARY_READ "Waiting for a read from a WAL summary file." +WAL_SUMMARY_WRITE "Waiting for a write to a WAL summary file." WAL_SYNC "Waiting for a WAL file to reach durable storage." WAL_SYNC_METHOD_ASSIGN "Waiting for data to reach durable storage while assigning a new WAL sync method." WAL_WRITE "Waiting for a write to a WAL file." diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c index 182d666852..94e7944748 100644 --- a/src/backend/utils/init/miscinit.c +++ b/src/backend/utils/init/miscinit.c @@ -306,6 +306,9 @@ GetBackendTypeDesc(BackendType backendType) case B_WAL_SENDER: backendDesc = "walsender"; break; + case B_WAL_SUMMARIZER: + backendDesc = "walsummarizer"; + break; case B_WAL_WRITER: backendDesc = "walwriter"; break; diff --git a/src/backend/utils/misc/guc_tables.c b/src/backend/utils/misc/guc_tables.c index 4c58574166..faf42bdbfb 100644 --- a/src/backend/utils/misc/guc_tables.c +++ b/src/backend/utils/misc/guc_tables.c @@ -63,6 +63,7 @@ #include "postmaster/postmaster.h" #include "postmaster/startup.h" #include "postmaster/syslogger.h" +#include "postmaster/walsummarizer.h" #include "postmaster/walwriter.h" #include "replication/logicallauncher.h" #include "replication/slot.h" @@ -704,6 +705,8 @@ const char *const config_group_names[] = gettext_noop("Write-Ahead Log / Archive Recovery"), /* WAL_RECOVERY_TARGET */ gettext_noop("Write-Ahead Log / Recovery Target"), + /* WAL_SUMMARIZATION */ + gettext_noop("Write-Ahead Log / Summarization"), /* REPLICATION_SENDING */ gettext_noop("Replication / Sending Servers"), /* REPLICATION_PRIMARY */ @@ -3181,6 +3184,32 @@ struct config_int ConfigureNamesInt[] = check_wal_segment_size, NULL, NULL }, + { + {"wal_summarize_mb", PGC_SIGHUP, WAL_SUMMARIZATION, + gettext_noop("Number of bytes of WAL per summary file."), + gettext_noop("Smaller values minimize extra work performed by incremental backup, but increase the number of files on disk."), + GUC_UNIT_MB, + }, + &wal_summarize_mb, + 256, + 0, + INT_MAX, + NULL, NULL, NULL + }, + + { + {"wal_summarize_keep_time", PGC_SIGHUP, WAL_SUMMARIZATION, + gettext_noop("Time for which WAL summary files should be kept."), + NULL, + GUC_UNIT_MIN, + }, + &wal_summarize_keep_time, + 7 * 24 * 60, /* 1 week */ + 0, + INT_MAX, + NULL, NULL, NULL + }, + { {"autovacuum_naptime", PGC_SIGHUP, AUTOVACUUM, gettext_noop("Time to sleep between autovacuum runs."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index d08d55c3fe..4736606ac1 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -299,6 +299,11 @@ #recovery_target_action = 'pause' # 'pause', 'promote', 'shutdown' # (change requires restart) +# - WAL Summarization - + +#wal_summarize_mb = 256 # MB of WAL per summary file, 0 disables +#wal_summarize_keep_time = '7d' # when to remove old summary files, 0 = never + #------------------------------------------------------------------------------ # REPLICATION diff --git a/src/bin/Makefile b/src/bin/Makefile index 373077bf52..aa2210925e 100644 --- a/src/bin/Makefile +++ b/src/bin/Makefile @@ -19,6 +19,7 @@ SUBDIRS = \ pg_archivecleanup \ pg_basebackup \ pg_checksums \ + pg_combinebackup \ pg_config \ pg_controldata \ pg_ctl \ diff --git a/src/bin/initdb/initdb.c b/src/bin/initdb/initdb.c index 0c6f5ceb0a..e68b40d2b5 100644 --- a/src/bin/initdb/initdb.c +++ b/src/bin/initdb/initdb.c @@ -227,6 +227,7 @@ static char *extra_options = ""; static const char *const subdirs[] = { "global", "pg_wal/archive_status", + "pg_wal/summaries", "pg_commit_ts", "pg_dynshmem", "pg_notify", diff --git a/src/bin/meson.build b/src/bin/meson.build index 67cb50630c..4cb6fd59bb 100644 --- a/src/bin/meson.build +++ b/src/bin/meson.build @@ -5,6 +5,7 @@ subdir('pg_amcheck') subdir('pg_archivecleanup') subdir('pg_basebackup') subdir('pg_checksums') +subdir('pg_combinebackup') subdir('pg_config') subdir('pg_controldata') subdir('pg_ctl') diff --git a/src/bin/pg_basebackup/bbstreamer_file.c b/src/bin/pg_basebackup/bbstreamer_file.c index 45f32974ff..6b78ee283d 100644 --- a/src/bin/pg_basebackup/bbstreamer_file.c +++ b/src/bin/pg_basebackup/bbstreamer_file.c @@ -296,6 +296,7 @@ should_allow_existing_directory(const char *pathname) if (strcmp(filename, "pg_wal") == 0 || strcmp(filename, "pg_xlog") == 0 || strcmp(filename, "archive_status") == 0 || + strcmp(filename, "summaries") == 0 || strcmp(filename, "pg_tblspc") == 0) return true; diff --git a/src/bin/pg_basebackup/pg_basebackup.c b/src/bin/pg_basebackup/pg_basebackup.c index 1a8cef345d..33416b11cf 100644 --- a/src/bin/pg_basebackup/pg_basebackup.c +++ b/src/bin/pg_basebackup/pg_basebackup.c @@ -101,6 +101,11 @@ typedef void (*WriteDataCallback) (size_t nbytes, char *buf, */ #define MINIMUM_VERSION_FOR_TERMINATED_TARFILE 150000 +/* + * pg_wal/summaries exists beginning with version 17. + */ +#define MINIMUM_VERSION_FOR_WAL_SUMMARIES 170000 + /* * Different ways to include WAL */ @@ -217,7 +222,8 @@ static void ReceiveBackupManifestInMemoryChunk(size_t r, char *copybuf, void *callback_data); static void BaseBackup(char *compression_algorithm, char *compression_detail, CompressionLocation compressloc, - pg_compress_specification *client_compress); + pg_compress_specification *client_compress, + char *incremental_manifest); static bool reached_end_position(XLogRecPtr segendpos, uint32 timeline, bool segment_finished); @@ -390,6 +396,8 @@ usage(void) printf(_("\nOptions controlling the output:\n")); printf(_(" -D, --pgdata=DIRECTORY receive base backup into directory\n")); printf(_(" -F, --format=p|t output format (plain (default), tar)\n")); + printf(_(" -i, --incremental=OLDMANIFEST\n")); + printf(_(" take incremental or differential backup\n")); printf(_(" -r, --max-rate=RATE maximum transfer rate to transfer data directory\n" " (in kB/s, or use suffix \"k\" or \"M\")\n")); printf(_(" -R, --write-recovery-conf\n" @@ -688,6 +696,23 @@ StartLogStreamer(char *startpos, uint32 timeline, char *sysidentifier, if (pg_mkdir_p(statusdir, pg_dir_create_mode) != 0 && errno != EEXIST) pg_fatal("could not create directory \"%s\": %m", statusdir); + + /* + * For newer server versions, likewise create pg_wal/summaries + */ + if (PQserverVersion(conn) < MINIMUM_VERSION_FOR_WAL_SUMMARIES) + { + char summarydir[MAXPGPATH]; + + snprintf(summarydir, sizeof(summarydir), "%s/%s/summaries", + basedir, + PQserverVersion(conn) < MINIMUM_VERSION_FOR_PG_WAL ? + "pg_xlog" : "pg_wal"); + + if (pg_mkdir_p(statusdir, pg_dir_create_mode) != 0 && + errno != EEXIST) + pg_fatal("could not create directory \"%s\": %m", summarydir); + } } /* @@ -1728,7 +1753,9 @@ ReceiveBackupManifestInMemoryChunk(size_t r, char *copybuf, static void BaseBackup(char *compression_algorithm, char *compression_detail, - CompressionLocation compressloc, pg_compress_specification *client_compress) + CompressionLocation compressloc, + pg_compress_specification *client_compress, + char *incremental_manifest) { PGresult *res; char *sysidentifier; @@ -1794,7 +1821,74 @@ BaseBackup(char *compression_algorithm, char *compression_detail, exit(1); /* - * Start the actual backup + * If the user wants an incremental backup, we must upload the manifest + * for the previous backup upon which it is to be based. + */ + if (incremental_manifest != NULL) + { + int fd; + char mbuf[65536]; + int nbytes; + + /* XXX add a server version check here */ + + /* Open the file. */ + fd = open(incremental_manifest, O_RDONLY | PG_BINARY, 0); + if (fd < 0) + pg_fatal("could not open file \"%s\": %m", incremental_manifest); + + /* Tell the server what we want to do. */ + if (PQsendQuery(conn, "UPLOAD_MANIFEST") == 0) + pg_fatal("could not send replication command \"%s\": %s", + "UPLOAD_MANIFEST", PQerrorMessage(conn)); + res = PQgetResult(conn); + if (PQresultStatus(res) != PGRES_COPY_IN) + { + if (PQresultStatus(res) == PGRES_FATAL_ERROR) + pg_fatal("could not upload manifest: %s", + PQerrorMessage(conn)); + else + pg_fatal("could not upload manifest: unexpected status %s", + PQresStatus(PQresultStatus(res))); + } + + /* Loop, reading from the file and sending the data to the server. */ + while ((nbytes = read(fd, mbuf, sizeof mbuf)) > 0) + { + if (PQputCopyData(conn, mbuf, nbytes) < 0) + pg_fatal("could not send COPY data: %s", + PQerrorMessage(conn)); + } + + /* Bail out if we exited the loop due to an error. */ + if (nbytes < 0) + pg_fatal("could not read file \"%s\": %m", incremental_manifest); + + /* End the COPY operation. */ + if (PQputCopyEnd(conn, NULL) < 0) + pg_fatal("could not send end-of-COPY: %s", + PQerrorMessage(conn)); + + /* See whether the server is happy with what we sent. */ + res = PQgetResult(conn); + if (PQresultStatus(res) == PGRES_FATAL_ERROR) + pg_fatal("could not upload manifest: %s", + PQerrorMessage(conn)); + else if (PQresultStatus(res) != PGRES_COMMAND_OK) + pg_fatal("could not upload manifest: unexpected status %s", + PQresStatus(PQresultStatus(res))); + + /* Consume ReadyForQuery message from server. */ + res = PQgetResult(conn); + if (res != NULL) + pg_fatal("unexpected extra result while sending manifest"); + + /* Add INCREMENTAL option to BASE_BACKUP command. */ + AppendPlainCommandOption(&buf, use_new_option_syntax, "INCREMENTAL"); + } + + /* + * Continue building up the options list for the BASE_BACKUP command. */ AppendStringCommandOption(&buf, use_new_option_syntax, "LABEL", label); if (estimatesize) @@ -1901,6 +1995,7 @@ BaseBackup(char *compression_algorithm, char *compression_detail, else basebkp = psprintf("BASE_BACKUP %s", buf.data); + /* OK, try to start the backup. */ if (PQsendQuery(conn, basebkp) == 0) pg_fatal("could not send replication command \"%s\": %s", "BASE_BACKUP", PQerrorMessage(conn)); @@ -2256,6 +2351,7 @@ main(int argc, char **argv) {"version", no_argument, NULL, 'V'}, {"pgdata", required_argument, NULL, 'D'}, {"format", required_argument, NULL, 'F'}, + {"incremental", required_argument, NULL, 'i'}, {"checkpoint", required_argument, NULL, 'c'}, {"create-slot", no_argument, NULL, 'C'}, {"max-rate", required_argument, NULL, 'r'}, @@ -2293,6 +2389,7 @@ main(int argc, char **argv) int option_index; char *compression_algorithm = "none"; char *compression_detail = NULL; + char *incremental_manifest = NULL; CompressionLocation compressloc = COMPRESS_LOCATION_UNSPECIFIED; pg_compress_specification client_compress; @@ -2317,7 +2414,7 @@ main(int argc, char **argv) atexit(cleanup_directories_atexit); - while ((c = getopt_long(argc, argv, "c:Cd:D:F:h:l:nNp:Pr:Rs:S:t:T:U:vwWX:zZ:", + while ((c = getopt_long(argc, argv, "c:Cd:D:F:h:i:l:nNp:Pr:Rs:S:t:T:U:vwWX:zZ:", long_options, &option_index)) != -1) { switch (c) @@ -2352,6 +2449,9 @@ main(int argc, char **argv) case 'h': dbhost = pg_strdup(optarg); break; + case 'i': + incremental_manifest = pg_strdup(optarg); + break; case 'l': label = pg_strdup(optarg); break; @@ -2765,7 +2865,7 @@ main(int argc, char **argv) } BaseBackup(compression_algorithm, compression_detail, compressloc, - &client_compress); + &client_compress, incremental_manifest); success = true; return 0; diff --git a/src/bin/pg_basebackup/t/010_pg_basebackup.pl b/src/bin/pg_basebackup/t/010_pg_basebackup.pl index b9f5e1266b..bf765291e7 100644 --- a/src/bin/pg_basebackup/t/010_pg_basebackup.pl +++ b/src/bin/pg_basebackup/t/010_pg_basebackup.pl @@ -223,10 +223,10 @@ SKIP: "check backup dir permissions"); } -# Only archive_status directory should be copied in pg_wal/. +# Only archive_status and summaries directories should be copied in pg_wal/. is_deeply( [ sort(slurp_dir("$tempdir/backup/pg_wal/")) ], - [ sort qw(. .. archive_status) ], + [ sort qw(. .. archive_status summaries) ], 'no WAL files copied'); # Contents of these directories should not be copied. diff --git a/src/bin/pg_combinebackup/.gitignore b/src/bin/pg_combinebackup/.gitignore new file mode 100644 index 0000000000..d7e617438c --- /dev/null +++ b/src/bin/pg_combinebackup/.gitignore @@ -0,0 +1 @@ +pg_combinebackup diff --git a/src/bin/pg_combinebackup/Makefile b/src/bin/pg_combinebackup/Makefile new file mode 100644 index 0000000000..78ba05e624 --- /dev/null +++ b/src/bin/pg_combinebackup/Makefile @@ -0,0 +1,52 @@ +#------------------------------------------------------------------------- +# +# Makefile for src/bin/pg_combinebackup +# +# Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group +# Portions Copyright (c) 1994, Regents of the University of California +# +# src/bin/pg_combinebackup/Makefile +# +#------------------------------------------------------------------------- + +PGFILEDESC = "pg_combinebackup - combine incremental backups" +PGAPPICON=win32 + +subdir = src/bin/pg_combinebackup +top_builddir = ../../.. +include $(top_builddir)/src/Makefile.global + +override CPPFLAGS := -I$(libpq_srcdir) $(CPPFLAGS) +LDFLAGS_INTERNAL += -L$(top_builddir)/src/fe_utils -lpgfeutils + +OBJS = \ + $(WIN32RES) \ + pg_combinebackup.o \ + backup_label.o \ + copy_file.o \ + load_manifest.o \ + reconstruct.o \ + write_manifest.o + +all: pg_combinebackup + +pg_combinebackup: $(OBJS) | submake-libpgport submake-libpgfeutils + $(CC) $(CFLAGS) $^ $(LDFLAGS) $(LDFLAGS_EX) $(LIBS) -o $@$(X) + +install: all installdirs + $(INSTALL_PROGRAM) pg_combinebackup$(X) '$(DESTDIR)$(bindir)/pg_combinebackup$(X)' + +installdirs: + $(MKDIR_P) '$(DESTDIR)$(bindir)' + +uninstall: + rm -f '$(DESTDIR)$(bindir)/pg_combinebackup$(X)' + +clean distclean maintainer-clean: + rm -f pg_combinebackup$(X) $(OBJS) + +check: + $(prove_check) + +installcheck: + $(prove_installcheck) diff --git a/src/bin/pg_combinebackup/backup_label.c b/src/bin/pg_combinebackup/backup_label.c new file mode 100644 index 0000000000..2a62aa6fad --- /dev/null +++ b/src/bin/pg_combinebackup/backup_label.c @@ -0,0 +1,281 @@ +/*------------------------------------------------------------------------- + * + * Read and manipulate backup label files + * + * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group + * Portions Copyright (c) 1994, Regents of the University of California + * + * src/bin/pg_combinebackup/backup_label.c + * + *------------------------------------------------------------------------- + */ +#include "postgres_fe.h" + +#include <unistd.h> + +#include "access/xlogdefs.h" +#include "backup_label.h" +#include "common/logging.h" +#include "common/file_perm.h" +#include "write_manifest.h" + +static int get_eol_offset(StringInfo buf); +static bool line_starts_with(char *s, char *e, char *match, char **sout); +static bool parse_lsn(char *s, char *e, XLogRecPtr *lsn, char **c); +static bool parse_tli(char *s, char *e, TimeLineID *tli); + +/* + * Parse a backup label file, starting at buf->cursor. + * + * We expect to find a START WAL LOCATION line, followed by a LSN, followed + * by a space; the resulting LSN is stored into *start_lsn. + * + * We expect to find a START TIMELINE line, followed by a TLI, followed by + * a newline; the resulting TLI is stored into *start_tli. + * + * We expect to find either both INCREMENTAL FROM LSN and INCREMENTAL FROM TLI + * or neither. If these are found, they should be followed by an LSN or TLI + * respectively and then by a newline, and the values will be stored into + * *previous_lsn and *previous_tli, respectively. + * + * Other lines in the provided backup_label data are ignored. filename is used + * for error reporting; errors are fatal. + */ +void +parse_backup_label(char *filename, StringInfo buf, + TimeLineID *start_tli, XLogRecPtr *start_lsn, + TimeLineID *previous_tli, XLogRecPtr *previous_lsn) +{ + int found = 0; + + *start_tli = 0; + *start_lsn = InvalidXLogRecPtr; + *previous_tli = 0; + *previous_lsn = InvalidXLogRecPtr; + + while (buf->cursor < buf->len) + { + char *s = &buf->data[buf->cursor]; + int eo = get_eol_offset(buf); + char *e = &buf->data[eo]; + char *c; + + if (line_starts_with(s, e, "START WAL LOCATION: ", &s)) + { + if (!parse_lsn(s, e, start_lsn, &c)) + pg_fatal("%s: could not parse START WAL LOCATION", + filename); + if (c >= e || *c != ' ') + pg_fatal("%s: improper terminator for START WAL LOCATION", + filename); + found |= 1; + } + else if (line_starts_with(s, e, "START TIMELINE: ", &s)) + { + if (!parse_tli(s, e, start_tli)) + pg_fatal("%s: could not parse TLI for START TIMELINE", + filename); + if (*start_tli == 0) + pg_fatal("%s: invalid TLI", filename); + found |= 2; + } + else if (line_starts_with(s, e, "INCREMENTAL FROM LSN: ", &s)) + { + if (!parse_lsn(s, e, previous_lsn, &c)) + pg_fatal("%s: could not parse INCREMENTAL FROM LSN", + filename); + if (c >= e || *c != '\n') + pg_fatal("%s: improper terminator for INCREMENTAL FROM LSN", + filename); + found |= 4; + } + else if (line_starts_with(s, e, "INCREMENTAL FROM TLI: ", &s)) + { + if (!parse_tli(s, e, previous_tli)) + pg_fatal("%s: could not parse INCREMENTAL FROM TLI", + filename); + if (*previous_tli == 0) + pg_fatal("%s: invalid TLI", filename); + found |= 8; + } + + buf->cursor = eo; + } + + if ((found & 1) == 0) + pg_fatal("%s: could not find START WAL LOCATION", filename); + if ((found & 2) == 0) + pg_fatal("%s: could not find START TIMELINE", filename); + if ((found & 4) != 0 && (found & 8) == 0) + pg_fatal("%s: INCREMENTAL FROM LSN requires INCREMENTAL FROM TLI", filename); + if ((found & 8) != 0 && (found & 4) == 0) + pg_fatal("%s: INCREMENTAL FROM TLI requires INCREMENTAL FROM LSN", filename); +} + +/* + * Write a backup label file to the output directory. + * + * This will be identical to the provided backup_label file, except that the + * INCREMENTAL FROM LSN and INCREMENTAL FROM TLI lines will be omitted. + * + * The new file will be checksummed using the specified algorithm. If + * mwriter != NULL, it will be added to the manifest. + */ +void +write_backup_label(char *output_directory, StringInfo buf, + pg_checksum_type checksum_type, manifest_writer *mwriter) +{ + char output_filename[MAXPGPATH]; + int output_fd; + pg_checksum_context checksum_ctx; + uint8 checksum_payload[PG_CHECKSUM_MAX_LENGTH]; + int checksum_length; + + pg_checksum_init(&checksum_ctx, checksum_type); + + snprintf(output_filename, MAXPGPATH, "%s/backup_label", output_directory); + + if ((output_fd = open(output_filename, + O_WRONLY | O_CREAT | O_EXCL | PG_BINARY, + pg_file_create_mode)) < 0) + pg_fatal("could not open file \"%s\": %m", output_filename); + + while (buf->cursor < buf->len) + { + char *s = &buf->data[buf->cursor]; + int eo = get_eol_offset(buf); + char *e = &buf->data[eo]; + + if (!line_starts_with(s, e, "INCREMENTAL FROM LSN: ", NULL) && + !line_starts_with(s, e, "INCREMENTAL FROM TLI: ", NULL)) + { + ssize_t wb; + + wb = write(output_fd, s, e - s); + if (wb != e - s) + { + if (wb < 0) + pg_fatal("could not write file \"%s\": %m", output_filename); + else + pg_fatal("could not write file \"%s\": wrote only %d of %d bytes", + output_filename, (int) wb, (int) (e - s)); + } + if (pg_checksum_update(&checksum_ctx, (uint8 *) s, e - s) < 0) + pg_fatal("could not update checksum of file \"%s\"", + output_filename); + } + + buf->cursor = eo; + } + + if (close(output_fd) != 0) + pg_fatal("could not close \"%s\": %m", output_filename); + + checksum_length = pg_checksum_final(&checksum_ctx, checksum_payload); + + if (mwriter != NULL) + { + struct stat sb; + + /* + * We could track the length ourselves, but must stat() to get the + * mtime. + */ + if (stat(output_filename, &sb) < 0) + pg_fatal("could not stat file \"%s\": %m", output_filename); + add_file_to_manifest(mwriter, "backup_label", sb.st_size, + sb.st_mtime, checksum_type, + checksum_length, checksum_payload); + } +} + +/* + * Return the offset at which the next line in the buffer starts, or there + * is none, the offset at which the buffer ends. + * + * The search begins at buf->cursor. + */ +static int +get_eol_offset(StringInfo buf) +{ + int eo = buf->cursor; + + while (eo < buf->len) + { + if (buf->data[eo] == '\n') + return eo + 1; + ++eo; + } + + return eo; +} + +/* + * Test whether the line that runs from s to e (inclusive of *s, but not + * inclusive of *e) starts with the match string provided, and return true + * or false according to whether or not this is the case. + * + * If the function returns true and if *sout != NULL, stores a pointer to the + * byte following the match into *sout. + */ +static bool +line_starts_with(char *s, char *e, char *match, char **sout) +{ + while (s < e && *match != '\0' && *s == *match) + ++s, ++match; + + if (*match == '\0' && sout != NULL) + *sout = s; + + return (*match == '\0'); +} + +/* + * Parse an LSN starting at s and not stopping at or before e. The return value + * is true on success and otherwise false. On success, stores the result into + * *lsn and sets *c to the first character that is not part of the LSN. + */ +static bool +parse_lsn(char *s, char *e, XLogRecPtr *lsn, char **c) +{ + char save = *e; + int nchars; + bool success; + unsigned hi; + unsigned lo; + + *e = '\0'; + success = (sscanf(s, "%X/%X%n", &hi, &lo, &nchars) == 2); + *e = save; + + if (success) + { + *lsn = ((XLogRecPtr) hi) << 32 | (XLogRecPtr) lo; + *c = s + nchars; + } + + return success; +} + +/* + * Parse a TLI starting at s and stopping at or before e. The return value is + * true on success and otherwise false. On success, stores the result into + * *tli. If the first character that is not part of the TLI is anything other + * than a newline, that is deemed a failure. + */ +static bool +parse_tli(char *s, char *e, TimeLineID *tli) +{ + char save = *e; + int nchars; + bool success; + + *e = '\0'; + success = (sscanf(s, "%u%n", tli, &nchars) == 1); + *e = save; + + if (success && s[nchars] != '\n') + success = false; + + return success; +} diff --git a/src/bin/pg_combinebackup/backup_label.h b/src/bin/pg_combinebackup/backup_label.h new file mode 100644 index 0000000000..08d6ed67a9 --- /dev/null +++ b/src/bin/pg_combinebackup/backup_label.h @@ -0,0 +1,29 @@ +/*------------------------------------------------------------------------- + * + * Read and manipulate backup label files + * + * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group + * Portions Copyright (c) 1994, Regents of the University of California + * + * src/bin/pg_combinebackup/backup_label.h + * + *------------------------------------------------------------------------- + */ +#ifndef BACKUP_LABEL_H +#define BACKUP_LABEL_H + +#include "common/checksum_helper.h" +#include "lib/stringinfo.h" + +struct manifest_writer; + +extern void parse_backup_label(char *filename, StringInfo buf, + TimeLineID *start_tli, + XLogRecPtr *start_lsn, + TimeLineID *previous_tli, + XLogRecPtr *previous_lsn); +extern void write_backup_label(char *output_directory, StringInfo buf, + pg_checksum_type checksum_type, + struct manifest_writer *mwriter); + +#endif /* BACKUP_LABEL_H */ diff --git a/src/bin/pg_combinebackup/copy_file.c b/src/bin/pg_combinebackup/copy_file.c new file mode 100644 index 0000000000..8ba6cc09e4 --- /dev/null +++ b/src/bin/pg_combinebackup/copy_file.c @@ -0,0 +1,169 @@ +/* + * Copy entire files. + * + * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group + * Portions Copyright (c) 1994, Regents of the University of California + * + * src/bin/pg_combinebackup/copy_file.h + * + *------------------------------------------------------------------------- + */ +#include "postgres_fe.h" + +#ifdef HAVE_COPYFILE_H +#include <copyfile.h> +#endif +#include <fcntl.h> +#include <sys/stat.h> +#include <unistd.h> + +#include "common/file_perm.h" +#include "common/logging.h" +#include "copy_file.h" + +static void copy_file_blocks(const char *src, const char *dst, + pg_checksum_context *checksum_ctx); + +#ifdef WIN32 +static void copy_file_copyfile(const char *src, const char *dst); +#endif + +/* + * Copy a regular file, optionally computing a checksum, and emitting + * appropriate debug messages. But if we're in dry-run mode, then just emit + * the messages and don't copy anything. + */ +void +copy_file(const char *src, const char *dst, + pg_checksum_context *checksum_ctx, bool dry_run) +{ + /* + * In dry-run mode, we don't actually copy anything, nor do we read any + * data from the source file, but we do verify that we can open it. + */ + if (dry_run) + { + int fd; + + if ((fd = open(src, O_RDONLY | PG_BINARY)) < 0) + pg_fatal("could not open \"%s\": %m", src); + if (close(fd) < 0) + pg_fatal("could not close \"%s\": %m", src); + } + + /* + * If we don't need to compute a checksum, then we can use any special + * operating system primitives that we know about to copy the file; this + * may be quicker than a naive block copy. + */ + if (checksum_ctx->type != CHECKSUM_TYPE_NONE) + { + char *strategy_name = NULL; + void (*strategy_implementation) (const char *, const char *) = NULL; + +#ifdef WIN32 + strategy_name = "CopyFile"; + strategy_implementation = copy_file_copyfile; +#endif + + if (strategy_name != NULL) + { + if (dry_run) + pg_log_debug("would copy \"%s\" to \"%s\" using strategy %s", + src, dst, strategy_name); + else + { + pg_log_debug("copying \"%s\" to \"%s\" using strategy %s", + src, dst, strategy_name); + (*strategy_implementation) (src, dst); + } + return; + } + } + + /* + * Fall back to the simple approach of reading and writing all the blocks, + * feeding them into the checksum context as we go. + */ + if (dry_run) + { + if (checksum_ctx->type == CHECKSUM_TYPE_NONE) + pg_log_debug("would copy \"%s\" to \"%s\"", + src, dst); + else + pg_log_debug("would copy \"%s\" to \"%s\" and checksum with %s", + src, dst, pg_checksum_type_name(checksum_ctx->type)); + } + else + { + if (checksum_ctx->type == CHECKSUM_TYPE_NONE) + pg_log_debug("copying \"%s\" to \"%s\"", + src, dst); + else + pg_log_debug("copying \"%s\" to \"%s\" and checksumming with %s", + src, dst, pg_checksum_type_name(checksum_ctx->type)); + copy_file_blocks(src, dst, checksum_ctx); + } +} + +/* + * Copy a file block by block, and optionally compute a checksum as we go. + */ +static void +copy_file_blocks(const char *src, const char *dst, + pg_checksum_context *checksum_ctx) +{ + int src_fd; + int dest_fd; + uint8 *buffer; + const int buffer_size = 50 * BLCKSZ; + ssize_t rb; + unsigned offset = 0; + + if ((src_fd = open(src, O_RDONLY | PG_BINARY, 0)) < 0) + pg_fatal("could not open file \"%s\": %m", src); + + if ((dest_fd = open(dst, O_WRONLY | O_CREAT | O_EXCL | PG_BINARY, + pg_file_create_mode)) < 0) + pg_fatal("could not open file \"%s\": %m", dst); + + buffer = pg_malloc(buffer_size); + + while ((rb = read(src_fd, buffer, buffer_size)) > 0) + { + ssize_t wb; + + if ((wb = write(dest_fd, buffer, rb)) != rb) + { + if (wb < 0) + pg_fatal("could not write file \"%s\": %m", dst); + else + pg_fatal("could not write file \"%s\": wrote only %d of %d bytes at offset %u", + dst, (int) wb, (int) rb, offset); + } + + if (pg_checksum_update(checksum_ctx, buffer, rb) < 0) + pg_fatal("could not update checksum of file \"%s\"", dst); + + offset += rb; + } + + if (rb < 0) + pg_fatal("could not read file \"%s\": %m", dst); + + pg_free(buffer); + close(src_fd); + close(dest_fd); +} + +#ifdef WIN32 +static void +copy_file_copyfile(const char *src, const char *dst) +{ + if (CopyFile(src, dst, true) == 0) + { + _dosmaperr(GetLastError()); + pg_fatal("could not copy \"%s\" to \"%s\": %m", src, dst); + } +} +#endif /* WIN32 */ diff --git a/src/bin/pg_combinebackup/copy_file.h b/src/bin/pg_combinebackup/copy_file.h new file mode 100644 index 0000000000..031030bacb --- /dev/null +++ b/src/bin/pg_combinebackup/copy_file.h @@ -0,0 +1,19 @@ +/* + * Copy entire files. + * + * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group + * Portions Copyright (c) 1994, Regents of the University of California + * + * src/bin/pg_combinebackup/copy_file.h + * + *------------------------------------------------------------------------- + */ +#ifndef COPY_FILE_H +#define COPY_FILE_H + +#include "common/checksum_helper.h" + +extern void copy_file(const char *src, const char *dst, + pg_checksum_context *checksum_ctx, bool dry_run); + +#endif /* COPY_FILE_H */ diff --git a/src/bin/pg_combinebackup/load_manifest.c b/src/bin/pg_combinebackup/load_manifest.c new file mode 100644 index 0000000000..d0b8de7912 --- /dev/null +++ b/src/bin/pg_combinebackup/load_manifest.c @@ -0,0 +1,245 @@ +/*------------------------------------------------------------------------- + * + * Load data from a backup manifest into memory. + * + * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group + * Portions Copyright (c) 1994, Regents of the University of California + * + * src/bin/pg_combinebackup/load_manifest.c + * + *------------------------------------------------------------------------- + */ + +#include "postgres_fe.h" + +#include <sys/stat.h> +#include <unistd.h> + +#include "common/hashfn.h" +#include "common/logging.h" +#include "common/parse_manifest.h" +#include "load_manifest.h" + +/* + * For efficiency, we'd like our hash table containing information about the + * manifest to start out with approximately the correct number of entries. + * There's no way to know the exact number of entries without reading the whole + * file, but we can get an estimate by dividing the file size by the estimated + * number of bytes per line. + * + * This could be off by about a factor of two in either direction, because the + * checksum algorithm has a big impact on the line lengths; e.g. a SHA512 + * checksum is 128 hex bytes, whereas a CRC-32C value is only 8, and there + * might be no checksum at all. + */ +#define ESTIMATED_BYTES_PER_MANIFEST_LINE 100 + +/* + * Define a hash table which we can use to store information about the files + * mentioned in the backup manifest. + */ +static uint32 hash_string_pointer(char *s); +#define SH_PREFIX manifest_files +#define SH_ELEMENT_TYPE manifest_file +#define SH_KEY_TYPE char * +#define SH_KEY pathname +#define SH_HASH_KEY(tb, key) hash_string_pointer(key) +#define SH_EQUAL(tb, a, b) (strcmp(a, b) == 0) +#define SH_SCOPE extern +#define SH_RAW_ALLOCATOR pg_malloc0 +#define SH_DEFINE +#include "lib/simplehash.h" + +static void record_manifest_details_for_file(JsonManifestParseContext *context, + char *pathname, size_t size, + pg_checksum_type checksum_type, + int checksum_length, + uint8 *checksum_payload); +static void record_manifest_details_for_wal_range(JsonManifestParseContext *context, + TimeLineID tli, + XLogRecPtr start_lsn, + XLogRecPtr end_lsn); +static void report_manifest_error(JsonManifestParseContext *context, + const char *fmt,...) + pg_attribute_printf(2, 3) pg_attribute_noreturn(); + +/* + * Load backup_manifest files from an array of backups and produces an array + * of manifest_data objects. + * + * NB: Since load_backup_manifest() can return NULL, the resulting array could + * contain NULL entries. + */ +manifest_data ** +load_backup_manifests(int n_backups, char **backup_directories) +{ + manifest_data **result; + int i; + + result = pg_malloc(sizeof(manifest_data *) * n_backups); + for (i = 0; i < n_backups; ++i) + result[i] = load_backup_manifest(backup_directories[i]); + + return result; +} + +/* + * Parse the backup_manifest file in the named backup directory. Construct a + * hash table with information about all the files it mentions, and a linked + * list of all the WAL ranges it mentions. + * + * If the backup_manifest file simply doesn't exist, logs a warning and returns + * NULL. Any other error, or any error parsing the contents of the file, is + * fatal. + */ +manifest_data * +load_backup_manifest(char *backup_directory) +{ + char pathname[MAXPGPATH]; + int fd; + struct stat statbuf; + off_t estimate; + uint32 initial_size; + manifest_files_hash *ht; + char *buffer; + int rc; + JsonManifestParseContext context; + manifest_data *result; + + /* Open the manifest file. */ + snprintf(pathname, MAXPGPATH, "%s/backup_manifest", backup_directory); + if ((fd = open(pathname, O_RDONLY | PG_BINARY, 0)) < 0) + { + if (errno == EEXIST) + { + pg_log_warning("\"%s\" does not exist", pathname); + return NULL; + } + pg_fatal("could not open file \"%s\": %m", pathname); + } + + /* Figure out how big the manifest is. */ + if (fstat(fd, &statbuf) != 0) + pg_fatal("could not stat file \"%s\": %m", pathname); + + /* Guess how large to make the hash table based on the manifest size. */ + estimate = statbuf.st_size / ESTIMATED_BYTES_PER_MANIFEST_LINE; + initial_size = Min(PG_UINT32_MAX, Max(estimate, 256)); + + /* Create the hash table. */ + ht = manifest_files_create(initial_size, NULL); + + /* + * Slurp in the whole file. + * + * This is not ideal, but there's currently no way to get pg_parse_json() + * to perform incremental parsing. + */ + buffer = pg_malloc(statbuf.st_size); + rc = read(fd, buffer, statbuf.st_size); + if (rc != statbuf.st_size) + { + if (rc < 0) + pg_fatal("could not read file \"%s\": %m", pathname); + else + pg_fatal("could not read file \"%s\": read %d of %lld", + pathname, rc, (long long int) statbuf.st_size); + } + + /* Close the manifest file. */ + close(fd); + + /* Parse the manifest. */ + result = pg_malloc0(sizeof(manifest_data)); + result->files = ht; + context.private_data = result; + context.perfile_cb = record_manifest_details_for_file; + context.perwalrange_cb = record_manifest_details_for_wal_range; + context.error_cb = report_manifest_error; + json_parse_manifest(&context, buffer, statbuf.st_size); + + /* All done. */ + pfree(buffer); + return result; +} + +/* + * Report an error while parsing the manifest. + * + * We consider all such errors to be fatal errors. The manifest parser + * expects this function not to return. + */ +static void +report_manifest_error(JsonManifestParseContext *context, const char *fmt,...) +{ + va_list ap; + + va_start(ap, fmt); + pg_log_generic_v(PG_LOG_ERROR, PG_LOG_PRIMARY, gettext(fmt), ap); + va_end(ap); + + exit(1); +} + +/* + * Record details extracted from the backup manifest for one file. + */ +static void +record_manifest_details_for_file(JsonManifestParseContext *context, + char *pathname, size_t size, + pg_checksum_type checksum_type, + int checksum_length, uint8 *checksum_payload) +{ + manifest_data *manifest = context->private_data; + manifest_file *m; + bool found; + + /* Make a new entry in the hash table for this file. */ + m = manifest_files_insert(manifest->files, pathname, &found); + if (found) + pg_fatal("duplicate path name in backup manifest: \"%s\"", pathname); + + /* Initialize the entry. */ + m->size = size; + m->checksum_type = checksum_type; + m->checksum_length = checksum_length; + m->checksum_payload = checksum_payload; +} + +/* + * Record details extracted from the backup manifest for one WAL range. + */ +static void +record_manifest_details_for_wal_range(JsonManifestParseContext *context, + TimeLineID tli, + XLogRecPtr start_lsn, XLogRecPtr end_lsn) +{ + manifest_data *manifest = context->private_data; + manifest_wal_range *range; + + /* Allocate and initialize a struct describing this WAL range. */ + range = palloc(sizeof(manifest_wal_range)); + range->tli = tli; + range->start_lsn = start_lsn; + range->end_lsn = end_lsn; + range->prev = manifest->last_wal_range; + range->next = NULL; + + /* Add it to the end of the list. */ + if (manifest->first_wal_range == NULL) + manifest->first_wal_range = range; + else + manifest->last_wal_range->next = range; + manifest->last_wal_range = range; +} + +/* + * Helper function for manifest_files hash table. + */ +static uint32 +hash_string_pointer(char *s) +{ + unsigned char *ss = (unsigned char *) s; + + return hash_bytes(ss, strlen(s)); +} diff --git a/src/bin/pg_combinebackup/load_manifest.h b/src/bin/pg_combinebackup/load_manifest.h new file mode 100644 index 0000000000..2bfeeff156 --- /dev/null +++ b/src/bin/pg_combinebackup/load_manifest.h @@ -0,0 +1,67 @@ +/*------------------------------------------------------------------------- + * + * Load data from a backup manifest into memory. + * + * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group + * Portions Copyright (c) 1994, Regents of the University of California + * + * src/bin/pg_combinebackup/load_manifest.h + * + *------------------------------------------------------------------------- + */ +#ifndef LOAD_MANIFEST_H +#define LOAD_MANIFEST_H + +#include "access/xlogdefs.h" +#include "common/checksum_helper.h" + +/* + * Each file described by the manifest file is parsed to produce an object + * like this. + */ +typedef struct manifest_file +{ + uint32 status; /* hash status */ + char *pathname; + size_t size; + pg_checksum_type checksum_type; + int checksum_length; + uint8 *checksum_payload; +} manifest_file; + +#define SH_PREFIX manifest_files +#define SH_ELEMENT_TYPE manifest_file +#define SH_KEY_TYPE char * +#define SH_SCOPE extern +#define SH_RAW_ALLOCATOR pg_malloc0 +#define SH_DECLARE +#include "lib/simplehash.h" + +/* + * Each WAL range described by the manifest file is parsed to produce an + * object like this. + */ +typedef struct manifest_wal_range +{ + TimeLineID tli; + XLogRecPtr start_lsn; + XLogRecPtr end_lsn; + struct manifest_wal_range *next; + struct manifest_wal_range *prev; +} manifest_wal_range; + +/* + * All the data parsed from a backup_manifest file. + */ +typedef struct manifest_data +{ + manifest_files_hash *files; + manifest_wal_range *first_wal_range; + manifest_wal_range *last_wal_range; +} manifest_data; + +extern manifest_data *load_backup_manifest(char *backup_directory); +extern manifest_data **load_backup_manifests(int n_backups, + char **backup_directories); + +#endif /* LOAD_MANIFEST_H */ diff --git a/src/bin/pg_combinebackup/meson.build b/src/bin/pg_combinebackup/meson.build new file mode 100644 index 0000000000..a6036dea74 --- /dev/null +++ b/src/bin/pg_combinebackup/meson.build @@ -0,0 +1,35 @@ +# Copyright (c) 2022-2023, PostgreSQL Global Development Group + +pg_combinebackup_sources = files( + 'pg_combinebackup.c', + 'backup_label.c', + 'copy_file.c', + 'load_manifest.c', + 'reconstruct.c', + 'write_manifest.c', +) + +if host_system == 'windows' + pg_combinebackup_sources += rc_bin_gen.process(win32ver_rc, extra_args: [ + '--NAME', 'pg_combinebackup', + '--FILEDESC', 'pg_combinebackup - combine incremental backups',]) +endif + +pg_combinebackup = executable('pg_combinebackup', + pg_combinebackup_sources, + dependencies: [frontend_code], + kwargs: default_bin_args, +) +bin_targets += pg_combinebackup + +tests += { + 'name': 'pg_combinebackup', + 'sd': meson.current_source_dir(), + 'bd': meson.current_build_dir(), + 'tap': { + 'tests': [ + 't/001_basic.pl', + 't/002_compare_backups.pl', + ], + } +} diff --git a/src/bin/pg_combinebackup/pg_combinebackup.c b/src/bin/pg_combinebackup/pg_combinebackup.c new file mode 100644 index 0000000000..32d2846433 --- /dev/null +++ b/src/bin/pg_combinebackup/pg_combinebackup.c @@ -0,0 +1,1270 @@ +/*------------------------------------------------------------------------- + * + * pg_combinebackup.c + * Combine incremental backups with prior backups. + * + * Copyright (c) 2017-2023, PostgreSQL Global Development Group + * + * IDENTIFICATION + * src/bin/pg_combinebackup/pg_combinebackup.c + * + *------------------------------------------------------------------------- + */ +#include "postgres_fe.h" + +#include <dirent.h> +#include <fcntl.h> +#include <limits.h> + +#include "access/xlogdefs.h" +#include "backup_label.h" +#include "common/blkreftable.h" +#include "common/checksum_helper.h" +#include "common/controldata_utils.h" +#include "common/file_perm.h" +#include "common/file_utils.h" +#include "common/logging.h" +#include "copy_file.h" +#include "fe_utils/option_utils.h" +#include "lib/stringinfo.h" +#include "load_manifest.h" +#include "getopt_long.h" +#include "reconstruct.h" +#include "write_manifest.h" + +/* Incremental file naming convention. */ +#define INCREMENTAL_PREFIX "INCREMENTAL." +#define INCREMENTAL_PREFIX_LENGTH 12 + +/* + * Tracking for directories that need to be removed, or have their contents + * removed, if the operation fails. + */ +typedef struct cb_cleanup_dir +{ + char *target_path; + bool rmtopdir; + struct cb_cleanup_dir *next; +} cb_cleanup_dir; + +/* + * Stores a tablespace mapping provided using -T, --tablespace-mapping. + */ +typedef struct cb_tablespace_mapping +{ + char old_dir[MAXPGPATH]; + char new_dir[MAXPGPATH]; + struct cb_tablespace_mapping *next; +} cb_tablespace_mapping; + +/* + * Stores data parsed from all command-line options. + */ +typedef struct cb_options +{ + bool debug; + char *output; + bool dry_run; + bool no_sync; + cb_tablespace_mapping *tsmappings; + pg_checksum_type manifest_checksums; + bool no_manifest; + DataDirSyncMethod sync_method; +} cb_options; + +/* + * Data about a tablespace. + * + * Every normal tablespace needs a tablespace mapping, but in-place tablespaces + * don't, so the list of tablespaces can contain more entries than the list of + * tablespace mappings. + */ +typedef struct cb_tablespace +{ + Oid oid; + bool in_place; + char old_dir[MAXPGPATH]; + char new_dir[MAXPGPATH]; + struct cb_tablespace *next; +} cb_tablespace; + +/* Directories to be removed if we exit uncleanly. */ +cb_cleanup_dir *cleanup_dir_list = NULL; + +static void add_tablespace_mapping(cb_options *opt, char *arg); +static StringInfo check_backup_label_files(int n_backups, char **backup_dirs); +static void check_control_files(int n_backups, char **backup_dirs); +static void check_input_dir_permissions(char *dir); +static void cleanup_directories_atexit(void); +static void create_output_directory(char *dirname, cb_options *opt); +static void help(const char *progname); +static bool parse_oid(char *s, Oid *result); +static void process_directory_recursively(Oid tsoid, + char *input_directory, + char *output_directory, + char *relative_path, + int n_prior_backups, + char **prior_backup_dirs, + manifest_data **manifests, + manifest_writer *mwriter, + cb_options *opt); +static int read_pg_version_file(char *directory); +static void remember_to_cleanup_directory(char *target_path, bool rmtopdir); +static void reset_directory_cleanup_list(void); +static cb_tablespace *scan_for_existing_tablespaces(char *pathname, + cb_options *opt); +static void slurp_file(int fd, char *filename, StringInfo buf, int maxlen); + +/* + * Main program. + */ +int +main(int argc, char *argv[]) +{ + static struct option long_options[] = { + {"debug", no_argument, NULL, 'd'}, + {"dry-run", no_argument, NULL, 'n'}, + {"no-sync", no_argument, NULL, 'N'}, + {"output", required_argument, NULL, 'o'}, + {"tablespace-mapping", no_argument, NULL, 'T'}, + {"manifest-checksums", required_argument, NULL, 1}, + {"no-manifest", no_argument, NULL, 2}, + {"sync-method", required_argument, NULL, 3}, + {NULL, 0, NULL, 0} + }; + + const char *progname; + char *last_input_dir; + int optindex; + int c; + int n_backups; + int n_prior_backups; + int version; + char **prior_backup_dirs; + cb_options opt; + cb_tablespace *tablespaces; + cb_tablespace *ts; + StringInfo last_backup_label; + manifest_data **manifests; + manifest_writer *mwriter; + + pg_logging_init(argv[0]); + progname = get_progname(argv[0]); + handle_help_version_opts(argc, argv, progname, help); + + memset(&opt, 0, sizeof(opt)); + opt.manifest_checksums = CHECKSUM_TYPE_CRC32C; + opt.sync_method = DATA_DIR_SYNC_METHOD_FSYNC; + + /* process command-line options */ + while ((c = getopt_long(argc, argv, "do:nNPT:", + long_options, &optindex)) != -1) + { + switch (c) + { + case 'd': + opt.debug = true; + pg_logging_increase_verbosity(); + break; + case 'o': + opt.output = optarg; + break; + case 'n': + opt.dry_run = true; + break; + case 'N': + opt.no_sync = true; + break; + case 'T': + add_tablespace_mapping(&opt, optarg); + break; + case 1: + if (!pg_checksum_parse_type(optarg, + &opt.manifest_checksums)) + pg_fatal("unrecognized checksum algorithm: \"%s\"", + optarg); + break; + case 2: + opt.no_manifest = true; + break; + case 3: + if (!parse_sync_method(optarg, &opt.sync_method)) + exit(1); + break; + default: + /* getopt_long already emitted a complaint */ + pg_log_error_hint("Try \"%s --help\" for more information.", progname); + exit(1); + } + } + + if (optind >= argc) + { + pg_log_error("%s: no input directories specified", progname); + pg_log_error_hint("Try \"%s --help\" for more information.", progname); + exit(1); + } + + if (opt.output == NULL) + pg_fatal("no output directory specified"); + + /* If no manifest is needed, no checksums are needed, either. */ + if (opt.no_manifest) + opt.manifest_checksums = CHECKSUM_TYPE_NONE; + + /* Read the server version from the final backup. */ + version = read_pg_version_file(argv[argc - 1]); + + /* Sanity-check control files. */ + n_backups = argc - optind; + check_control_files(n_backups, argv + optind); + + /* Sanity-check backup_label files, and get the contents of the last one. */ + last_backup_label = check_backup_label_files(n_backups, argv + optind); + + /* Load backup manifests. */ + manifests = load_backup_manifests(n_backups, argv + optind); + + /* Figure out which tablespaces are going to be included in the output. */ + last_input_dir = argv[argc - 1]; + check_input_dir_permissions(last_input_dir); + tablespaces = scan_for_existing_tablespaces(last_input_dir, &opt); + + /* + * Create output directories. + * + * We create one output directory for the main data directory plus one for + * each non-in-place tablespace. create_output_directory() will arrange + * for those directories to be cleaned up on failure. In-place tablespaces + * aren't handled at this stage because they're located beneath the main + * output directory, and thus the cleanup of that directory will get rid + * of them. Plus, the pg_tblspc directory that needs to contain them + * doesn't exist yet. + */ + atexit(cleanup_directories_atexit); + create_output_directory(opt.output, &opt); + for (ts = tablespaces; ts != NULL; ts = ts->next) + if (!ts->in_place) + create_output_directory(ts->new_dir, &opt); + + /* If we need to write a backup_manifest, prepare to do so. */ + if (!opt.dry_run && !opt.no_manifest) + mwriter = create_manifest_writer(opt.output); + else + mwriter = NULL; + + /* Write backup label into output directory. */ + if (opt.dry_run) + pg_log_debug("would generate \"%s/backup_label\"", opt.output); + else + { + pg_log_debug("generating \"%s/backup_label\"", opt.output); + last_backup_label->cursor = 0; + write_backup_label(opt.output, last_backup_label, + opt.manifest_checksums, mwriter); + } + + /* + * We'll need the pathnames to the prior backups. By "prior" we mean all + * but the last one listed on the command line. + */ + n_prior_backups = argc - optind - 1; + prior_backup_dirs = argv + optind; + + /* Process everything that's not part of a user-defined tablespace. */ + pg_log_debug("processing backup directory \"%s\"", last_input_dir); + process_directory_recursively(InvalidOid, last_input_dir, opt.output, + NULL, n_prior_backups, prior_backup_dirs, + manifests, mwriter, &opt); + + /* Process user-defined tablespaces. */ + for (ts = tablespaces; ts != NULL; ts = ts->next) + { + pg_log_debug("processing tablespace directory \"%s\"", ts->old_dir); + + /* + * If it's a normal tablespace, we need to set up a symbolic link from + * pg_tblspc/${OID} to the target directory; if it's an in-place + * tablespace, we need to create a directory at pg_tblspc/${OID}. + */ + if (!ts->in_place) + { + char linkpath[MAXPGPATH]; + + snprintf(linkpath, MAXPGPATH, "%s/pg_tblspc/%u", opt.output, + ts->oid); + + if (opt.dry_run) + pg_log_debug("would create symbolic link from \"%s\" to \"%s\"", + linkpath, ts->new_dir); + else + { + pg_log_debug("creating symbolic link from \"%s\" to \"%s\"", + linkpath, ts->new_dir); + if (symlink(ts->new_dir, linkpath) != 0) + pg_fatal("could not create symbolic link from \"%s\" to \"%s\": %m", + linkpath, ts->new_dir); + } + } + else + { + if (opt.dry_run) + pg_log_debug("would create directory \"%s\"", ts->new_dir); + else + { + pg_log_debug("creating directory \"%s\"", ts->new_dir); + if (pg_mkdir_p(ts->new_dir, pg_dir_create_mode) == -1) + pg_fatal("could not create directory \"%s\": %m", + ts->new_dir); + } + } + + /* OK, now handle the directory contents. */ + process_directory_recursively(ts->oid, ts->old_dir, ts->new_dir, + NULL, n_prior_backups, prior_backup_dirs, + manifests, mwriter, &opt); + } + + /* Finalize the backup_manifest, if we're generating one. */ + if (mwriter != NULL) + finalize_manifest(mwriter, + manifests[n_prior_backups]->first_wal_range); + + /* fsync that output directory unless we've been told not to do so */ + if (!opt.no_sync) + { + if (opt.dry_run) + pg_log_debug("would recursively fsync \"%s\"", opt.output); + else + { + pg_log_debug("recursively fsyncing \"%s\"", opt.output); + sync_pgdata(opt.output, version * 10000, opt.sync_method); + } + } + + /* It's a success, so don't remove the output directories. */ + reset_directory_cleanup_list(); + exit(0); +} + +/* + * Process the option argument for the -T, --tablespace-mapping switch. + */ +static void +add_tablespace_mapping(cb_options *opt, char *arg) +{ + cb_tablespace_mapping *tsmap = pg_malloc0(sizeof(cb_tablespace_mapping)); + char *dst; + char *dst_ptr; + char *arg_ptr; + + /* + * Basically, we just want to copy everything before the equals sign to + * tsmap->old_dir and everything afterwards to tsmap->new_dir, but if + * there's more or less than one equals sign, that's an error, and if + * there's an equals sign preceded by a backslash, don't treat it as a + * field separator but instead copy a literal equals sign. + */ + dst_ptr = dst = tsmap->old_dir; + for (arg_ptr = arg; *arg_ptr != '\0'; arg_ptr++) + { + if (dst_ptr - dst >= MAXPGPATH) + pg_fatal("directory name too long"); + + if (*arg_ptr == '\\' && *(arg_ptr + 1) == '=') + ; /* skip backslash escaping = */ + else if (*arg_ptr == '=' && (arg_ptr == arg || *(arg_ptr - 1) != '\\')) + { + if (tsmap->new_dir[0] != '\0') + pg_fatal("multiple \"=\" signs in tablespace mapping"); + else + dst = dst_ptr = tsmap->new_dir; + } + else + *dst_ptr++ = *arg_ptr; + } + if (!tsmap->old_dir[0] || !tsmap->new_dir[0]) + pg_fatal("invalid tablespace mapping format \"%s\", must be \"OLDDIR=NEWDIR\"", arg); + + /* + * All tablespaces are created with absolute directories, so specifying a + * non-absolute path here would never match, possibly confusing users. + * + * In contrast to pg_basebackup, both the old and new directories are on + * the local machine, so the local machine's definition of an absolute + * path is the only relevant one. + */ + if (!is_absolute_path(tsmap->old_dir)) + pg_fatal("old directory is not an absolute path in tablespace mapping: %s", + tsmap->old_dir); + + if (!is_absolute_path(tsmap->new_dir)) + pg_fatal("old directory is not an absolute path in tablespace mapping: %s", + tsmap->new_dir); + + /* Canonicalize paths to avoid spurious failures when comparing. */ + canonicalize_path(tsmap->old_dir); + canonicalize_path(tsmap->new_dir); + + /* Add it to the list. */ + tsmap->next = opt->tsmappings; + opt->tsmappings = tsmap; +} + +/* + * Check that the backup_label files form a coherent backup chain, and return + * the contents of the backup_label file from the latest backup. + */ +static StringInfo +check_backup_label_files(int n_backups, char **backup_dirs) +{ + StringInfo buf = makeStringInfo(); + StringInfo lastbuf = buf; + int i; + TimeLineID check_tli = 0; + XLogRecPtr check_lsn = InvalidXLogRecPtr; + + /* Try to read each backup_label file in turn, last to first. */ + for (i = n_backups - 1; i >= 0; --i) + { + char pathbuf[MAXPGPATH]; + int fd; + TimeLineID start_tli; + TimeLineID previous_tli; + XLogRecPtr start_lsn; + XLogRecPtr previous_lsn; + + /* Open the backup_label file. */ + snprintf(pathbuf, MAXPGPATH, "%s/backup_label", backup_dirs[i]); + pg_log_debug("reading \"%s\"", pathbuf); + if ((fd = open(pathbuf, O_RDONLY, 0)) < 0) + pg_fatal("could not open file \"%s\": %m", pathbuf); + + /* + * Slurp the whole file into memory. + * + * The exact size limit that we impose here doesn't really matter -- + * most of what's supposed to be in the file is fixed size and quite + * short. However, the length of the backup_label is limited (at least + * by some parts of the code) to MAXGPATH, so include that value in + * the maximum length that we tolerate. + */ + slurp_file(fd, pathbuf, buf, 10000 + MAXPGPATH); + + /* Close the file. */ + if (close(fd) != 0) + pg_fatal("could not close \"%s\": %m", pathbuf); + + /* Parse the file contents. */ + parse_backup_label(pathbuf, buf, &start_tli, &start_lsn, + &previous_tli, &previous_lsn); + + /* + * Sanity checks. + * + * XXX. It's actually not required that start_lsn == check_lsn. It + * would be OK if start_lsn > check_lsn provided that start_lsn is + * less than or equal to the relevant switchpoint. But at the moment + * we don't have that information. + */ + if (i > 0 && previous_tli == 0) + pg_fatal("backup at \"%s\" is a full backup, but only the first backup should be a full backup", + backup_dirs[i]); + if (i == 0 && previous_tli != 0) + pg_fatal("backup at \"%s\" is an incremental backup, but the first backup should be a full backup", + backup_dirs[i]); + if (i < n_backups - 1 && start_tli != check_tli) + pg_fatal("backup at \"%s\" starts on timeline %u, but expected %u", + backup_dirs[i], start_tli, check_tli); + if (i < n_backups - 1 && start_lsn != check_lsn) + pg_fatal("backup at \"%s\" starts at LSN %X/%X, but expected %X/%X", + backup_dirs[i], + LSN_FORMAT_ARGS(start_lsn), + LSN_FORMAT_ARGS(check_lsn)); + check_tli = previous_tli; + check_lsn = previous_lsn; + + /* + * The last backup label in the chain needs to be saved for later use, + * while the others are only needed within this loop. + */ + if (lastbuf == buf) + buf = makeStringInfo(); + else + resetStringInfo(buf); + } + + /* Free memory that we don't need any more. */ + if (lastbuf != buf) + { + pfree(buf->data); + pfree(buf); + } + + /* + * Return the data from the first backup_info that we read (which is the + * backup_label from the last directory specified on the command line). + */ + return lastbuf; +} + +/* + * Sanity check control files. + */ +static void +check_control_files(int n_backups, char **backup_dirs) +{ + int i; + uint64 system_identifier; + + /* Try to read each control file in turn, last to first. */ + for (i = n_backups - 1; i >= 0; --i) + { + ControlFileData *control_file; + bool crc_ok; + + pg_log_debug("reading \"%s/global/pg_control\"", backup_dirs[i]); + control_file = get_controlfile(backup_dirs[i], &crc_ok); + + /* Control file contents not meaningful if CRC is bad. */ + if (!crc_ok) + pg_fatal("%s/global/pg_control: crc is incorrect", backup_dirs[i]); + + /* Can't interpret control file if not current version. */ + if (control_file->pg_control_version != PG_CONTROL_VERSION) + pg_fatal("%s/global/pg_control: unexpected control file version", + backup_dirs[i]); + + /* System identifiers should all match. */ + if (i == n_backups - 1) + system_identifier = control_file->system_identifier; + else if (system_identifier != control_file->system_identifier) + pg_fatal("%s/global/pg_control: expected system identifier %llu, but found %llu", + backup_dirs[i], (unsigned long long) system_identifier, + (unsigned long long) control_file->system_identifier); + + /* Release memory. */ + pfree(control_file); + } + + /* + * If debug output is enabled, make a note of the system identifier that + * we found in all of the relevant control files. + */ + pg_log_debug("system identifier is %llu", + (unsigned long long) system_identifier); +} + +/* + * Set default permissions for new files and directories based on the + * permissions of the given directory. The intent here is that the output + * directory should use the same permissions scheme as the final input + * directory. + */ +static void +check_input_dir_permissions(char *dir) +{ + struct stat st; + + if (stat(dir, &st) != 0) + pg_fatal("could not stat \"%s\": %m", dir); + + SetDataDirectoryCreatePerm(st.st_mode); +} + +/* + * Clean up output directories before exiting. + */ +static void +cleanup_directories_atexit(void) +{ + while (cleanup_dir_list != NULL) + { + cb_cleanup_dir *dir = cleanup_dir_list; + + if (dir->rmtopdir) + { + pg_log_info("removing output directory \"%s\"", dir->target_path); + if (!rmtree(dir->target_path, dir->rmtopdir)) + pg_log_error("failed to remove output directory"); + } + else + { + pg_log_info("removing contents of output directory \"%s\"", + dir->target_path); + if (!rmtree(dir->target_path, dir->rmtopdir)) + pg_log_error("failed to remove contents of output directory"); + } + + cleanup_dir_list = cleanup_dir_list->next; + pfree(dir); + } +} + +/* + * Create the named output directory, unless it already exists or we're in + * dry-run mode. If it already exists but is not empty, that's a fatal error. + * + * Adds the created directory to the list of directories to be cleaned up + * at process exit. + */ +static void +create_output_directory(char *dirname, cb_options *opt) +{ + switch (pg_check_dir(dirname)) + { + case 0: + if (opt->dry_run) + { + pg_log_debug("would create directory \"%s\"", dirname); + return; + } + pg_log_debug("creating directory \"%s\"", dirname); + if (pg_mkdir_p(dirname, pg_dir_create_mode) == -1) + pg_fatal("could not create directory \"%s\": %m", dirname); + remember_to_cleanup_directory(dirname, true); + break; + + case 1: + pg_log_debug("using existing directory \"%s\"", dirname); + remember_to_cleanup_directory(dirname, false); + break; + + case 2: + case 3: + case 4: + pg_fatal("directory \"%s\" exists but is not empty", dirname); + + case -1: + pg_fatal("could not access directory \"%s\": %m", dirname); + } +} + +/* + * help + * + * Prints help page for the program + * + * progname: the name of the executed program, such as "pg_combinebackup" + */ +static void +help(const char *progname) +{ + printf(_("%s combines incremental backups.\n\n"), progname); + printf(_("Usage:\n")); + printf(_(" %s [OPTION]... DIRECTORY...\n"), progname); + printf(_("\nOptions:\n")); + printf(_(" -d, --debug generate lots of debugging output\n")); + printf(_(" -n, --dry-run don't actually do anything\n")); + printf(_(" -N, --no-sync do not wait for changes to be written safely to disk\n")); + printf(_(" -o, --output output directory\n")); + printf(_(" -T, --tablespace-mapping=OLDDIR=NEWDIR\n")); + printf(_(" relocate tablespace in OLDDIR to NEWDIR\n")); + printf(_(" --manifest-checksums=SHA{224,256,384,512}|CRC32C|NONE\n" + " use algorithm for manifest checksums\n")); + printf(_(" --no-manifest suppress generation of backup manifest\n")); + printf(_(" --sync-method=METHOD set method for syncing files to disk\n")); + printf(_(" -?, --help show this help, then exit\n")); + + printf(_("\nReport bugs to <%s>.\n"), PACKAGE_BUGREPORT); + printf(_("%s home page: <%s>\n"), PACKAGE_NAME, PACKAGE_URL); +} + +/* + * Try to parse a string as a non-zero OID without leading zeroes. + * + * If it works, return true and set *result to the answer, else return false. + */ +static bool +parse_oid(char *s, Oid *result) +{ + Oid oid; + char *ep; + + errno = 0; + oid = strtoul(s, &ep, 10); + if (errno != 0 || *ep != '\0' || oid < 1 || oid > PG_UINT32_MAX) + return false; + + *result = oid; + return true; +} + +/* + * Copy files from the input directory to the output directory, reconstructing + * full files from incremental files as required. + * + * If processing is a user-defined tablespace, the tsoid should be the OID + * of that tablespace and input_directory and output_directory should be the + * toplevel input and output directories for that tablespace. Otherwise, + * tsoid should be InvalidOid and input_directory and output_directory should + * be the main input and output directories. + * + * relative_path is the path beneath the given input and output directories + * that we are currently processing. If NULL, it indicates that we're + * processing the input and output directories themselves. + * + * n_prior_backups is the number of prior backups that we have available. + * This doesn't count the very last backup, which is referenced by + * output_directory, just the older ones. prior_backup_dirs is an array of + * the locations of those previous backups. + */ +static void +process_directory_recursively(Oid tsoid, + char *input_directory, + char *output_directory, + char *relative_path, + int n_prior_backups, + char **prior_backup_dirs, + manifest_data **manifests, + manifest_writer *mwriter, + cb_options *opt) +{ + char ifulldir[MAXPGPATH]; + char ofulldir[MAXPGPATH]; + char manifest_prefix[MAXPGPATH]; + DIR *dir; + struct dirent *de; + bool is_pg_tblspc; + bool is_pg_wal; + manifest_data *latest_manifest = manifests[n_prior_backups]; + pg_checksum_type checksum_type; + + StaticAssertStmt(strlen(INCREMENTAL_PREFIX) == INCREMENTAL_PREFIX_LENGTH, + "INCREMENTAL_PREFIX_LENGTH is incorrect"); + + /* + * pg_tblspc and pg_wal are special cases, so detect those here. + * + * pg_tblspc is only special at the top level, but subdirectories of + * pg_wal are just as special as the top level directory. + * + * Since incremental backup does not exist in pre-v10 versions, we don't + * have to worry about the old pg_xlog naming. + */ + is_pg_tblspc = !OidIsValid(tsoid) && relative_path != NULL && + strcmp(relative_path, "pg_tblspc") == 0; + is_pg_wal = !OidIsValid(tsoid) && relative_path != NULL && + (strcmp(relative_path, "pg_wal") == 0 || + strncmp(relative_path, "pg_wal/", 7) == 0); + + /* + * If we're under pg_wal, then we don't need checksums, because these + * files aren't included in the backup manifest. Otherwise use whatever + * type of checksum is configured. + */ + if (!is_pg_wal) + checksum_type = opt->manifest_checksums; + else + checksum_type = CHECKSUM_TYPE_NONE; + + /* + * Append the relative path to the input and output directories, and + * figure out the appropriate prefix to add to files in this directory + * when looking them up in a backup manifest. + */ + if (relative_path == NULL) + { + strncpy(ifulldir, input_directory, MAXPGPATH); + strncpy(ofulldir, output_directory, MAXPGPATH); + if (OidIsValid(tsoid)) + snprintf(manifest_prefix, MAXPGPATH, "pg_tblspc/%u/", tsoid); + else + manifest_prefix[0] = '\0'; + } + else + { + snprintf(ifulldir, MAXPGPATH, "%s/%s", input_directory, + relative_path); + snprintf(ofulldir, MAXPGPATH, "%s/%s", output_directory, + relative_path); + if (OidIsValid(tsoid)) + snprintf(manifest_prefix, MAXPGPATH, "pg_tblspc/%u/%s/", + tsoid, relative_path); + else + snprintf(manifest_prefix, MAXPGPATH, "%s/", relative_path); + } + + /* + * Toplevel output directories have already been created by the time this + * function is called, but any subdirectories are our responsibility. + */ + if (relative_path != NULL) + { + if (opt->dry_run) + pg_log_debug("would create directory \"%s\"", ofulldir); + else + { + pg_log_debug("creating directory \"%s\"", ofulldir); + if (mkdir(ofulldir, pg_dir_create_mode) == -1) + pg_fatal("could not create directory \"%s\": %m", ofulldir); + } + } + + /* It's time to scan the directory. */ + if ((dir = opendir(ifulldir)) == NULL) + pg_fatal("could not open directory \"%s\": %m", ifulldir); + while (errno = 0, (de = readdir(dir)) != NULL) + { + PGFileType type; + char ifullpath[MAXPGPATH]; + char ofullpath[MAXPGPATH]; + char manifest_path[MAXPGPATH]; + Oid oid = InvalidOid; + int checksum_length = 0; + uint8 *checksum_payload = NULL; + pg_checksum_context checksum_ctx; + + /* Ignore "." and ".." entries. */ + if (strcmp(de->d_name, ".") == 0 || + strcmp(de->d_name, "..") == 0) + continue; + + /* Construct input path. */ + snprintf(ifullpath, MAXPGPATH, "%s/%s", ifulldir, de->d_name); + + /* Figure out what kind of directory entry this is. */ + type = get_dirent_type(ifullpath, de, false, PG_LOG_ERROR); + if (type == PGFILETYPE_ERROR) + exit(1); + + /* + * If we're processing pg_tblspc, then check whether the filename + * looks like it could be a tablespace OID. If so, and if the + * directory entry is a symbolic link or a directory, skip it. + * + * Our goal here is to ignore anything that would have been considered + * by scan_for_existing_tablespaces to be a tablespace. + */ + if (is_pg_tblspc && parse_oid(de->d_name, &oid) && + (type == PGFILETYPE_LNK || type == PGFILETYPE_DIR)) + continue; + + /* If it's a directory, recurse. */ + if (type == PGFILETYPE_DIR) + { + char new_relative_path[MAXPGPATH]; + + /* Append new pathname component to relative path. */ + if (relative_path == NULL) + strncpy(new_relative_path, de->d_name, MAXPGPATH); + else + snprintf(new_relative_path, MAXPGPATH, "%s/%s", relative_path, + de->d_name); + + /* And recurse. */ + process_directory_recursively(tsoid, + input_directory, output_directory, + new_relative_path, + n_prior_backups, prior_backup_dirs, + manifests, mwriter, opt); + continue; + } + + /* Skip anything that's not a regular file. */ + if (type != PGFILETYPE_REG) + { + if (type == PGFILETYPE_LNK) + pg_log_warning("skipping symbolic link \"%s\"", ifullpath); + else + pg_log_warning("skipping special file \"%s\"", ifullpath); + continue; + } + + /* + * Skip the backup_label and backup_manifest files; they require + * special handling and are handled elsewhere. + */ + if (relative_path == NULL && + (strcmp(de->d_name, "backup_label") == 0 || + strcmp(de->d_name, "backup_manifest") == 0)) + continue; + + /* + * If it's an incremental file, hand it off to the reconstruction + * code, which will figure out what to do. + */ + if (strncmp(de->d_name, INCREMENTAL_PREFIX, + INCREMENTAL_PREFIX_LENGTH) == 0) + { + /* Output path should not include "INCREMENTAL." prefix. */ + snprintf(ofullpath, MAXPGPATH, "%s/%s", ofulldir, + de->d_name + INCREMENTAL_PREFIX_LENGTH); + + + /* Manifest path likewise omits incremental prefix. */ + snprintf(manifest_path, MAXPGPATH, "%s%s", manifest_prefix, + de->d_name + INCREMENTAL_PREFIX_LENGTH); + + /* Reconstruction logic will do the rest. */ + reconstruct_from_incremental_file(ifullpath, ofullpath, + relative_path, + de->d_name + INCREMENTAL_PREFIX_LENGTH, + n_prior_backups, + prior_backup_dirs, + manifests, + manifest_path, + checksum_type, + &checksum_length, + &checksum_payload, + opt->dry_run); + } + else + { + /* Construct the path that the backup_manifest will use. */ + snprintf(manifest_path, MAXPGPATH, "%s%s", manifest_prefix, + de->d_name); + + /* + * It's not an incremental file, so we need to copy the entire + * file to the output directory. + * + * If a checksum of the required type already exists in the + * backup_manifest for the final input directory, we can save some + * work by reusing that checksum instead of computing a new one. + */ + if (checksum_type != CHECKSUM_TYPE_NONE && + latest_manifest != NULL) + { + manifest_file *mfile; + + mfile = manifest_files_lookup(latest_manifest->files, + manifest_path); + if (mfile == NULL) + { + /* + * The directory is out of sync with the backup_manifest, + * so emit a warning. + */ + pg_log_warning("\"%s/backup_manifest\" contains no entry for \"%s\"", + input_directory, manifest_path); + } + else if (mfile->checksum_type == checksum_type) + { + checksum_length = mfile->checksum_length; + checksum_payload = mfile->checksum_payload; + } + } + + /* + * If we're reusing a checksum, then we don't need copy_file() to + * compute one for us, but otherwise, it needs to compute whatever + * type of checksum we need. + */ + if (checksum_length != 0) + pg_checksum_init(&checksum_ctx, CHECKSUM_TYPE_NONE); + else + pg_checksum_init(&checksum_ctx, checksum_type); + + /* Actually copy the file. */ + snprintf(ofullpath, MAXPGPATH, "%s/%s", ofulldir, de->d_name); + copy_file(ifullpath, ofullpath, &checksum_ctx, opt->dry_run); + + /* + * If copy_file() performed a checksum calculation for us, then + * save the results (except in dry-run mode, when there's no + * point). + */ + if (checksum_ctx.type != CHECKSUM_TYPE_NONE && !opt->dry_run) + { + checksum_payload = pg_malloc(PG_CHECKSUM_MAX_LENGTH); + checksum_length = pg_checksum_final(&checksum_ctx, + checksum_payload); + } + } + + /* Generate manifest entry, if needed. */ + if (mwriter != NULL) + { + struct stat sb; + + /* + * In order to generate a manifest entry, we need the file size + * and mtime. We have no way to know the correct mtime except to + * stat() the file, so just do that and get the size as well. + * + * If we didn't need the mtime here, we could try to obtain the + * file size from the reconstruction or file copy process above, + * although that is actually not convenient in all cases. If we + * write the file ourselves then clearly we can keep a count of + * bytes, but if we use something like CopyFile() then it's + * trickier. Since we have to stat() anyway to get the mtime, + * there's no point in worrying about it. + */ + if (stat(ofullpath, &sb) < 0) + pg_fatal("could not stat file \"%s\": %m", ofullpath); + + /* OK, now do the work. */ + add_file_to_manifest(mwriter, manifest_path, + sb.st_size, sb.st_mtime, + checksum_type, checksum_length, + checksum_payload); + } + + /* Avoid leaking memory. */ + if (checksum_payload != NULL) + pfree(checksum_payload); + } + + closedir(dir); +} + +/* + * Read the version number from PG_VERSION and convert it to the usual server + * version number format. (e.g. If PG_VERSION contains "14\n" this function + * will return 140000) + */ +static int +read_pg_version_file(char *directory) +{ + char filename[MAXPGPATH]; + StringInfoData buf; + int fd; + int version; + char *ep; + + /* Construct pathname. */ + snprintf(filename, MAXPGPATH, "%s/PG_VERSION", directory); + + /* Open file. */ + if ((fd = open(filename, O_RDONLY, 0)) < 0) + pg_fatal("could not open file \"%s\": %m", filename); + + /* Read into memory. Length limit of 128 should be more than generous. */ + initStringInfo(&buf); + slurp_file(fd, filename, &buf, 128); + + /* Close the file. */ + if (close(fd) != 0) + pg_fatal("could not close \"%s\": %m", filename); + + /* Convert to integer. */ + errno = 0; + version = strtoul(buf.data, &ep, 10); + if (errno != 0 || *ep != '\n') + { + /* + * Incremental backup is not relevant to very old server versions that + * used multi-part version number (e.g. 9.6, or 8.4). So if we see + * what looks like the beginning of such a version number, just bail + * out. + */ + if (version < 10 && *ep == '.') + pg_fatal("%s: server version too old\n", filename); + pg_fatal("%s: could not parse version number\n", filename); + } + + /* Debugging output. */ + pg_log_debug("read server version %d from \"%s\"", version, filename); + + /* Release memory and return result. */ + pfree(buf.data); + return version * 10000; +} + +/* + * Add a directory to the list of output directories to clean up. + */ +static void +remember_to_cleanup_directory(char *target_path, bool rmtopdir) +{ + cb_cleanup_dir *dir = pg_malloc(sizeof(cb_cleanup_dir)); + + dir->target_path = target_path; + dir->rmtopdir = rmtopdir; + dir->next = cleanup_dir_list; + cleanup_dir_list = dir; +} + +/* + * Empty out the list of directories scheduled for cleanup a exit. + * + * We want to remove the output directories only on a failure, so call this + * function when we know that the operation has succeeded. + * + * Since we only expect this to be called when we're about to exit, we could + * just set cleanup_dir_list to NULL and be done with it, but we free the + * memory to be tidy. + */ +static void +reset_directory_cleanup_list(void) +{ + while (cleanup_dir_list != NULL) + { + cb_cleanup_dir *dir = cleanup_dir_list; + + cleanup_dir_list = cleanup_dir_list->next; + pfree(dir); + } +} + +/* + * Scan the pg_tblspc directory of the final input backup to get a canonical + * list of what tablespaces are part of the backup. + * + * 'pathname' should be the path to the toplevel backup directory for the + * final backup in the backup chain. + */ +static cb_tablespace * +scan_for_existing_tablespaces(char *pathname, cb_options *opt) +{ + char pg_tblspc[MAXPGPATH]; + DIR *dir; + struct dirent *de; + cb_tablespace *tslist = NULL; + + snprintf(pg_tblspc, MAXPGPATH, "%s/pg_tblspc", pathname); + pg_log_debug("scanning \"%s\"", pg_tblspc); + + if ((dir = opendir(pg_tblspc)) == NULL) + pg_fatal("could not open directory \"%s\": %m", pathname); + + while (errno = 0, (de = readdir(dir)) != NULL) + { + Oid oid; + char tblspcdir[MAXPGPATH]; + char link_target[MAXPGPATH]; + int link_length; + cb_tablespace *ts; + cb_tablespace *otherts; + PGFileType type; + + /* Silently ignore "." and ".." entries. */ + if (strcmp(de->d_name, ".") == 0 || strcmp(de->d_name, "..") == 0) + continue; + + /* Construct full pathname. */ + snprintf(tblspcdir, MAXPGPATH, "%s/%s", pg_tblspc, de->d_name); + + /* Ignore any file name that doesn't look like a proper OID. */ + if (!parse_oid(de->d_name, &oid)) + { + pg_log_debug("skipping \"%s\" because the filename is not a legal tablespace OID", + tblspcdir); + continue; + } + + /* Only symbolic links and directories are tablespaces. */ + type = get_dirent_type(tblspcdir, de, false, PG_LOG_ERROR); + if (type == PGFILETYPE_ERROR) + exit(1); + if (type != PGFILETYPE_LNK && type != PGFILETYPE_DIR) + { + pg_log_debug("skipping \"%s\" because it is neither a symbolic link nor a directory", + tblspcdir); + continue; + } + + /* Create a new tablespace object. */ + ts = pg_malloc0(sizeof(cb_tablespace)); + ts->oid = oid; + + /* + * If it's a link, it's not an in-place tablespace. Otherwise, it must + * be a directory, and thus an in-place tablespace. + */ + if (type == PGFILETYPE_LNK) + { + cb_tablespace_mapping *tsmap; + + /* Read the link target. */ + link_length = readlink(tblspcdir, link_target, sizeof(link_target)); + if (link_length < 0) + pg_fatal("could not read symbolic link \"%s\": %m", + tblspcdir); + if (link_length >= sizeof(link_target)) + pg_fatal("symbolic link \"%s\" is too long", tblspcdir); + link_target[link_length] = '\0'; + if (!is_absolute_path(link_target)) + pg_fatal("symbolic link \"%s\" is relative", tblspcdir); + + /* Caonicalize the link target. */ + canonicalize_path(link_target); + + /* + * Find the corresponding tablespace mapping and copy the relevant + * details into the new tablespace entry. + */ + for (tsmap = opt->tsmappings; tsmap != NULL; tsmap = tsmap->next) + { + if (strcmp(tsmap->old_dir, link_target) == 0) + { + strncpy(ts->old_dir, tsmap->old_dir, MAXPGPATH); + strncpy(ts->new_dir, tsmap->new_dir, MAXPGPATH); + ts->in_place = false; + break; + } + } + + /* Every non-in-place tablespace must be mapped. */ + if (tsmap == NULL) + pg_fatal("tablespace at \"%s\" has no tablespace mapping", + link_target); + } + else + { + /* + * For an in-place tablespace, there's no separate directory, so + * we just record the paths within the data directories. + */ + snprintf(ts->old_dir, MAXPGPATH, "%s/%s", pg_tblspc, de->d_name); + snprintf(ts->new_dir, MAXPGPATH, "%s/pg_tblpc/%s", opt->output, + de->d_name); + ts->in_place = true; + } + + /* Tablespaces should not share a directory. */ + for (otherts = tslist; otherts != NULL; otherts = otherts->next) + if (strcmp(ts->new_dir, otherts->new_dir) == 0) + pg_fatal("tablespaces with OIDs %u and %u both point at \"%s\"", + otherts->oid, oid, ts->new_dir); + + /* Add this tablespace to the list. */ + ts->next = tslist; + tslist = ts; + } + + return tslist; +} + +/* + * Read a file into a StringInfo. + * + * fd is used for the actual file I/O, filename for error reporting purposes. + * A file longer than maxlen is a fatal error. + */ +static void +slurp_file(int fd, char *filename, StringInfo buf, int maxlen) +{ + struct stat st; + ssize_t rb; + + /* Check file size, and complain if it's too large. */ + if (fstat(fd, &st) != 0) + pg_fatal("could not stat \"%s\": %m", filename); + if (st.st_size > maxlen) + pg_fatal("file \"%s\" is too large", filename); + + /* Make sure we have enough space. */ + enlargeStringInfo(buf, st.st_size); + + /* Read the data. */ + rb = read(fd, &buf->data[buf->len], st.st_size); + + /* + * We don't expect any concurrent changes, so we should read exactly the + * expected number of bytes. + */ + if (rb != st.st_size) + { + if (rb < 0) + pg_fatal("could not read file \"%s\": %m", filename); + else + pg_fatal("could not read file \"%s\": read only %d of %d bytes", + filename, (int) rb, (int) st.st_size); + } + + /* Adjust buffer length for new data and restore trailing-\0 invariant */ + buf->len += rb; + buf->data[buf->len] = '\0'; +} diff --git a/src/bin/pg_combinebackup/reconstruct.c b/src/bin/pg_combinebackup/reconstruct.c new file mode 100644 index 0000000000..c774bf1842 --- /dev/null +++ b/src/bin/pg_combinebackup/reconstruct.c @@ -0,0 +1,618 @@ +/*------------------------------------------------------------------------- + * + * reconstruct.c + * Reconstruct full file from incremental file and backup chain. + * + * Copyright (c) 2017-2023, PostgreSQL Global Development Group + * + * IDENTIFICATION + * src/bin/pg_combinebackup/reconstruct.c + * + *------------------------------------------------------------------------- + */ +#include "postgres_fe.h" + +#include <unistd.h> + +#include "backup/basebackup_incremental.h" +#include "common/logging.h" +#include "common/file_perm.h" +#include "copy_file.h" +#include "reconstruct.h" +#include "storage/block.h" + +/* + * An rfile stores the data that we need in order to be able to use some file + * on disk for reconstruction. For any given output file, we create one rfile + * per backup that we need to consult when we constructing that output file. + * + * If we find a full version of the file in the backup chain, then only + * filename and fd are initialized; the remaining fields are 0 or NULL. + * For an incremental file, header_length, num_blocks, relative_block_numbers, + * and truncation_block_length are also set. + * + * num_blocks_read and highest_offset_read always start out as 0. + */ +typedef struct rfile +{ + char *filename; + int fd; + size_t header_length; + unsigned num_blocks; + BlockNumber *relative_block_numbers; + unsigned truncation_block_length; + unsigned num_blocks_read; + off_t highest_offset_read; +} rfile; + +static void debug_reconstruction(int n_source, + rfile **sources, + bool dry_run); +static unsigned find_reconstructed_block_length(rfile *s); +static rfile *make_incremental_rfile(char *filename); +static rfile *make_rfile(char *filename, bool missing_ok); +static void write_reconstructed_file(char *input_filename, + char *output_filename, + unsigned block_length, + rfile **sourcemap, + off_t *offsetmap, + pg_checksum_context *checksum_ctx, + bool dry_run); +static void read_bytes(rfile *rf, void *buffer, unsigned length); + +/* + * Reconstruct a full file from an incremental file and a chain of prior + * backups. + * + * input_filename should be the path to the incremental file, and + * output_filename should be the path where the reconstructed file is to be + * written. + * + * relative_path should be the relative path to the directory containing this + * file. bare_file_name should be the name of the file within that directory, + * without "INCREMENTAL.". + * + * n_prior_backups is the number of prior backups, and prior_backup_dirs is + * an array of pathnames where those backups can be found. + */ +void +reconstruct_from_incremental_file(char *input_filename, + char *output_filename, + char *relative_path, + char *bare_file_name, + int n_prior_backups, + char **prior_backup_dirs, + manifest_data **manifests, + char *manifest_path, + pg_checksum_type checksum_type, + int *checksum_length, + uint8 **checksum_payload, + bool dry_run) +{ + rfile **source; + rfile *latest_source = NULL; + rfile **sourcemap; + off_t *offsetmap; + unsigned block_length; + unsigned num_missing_blocks; + unsigned i; + unsigned sidx = n_prior_backups; + bool full_copy_possible = true; + int copy_source_index = -1; + rfile *copy_source = NULL; + pg_checksum_context checksum_ctx; + + /* + * Every block must come either from the latest version of the file or + * from one of the prior backups. + */ + source = pg_malloc0(sizeof(rfile *) * (1 + n_prior_backups)); + + /* + * Use the information from the latest incremental file to figure out how + * long the reconstructed file should be. + */ + latest_source = make_incremental_rfile(input_filename); + source[n_prior_backups] = latest_source; + block_length = find_reconstructed_block_length(latest_source); + + /* + * For each block in the output file, we need to know from which file we + * need to obtain it and at what offset in that file it's stored. + * sourcemap gives us the first of these things, and offsetmap the latter. + */ + sourcemap = pg_malloc0(sizeof(rfile *) * block_length); + offsetmap = pg_malloc0(sizeof(off_t) * block_length); + + /* + * Blocks prior to the truncation_block_length threshold must be obtained + * from some prior backup, while those after that threshold are left as + * zeroes if not present in the newest incremental file. + * num_missing_blocks counts the number of blocks that we must be found + * somewhere in the backup chain, and is thus initially equal to + * truncation_block_length. + */ + num_missing_blocks = latest_source->truncation_block_length; + + /* + * Every block that is present in the newest incremental file should be + * sourced from that file. If it precedes the truncation_block_length, + * it's a block that we would otherwise have had to find in an older + * backup and thus reduces the number of blocks remaining to be found by + * one; otherwise, it's an extra block that needs to be included in the + * output but would not have needed to be found in an older backup if it + * had not been present. + */ + for (i = 0; i < latest_source->num_blocks; ++i) + { + BlockNumber b = latest_source->relative_block_numbers[i]; + + Assert(b < block_length); + sourcemap[b] = latest_source; + offsetmap[b] = latest_source->header_length + (i * BLCKSZ); + if (b < latest_source->truncation_block_length) + num_missing_blocks--; + + /* + * A full copy of a file from an earlier backup is only possible if no + * blocks are needed from any later incremental file. + */ + full_copy_possible = false; + } + + while (num_missing_blocks > 0) + { + char source_filename[MAXPGPATH]; + rfile *s; + + /* + * Move to the next backup in the chain. If there are no more, then + * something has gone wrong and reconstruction has failed. + */ + if (sidx == 0) + pg_fatal("reconstruction for file \"%s\" failed to find %u required blocks", + output_filename, num_missing_blocks); + --sidx; + + /* + * Look for the full file in the previous backup. If not found, then + * look for an incremental file instead. + */ + snprintf(source_filename, MAXPGPATH, "%s/%s/%s", + prior_backup_dirs[sidx], relative_path, bare_file_name); + if ((s = make_rfile(source_filename, true)) == NULL) + { + snprintf(source_filename, MAXPGPATH, "%s/%s/INCREMENTAL.%s", + prior_backup_dirs[sidx], relative_path, bare_file_name); + s = make_incremental_rfile(source_filename); + } + source[sidx] = s; + + /* + * If s->header_length == 0, then this is a full file; otherwise, it's + * an incremental file. + */ + if (s->header_length != 0) + { + /* + * Since we found another incremental file, source all blocks from + * it that we need but don't yet have. + */ + for (i = 0; i < s->num_blocks; ++i) + { + BlockNumber b = s->relative_block_numbers[i]; + + if (b < latest_source->truncation_block_length && + sourcemap[b] == NULL) + { + sourcemap[b] = s; + offsetmap[b] = s->header_length + (i * BLCKSZ); + + Assert(num_missing_blocks > 0); + --num_missing_blocks; + + /* + * A full copy of a file from an earlier backup is only + * possible if no blocks are needed from any later + * incremental file. + */ + full_copy_possible = false; + } + } + } + else + { + BlockNumber b; + + /* + * Since we found a full file, source all remaining required + * blocks from it. + */ + for (b = 0; b < latest_source->truncation_block_length; ++b) + { + if (sourcemap[b] == NULL) + { + sourcemap[b] = s; + offsetmap[b] = b * BLCKSZ; + + Assert(num_missing_blocks > 0); + --num_missing_blocks; + } + } + Assert(num_missing_blocks == 0); + + /* + * If a full copy looks possible, check whether the resulting file + * should be exactly as long as the source file is. If so, a full + * copy is acceptable, otherwise not. + */ + if (full_copy_possible) + { + struct stat sb; + uint64 expected_length; + + if (fstat(s->fd, &sb) < 0) + pg_fatal("could not stat \"%s\": %m", s->filename); + expected_length = + (uint64) latest_source->truncation_block_length; + expected_length *= BLCKSZ; + if (expected_length == sb.st_size) + { + copy_source = s; + copy_source_index = sidx; + } + } + } + } + + /* + * If a checksum of the required type already exists in the + * backup_manifest for the relevant input directory, we can save some work + * by reusing that checksum instead of computing a new one. + */ + if (copy_source_index >= 0 && manifests[copy_source_index] != NULL && + checksum_type != CHECKSUM_TYPE_NONE) + { + manifest_file *mfile; + + mfile = manifest_files_lookup(manifests[copy_source_index]->files, + manifest_path); + if (mfile == NULL) + { + /* + * The directory is out of sync with the backup_manifest, so emit + * a warning. + */ + pg_log_warning("\"%s/backup_manifest\" contains no entry for \"%s\"", + prior_backup_dirs[copy_source_index], + manifest_path); + } + else if (mfile->checksum_type == checksum_type) + { + *checksum_length = mfile->checksum_length; + *checksum_payload = pg_malloc(*checksum_length); + memcpy(*checksum_payload, mfile->checksum_payload, + *checksum_length); + checksum_type = CHECKSUM_TYPE_NONE; + } + } + + /* Prepare for checksum calculation, if required. */ + pg_checksum_init(&checksum_ctx, checksum_type); + + /* + * If the full file can be created by copying a file from an older backup + * in the chain without needing to overwrite any blocks or truncate the + * result, then forget about performing reconstruction and just copy that + * file in its entirety. + * + * Otherwise, reconstruct. + */ + if (copy_source != NULL) + copy_file(copy_source->filename, output_filename, + &checksum_ctx, dry_run); + else + { + write_reconstructed_file(input_filename, output_filename, + block_length, sourcemap, offsetmap, + &checksum_ctx, dry_run); + debug_reconstruction(n_prior_backups + 1, source, dry_run); + } + + /* Save results of checksum calculation. */ + if (checksum_type != CHECKSUM_TYPE_NONE) + { + *checksum_payload = pg_malloc(PG_CHECKSUM_MAX_LENGTH); + *checksum_length = pg_checksum_final(&checksum_ctx, + *checksum_payload); + } + + /* + * Close files and release memory. + */ + for (i = 0; i <= n_prior_backups; ++i) + { + rfile *s = source[i]; + + if (s == NULL) + continue; + if (close(s->fd) != 0) + pg_fatal("could not close \"%s\": %m", s->filename); + if (s->relative_block_numbers != NULL) + pfree(s->relative_block_numbers); + pg_free(s->filename); + } + pfree(sourcemap); + pfree(offsetmap); + pfree(source); +} + +/* + * Perform post-reconstruction logging and sanity checks. + */ +static void +debug_reconstruction(int n_source, rfile **sources, bool dry_run) +{ + unsigned i; + + for (i = 0; i < n_source; ++i) + { + rfile *s = sources[i]; + + /* Ignore source if not used. */ + if (s == NULL) + continue; + + /* If no data is needed from this file, we can ignore it. */ + if (s->num_blocks_read == 0) + continue; + + /* Debug logging. */ + if (dry_run) + pg_log_debug("would have read %u blocks from \"%s\"", + s->num_blocks_read, s->filename); + else + pg_log_debug("read %u blocks from \"%s\"", + s->num_blocks_read, s->filename); + + /* + * In dry-run mode, we don't actually try to read data from the file, + * but we do try to verify that the file is long enough that we could + * have read the data if we'd tried. + * + * If this fails, then it means that a non-dry-run attempt would fail, + * complaining of not being able to read the required bytes from the + * file. + */ + if (dry_run) + { + struct stat sb; + + if (fstat(s->fd, &sb) < 0) + pg_fatal("could not stat \"%s\": %m", s->filename); + if (sb.st_size < s->highest_offset_read) + pg_fatal("file \"%s\" is too short: expected %llu, found %llu", + s->filename, + (unsigned long long) s->highest_offset_read, + (unsigned long long) sb.st_size); + } + } +} + +/* + * When we perform reconstruction using an incremental file, the output file + * should be at least as long as the truncation_block_length. Any blocks + * present in the incremental file increase the output length as far as is + * necessary to include those blocks. + */ +static unsigned +find_reconstructed_block_length(rfile *s) +{ + unsigned block_length = s->truncation_block_length; + unsigned i; + + for (i = 0; i < s->num_blocks; ++i) + if (s->relative_block_numbers[i] >= block_length) + block_length = s->relative_block_numbers[i] + 1; + + return block_length; +} + +/* + * Initialize an incremental rfile, reading the header so that we know which + * blocks it contains. + */ +static rfile * +make_incremental_rfile(char *filename) +{ + rfile *rf; + unsigned magic; + + rf = make_rfile(filename, false); + + /* Read and validate magic number. */ + read_bytes(rf, &magic, sizeof(magic)); + if (magic != INCREMENTAL_MAGIC) + pg_fatal("file \"%s\" has bad incremental magic number (0x%x not 0x%x)", + filename, magic, INCREMENTAL_MAGIC); + + /* Read block count. */ + read_bytes(rf, &rf->num_blocks, sizeof(rf->num_blocks)); + if (rf->num_blocks > RELSEG_SIZE) + pg_fatal("file \"%s\" has block count %u in excess of segment size %u", + filename, rf->num_blocks, RELSEG_SIZE); + + /* Read truncation block length. */ + read_bytes(rf, &rf->truncation_block_length, + sizeof(rf->truncation_block_length)); + if (rf->truncation_block_length > RELSEG_SIZE) + pg_fatal("file \"%s\" has truncation block length %u in excess of segment size %u", + filename, rf->truncation_block_length, RELSEG_SIZE); + + /* Read block numbers if there are any. */ + if (rf->num_blocks > 0) + { + rf->relative_block_numbers = + pg_malloc0(sizeof(BlockNumber) * rf->num_blocks); + read_bytes(rf, rf->relative_block_numbers, + sizeof(BlockNumber) * rf->num_blocks); + } + + /* Remember length of header. */ + rf->header_length = sizeof(magic) + sizeof(rf->num_blocks) + + sizeof(rf->truncation_block_length) + + sizeof(BlockNumber) * rf->num_blocks; + + return rf; +} + +/* + * Allocate and perform basic initialization of an rfile. + */ +static rfile * +make_rfile(char *filename, bool missing_ok) +{ + rfile *rf; + + rf = pg_malloc0(sizeof(rfile)); + rf->filename = pstrdup(filename); + if ((rf->fd = open(filename, O_RDONLY | PG_BINARY, 0)) < 0) + { + if (missing_ok && errno == ENOENT) + { + pg_free(rf); + return NULL; + } + pg_fatal("could not open file \"%s\": %m", filename); + } + + return rf; +} + +/* + * Read the indicated number of bytes from an rfile into the buffer. + */ +static void +read_bytes(rfile *rf, void *buffer, unsigned length) +{ + unsigned rb = read(rf->fd, buffer, length); + + if (rb != length) + { + if (rb < 0) + pg_fatal("could not read file \"%s\": %m", rf->filename); + else + pg_fatal("could not read file \"%s\": read only %d of %d bytes", + rf->filename, (int) rb, length); + } +} + +/* + * Write out a reconstructed file. + */ +static void +write_reconstructed_file(char *input_filename, + char *output_filename, + unsigned block_length, + rfile **sourcemap, + off_t *offsetmap, + pg_checksum_context *checksum_ctx, + bool dry_run) +{ + int wfd = -1; + unsigned i; + unsigned zero_blocks = 0; + + /* Debugging output. */ + if (dry_run) + pg_log_debug("would reconstruct \"%s\" (%u blocks, checksum %s)", + output_filename, block_length, + pg_checksum_type_name(checksum_ctx->type)); + else + pg_log_debug("reconstructing \"%s\" (%u blocks, checksum %s)", + output_filename, block_length, + pg_checksum_type_name(checksum_ctx->type)); + + /* Open the output file, except in dry_run mode. */ + if (!dry_run && + (wfd = open(output_filename, + O_RDWR | PG_BINARY | O_CREAT | O_EXCL, + pg_file_create_mode)) < 0) + pg_fatal("could not open file \"%s\": %m", output_filename); + + /* Read and write the blocks as required. */ + for (i = 0; i < block_length; ++i) + { + uint8 buffer[BLCKSZ]; + rfile *s = sourcemap[i]; + unsigned wb; + + /* Update accounting information. */ + if (s == NULL) + ++zero_blocks; + else + { + s->num_blocks_read++; + s->highest_offset_read = Max(s->highest_offset_read, + offsetmap[i] + BLCKSZ); + } + + /* Skip the rest of this in dry-run mode. */ + if (dry_run) + continue; + + /* Read or zero-fill the block as appropriate. */ + if (s == NULL) + { + /* + * New block not mentioned in the WAL summary. Should have been an + * uninitialized block, so just zero-fill it. + */ + memset(buffer, 0, BLCKSZ); + } + else + { + unsigned rb; + + /* Read the block from the correct source, except if dry-run. */ + rb = pg_pread(s->fd, buffer, BLCKSZ, offsetmap[i]); + if (rb != BLCKSZ) + { + if (rb < 0) + pg_fatal("could not read file \"%s\": %m", s->filename); + else + pg_fatal("could not read file \"%s\": read only %d of %d bytes at offset %u", + s->filename, (int) rb, BLCKSZ, + (unsigned) offsetmap[i]); + } + } + + /* Write out the block. */ + if ((wb = write(wfd, buffer, BLCKSZ)) != BLCKSZ) + { + if (wb < 0) + pg_fatal("could not write file \"%s\": %m", output_filename); + else + pg_fatal("could not write file \"%s\": wrote only %d of %d bytes", + output_filename, (int) wb, BLCKSZ); + } + + /* Update the checksum computation. */ + if (pg_checksum_update(checksum_ctx, buffer, BLCKSZ) < 0) + pg_fatal("could not update checksum of file \"%s\"", + output_filename); + } + + /* Debugging output. */ + if (zero_blocks > 0) + { + if (dry_run) + pg_log_debug("would have zero-filled %u blocks", zero_blocks); + else + pg_log_debug("zero-filled %u blocks", zero_blocks); + } + + /* Close the output file. */ + if (wfd >= 0 && close(wfd) != 0) + pg_fatal("could not close \"%s\": %m", output_filename); +} diff --git a/src/bin/pg_combinebackup/reconstruct.h b/src/bin/pg_combinebackup/reconstruct.h new file mode 100644 index 0000000000..c599a70d42 --- /dev/null +++ b/src/bin/pg_combinebackup/reconstruct.h @@ -0,0 +1,32 @@ +/*------------------------------------------------------------------------- + * + * reconstruct.h + * Reconstruct full file from incremental file and backup chain. + * + * Copyright (c) 2017-2023, PostgreSQL Global Development Group + * + * IDENTIFICATION + * src/bin/pg_combinebackup/reconstruct.h + * + *------------------------------------------------------------------------- + */ +#ifndef RECONSTRUCT_H +#define RECONSTRUCT_H + +#include "common/checksum_helper.h" +#include "load_manifest.h" + +extern void reconstruct_from_incremental_file(char *input_filename, + char *output_filename, + char *relative_path, + char *bare_file_name, + int n_prior_backups, + char **prior_backup_dirs, + manifest_data **manifests, + char *manifest_path, + pg_checksum_type checksum_type, + int *checksum_length, + uint8 **checksum_payload, + bool dry_run); + +#endif diff --git a/src/bin/pg_combinebackup/t/001_basic.pl b/src/bin/pg_combinebackup/t/001_basic.pl new file mode 100644 index 0000000000..fb66075d1a --- /dev/null +++ b/src/bin/pg_combinebackup/t/001_basic.pl @@ -0,0 +1,23 @@ +# Copyright (c) 2021-2023, PostgreSQL Global Development Group + +use strict; +use warnings; +use PostgreSQL::Test::Utils; +use Test::More; + +my $tempdir = PostgreSQL::Test::Utils::tempdir; + +program_help_ok('pg_combinebackup'); +program_version_ok('pg_combinebackup'); +program_options_handling_ok('pg_combinebackup'); + +command_fails_like( + ['pg_combinebackup'], + qr/no input directories specified/, + 'input directories must be specified'); +command_fails_like( + [ 'pg_combinebackup', $tempdir ], + qr/no output directory specified/, + 'output directory must be specified'); + +done_testing(); diff --git a/src/bin/pg_combinebackup/t/002_compare_backups.pl b/src/bin/pg_combinebackup/t/002_compare_backups.pl new file mode 100644 index 0000000000..3d9238f366 --- /dev/null +++ b/src/bin/pg_combinebackup/t/002_compare_backups.pl @@ -0,0 +1,154 @@ +# Copyright (c) 2021-2023, PostgreSQL Global Development Group + +use strict; +use warnings; +use File::Compare; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Set up a new database instance. +my $primary = PostgreSQL::Test::Cluster->new('primary'); +$primary->init(has_archiving => 1, allows_streaming => 1); +$primary->append_conf('postgresql.conf', 'autovacuum = off'); +$primary->start; + +# Create some test tables, each containing one row of data, plus a whole +# extra database. +$primary->safe_psql('postgres', <<EOM); +CREATE TABLE will_change (a int, b text); +INSERT INTO will_change VALUES (1, 'initial test row'); +CREATE TABLE will_grow (a int, b text); +INSERT INTO will_grow VALUES (1, 'initial test row'); +CREATE TABLE will_shrink (a int, b text); +INSERT INTO will_shrink VALUES (1, 'initial test row'); +CREATE TABLE will_get_vacuumed (a int, b text); +INSERT INTO will_get_vacuumed VALUES (1, 'initial test row'); +CREATE TABLE will_get_dropped (a int, b text); +INSERT INTO will_get_dropped VALUES (1, 'initial test row'); +CREATE TABLE will_get_rewritten (a int, b text); +INSERT INTO will_get_rewritten VALUES (1, 'initial test row'); +CREATE DATABASE db_will_get_dropped; +EOM + +# Take a full backup. +my $backup1path = $primary->backup_dir . '/backup1'; +$primary->command_ok( + [ 'pg_basebackup', '-D', $backup1path, '--no-sync', '-cfast' ], + "full backup"); + +# Now make some database changes. +$primary->safe_psql('postgres', <<EOM); +UPDATE will_change SET b = 'modified value' WHERE a = 1; +INSERT INTO will_grow + SELECT g, 'additional row' FROM generate_series(2, 5000) g; +TRUNCATE will_shrink; +VACUUM will_get_vacuumed; +DROP TABLE will_get_dropped; +CREATE TABLE newly_created (a int, b text); +INSERT INTO newly_created VALUES (1, 'row for new table'); +VACUUM FULL will_get_rewritten; +DROP DATABASE db_will_get_dropped; +CREATE DATABASE db_newly_created; +EOM + +# Take an incremental backup. +my $backup2path = $primary->backup_dir . '/backup2'; +$primary->command_ok( + [ 'pg_basebackup', '-D', $backup2path, '--no-sync', '-cfast', + '--incremental', $backup1path . '/backup_manifest' ], + "incremental backup"); + +# Find an LSN to which either backup can be recovered. +my $lsn = $primary->safe_psql('postgres', "SELECT pg_current_wal_lsn();"); + +# Make sure that the WAL segment containing that LSN has been archived. +# PostgreSQL won't issue two consecutive XLOG_SWITCH records, and the backup +# just issued one, so call txid_current() to generate some WAL activity +# before calling pg_switch_wal(). +$primary->safe_psql('postgres', 'SELECT txid_current();'); +$primary->safe_psql('postgres', 'SELECT pg_switch_wal()'); + +# Now wait for the LSN we chose above to be archived. +my $archive_wait_query = + "SELECT pg_walfile_name('$lsn') <= last_archived_wal FROM pg_stat_archiver;"; +$primary->poll_query_until('postgres', $archive_wait_query) + or die "Timed out while waiting for WAL segment to be archived"; + +# Perform PITR from the full backup. Disable archive_mode so that the archive +# doesn't find out about the new timeline; that way, the later PITR below will +# choose the same timeline. +my $pitr1 = PostgreSQL::Test::Cluster->new('pitr1'); +$pitr1->init_from_backup($primary, 'backup1', + standby => 1, has_restoring => 1); +$pitr1->append_conf('postgresql.conf', qq{ +recovery_target_lsn = '$lsn' +recovery_target_action = 'promote' +archive_mode = 'off' +}); +$pitr1->start(); + +# Perform PITR to the same LSN from the incremental backup. Use the same +# basic configuration as before. +my $pitr2 = PostgreSQL::Test::Cluster->new('pitr2'); +$pitr2->init_from_backup($primary, 'backup2', + standby => 1, has_restoring => 1, + combine_with_prior => [ 'backup1' ]); +$pitr2->append_conf('postgresql.conf', qq{ +recovery_target_lsn = '$lsn' +recovery_target_action = 'promote' +archive_mode = 'off' +}); +$pitr2->start(); + +# Wait until both servers exit recovery. +$pitr1->poll_query_until('postgres', + "SELECT NOT pg_is_in_recovery();") + or die "Timed out while waiting apply to reach LSN $lsn"; +$pitr2->poll_query_until('postgres', + "SELECT NOT pg_is_in_recovery();") + or die "Timed out while waiting apply to reach LSN $lsn"; + +# Perform a logical dump of each server, and check that they match. +# It would be much nicer if we could physically compare the data files, but +# that doesn't really work. The contents of the page hole aren't guaranteed to +# be identical, and there can be other discrepancies as well. To make this work +# we'd need the equivalent of each AM's rm_mask functon written or at least +# callable from Perl, and that doesn't seem practical. +# +# NB: We're just using the primary's backup directory for scratch space here. +# This could equally well be any other directory we wanted to pick. +my $backupdir = $primary->backup_dir; +my $dump1 = $backupdir . '/pitr1.dump'; +my $dump2 = $backupdir . '/pitr2.dump'; +$pitr1->command_ok([ + 'pg_dumpall', '-f', $dump1, '--no-sync', '--no-unlogged-table-data', + '-d', $pitr1->connstr('postgres'), + ], + 'dump from PITR 1'); +$pitr1->command_ok([ + 'pg_dumpall', '-f', $dump2, '--no-sync', '--no-unlogged-table-data', + '-d', $pitr1->connstr('postgres'), + ], + 'dump from PITR 2'); + +# Compare the two dumps, there should be no differences. +my $compare_res = compare($dump1, $dump2); +note($dump1); +note($dump2); +is($compare_res, 0, "dumps are identical"); + +# Provide more context if the dumps do not match. +if ($compare_res != 0) +{ + my ($stdout, $stderr) = + run_command([ 'diff', '-u', $dump1, $dump2 ]); + print "=== diff of $dump1 and $dump2\n"; + print "=== stdout ===\n"; + print $stdout; + print "=== stderr ===\n"; + print $stderr; + print "=== EOF ===\n"; +} + +done_testing(); diff --git a/src/bin/pg_combinebackup/write_manifest.c b/src/bin/pg_combinebackup/write_manifest.c new file mode 100644 index 0000000000..82160134d8 --- /dev/null +++ b/src/bin/pg_combinebackup/write_manifest.c @@ -0,0 +1,293 @@ +/*------------------------------------------------------------------------- + * + * Write a new backup manifest. + * + * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group + * Portions Copyright (c) 1994, Regents of the University of California + * + * src/bin/pg_combinebackup/write_manifest.c + * + *------------------------------------------------------------------------- + */ + +#include "postgres_fe.h" + +#include <fcntl.h> +#include <time.h> +#include <unistd.h> + +#include "common/checksum_helper.h" +#include "common/file_perm.h" +#include "common/logging.h" +#include "lib/stringinfo.h" +#include "load_manifest.h" +#include "mb/pg_wchar.h" +#include "write_manifest.h" + +struct manifest_writer +{ + char pathname[MAXPGPATH]; + int fd; + StringInfoData buf; + bool first_file; + bool still_checksumming; + pg_checksum_context manifest_ctx; +}; + +static void escape_json(StringInfo buf, const char *str); +static void flush_manifest(manifest_writer *mwriter); +static size_t hex_encode(const uint8 *src, size_t len, char *dst); + +/* + * Create a new backup manifest writer. + * + * The backup manifest will be written into a file named backup_manifest + * in the specified directory. + */ +manifest_writer * +create_manifest_writer(char *directory) +{ + manifest_writer *mwriter = pg_malloc(sizeof(manifest_writer)); + + snprintf(mwriter->pathname, MAXPGPATH, "%s/backup_manifest", directory); + mwriter->fd = -1; + initStringInfo(&mwriter->buf); + mwriter->first_file = true; + mwriter->still_checksumming = true; + pg_checksum_init(&mwriter->manifest_ctx, CHECKSUM_TYPE_SHA256); + + appendStringInfo(&mwriter->buf, + "{ \"PostgreSQL-Backup-Manifest-Version\": 1,\n" + "\"Files\": ["); + + return mwriter; +} + +/* + * Add an entry for a file to a backup manifest. + * + * This is very similar to the backend's AddFileToBackupManifest, but + * various adjustments are required due to frontend/backend differences + * and other details. + */ +void +add_file_to_manifest(manifest_writer *mwriter, const char *manifest_path, + size_t size, pg_time_t mtime, + pg_checksum_type checksum_type, + int checksum_length, + uint8 *checksum_payload) +{ + int pathlen = strlen(manifest_path); + + if (mwriter->first_file) + { + appendStringInfoChar(&mwriter->buf, '\n'); + mwriter->first_file = false; + } + else + appendStringInfoString(&mwriter->buf, ",\n"); + + if (pg_encoding_verifymbstr(PG_UTF8, manifest_path, pathlen) == pathlen) + { + appendStringInfoString(&mwriter->buf, "{ \"Path\": "); + escape_json(&mwriter->buf, manifest_path); + appendStringInfoString(&mwriter->buf, ", "); + } + else + { + appendStringInfoString(&mwriter->buf, "{ \"Encoded-Path\": \""); + enlargeStringInfo(&mwriter->buf, 2 * pathlen); + mwriter->buf.len += hex_encode((const uint8 *) manifest_path, pathlen, + &mwriter->buf.data[mwriter->buf.len]); + appendStringInfoString(&mwriter->buf, "\", "); + } + + appendStringInfo(&mwriter->buf, "\"Size\": %zu, ", size); + + appendStringInfoString(&mwriter->buf, "\"Last-Modified\": \""); + enlargeStringInfo(&mwriter->buf, 128); + mwriter->buf.len += strftime(&mwriter->buf.data[mwriter->buf.len], 128, + "%Y-%m-%d %H:%M:%S %Z", + gmtime(&mtime)); + appendStringInfoChar(&mwriter->buf, '"'); + + if (mwriter->buf.len > 128 * 1024) + flush_manifest(mwriter); + + if (checksum_length > 0) + { + appendStringInfo(&mwriter->buf, + ", \"Checksum-Algorithm\": \"%s\", \"Checksum\": \"", + pg_checksum_type_name(checksum_type)); + + enlargeStringInfo(&mwriter->buf, 2 * checksum_length); + mwriter->buf.len += hex_encode(checksum_payload, checksum_length, + &mwriter->buf.data[mwriter->buf.len]); + + appendStringInfoChar(&mwriter->buf, '"'); + } + + appendStringInfoString(&mwriter->buf, " }"); + + if (mwriter->buf.len > 128 * 1024) + flush_manifest(mwriter); +} + +/* + * Finalize the backup_manifest. + */ +void +finalize_manifest(manifest_writer *mwriter, + manifest_wal_range *first_wal_range) +{ + uint8 checksumbuf[PG_SHA256_DIGEST_LENGTH]; + int len; + manifest_wal_range *wal_range; + + /* Terminate the list of files. */ + appendStringInfoString(&mwriter->buf, "\n],\n"); + + /* Start a list of LSN ranges. */ + appendStringInfoString(&mwriter->buf, "\"WAL-Ranges\": [\n"); + + for (wal_range = first_wal_range; wal_range != NULL; + wal_range = wal_range->next) + appendStringInfo(&mwriter->buf, + "%s{ \"Timeline\": %u, \"Start-LSN\": \"%X/%X\", \"End-LSN\": \"%X/%X\" }", + wal_range == first_wal_range ? "" : ",\n", + wal_range->tli, + LSN_FORMAT_ARGS(wal_range->start_lsn), + LSN_FORMAT_ARGS(wal_range->end_lsn)); + + /* Terminate the list of WAL ranges. */ + appendStringInfoString(&mwriter->buf, "\n],\n"); + + /* Flush accumulated data and update checksum calculation. */ + flush_manifest(mwriter); + + /* Checksum only includes data up to this point. */ + mwriter->still_checksumming = false; + + /* Compute and insert manifest checksum. */ + appendStringInfoString(&mwriter->buf, "\"Manifest-Checksum\": \""); + enlargeStringInfo(&mwriter->buf, 2 * PG_SHA256_DIGEST_STRING_LENGTH); + len = pg_checksum_final(&mwriter->manifest_ctx, checksumbuf); + Assert(len == PG_SHA256_DIGEST_LENGTH); + mwriter->buf.len += + hex_encode(checksumbuf, len, &mwriter->buf.data[mwriter->buf.len]); + appendStringInfoString(&mwriter->buf, "\"}\n"); + + /* Flush the last manifest checksum itself. */ + flush_manifest(mwriter); + + /* Close the file. */ + if (close(mwriter->fd) != 0) + pg_fatal("could not close \"%s\": %m", mwriter->pathname); + mwriter->fd = -1; +} + +/* + * Produce a JSON string literal, properly escaping characters in the text. + */ +static void +escape_json(StringInfo buf, const char *str) +{ + const char *p; + + appendStringInfoCharMacro(buf, '"'); + for (p = str; *p; p++) + { + switch (*p) + { + case '\b': + appendStringInfoString(buf, "\\b"); + break; + case '\f': + appendStringInfoString(buf, "\\f"); + break; + case '\n': + appendStringInfoString(buf, "\\n"); + break; + case '\r': + appendStringInfoString(buf, "\\r"); + break; + case '\t': + appendStringInfoString(buf, "\\t"); + break; + case '"': + appendStringInfoString(buf, "\\\""); + break; + case '\\': + appendStringInfoString(buf, "\\\\"); + break; + default: + if ((unsigned char) *p < ' ') + appendStringInfo(buf, "\\u%04x", (int) *p); + else + appendStringInfoCharMacro(buf, *p); + break; + } + } + appendStringInfoCharMacro(buf, '"'); +} + +/* + * Flush whatever portion of the backup manifest we have generated and + * buffered in memory out to a file on disk. + * + * The first call to this function will create the file. After that, we + * keep it open and just append more data. + */ +static void +flush_manifest(manifest_writer *mwriter) +{ + char pathname[MAXPGPATH]; + + if (mwriter->fd == -1 && + (mwriter->fd = open(mwriter->pathname, + O_WRONLY | O_CREAT | O_EXCL | PG_BINARY, + pg_file_create_mode)) < 0) + pg_fatal("could not open file \"%s\": %m", mwriter->pathname); + + if (mwriter->buf.len > 0) + { + ssize_t wb; + + wb = write(mwriter->fd, mwriter->buf.data, mwriter->buf.len); + if (wb != mwriter->buf.len) + { + if (wb < 0) + pg_fatal("could not write \"%s\": %m", mwriter->pathname); + else + pg_fatal("could not write file \"%s\": wrote only %d of %d bytes", + pathname, (int) wb, mwriter->buf.len); + } + + if (mwriter->still_checksumming) + pg_checksum_update(&mwriter->manifest_ctx, + (uint8 *) mwriter->buf.data, + mwriter->buf.len); + resetStringInfo(&mwriter->buf); + } +} + +/* + * Encode bytes using two hexademical digits for each one. + */ +static size_t +hex_encode(const uint8 *src, size_t len, char *dst) +{ + const uint8 *end = src + len; + + while (src < end) + { + unsigned n1 = (*src >> 4) & 0xF; + unsigned n2 = *src & 0xF; + + *dst++ = n1 < 10 ? '0' + n1 : 'a' + n1 - 10; + *dst++ = n2 < 10 ? '0' + n2 : 'a' + n2 - 10; + ++src; + } + + return len * 2; +} diff --git a/src/bin/pg_combinebackup/write_manifest.h b/src/bin/pg_combinebackup/write_manifest.h new file mode 100644 index 0000000000..8fd7fe02c8 --- /dev/null +++ b/src/bin/pg_combinebackup/write_manifest.h @@ -0,0 +1,33 @@ +/*------------------------------------------------------------------------- + * + * Write a new backup manifest. + * + * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group + * Portions Copyright (c) 1994, Regents of the University of California + * + * src/bin/pg_combinebackup/write_manifest.h + * + *------------------------------------------------------------------------- + */ +#ifndef WRITE_MANIFEST_H +#define WRITE_MANIFEST_H + +#include "common/checksum_helper.h" +#include "pgtime.h" + +struct manifest_wal_range; + +struct manifest_writer; +typedef struct manifest_writer manifest_writer; + +extern manifest_writer *create_manifest_writer(char *directory); +extern void add_file_to_manifest(manifest_writer *mwriter, + const char *manifest_path, + size_t size, pg_time_t mtime, + pg_checksum_type checksum_type, + int checksum_length, + uint8 *checksum_payload); +extern void finalize_manifest(manifest_writer *mwriter, + struct manifest_wal_range *first_wal_range); + +#endif /* WRITE_MANIFEST_H */ diff --git a/src/bin/pg_resetwal/pg_resetwal.c b/src/bin/pg_resetwal/pg_resetwal.c index 3ae3fc06df..5407f51a4e 100644 --- a/src/bin/pg_resetwal/pg_resetwal.c +++ b/src/bin/pg_resetwal/pg_resetwal.c @@ -85,6 +85,7 @@ static void RewriteControlFile(void); static void FindEndOfXLOG(void); static void KillExistingXLOG(void); static void KillExistingArchiveStatus(void); +static void KillExistingWALSummaries(void); static void WriteEmptyXLOG(void); static void usage(void); @@ -493,6 +494,7 @@ main(int argc, char *argv[]) RewriteControlFile(); KillExistingXLOG(); KillExistingArchiveStatus(); + KillExistingWALSummaries(); WriteEmptyXLOG(); printf(_("Write-ahead log reset\n")); @@ -1034,6 +1036,40 @@ KillExistingArchiveStatus(void) pg_fatal("could not close directory \"%s\": %m", ARCHSTATDIR); } +/* + * Remove existing WAL summary files + */ +static void +KillExistingWALSummaries(void) +{ +#define WALSUMMARYDIR XLOGDIR "/summaries" +#define WALSUMMARY_NHEXCHARS 40 + + DIR *xldir; + struct dirent *xlde; + char path[MAXPGPATH + sizeof(WALSUMMARYDIR)]; + + xldir = opendir(WALSUMMARYDIR); + if (xldir == NULL) + pg_fatal("could not open directory \"%s\": %m", WALSUMMARYDIR); + + while (errno = 0, (xlde = readdir(xldir)) != NULL) + { + if (strspn(xlde->d_name, "0123456789ABCDEF") == WALSUMMARY_NHEXCHARS && + strcmp(xlde->d_name + WALSUMMARY_NHEXCHARS, ".summary") == 0) + { + snprintf(path, sizeof(path), "%s/%s", WALSUMMARYDIR, xlde->d_name); + if (unlink(path) < 0) + pg_fatal("could not delete file \"%s\": %m", path); + } + } + + if (errno) + pg_fatal("could not read directory \"%s\": %m", WALSUMMARYDIR); + + if (closedir(xldir)) + pg_fatal("could not close directory \"%s\": %m", ARCHSTATDIR); +} /* * Write an empty XLOG file, containing only the checkpoint record diff --git a/src/common/Makefile b/src/common/Makefile index 3c8effc533..2b41dd1839 100644 --- a/src/common/Makefile +++ b/src/common/Makefile @@ -49,6 +49,7 @@ OBJS_COMMON = \ archive.o \ base64.o \ binaryheap.o \ + blkreftable.o \ checksum_helper.o \ compression.o \ config_info.o \ diff --git a/src/common/blkreftable.c b/src/common/blkreftable.c new file mode 100644 index 0000000000..012a443584 --- /dev/null +++ b/src/common/blkreftable.c @@ -0,0 +1,1309 @@ +/*------------------------------------------------------------------------- + * + * blkreftable.c + * Block reference tables. + * + * A block reference table is used to keep track of which blocks have + * been modified by WAL records within a certain LSN range. + * + * For each relation fork, we keep track of all blocks that have appeared + * in block reference in the WAL. We also keep track of the "limit block", + * which is the smallest relation length in blocks known to have occurred + * during that range of WAL records. This should be set to 0 if the relation + * fork is created or destroyed, and to the post-truncation length if + * truncated. + * + * Whenever we set the limit block, we also forget about any modified blocks + * beyond that point. Those blocks don't exist any more. Such blocks can + * later be marked as modified again; if that happens, it means the relation + * was re-extended. + * + * Portions Copyright (c) 2010-2022, PostgreSQL Global Development Group + * + * src/common/blkreftable.c + * + *------------------------------------------------------------------------- + */ + + +#ifndef FRONTEND +#include "postgres.h" +#else +#include "postgres_fe.h" +#endif + +#ifdef FRONTEND +#include "common/logging.h" +#endif + +#include "common/blkreftable.h" +#include "common/hashfn.h" +#include "port/pg_crc32c.h" + +/* + * A block reference table keeps track of the status of each relation + * fork individually. + */ +typedef struct BlockRefTableKey +{ + RelFileLocator rlocator; + ForkNumber forknum; +} BlockRefTableKey; + +/* + * We could need to store data either for a relation in which only a + * tiny fraction of the blocks have been modified or for a relation in + * which nearly every block has been modified, and we want a + * space-efficient representation in both cases. To accomplish this, + * we divide the relation into chunks of 2^16 blocks and choose between + * an array representation and a bitmap representation for each chunk. + * + * When the number of modified blocks in a given chunk is small, we + * essentially store an array of block numbers, but we need not store the + * entire block number: instead, we store each block number as a 2-byte + * offset from the start of the chunk. + * + * When the number of modified blocks in a given chunk is large, we switch + * to a bitmap representation. + * + * These same basic representational choices are used both when a block + * reference table is stored in memory and when it is serialized to disk. + * + * In the in-memory representation, we initially allocate each chunk with + * space for a number of entries given by INITIAL_ENTRIES_PER_CHUNK and + * increase that as necessary until we reach MAX_ENTRIES_PER_CHUNK. + * Any chunk whose allocated size reaches MAX_ENTRIES_PER_CHUNK is converted + * to a bitmap, and thus never needs to grow further. + */ +#define BLOCKS_PER_CHUNK (1 << 16) +#define BLOCKS_PER_ENTRY (BITS_PER_BYTE * sizeof(uint16)) +#define MAX_ENTRIES_PER_CHUNK (BLOCKS_PER_CHUNK / BLOCKS_PER_ENTRY) +#define INITIAL_ENTRIES_PER_CHUNK 16 +typedef uint16 *BlockRefTableChunk; + +/* + * State for one relation fork. + * + * 'rlocator' and 'forknum' identify the relation fork to which this entry + * pertains. + * + * 'limit_block' is the shortest known length of the relation in blocks + * within the LSN range covered by a particular block reference table. + * It should be set to 0 if the relation fork is created or dropped. If the + * relation fork is truncated, it should be set to the number of blocks that + * remain after truncation. + * + * 'nchunks' is the allocated length of each of the three arrays that follow. + * We can only represent the status of block numbers less than nchunks * + * BLOCKS_PER_CHUNK. + * + * 'chunk_size' is an array storing the allocated size of each chunk. + * + * 'chunk_usage' is an array storing the number of elements used in each + * chunk. If that value is less than MAX_ENTRIES_PER_CHUNK, the corresonding + * chunk is used as an array; else the corresponding chunk is used as a bitmap. + * When used as a bitmap, the least significant bit of the first array element + * is the status of the lowest-numbered block covered by this chunk. + * + * 'chunk_data' is the array of chunks. + */ +struct BlockRefTableEntry +{ + BlockRefTableKey key; + BlockNumber limit_block; + char status; + uint32 nchunks; + uint16 *chunk_size; + uint16 *chunk_usage; + BlockRefTableChunk *chunk_data; +}; + +/* Declare and define a hash table over type BlockRefTableEntry. */ +#define SH_PREFIX blockreftable +#define SH_ELEMENT_TYPE BlockRefTableEntry +#define SH_KEY_TYPE BlockRefTableKey +#define SH_KEY key +#define SH_HASH_KEY(tb, key) \ + hash_bytes((const unsigned char *) &key, sizeof(BlockRefTableKey)) +#define SH_EQUAL(tb, a, b) memcmp(&a, &b, sizeof(BlockRefTableKey)) == 0 +#define SH_SCOPE static inline +#ifdef FRONTEND +#define SH_RAW_ALLOCATOR pg_malloc0 +#endif +#define SH_DEFINE +#define SH_DECLARE +#include "lib/simplehash.h" + +/* + * A block reference table is basically just the hash table, but we don't + * want to expose that to outside callers. + * + * We keep track of the memory context in use explicitly too, so that it's + * easy to place all of our allocations in the same context. + */ +struct BlockRefTable +{ + blockreftable_hash *hash; +#ifndef FRONTEND + MemoryContext mcxt; +#endif +}; + +/* + * On-disk serialization format for block reference table entries. + */ +typedef struct BlockRefTableSerializedEntry +{ + RelFileLocator rlocator; + ForkNumber forknum; + BlockNumber limit_block; + uint32 nchunks; +} BlockRefTableSerializedEntry; + +/* + * Buffer size, so that we avoid doing many small I/Os. + */ +#define BUFSIZE 65536 + +/* + * Ad-hoc buffer for file I/O. + */ +typedef struct BlockRefTableBuffer +{ + io_callback_fn io_callback; + void *io_callback_arg; + char data[BUFSIZE]; + int used; + int cursor; + pg_crc32c crc; +} BlockRefTableBuffer; + +/* + * State for keeping track of progress while incrementally reading a block + * table reference file from disk. + * + * total_chunks means the number of chunks for the RelFileLocator/ForkNumber + * combination that is curently being read, and consumed_chunks is the number + * of those that have been read. (We always read all the information for + * a single chunk at one time, so we don't need to be able to represent the + * state where a chunk has been partially read.) + * + * chunk_size is the array of chunk sizes. The length is given by total_chunks. + * + * chunk_data holds the current chunk. + * + * chunk_position helps us figure out how much progress we've made in returning + * the block numbers for the current chunk to the caller. If the chunk is a + * bitmap, it's the number of bits we've scanned; otherwise, it's the number + * of chunk entries we've scanned. + */ +struct BlockRefTableReader +{ + BlockRefTableBuffer buffer; + char *error_filename; + report_error_fn error_callback; + void *error_callback_arg; + uint32 total_chunks; + uint32 consumed_chunks; + uint16 *chunk_size; + uint16 chunk_data[MAX_ENTRIES_PER_CHUNK]; + uint32 chunk_position; +}; + +/* + * State for keeping track of progress while incrementally writing a block + * reference table file to disk. + */ +struct BlockRefTableWriter +{ + BlockRefTableBuffer buffer; +}; + +/* Function prototypes. */ +static int BlockRefTableComparator(const void *a, const void *b); +static void BlockRefTableFlush(BlockRefTableBuffer *buffer); +static void BlockRefTableRead(BlockRefTableReader *reader, void *data, + int length); +static void BlockRefTableWrite(BlockRefTableBuffer *buffer, void *data, + int length); +static void BlockRefTableFileTerminate(BlockRefTableBuffer *buffer); + +/* + * Create an empty block reference table. + */ +BlockRefTable * +CreateEmptyBlockRefTable(void) +{ + BlockRefTable *brtab = palloc(sizeof(BlockRefTable)); + + /* + * Even completely empty database has a few hundred relation forks, so it + * seems best to size the hash on the assumption that we're going to have + * at least a few thousand entries. + */ +#ifdef FRONTEND + brtab->hash = blockreftable_create(4096, NULL); +#else + brtab->mcxt = CurrentMemoryContext; + brtab->hash = blockreftable_create(brtab->mcxt, 4096, NULL); +#endif + + return brtab; +} + +/* + * Set the "limit block" for a relation fork and forget any modified blocks + * with equal or higher block numbers. + * + * The "limit block" is the shortest known length of the relation within the + * range of WAL records covered by this block reference table. + */ +void +BlockRefTableSetLimitBlock(BlockRefTable *brtab, + const RelFileLocator *rlocator, + ForkNumber forknum, + BlockNumber limit_block) +{ + BlockRefTableEntry *brtentry; + BlockRefTableKey key; + bool found; + + memcpy(&key.rlocator, rlocator, sizeof(RelFileLocator)); + key.forknum = forknum; + brtentry = blockreftable_insert(brtab->hash, key, &found); + + if (!found) + { + /* + * We have no existing data about this relation fork, so just record + * the limit_block value supplied by the caller, and make sure other + * parts of the entry are properly initialized. + */ + brtentry->limit_block = limit_block; + brtentry->nchunks = 0; + brtentry->chunk_size = NULL; + brtentry->chunk_usage = NULL; + brtentry->chunk_data = NULL; + return; + } + + BlockRefTableEntrySetLimitBlock(brtentry, limit_block); +} + +/* + * Mark a block in a given relation fork as known to have been modified. + */ +void +BlockRefTableMarkBlockModified(BlockRefTable *brtab, + const RelFileLocator *rlocator, + ForkNumber forknum, + BlockNumber blknum) +{ + BlockRefTableEntry *brtentry; + BlockRefTableKey key; + bool found; +#ifndef FRONTEND + MemoryContext oldcontext = MemoryContextSwitchTo(brtab->mcxt); +#endif + + memcpy(&key.rlocator, rlocator, sizeof(RelFileLocator)); + key.forknum = forknum; + brtentry = blockreftable_insert(brtab->hash, key, &found); + + if (!found) + { + /* + * We want to set the initial limit block value to something higher + * than any legal block number. InvalidBlockNumber fits the bill. + */ + brtentry->limit_block = InvalidBlockNumber; + brtentry->nchunks = 0; + brtentry->chunk_size = NULL; + brtentry->chunk_usage = NULL; + brtentry->chunk_data = NULL; + } + + BlockRefTableEntryMarkBlockModified(brtentry, forknum, blknum); + +#ifndef FRONTEND + MemoryContextSwitchTo(oldcontext); +#endif +} + +/* + * Get an entry from a block reference table. + * + * If the entry does not exist, this function returns NULL. Otherwise, it + * returns the entry and sets *limit_block to the value from the entry. + */ +BlockRefTableEntry * +BlockRefTableGetEntry(BlockRefTable *brtab, const RelFileLocator *rlocator, + ForkNumber forknum, BlockNumber *limit_block) +{ + BlockRefTableKey key; + BlockRefTableEntry *entry; + + Assert(limit_block != NULL); + + memcpy(&key.rlocator, rlocator, sizeof(RelFileLocator)); + key.forknum = forknum; + entry = blockreftable_lookup(brtab->hash, key); + + if (entry != NULL) + *limit_block = entry->limit_block; + + return entry; +} + +/* + * Get block numbers from a table entry. + * + * 'blocks' must point to enough space to hold at least 'nblocks' block + * numbers, and any block numbers we manage to get will be written there. + * The return value is the number of block numbers actually written. + * + * We do not return block numbers unless they are greater than or equal to + * start_blkno and strictly less than stop_blkno. + */ +int +BlockRefTableEntryGetBlocks(BlockRefTableEntry *entry, + BlockNumber start_blkno, + BlockNumber stop_blkno, + BlockNumber *blocks, + int nblocks) +{ + uint32 start_chunkno; + uint32 stop_chunkno; + uint32 chunkno; + int nresults = 0; + + Assert(entry != NULL); + + /* + * Figure out which chunks could potentially contain blocks of interest. + * + * We need to be careful about overflow here, because stop_blkno could be + * InvalidBlockNumber or something very close to it. + */ + start_chunkno = start_blkno / BLOCKS_PER_CHUNK; + stop_chunkno = stop_blkno / BLOCKS_PER_CHUNK; + if ((stop_blkno % BLOCKS_PER_CHUNK) != 0) + ++stop_chunkno; + if (stop_chunkno > entry->nchunks) + stop_chunkno = entry->nchunks; + + /* + * Loop over chunks. + */ + for (chunkno = start_chunkno; chunkno < stop_chunkno; ++chunkno) + { + uint16 chunk_usage = entry->chunk_usage[chunkno]; + BlockRefTableChunk chunk_data = entry->chunk_data[chunkno]; + unsigned start_offset = 0; + unsigned stop_offset = BLOCKS_PER_CHUNK; + + /* + * If the start and/or stop block number falls within this chunk, the + * whole chunk may not be of interest. Figure out which portion we + * care about, if it's not the whole thing. + */ + if (chunkno == start_chunkno) + start_offset = start_blkno % BLOCKS_PER_CHUNK; + if (chunkno == stop_chunkno) + stop_offset = stop_blkno % BLOCKS_PER_CHUNK; + + /* + * Handling differs depending on whether this is an array of offsets + * or a bitmap. + */ + if (chunk_usage == MAX_ENTRIES_PER_CHUNK) + { + unsigned i; + + /* It's a bitmap, so test every relevant bit. */ + for (i = start_offset; i < BLOCKS_PER_CHUNK; ++i) + { + uint16 w = chunk_data[i / BLOCKS_PER_ENTRY]; + + if ((w & (1 << (i % BLOCKS_PER_ENTRY))) != 0) + { + BlockNumber blkno = chunkno * BLOCKS_PER_CHUNK + i; + + blocks[nresults++] = blkno; + + /* Early exit if we run out of output space. */ + if (nresults == nblocks) + return nresults; + } + } + } + else + { + unsigned i; + + /* It's an array of offsets, so check each one. */ + for (i = 0; i < chunk_usage; ++i) + { + uint16 offset = chunk_data[i]; + + if (offset >= start_offset && offset < stop_offset) + { + BlockNumber blkno = chunkno * BLOCKS_PER_CHUNK + offset; + + blocks[nresults++] = blkno; + + /* Early exit if we run out of output space. */ + if (nresults == nblocks) + return nresults; + } + } + } + } + + return nresults; +} + +/* + * Serialize a block reference table to a file. + */ +void +WriteBlockRefTable(BlockRefTable *brtab, + io_callback_fn write_callback, + void *write_callback_arg) +{ + BlockRefTableSerializedEntry *sdata = NULL; + BlockRefTableBuffer buffer; + uint32 magic = BLOCKREFTABLE_MAGIC; + + /* Prepare buffer. */ + memset(&buffer, 0, sizeof(BlockRefTableBuffer)); + buffer.io_callback = write_callback; + buffer.io_callback_arg = write_callback_arg; + INIT_CRC32C(buffer.crc); + + /* Write magic number. */ + BlockRefTableWrite(&buffer, &magic, sizeof(uint32)); + + /* Write the entries, assuming there are some. */ + if (brtab->hash->members > 0) + { + unsigned i = 0; + blockreftable_iterator it; + BlockRefTableEntry *brtentry; + + /* Extract entries into serializable format and sort them. */ + sdata = + palloc(brtab->hash->members * sizeof(BlockRefTableSerializedEntry)); + blockreftable_start_iterate(brtab->hash, &it); + while ((brtentry = blockreftable_iterate(brtab->hash, &it)) != NULL) + { + BlockRefTableSerializedEntry *sentry = &sdata[i++]; + + sentry->rlocator = brtentry->key.rlocator; + sentry->forknum = brtentry->key.forknum; + sentry->limit_block = brtentry->limit_block; + sentry->nchunks = brtentry->nchunks; + + /* trim trailing zero entries */ + while (sentry->nchunks > 0 && + brtentry->chunk_usage[sentry->nchunks - 1] == 0) + sentry->nchunks--; + } + Assert(i == brtab->hash->members); + qsort(sdata, i, sizeof(BlockRefTableSerializedEntry), + BlockRefTableComparator); + + /* Loop over entries in sorted order and serialize each one. */ + for (i = 0; i < brtab->hash->members; ++i) + { + BlockRefTableSerializedEntry *sentry = &sdata[i]; + BlockRefTableEntry *brtentry; + BlockRefTableKey key; + unsigned j; + + /* Write the serialized entry itself. */ + BlockRefTableWrite(&buffer, sentry, + sizeof(BlockRefTableSerializedEntry)); + + /* Look up the original entry so we can access the chunks. */ + memcpy(&key.rlocator, &sentry->rlocator, sizeof(RelFileLocator)); + key.forknum = sentry->forknum; + brtentry = blockreftable_lookup(brtab->hash, key); + Assert(brtentry != NULL); + + /* Write the untruncated portion of the chunk length array. */ + if (sentry->nchunks != 0) + BlockRefTableWrite(&buffer, brtentry->chunk_usage, + sentry->nchunks * sizeof(uint16)); + + /* Write the contents of each chunk. */ + for (j = 0; j < brtentry->nchunks; ++j) + { + if (brtentry->chunk_usage[j] == 0) + continue; + BlockRefTableWrite(&buffer, brtentry->chunk_data[j], + brtentry->chunk_usage[j] * sizeof(uint16)); + } + } + } + + /* Write out appropriate terminator and CRC and flush buffer. */ + BlockRefTableFileTerminate(&buffer); +} + +/* + * Prepare to incrementally read a block reference table file. + * + * 'read_callback' is a function that can be called to read data from the + * underlying file (or other data source) into our internal buffer. + * + * 'read_callback_arg' is an opaque argument to be passed to read_callback. + * + * 'error_filename' is the filename that should be included in error messages + * if the file is found to be malformed. The value is not copied, so the + * caller should ensure that it remains valid until done with this + * BlockRefTableReader. + * + * 'error_callback' is a function to be called if the file is found to be + * malformed. This is not used for I/O errors, which must be handled internally + * by read_callback. + * + * 'error_callback_arg' is an opaque arguent to be passed to error_callback. + */ +BlockRefTableReader * +CreateBlockRefTableReader(io_callback_fn read_callback, + void *read_callback_arg, + char *error_filename, + report_error_fn error_callback, + void *error_callback_arg) +{ + BlockRefTableReader *reader; + uint32 magic; + + /* Initialize data structure. */ + reader = palloc0(sizeof(BlockRefTableReader)); + reader->buffer.io_callback = read_callback; + reader->buffer.io_callback_arg = read_callback_arg; + reader->error_filename = error_filename; + reader->error_callback = error_callback; + reader->error_callback_arg = error_callback_arg; + INIT_CRC32C(reader->buffer.crc); + + /* Verify magic number. */ + BlockRefTableRead(reader, &magic, sizeof(uint32)); + if (magic != BLOCKREFTABLE_MAGIC) + error_callback(error_callback_arg, + "file \"%s\" has wrong magic number: expected %u, found %u", + error_filename, + BLOCKREFTABLE_MAGIC, magic); + + return reader; +} + +/* + * Read next relation fork covered by this block reference table file. + * + * After calling this function, you must call BlockRefTableReaderGetBlocks + * until it returns 0 before calling it again. + */ +bool +BlockRefTableReaderNextRelation(BlockRefTableReader *reader, + RelFileLocator *rlocator, + ForkNumber *forknum, + BlockNumber *limit_block) +{ + BlockRefTableSerializedEntry sentry; + BlockRefTableSerializedEntry zentry = {0}; + + /* + * Sanity check: caller must read all blocks from all chunks before moving + * on to the next relation. + */ + Assert(reader->total_chunks == reader->consumed_chunks); + + /* Read serialized entry. */ + BlockRefTableRead(reader, &sentry, + sizeof(BlockRefTableSerializedEntry)); + + /* + * If we just read the sentinel entry indicating that we've reached the + * end, read and check the CRC. + */ + if (memcmp(&sentry, &zentry, sizeof(BlockRefTableSerializedEntry)) == 0) + { + pg_crc32c expected_crc; + pg_crc32c actual_crc; + + /* + * We want to know the CRC of the file excluding the 4-byte CRC + * itself, so copy the current value of the CRC accumulator before + * reading those bytes, and use the copy to finalize the calculation. + */ + expected_crc = reader->buffer.crc; + FIN_CRC32C(expected_crc); + + /* Now we can read the actual value. */ + BlockRefTableRead(reader, &actual_crc, sizeof(pg_crc32c)); + + /* Throw an error if there is a mismatch. */ + if (!EQ_CRC32C(expected_crc, actual_crc)) + reader->error_callback(reader->error_callback_arg, + "file \"%s\" has wrong checksum: expected %08X, found %08X", + reader->error_filename, expected_crc, actual_crc); + + return false; + } + + /* Read chunk size array. */ + if (reader->chunk_size != NULL) + pfree(reader->chunk_size); + reader->chunk_size = palloc(sentry.nchunks * sizeof(uint16)); + BlockRefTableRead(reader, reader->chunk_size, + sentry.nchunks * sizeof(uint16)); + + /* Set up for chunk scan. */ + reader->total_chunks = sentry.nchunks; + reader->consumed_chunks = 0; + + /* Return data to caller. */ + memcpy(rlocator, &sentry.rlocator, sizeof(RelFileLocator)); + *forknum = sentry.forknum; + *limit_block = sentry.limit_block; + return true; +} + +/* + * Get modified blocks associated with the relation fork returned by + * the most recent call to BlockRefTableReaderNextRelation. + * + * On return, block numbers will be written into the 'blocks' array, whose + * length should be passed via 'nblocks'. The return value is the number of + * entries actually written into the 'blocks' array, which may be less than + * 'nblocks' if we run out of modified blocks in the relation fork before + * we run out of room in the array. + */ +unsigned +BlockRefTableReaderGetBlocks(BlockRefTableReader *reader, + BlockNumber *blocks, + int nblocks) +{ + unsigned blocks_found = 0; + + /* Must provide space for at least one block number to be returned. */ + Assert(nblocks > 0); + + /* Loop collecting blocks to return to caller. */ + for (;;) + { + uint16 next_chunk_size; + + /* + * If we've read at least one chunk, maybe it contains some block + * numbers that could satisfy caller's request. + */ + if (reader->consumed_chunks > 0) + { + uint32 chunkno = reader->consumed_chunks - 1; + uint16 chunk_size = reader->chunk_size[chunkno]; + + if (chunk_size == MAX_ENTRIES_PER_CHUNK) + { + /* Bitmap format, so search for bits that are set. */ + while (reader->chunk_position < BLOCKS_PER_CHUNK && + blocks_found < nblocks) + { + uint16 chunkoffset = reader->chunk_position; + uint16 w; + + w = reader->chunk_data[chunkoffset / BLOCKS_PER_ENTRY]; + if ((w & (1u << (chunkoffset % BLOCKS_PER_ENTRY))) != 0) + blocks[blocks_found++] = + chunkno * BLOCKS_PER_CHUNK + chunkoffset; + ++reader->chunk_position; + } + } + else + { + /* Not in bitmap format, so each entry is a 2-byte offset. */ + while (reader->chunk_position < chunk_size && + blocks_found < nblocks) + { + blocks[blocks_found++] = chunkno * BLOCKS_PER_CHUNK + + reader->chunk_data[reader->chunk_position]; + ++reader->chunk_position; + } + } + } + + /* We found enough blocks, so we're done. */ + if (blocks_found >= nblocks) + break; + + /* + * We didn't find enough blocks, so we must need the next chunk. If + * there are none left, though, then we're done anyway. + */ + if (reader->consumed_chunks == reader->total_chunks) + break; + + /* + * Read data for next chunk and reset scan position to beginning of + * chunk. Note that the next chunk might be empty, in which case we + * consume the chunk without actually consuming any bytes from the + * underlying file. + */ + next_chunk_size = reader->chunk_size[reader->consumed_chunks]; + if (next_chunk_size > 0) + BlockRefTableRead(reader, reader->chunk_data, + next_chunk_size * sizeof(uint16)); + ++reader->consumed_chunks; + reader->chunk_position = 0; + } + + return blocks_found; +} + +/* + * Release memory used while reading a block reference table from a file. + */ +void +DestroyBlockRefTableReader(BlockRefTableReader *reader) +{ + if (reader->chunk_size != NULL) + { + pfree(reader->chunk_size); + reader->chunk_size = NULL; + } + pfree(reader); +} + +/* + * Prepare to write a block reference table file incrementally. + * + * Caller must be able to supply BlockRefTableEntry objects sorted in the + * appropriate order. + */ +BlockRefTableWriter * +CreateBlockRefTableWriter(io_callback_fn write_callback, + void *write_callback_arg) +{ + BlockRefTableWriter *writer; + uint32 magic = BLOCKREFTABLE_MAGIC; + + /* Prepare buffer and CRC check and save callbacks. */ + writer = palloc0(sizeof(BlockRefTableWriter)); + writer->buffer.io_callback = write_callback; + writer->buffer.io_callback_arg = write_callback_arg; + INIT_CRC32C(writer->buffer.crc); + + /* Write magic number. */ + BlockRefTableWrite(&writer->buffer, &magic, sizeof(uint32)); + + return writer; +} + +/* + * Append one entry to a block reference table file. + * + * Note that entries must be written in the proper order, that is, sorted by + * tablespace, then database, then relfilenumber, then fork number. Caller + * is responsible for supplying data in the correct order. If that seems hard, + * use an in-memory BlockRefTable instead. + */ +void +BlockRefTableWriteEntry(BlockRefTableWriter *writer, BlockRefTableEntry *entry) +{ + BlockRefTableSerializedEntry sentry; + unsigned j; + + /* Convert to serialized entry format. */ + sentry.rlocator = entry->key.rlocator; + sentry.forknum = entry->key.forknum; + sentry.limit_block = entry->limit_block; + sentry.nchunks = entry->nchunks; + + /* Trim trailing zero entries. */ + while (sentry.nchunks > 0 && entry->chunk_usage[sentry.nchunks - 1] == 0) + sentry.nchunks--; + + /* Write the serialized entry itself. */ + BlockRefTableWrite(&writer->buffer, &sentry, + sizeof(BlockRefTableSerializedEntry)); + + /* Write the untruncated portion of the chunk length array. */ + if (sentry.nchunks != 0) + BlockRefTableWrite(&writer->buffer, entry->chunk_usage, + sentry.nchunks * sizeof(uint16)); + + /* Write the contents of each chunk. */ + for (j = 0; j < entry->nchunks; ++j) + { + if (entry->chunk_usage[j] == 0) + continue; + BlockRefTableWrite(&writer->buffer, entry->chunk_data[j], + entry->chunk_usage[j] * sizeof(uint16)); + } +} + +/* + * Finalize an incremental write of a block reference table file. + */ +void +DestroyBlockRefTableWriter(BlockRefTableWriter *writer) +{ + BlockRefTableFileTerminate(&writer->buffer); + pfree(writer); +} + +/* + * Allocate a standalone BlockRefTableEntry. + * + * When we're manipulating a full in-memory BlockRefTable, the entries are + * part of the hash table and are allocated by simplehash. This routine is + * used by callers that want to write out a BlockRefTable to a file without + * needing to store the whole thing in memory at once. + * + * Entries allocated by this function can be manipulated using the functions + * BlockRefTableEntrySetLimitBlock and BlockRefTableEntryMarkBlockModified + * and then written using BlockRefTableWriteEntry and freed using + * BlockRefTableFreeEntry. + */ +BlockRefTableEntry * +CreateBlockRefTableEntry(RelFileLocator rlocator, ForkNumber forknum) +{ + BlockRefTableEntry *entry = palloc0(sizeof(BlockRefTableEntry)); + + memcpy(&entry->key.rlocator, &rlocator, sizeof(RelFileLocator)); + entry->key.forknum = forknum; + entry->limit_block = InvalidBlockNumber; + + return entry; +} + +/* + * Update a BlockRefTableEntry with a new value for the "limit block" and + * forget any equal-or-higher-numbered modified blocks. + * + * The "limit block" is the shortest known length of the relation within the + * range of WAL records covered by this block reference table. + */ +void +BlockRefTableEntrySetLimitBlock(BlockRefTableEntry *entry, + BlockNumber limit_block) +{ + unsigned chunkno; + unsigned limit_chunkno; + unsigned limit_chunkoffset; + BlockRefTableChunk limit_chunk; + + /* If we already have an equal or lower limit block, do nothing. */ + if (limit_block >= entry->limit_block) + return; + + /* Record the new limit block value. */ + entry->limit_block = limit_block; + + /* + * Figure out which chunk would store the state of the new limit block, + * and which offset within that chunk. + */ + limit_chunkno = limit_block / BLOCKS_PER_CHUNK; + limit_chunkoffset = limit_block % BLOCKS_PER_CHUNK; + + /* + * If the number of chunks is not large enough for any blocks with equal + * or higher block numbers to exist, then there is nothing further to do. + */ + if (limit_chunkno >= entry->nchunks) + return; + + /* Discard entire contents of any higher-numbered chunks. */ + for (chunkno = limit_chunkno + 1; chunkno < entry->nchunks; ++chunkno) + entry->chunk_usage[chunkno] = 0; + + /* + * Next, we need to discard any offsets within the chunk that would + * contain the limit_block. We must handle this differenly depending on + * whether the chunk that would contain limit_block is a bitmap or an + * array of offsets. + */ + limit_chunk = entry->chunk_data[limit_chunkno]; + if (entry->chunk_usage[limit_chunkno] == MAX_ENTRIES_PER_CHUNK) + { + unsigned chunkoffset; + + /* It's a bitmap. Unset bits. */ + for (chunkoffset = limit_chunkoffset; chunkoffset < BLOCKS_PER_CHUNK; + ++chunkoffset) + limit_chunk[chunkoffset / BLOCKS_PER_ENTRY] &= + ~(1 << (chunkoffset % BLOCKS_PER_ENTRY)); + } + else + { + unsigned i, + j = 0; + + /* It's an offset array. Filter out large offsets. */ + for (i = 0; i < entry->chunk_usage[limit_chunkno]; ++i) + { + Assert(j <= i); + if (limit_chunk[i] < limit_chunkoffset) + limit_chunk[j++] = limit_chunk[i]; + } + Assert(j <= entry->chunk_usage[limit_chunkno]); + entry->chunk_usage[limit_chunkno] = j; + } +} + +/* + * Mark a block in a given BlkRefTableEntry as known to have been modified. + */ +void +BlockRefTableEntryMarkBlockModified(BlockRefTableEntry *entry, + ForkNumber forknum, + BlockNumber blknum) +{ + unsigned chunkno; + unsigned chunkoffset; + unsigned i; + + /* + * Which chunk should store the state of this block? And what is the + * offset of this block relative to the start of that chunk? + */ + chunkno = blknum / BLOCKS_PER_CHUNK; + chunkoffset = blknum % BLOCKS_PER_CHUNK; + + /* + * If 'nchunks' isn't big enough for us to be able to represent the state + * of this block, we need to enlarge our arrays. + */ + if (chunkno >= entry->nchunks) + { + unsigned max_chunks; + unsigned extra_chunks; + + /* + * New array size is a power of 2, at least 16, big enough so that + * chunkno will be a valid array index. + */ + max_chunks = Max(16, entry->nchunks); + while (max_chunks < chunkno + 1) + chunkno *= 2; + Assert(max_chunks > chunkno); + extra_chunks = max_chunks - entry->nchunks; + + if (entry->nchunks == 0) + { + entry->chunk_size = palloc0(sizeof(uint16) * max_chunks); + entry->chunk_usage = palloc0(sizeof(uint16) * max_chunks); + entry->chunk_data = + palloc0(sizeof(BlockRefTableChunk) * max_chunks); + } + else + { + entry->chunk_size = repalloc(entry->chunk_size, + sizeof(uint16) * max_chunks); + memset(&entry->chunk_size[entry->nchunks], 0, + extra_chunks * sizeof(uint16)); + entry->chunk_usage = repalloc(entry->chunk_usage, + sizeof(uint16) * max_chunks); + memset(&entry->chunk_usage[entry->nchunks], 0, + extra_chunks * sizeof(uint16)); + entry->chunk_data = repalloc(entry->chunk_data, + sizeof(BlockRefTableChunk) * max_chunks); + memset(&entry->chunk_data[entry->nchunks], 0, + extra_chunks * sizeof(BlockRefTableChunk)); + } + entry->nchunks = max_chunks; + } + + /* + * If the chunk that covers this block number doesn't exist yet, create it + * as an array and add the appropriate offset to it. We make it pretty + * small initially, because there might only be 1 or a few block + * references in this chunk and we don't want to use up too much memory. + */ + if (entry->chunk_size[chunkno] == 0) + { + entry->chunk_data[chunkno] = + palloc(sizeof(uint16) * INITIAL_ENTRIES_PER_CHUNK); + entry->chunk_size[chunkno] = INITIAL_ENTRIES_PER_CHUNK; + entry->chunk_data[chunkno][0] = chunkoffset; + entry->chunk_usage[chunkno] = 1; + return; + } + + /* + * If the number of entries in this chunk is already maximum, it must be a + * bitmap. Just set the appropriate bit. + */ + if (entry->chunk_usage[chunkno] == MAX_ENTRIES_PER_CHUNK) + { + BlockRefTableChunk chunk = entry->chunk_data[chunkno]; + + chunk[chunkoffset / BLOCKS_PER_ENTRY] |= + 1 << (chunkoffset % BLOCKS_PER_ENTRY); + return; + } + + /* + * There is an existing chunk and it's in array format. Let's find out + * whether it already has an entry for this block. If so, we do not need + * to do anything. + */ + for (i = 0; i < entry->chunk_usage[chunkno]; ++i) + { + if (entry->chunk_data[chunkno][i] == chunkoffset) + return; + } + + /* + * If the number of entries currently used is one less than the maximum, + * it's time to convert to bitmap format. + */ + if (entry->chunk_usage[chunkno] == MAX_ENTRIES_PER_CHUNK - 1) + { + BlockRefTableChunk newchunk; + unsigned j; + + /* Allocate a new chunk. */ + newchunk = palloc0(MAX_ENTRIES_PER_CHUNK * sizeof(uint16)); + + /* Set the bit for each existing entry. */ + for (j = 0; j < entry->chunk_usage[chunkno]; ++j) + { + unsigned coff = entry->chunk_data[chunkno][j]; + + newchunk[coff / BLOCKS_PER_ENTRY] |= + 1 << (coff % BLOCKS_PER_ENTRY); + } + + /* Set the bit for the new entry. */ + newchunk[chunkoffset / BLOCKS_PER_ENTRY] |= + 1 << (chunkoffset % BLOCKS_PER_ENTRY); + + /* Swap the new chunk into place and update metadata. */ + pfree(entry->chunk_data[chunkno]); + entry->chunk_data[chunkno] = newchunk; + entry->chunk_size[chunkno] = MAX_ENTRIES_PER_CHUNK; + entry->chunk_usage[chunkno] = MAX_ENTRIES_PER_CHUNK; + return; + } + + /* + * OK, we currently have an array, and we don't need to convert to a + * bitmap, but we do need to add a new element. If there's not enough + * room, we'll have to expand the array. + */ + if (entry->chunk_usage[chunkno] == entry->chunk_size[chunkno]) + { + unsigned newsize = entry->chunk_size[chunkno] * 2; + + Assert(newsize <= MAX_ENTRIES_PER_CHUNK); + entry->chunk_data[chunkno] = repalloc(entry->chunk_data[chunkno], + newsize * sizeof(uint16)); + entry->chunk_size[chunkno] = newsize; + } + + /* Now we can add the new entry. */ + entry->chunk_data[chunkno][entry->chunk_usage[chunkno]] = + chunkoffset; + entry->chunk_usage[chunkno]++; +} + +/* + * Release memory for a BlockRefTablEntry that was created by + * CreateBlockRefTableEntry. + */ +void +BlockRefTableFreeEntry(BlockRefTableEntry *entry) +{ + if (entry->chunk_size != NULL) + { + pfree(entry->chunk_size); + entry->chunk_size = NULL; + } + + if (entry->chunk_usage != NULL) + { + pfree(entry->chunk_usage); + entry->chunk_usage = NULL; + } + + if (entry->chunk_data != NULL) + { + pfree(entry->chunk_data); + entry->chunk_data = NULL; + } + + pfree(entry); +} + +/* + * Comparator for BlockRefTableSerializedEntry objects. + * + * We make the tablespace OID the first column of the sort key to match + * the on-disk tree structure. + */ +static int +BlockRefTableComparator(const void *a, const void *b) +{ + const BlockRefTableSerializedEntry *sa = a; + const BlockRefTableSerializedEntry *sb = b; + + if (sa->rlocator.spcOid > sb->rlocator.spcOid) + return 1; + if (sa->rlocator.spcOid < sb->rlocator.spcOid) + return -1; + + if (sa->rlocator.dbOid > sb->rlocator.dbOid) + return 1; + if (sa->rlocator.dbOid < sb->rlocator.dbOid) + return -1; + + if (sa->rlocator.relNumber > sb->rlocator.relNumber) + return 1; + if (sa->rlocator.relNumber < sb->rlocator.relNumber) + return -1; + + if (sa->forknum > sb->forknum) + return 1; + if (sa->forknum < sb->forknum) + return -1; + + return 0; +} + +/* + * Flush any buffered data out of a BlockRefTableBuffer. + */ +static void +BlockRefTableFlush(BlockRefTableBuffer *buffer) +{ + buffer->io_callback(buffer->io_callback_arg, buffer->data, buffer->used); + buffer->used = 0; +} + +/* + * Read data from a BlockRefTableBuffer, and update the running CRC + * calculation for the returned data (but not any data that we may have + * buffered but not yet actually returned). + */ +static void +BlockRefTableRead(BlockRefTableReader *reader, void *data, int length) +{ + BlockRefTableBuffer *buffer = &reader->buffer; + + /* Loop until read is fully satisfied. */ + while (length > 0) + { + if (buffer->cursor < buffer->used) + { + /* + * If any buffered data is available, use that to satisfy as much + * of the request as possible. + */ + int bytes_to_copy = Min(length, buffer->used - buffer->cursor); + + memcpy(data, &buffer->data[buffer->cursor], bytes_to_copy); + COMP_CRC32C(buffer->crc, &buffer->data[buffer->cursor], + bytes_to_copy); + buffer->cursor += bytes_to_copy; + data = ((char *) data) + bytes_to_copy; + length -= bytes_to_copy; + } + else if (length >= BUFSIZE) + { + /* + * If the request length is long, read directly into caller's + * buffer. + */ + int bytes_read; + + bytes_read = buffer->io_callback(buffer->io_callback_arg, + data, length); + COMP_CRC32C(buffer->crc, data, bytes_read); + data = ((char *) data) + bytes_read; + length -= bytes_read; + + /* If we didn't get anything, that's bad. */ + if (bytes_read == 0) + reader->error_callback(reader->error_callback_arg, + "file \"%s\" ends unexpectedly", + reader->error_filename); + } + else + { + /* + * Refill our buffer. + */ + buffer->used = buffer->io_callback(buffer->io_callback_arg, + buffer->data, BUFSIZE); + buffer->cursor = 0; + + /* If we didn't get anything, that's bad. */ + if (buffer->used == 0) + reader->error_callback(reader->error_callback_arg, + "file \"%s\" ends unexpectedly", + reader->error_filename); + } + } +} + +/* + * Supply data to a BlockRefTableBuffer for write to the underlying File, + * and update the running CRC calculation for that data. + */ +static void +BlockRefTableWrite(BlockRefTableBuffer *buffer, void *data, int length) +{ + /* Update running CRC calculation. */ + COMP_CRC32C(buffer->crc, data, length); + + /* If the new data can't fit into the buffer, flush the buffer. */ + if (buffer->used + length > BUFSIZE) + { + buffer->io_callback(buffer->io_callback_arg, buffer->data, + buffer->used); + buffer->used = 0; + } + + /* If the new data would fill the buffer, or more, write it directly. */ + if (length >= BUFSIZE) + { + buffer->io_callback(buffer->io_callback_arg, data, length); + return; + } + + /* Otherwise, copy the new data into the buffer. */ + memcpy(&buffer->data[buffer->used], data, length); + buffer->used += length; + Assert(buffer->used <= BUFSIZE); +} + +/* + * Generate the sentinel and CRC required at the end of a block reference + * table file and flush them out of our internal buffer. + */ +static void +BlockRefTableFileTerminate(BlockRefTableBuffer *buffer) +{ + BlockRefTableSerializedEntry zentry = {0}; + pg_crc32c crc; + + /* Write a sentinel indicating that there are no more entries. */ + BlockRefTableWrite(buffer, &zentry, + sizeof(BlockRefTableSerializedEntry)); + + /* + * Writing the checksum will perturb the ongoing checksum calculation, so + * copy the state first and finalize the computation using the copy. + */ + crc = buffer->crc; + FIN_CRC32C(crc); + BlockRefTableWrite(buffer, &crc, sizeof(pg_crc32c)); + + /* Flush any leftover data out of our buffer. */ + BlockRefTableFlush(buffer); +} diff --git a/src/common/meson.build b/src/common/meson.build index aa646f96a3..6348d60ec4 100644 --- a/src/common/meson.build +++ b/src/common/meson.build @@ -4,6 +4,7 @@ common_sources = files( 'archive.c', 'base64.c', 'binaryheap.c', + 'blkreftable.c', 'checksum_helper.c', 'compression.c', 'controldata_utils.c', diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 4ad572cb87..9d1e4ab57b 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -209,6 +209,7 @@ extern int XLogFileOpen(XLogSegNo segno, TimeLineID tli); extern void CheckXLogRemoved(XLogSegNo segno, TimeLineID tli); extern XLogSegNo XLogGetLastRemovedSegno(void); +extern XLogSegNo XLogGetOldestSegno(TimeLineID tli); extern void XLogSetAsyncXactLSN(XLogRecPtr asyncXactLSN); extern void XLogSetReplicationSlotMinimumLSN(XLogRecPtr lsn); diff --git a/src/include/access/xlogbackup.h b/src/include/access/xlogbackup.h index 1611358137..90e04cad56 100644 --- a/src/include/access/xlogbackup.h +++ b/src/include/access/xlogbackup.h @@ -28,6 +28,8 @@ typedef struct BackupState XLogRecPtr checkpointloc; /* last checkpoint location */ pg_time_t starttime; /* backup start time */ bool started_in_recovery; /* backup started in recovery? */ + XLogRecPtr istartpoint; /* incremental based on backup at this LSN */ + TimeLineID istarttli; /* incremental based on backup on this TLI */ /* Fields saved at the end of backup */ XLogRecPtr stoppoint; /* backup stop WAL location */ diff --git a/src/include/backup/basebackup.h b/src/include/backup/basebackup.h index 1432d9c206..345bd22534 100644 --- a/src/include/backup/basebackup.h +++ b/src/include/backup/basebackup.h @@ -34,6 +34,9 @@ typedef struct int64 size; /* total size as sent; -1 if not known */ } tablespaceinfo; -extern void SendBaseBackup(BaseBackupCmd *cmd); +struct IncrementalBackupInfo; + +extern void SendBaseBackup(BaseBackupCmd *cmd, + struct IncrementalBackupInfo *ib); #endif /* _BASEBACKUP_H */ diff --git a/src/include/backup/basebackup_incremental.h b/src/include/backup/basebackup_incremental.h new file mode 100644 index 0000000000..c300235a2f --- /dev/null +++ b/src/include/backup/basebackup_incremental.h @@ -0,0 +1,56 @@ +/*------------------------------------------------------------------------- + * + * basebackup_incremental.h + * API for incremental backup support + * + * Portions Copyright (c) 2010-2022, PostgreSQL Global Development Group + * + * src/include/backup/basebackup_incremental.h + * + *------------------------------------------------------------------------- + */ +#ifndef BASEBACKUP_INCREMENTAL_H +#define BASEBACKUP_INCREMENTAL_H + +#include "access/xlogbackup.h" +#include "common/relpath.h" +#include "storage/block.h" +#include "utils/palloc.h" + +#define INCREMENTAL_MAGIC 0xd3ae1f0d + +typedef enum +{ + BACK_UP_FILE_FULLY, + BACK_UP_FILE_INCREMENTALLY, + DO_NOT_BACK_UP_FILE +} FileBackupMethod; + +struct IncrementalBackupInfo; +typedef struct IncrementalBackupInfo IncrementalBackupInfo; + +extern IncrementalBackupInfo *CreateIncrementalBackupInfo(MemoryContext); + +extern void AppendIncrementalManifestData(IncrementalBackupInfo *ib, + const char *data, + int len); +extern void FinalizeIncrementalManifest(IncrementalBackupInfo *ib); + +extern void PrepareForIncrementalBackup(IncrementalBackupInfo *ib, + BackupState *backup_state); + +extern char *GetIncrementalFilePath(Oid dboid, Oid spcoid, + RelFileNumber relfilenumber, + ForkNumber forknum, unsigned segno); +extern FileBackupMethod GetFileBackupMethod(IncrementalBackupInfo *ib, + char *path, + Oid dboid, Oid spcoid, + RelFileNumber relfilenumber, + ForkNumber forknum, + unsigned segno, size_t size, + unsigned *num_blocks_required, + BlockNumber *relative_block_numbers, + unsigned *truncation_block_length); +extern size_t GetIncrementalFileSize(unsigned num_blocks_required); + +#endif diff --git a/src/include/backup/walsummary.h b/src/include/backup/walsummary.h new file mode 100644 index 0000000000..d086e64019 --- /dev/null +++ b/src/include/backup/walsummary.h @@ -0,0 +1,49 @@ +/*------------------------------------------------------------------------- + * + * walsummary.h + * WAL summary management + * + * Portions Copyright (c) 2010-2022, PostgreSQL Global Development Group + * + * src/include/backup/walsummary.h + * + *------------------------------------------------------------------------- + */ +#ifndef WALSUMMARY_H +#define WALSUMMARY_H + +#include <time.h> + +#include "access/xlogdefs.h" +#include "nodes/pg_list.h" +#include "storage/fd.h" + +typedef struct WalSummaryIO +{ + File file; + off_t filepos; +} WalSummaryIO; + +typedef struct WalSummaryFile +{ + XLogRecPtr start_lsn; + XLogRecPtr end_lsn; + TimeLineID tli; +} WalSummaryFile; + +extern List *GetWalSummaries(TimeLineID tli, XLogRecPtr start_lsn, + XLogRecPtr end_lsn); +extern List *FilterWalSummaries(List *wslist, TimeLineID tli, + XLogRecPtr start_lsn, XLogRecPtr end_lsn); +extern bool WalSummariesAreComplete(List *wslist, + XLogRecPtr start_lsn, XLogRecPtr end_lsn, + XLogRecPtr *missing_lsn); +extern File OpenWalSummaryFile(WalSummaryFile *ws, bool missing_ok); +extern void RemoveWalSummaryIfOlderThan(WalSummaryFile *ws, + time_t cutoff_time); + +extern int ReadWalSummary(void *wal_summary_io, void *data, int length); +extern int WriteWalSummary(void *wal_summary_io, void *data, int length); +extern void ReportWalSummaryError(void *callback_arg, char *fmt,...); + +#endif /* WALSUMMARY_H */ diff --git a/src/include/catalog/pg_proc.dat b/src/include/catalog/pg_proc.dat index c92d0631a0..9717c4630e 100644 --- a/src/include/catalog/pg_proc.dat +++ b/src/include/catalog/pg_proc.dat @@ -12071,4 +12071,23 @@ proname => 'any_value_transfn', prorettype => 'anyelement', proargtypes => 'anyelement anyelement', prosrc => 'any_value_transfn' }, +{ oid => '8436', + descr => 'list of available WAL summary files', + proname => 'pg_available_wal_summaries', prorows => '100', + proretset => 't', provolatile => 'v', proparallel => 's', + prorettype => 'record', proargtypes => '', + proallargtypes => '{int8,pg_lsn,pg_lsn}', + proargmodes => '{o,o,o}', + proargnames => '{tli,start_lsn,end_lsn}', + prosrc => 'pg_available_wal_summaries' }, +{ oid => '8437', + descr => 'contents of a WAL sumamry file', + proname => 'pg_wal_summary_contents', prorows => '100', + proretset => 't', provolatile => 'v', proparallel => 's', + prorettype => 'record', proargtypes => 'int8 pg_lsn pg_lsn', + proallargtypes => '{int8,pg_lsn,pg_lsn,oid,oid,oid,int2,int8,bool}', + proargmodes => '{i,i,i,o,o,o,o,o,o}', + proargnames => '{tli,start_lsn,end_lsn,relfilenode,reltablespace,reldatabase,relforknumber,relblocknumber,is_limit_block}', + prosrc => 'pg_wal_summary_contents' }, + ] diff --git a/src/include/common/blkreftable.h b/src/include/common/blkreftable.h new file mode 100644 index 0000000000..22d9883dc5 --- /dev/null +++ b/src/include/common/blkreftable.h @@ -0,0 +1,120 @@ +/*------------------------------------------------------------------------- + * + * blkreftable.h + * Block reference tables. + * + * A block reference table is used to keep track of which blocks have + * been modified by WAL records within a certain LSN range. + * + * For each relation fork, there is a "limit block number". All existing + * blocks greater than or equal to the limit block number must be + * considered modified; for those less than the limit block number, + * we maintain a bitmap. When a relation fork is created or dropped, + * the limit block number should be set to 0. When it's truncated, + * the limit block number should be set to the length in blocks to + * which it was truncated. + * + * Portions Copyright (c) 2010-2022, PostgreSQL Global Development Group + * + * src/include/common/blkreftable.h + * + *------------------------------------------------------------------------- + */ +#ifndef BLKREFTABLE_H +#define BLKREFTABLE_H + +#include "storage/block.h" +#include "storage/relfilelocator.h" + +/* Magic number for serialization file format. */ +#define BLOCKREFTABLE_MAGIC 0x652b137b + +struct BlockRefTable; +struct BlockRefTableEntry; +struct BlockRefTableReader; +struct BlockRefTableWriter; +typedef struct BlockRefTable BlockRefTable; +typedef struct BlockRefTableEntry BlockRefTableEntry; +typedef struct BlockRefTableReader BlockRefTableReader; +typedef struct BlockRefTableWriter BlockRefTableWriter; + +/* + * The return value of io_callback_fn should be the number of bytes read + * or written. If an error occurs, the functions should report it and + * not return. When used as a write callback, short writes should be retried + * or treated as errors, so that if the callback returns, the return value + * is always the request length. + * + * report_error_fn should not return. + */ +typedef int (*io_callback_fn) (void *callback_arg, void *data, int length); +typedef void (*report_error_fn) (void *calblack_arg, char *msg,...); + + +/* + * Functions for manipulating an entire in-memory block reference table. + */ +extern BlockRefTable *CreateEmptyBlockRefTable(void); +extern void BlockRefTableSetLimitBlock(BlockRefTable *brtab, + const RelFileLocator *rlocator, + ForkNumber forknum, + BlockNumber limit_block); +extern void BlockRefTableMarkBlockModified(BlockRefTable *brtab, + const RelFileLocator *rlocator, + ForkNumber forknum, + BlockNumber blknum); +extern void WriteBlockRefTable(BlockRefTable *brtab, + io_callback_fn write_callback, + void *write_callback_arg); + +extern BlockRefTableEntry *BlockRefTableGetEntry(BlockRefTable *brtab, + const RelFileLocator *rlocator, + ForkNumber forknum, + BlockNumber *limit_block); +extern int BlockRefTableEntryGetBlocks(BlockRefTableEntry *entry, + BlockNumber start_blkno, + BlockNumber stop_blkno, + BlockNumber *blocks, + int nblocks); + +/* + * Functions for reading a block reference table incrementally from disk. + */ +extern BlockRefTableReader *CreateBlockRefTableReader(io_callback_fn read_callback, + void *read_callback_arg, + char *error_filename, + report_error_fn error_callback, + void *error_callback_arg); +extern bool BlockRefTableReaderNextRelation(BlockRefTableReader *reader, + RelFileLocator *rlocator, + ForkNumber *forknum, + BlockNumber *limit_block); +extern unsigned BlockRefTableReaderGetBlocks(BlockRefTableReader *reader, + BlockNumber *blocks, + int nblocks); +extern void DestroyBlockRefTableReader(BlockRefTableReader *reader); + +/* + * Functions for writing a block reference table incrementally to disk. + * + * Note that entries must be written in the proper order, that is, sorted by + * database, then tablespace, then relfilenumber, then fork number. Caller + * is responsible for supplying data in the correct order. If that seems hard, + * use an in-memory BlockRefTable instead. + */ +extern BlockRefTableWriter *CreateBlockRefTableWriter(io_callback_fn write_callback, + void *write_callback_arg); +extern void BlockRefTableWriteEntry(BlockRefTableWriter *writer, + BlockRefTableEntry *entry); +extern void DestroyBlockRefTableWriter(BlockRefTableWriter *writer); + +extern BlockRefTableEntry *CreateBlockRefTableEntry(RelFileLocator rlocator, + ForkNumber forknum); +extern void BlockRefTableEntrySetLimitBlock(BlockRefTableEntry *entry, + BlockNumber limit_block); +extern void BlockRefTableEntryMarkBlockModified(BlockRefTableEntry *entry, + ForkNumber forknum, + BlockNumber blknum); +extern void BlockRefTableFreeEntry(BlockRefTableEntry *entry); + +#endif /* BLKREFTABLE_H */ diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 7232b03e37..042fdc6ca1 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -340,6 +340,7 @@ typedef enum BackendType B_STARTUP, B_WAL_RECEIVER, B_WAL_SENDER, + B_WAL_SUMMARIZER, B_WAL_WRITER, } BackendType; @@ -446,6 +447,7 @@ typedef enum CheckpointerProcess, WalWriterProcess, WalReceiverProcess, + WalSummarizerProcess, NUM_AUXPROCTYPES /* Must be last! */ } AuxProcType; @@ -458,6 +460,7 @@ extern PGDLLIMPORT AuxProcType MyAuxProcType; #define AmCheckpointerProcess() (MyAuxProcType == CheckpointerProcess) #define AmWalWriterProcess() (MyAuxProcType == WalWriterProcess) #define AmWalReceiverProcess() (MyAuxProcType == WalReceiverProcess) +#define AmWalSummarizerProcess() (MyAuxProcType == WalSummarizerProcess) /***************************************************************************** diff --git a/src/include/nodes/replnodes.h b/src/include/nodes/replnodes.h index 4321ba8f86..856491eecd 100644 --- a/src/include/nodes/replnodes.h +++ b/src/include/nodes/replnodes.h @@ -108,4 +108,13 @@ typedef struct TimeLineHistoryCmd TimeLineID timeline; } TimeLineHistoryCmd; +/* ---------------------- + * UPLOAD_MANIFEST command + * ---------------------- + */ +typedef struct UploadManifestCmd +{ + NodeTag type; +} UploadManifestCmd; + #endif /* REPLNODES_H */ diff --git a/src/include/postmaster/walsummarizer.h b/src/include/postmaster/walsummarizer.h new file mode 100644 index 0000000000..7584cb69a7 --- /dev/null +++ b/src/include/postmaster/walsummarizer.h @@ -0,0 +1,31 @@ +/*------------------------------------------------------------------------- + * + * walsummarizer.h + * + * Header file for background WAL summarization process. + * + * Portions Copyright (c) 1996-2022, PostgreSQL Global Development Group + * + * IDENTIFICATION + * src/include/postmaster/walsummarizer.h + * + *------------------------------------------------------------------------- + */ +#ifndef WALSUMMARIZER_H +#define WALSUMMARIZER_H + +#include "access/xlogdefs.h" + +extern int wal_summarize_mb; +extern int wal_summarize_keep_time; + +extern Size WalSummarizerShmemSize(void); +extern void WalSummarizerShmemInit(void); +extern void WalSummarizerMain(void) pg_attribute_noreturn(); + +extern XLogRecPtr GetOldestUnsummarizedLSN(TimeLineID *tli, + bool *lsn_is_exact); +extern void SetWalSummarizerLatch(void); +extern XLogRecPtr WaitForWalSummarization(XLogRecPtr lsn, long timeout); + +#endif diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index ef74f32693..ee55008082 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -417,11 +417,12 @@ extern PGDLLIMPORT PGPROC *PreparedXactProcs; * We set aside some extra PGPROC structures for auxiliary processes, * ie things that aren't full-fledged backends but need shmem access. * - * Background writer, checkpointer, WAL writer and archiver run during normal - * operation. Startup process and WAL receiver also consume 2 slots, but WAL - * writer is launched only after startup has exited, so we only need 5 slots. + * Background writer, checkpointer, WAL writer, WAL summarizer, and archiver + * run during normal operation. Startup process and WAL receiver also consume + * 2 slots, but WAL writer is launched only after startup has exited, so we + * only need 6 slots. */ -#define NUM_AUXILIARY_PROCS 5 +#define NUM_AUXILIARY_PROCS 6 /* configurable options */ extern PGDLLIMPORT int DeadlockTimeout; diff --git a/src/include/utils/guc_tables.h b/src/include/utils/guc_tables.h index d5a0880678..7d3bc0f671 100644 --- a/src/include/utils/guc_tables.h +++ b/src/include/utils/guc_tables.h @@ -72,6 +72,7 @@ enum config_group WAL_RECOVERY, WAL_ARCHIVE_RECOVERY, WAL_RECOVERY_TARGET, + WAL_SUMMARIZATION, REPLICATION_SENDING, REPLICATION_PRIMARY, REPLICATION_STANDBY, diff --git a/src/test/perl/PostgreSQL/Test/Cluster.pm b/src/test/perl/PostgreSQL/Test/Cluster.pm index c3d46c7c70..b711d60fc4 100644 --- a/src/test/perl/PostgreSQL/Test/Cluster.pm +++ b/src/test/perl/PostgreSQL/Test/Cluster.pm @@ -779,6 +779,10 @@ a tar-format backup, pass the name of the tar program to use in the keyword parameter tar_program. Note that tablespace tar files aren't handled here. +To restore from an incremental backup, pass the parameter combine_with_prior +as a reference to an array of prior backup names with which this backup +is to be combined using pg_combinebackup. + Streaming replication can be enabled on this node by passing the keyword parameter has_streaming => 1. This is disabled by default. @@ -816,7 +820,22 @@ sub init_from_backup mkdir $self->archive_dir; my $data_path = $self->data_dir; - if (defined $params{tar_program}) + if (defined $params{combine_with_prior}) + { + my @prior_backups = @{$params{combine_with_prior}}; + my @prior_backup_path; + + for my $prior_backup_name (@prior_backups) + { + push @prior_backup_path, + $root_node->backup_dir . '/' . $prior_backup_name; + } + + local %ENV = $self->_get_env(); + PostgreSQL::Test::Utils::system_or_bail('pg_combinebackup', + @prior_backup_path, $backup_path, '-o', $data_path); + } + elsif (defined $params{tar_program}) { mkdir($data_path); PostgreSQL::Test::Utils::system_or_bail($params{tar_program}, 'xf', diff --git a/src/test/recovery/t/001_stream_rep.pl b/src/test/recovery/t/001_stream_rep.pl index 95f9b0d772..ad11be4664 100644 --- a/src/test/recovery/t/001_stream_rep.pl +++ b/src/test/recovery/t/001_stream_rep.pl @@ -15,6 +15,8 @@ my $node_primary = PostgreSQL::Test::Cluster->new('primary'); $node_primary->init( allows_streaming => 1, auth_extra => [ '--create-role', 'repl_role' ]); +# WAL summarization can postpone WAL recycling, leading to test failures +$node_primary->append_conf('postgresql.conf', "wal_summarize_mb = 0"); $node_primary->start; my $backup_name = 'my_backup'; diff --git a/src/test/recovery/t/019_replslot_limit.pl b/src/test/recovery/t/019_replslot_limit.pl index 7d94f15778..4f52ddbe79 100644 --- a/src/test/recovery/t/019_replslot_limit.pl +++ b/src/test/recovery/t/019_replslot_limit.pl @@ -22,6 +22,7 @@ $node_primary->append_conf( min_wal_size = 2MB max_wal_size = 4MB log_checkpoints = yes +wal_summarize_mb = 0 )); $node_primary->start; $node_primary->safe_psql('postgres', @@ -256,6 +257,7 @@ $node_primary2->append_conf( min_wal_size = 32MB max_wal_size = 32MB log_checkpoints = yes +wal_summarize_mb = 0 )); $node_primary2->start; $node_primary2->safe_psql('postgres', @@ -310,6 +312,7 @@ $node_primary3->append_conf( max_wal_size = 2MB log_checkpoints = yes max_slot_wal_keep_size = 1MB + wal_summarize_mb = 0 )); $node_primary3->start; $node_primary3->safe_psql('postgres', diff --git a/src/test/recovery/t/035_standby_logical_decoding.pl b/src/test/recovery/t/035_standby_logical_decoding.pl index 480e6d6caa..a91437dfa7 100644 --- a/src/test/recovery/t/035_standby_logical_decoding.pl +++ b/src/test/recovery/t/035_standby_logical_decoding.pl @@ -250,6 +250,7 @@ $node_primary->append_conf( wal_level = 'logical' max_replication_slots = 4 max_wal_senders = 4 +wal_summarize_mb = 0 }); $node_primary->dump_info; $node_primary->start; diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index e69bb671bf..2ae238bf81 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -3992,3 +3992,26 @@ yyscan_t z_stream z_streamp zic_t +BlockRefTable +BlockRefTableBuffer +BlockRefTableEntry +BlockRefTableKey +BlockRefTableReader +BlockRefTableSerializedEntry +BlockRefTableWriter +FileBackupMethod +IncrementalBackupInfo +SummarizerReadLocalXLogPrivate +UploadManifestCmd +WalSummarizerData +WalSummaryFile +WalSummaryIO +backup_file_entry +backup_wal_range +cb_cleanup_dir +cb_options +cb_tablespace +cb_tablespace_mapping +manifest_data +manifest_writer +rfile -- 2.37.1 (Apple Git-137.1) [application/octet-stream] v5-0001-Refactor-parse_filename_for_nontemp_relation-to-p.patch (11.9K, ../../CA+TgmoZNVeBzoqDL8xvr-nkaepq815jtDR4nJzPew7=3iEuM1g@mail.gmail.com/7-v5-0001-Refactor-parse_filename_for_nontemp_relation-to-p.patch) download | inline diff: From 22ffb20913efc6d9ffbe77b78444aeb8ba54217b Mon Sep 17 00:00:00 2001 From: Robert Haas <[email protected]> Date: Tue, 3 Oct 2023 13:30:44 -0400 Subject: [PATCH v5 1/6] Refactor parse_filename_for_nontemp_relation to parse more. Instead of returning the number of characters in the RelFileNumber, return the RelFileNumber itself. Continue to return the fork number, as before, and additionally return the segment number. parse_filename_for_nontemp_relation now rejects a RelFileNumber or segment number that begins with a leading zero. Before, we accepted such cases as relation filenames, but if we continued to do so after this change, the function might return the same values for two different files (e.g. 1234.5 and 001234.5 or 1234.005) which could be annoying for callers. Since we don't actually ever generate filenames with leading zeroes in the names, any such files that we find must have been created by something other than PostgreSQL, and it is therefore reasonable to treat them as non-relation files. Along the way, change unlogged_relation_entry to store a RelFileNumber rather than an OID. This update should have been made in 851f4cc75cdd8c831f1baa9a7abf8c8248b65890, but it was overlooked. It's trivial to make the update as part of this commit, perhaps more trivial than it would have been without it, so do that. --- src/backend/backup/basebackup.c | 15 ++-- src/backend/storage/file/reinit.c | 137 ++++++++++++++++++------------ src/include/storage/reinit.h | 5 +- 3 files changed, 93 insertions(+), 64 deletions(-) diff --git a/src/backend/backup/basebackup.c b/src/backend/backup/basebackup.c index 7d025bcf38..b126d9c890 100644 --- a/src/backend/backup/basebackup.c +++ b/src/backend/backup/basebackup.c @@ -1197,9 +1197,9 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, { int excludeIdx; bool excludeFound; - ForkNumber relForkNum; /* Type of fork if file is a relation */ - int relnumchars; /* Chars in filename that are the - * relnumber */ + RelFileNumber relNumber; + ForkNumber relForkNum; + unsigned segno; /* Skip special stuff */ if (strcmp(de->d_name, ".") == 0 || strcmp(de->d_name, "..") == 0) @@ -1249,23 +1249,20 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, /* Exclude all forks for unlogged tables except the init fork */ if (isDbDir && - parse_filename_for_nontemp_relation(de->d_name, &relnumchars, - &relForkNum)) + parse_filename_for_nontemp_relation(de->d_name, &relNumber, + &relForkNum, &segno)) { /* Never exclude init forks */ if (relForkNum != INIT_FORKNUM) { char initForkFile[MAXPGPATH]; - char relNumber[OIDCHARS + 1]; /* * If any other type of fork, check if there is an init fork * with the same RelFileNumber. If so, the file can be * excluded. */ - memcpy(relNumber, de->d_name, relnumchars); - relNumber[relnumchars] = '\0'; - snprintf(initForkFile, sizeof(initForkFile), "%s/%s_init", + snprintf(initForkFile, sizeof(initForkFile), "%s/%u_init", path, relNumber); if (lstat(initForkFile, &statbuf) == 0) diff --git a/src/backend/storage/file/reinit.c b/src/backend/storage/file/reinit.c index fb55371b1b..5df2517b46 100644 --- a/src/backend/storage/file/reinit.c +++ b/src/backend/storage/file/reinit.c @@ -31,7 +31,7 @@ static void ResetUnloggedRelationsInDbspaceDir(const char *dbspacedirname, typedef struct { - Oid reloid; /* hash key */ + RelFileNumber relnumber; /* hash key */ } unlogged_relation_entry; /* @@ -195,12 +195,13 @@ ResetUnloggedRelationsInDbspaceDir(const char *dbspacedirname, int op) while ((de = ReadDir(dbspace_dir, dbspacedirname)) != NULL) { ForkNumber forkNum; - int relnumchars; + unsigned segno; unlogged_relation_entry ent; /* Skip anything that doesn't look like a relation data file. */ - if (!parse_filename_for_nontemp_relation(de->d_name, &relnumchars, - &forkNum)) + if (!parse_filename_for_nontemp_relation(de->d_name, + &ent.relnumber, + &forkNum, &segno)) continue; /* Also skip it unless this is the init fork. */ @@ -208,10 +209,8 @@ ResetUnloggedRelationsInDbspaceDir(const char *dbspacedirname, int op) continue; /* - * Put the OID portion of the name into the hash table, if it - * isn't already. + * Put the RelFileNumber into the hash table, if it isn't already. */ - ent.reloid = atooid(de->d_name); (void) hash_search(hash, &ent, HASH_ENTER, NULL); } @@ -235,12 +234,13 @@ ResetUnloggedRelationsInDbspaceDir(const char *dbspacedirname, int op) while ((de = ReadDir(dbspace_dir, dbspacedirname)) != NULL) { ForkNumber forkNum; - int relnumchars; + unsigned segno; unlogged_relation_entry ent; /* Skip anything that doesn't look like a relation data file. */ - if (!parse_filename_for_nontemp_relation(de->d_name, &relnumchars, - &forkNum)) + if (!parse_filename_for_nontemp_relation(de->d_name, + &ent.relnumber, + &forkNum, &segno)) continue; /* We never remove the init fork. */ @@ -251,7 +251,6 @@ ResetUnloggedRelationsInDbspaceDir(const char *dbspacedirname, int op) * See whether the OID portion of the name shows up in the hash * table. If so, nuke it! */ - ent.reloid = atooid(de->d_name); if (hash_search(hash, &ent, HASH_FIND, NULL)) { snprintf(rm_path, sizeof(rm_path), "%s/%s", @@ -285,14 +284,14 @@ ResetUnloggedRelationsInDbspaceDir(const char *dbspacedirname, int op) while ((de = ReadDir(dbspace_dir, dbspacedirname)) != NULL) { ForkNumber forkNum; - int relnumchars; - char relnumbuf[OIDCHARS + 1]; + RelFileNumber relNumber; + unsigned segno; char srcpath[MAXPGPATH * 2]; char dstpath[MAXPGPATH]; /* Skip anything that doesn't look like a relation data file. */ - if (!parse_filename_for_nontemp_relation(de->d_name, &relnumchars, - &forkNum)) + if (!parse_filename_for_nontemp_relation(de->d_name, &relNumber, + &forkNum, &segno)) continue; /* Also skip it unless this is the init fork. */ @@ -304,11 +303,12 @@ ResetUnloggedRelationsInDbspaceDir(const char *dbspacedirname, int op) dbspacedirname, de->d_name); /* Construct destination pathname. */ - memcpy(relnumbuf, de->d_name, relnumchars); - relnumbuf[relnumchars] = '\0'; - snprintf(dstpath, sizeof(dstpath), "%s/%s%s", - dbspacedirname, relnumbuf, de->d_name + relnumchars + 1 + - strlen(forkNames[INIT_FORKNUM])); + if (segno == 0) + snprintf(dstpath, sizeof(dstpath), "%s/%u", + dbspacedirname, relNumber); + else + snprintf(dstpath, sizeof(dstpath), "%s/%u.%u", + dbspacedirname, relNumber, segno); /* OK, we're ready to perform the actual copy. */ elog(DEBUG2, "copying %s to %s", srcpath, dstpath); @@ -327,14 +327,14 @@ ResetUnloggedRelationsInDbspaceDir(const char *dbspacedirname, int op) dbspace_dir = AllocateDir(dbspacedirname); while ((de = ReadDir(dbspace_dir, dbspacedirname)) != NULL) { + RelFileNumber relNumber; ForkNumber forkNum; - int relnumchars; - char relnumbuf[OIDCHARS + 1]; + unsigned segno; char mainpath[MAXPGPATH]; /* Skip anything that doesn't look like a relation data file. */ - if (!parse_filename_for_nontemp_relation(de->d_name, &relnumchars, - &forkNum)) + if (!parse_filename_for_nontemp_relation(de->d_name, &relNumber, + &forkNum, &segno)) continue; /* Also skip it unless this is the init fork. */ @@ -342,11 +342,12 @@ ResetUnloggedRelationsInDbspaceDir(const char *dbspacedirname, int op) continue; /* Construct main fork pathname. */ - memcpy(relnumbuf, de->d_name, relnumchars); - relnumbuf[relnumchars] = '\0'; - snprintf(mainpath, sizeof(mainpath), "%s/%s%s", - dbspacedirname, relnumbuf, de->d_name + relnumchars + 1 + - strlen(forkNames[INIT_FORKNUM])); + if (segno == 0) + snprintf(mainpath, sizeof(mainpath), "%s/%u", + dbspacedirname, relNumber); + else + snprintf(mainpath, sizeof(mainpath), "%s/%u.%u", + dbspacedirname, relNumber, segno); fsync_fname(mainpath, false); } @@ -371,52 +372,82 @@ ResetUnloggedRelationsInDbspaceDir(const char *dbspacedirname, int op) * This function returns true if the file appears to be in the correct format * for a non-temporary relation and false otherwise. * - * NB: If this function returns true, the caller is entitled to assume that - * *relnumchars has been set to a value no more than OIDCHARS, and thus - * that a buffer of OIDCHARS+1 characters is sufficient to hold the - * RelFileNumber portion of the filename. This is critical to protect against - * a possible buffer overrun. + * If it returns true, it sets *relnumber, *fork, and *segno to the values + * extracted from the filename. If it returns false, these values are set to + * InvalidRelFileNumber, InvalidForkNumber, and 0, respectively. */ bool -parse_filename_for_nontemp_relation(const char *name, int *relnumchars, - ForkNumber *fork) +parse_filename_for_nontemp_relation(const char *name, RelFileNumber *relnumber, + ForkNumber *fork, unsigned *segno) { - int pos; + unsigned long n, + s; + ForkNumber f; + char *endp; - /* Look for a non-empty string of digits (that isn't too long). */ - for (pos = 0; isdigit((unsigned char) name[pos]); ++pos) - ; - if (pos == 0 || pos > OIDCHARS) + *relnumber = InvalidRelFileNumber; + *fork = InvalidForkNumber; + *segno = 0; + + /* + * Relation filenames should begin with a digit that is not a zero. By + * rejecting cases involving leading zeroes, the caller can assume that + * there's only one possible string of characters that could have produced + * any given value for *relnumber. + * + * (To be clear, we don't expect files with names like 0017.3 to exist at + * all -- but if 0017.3 does exist, it's a non-relation file, not part of + * the main fork for relfilenode 17.) + */ + if (name[0] < '1' || name[0] > '9') + return false; + + /* + * Parse the leading digit string. If the value is out of range, we + * conclude that this isn't a relation file at all. + */ + errno = 0; + n = strtoul(name, &endp, 10); + if (errno || name == endp || n <= 0 || n > PG_UINT32_MAX) return false; - *relnumchars = pos; + name = endp; /* Check for a fork name. */ - if (name[pos] != '_') - *fork = MAIN_FORKNUM; + if (*name != '_') + f = MAIN_FORKNUM; else { int forkchar; - forkchar = forkname_chars(&name[pos + 1], fork); + forkchar = forkname_chars(name + 1, &f); if (forkchar <= 0) return false; - pos += forkchar + 1; + name += forkchar + 1; } /* Check for a segment number. */ - if (name[pos] == '.') + if (*name != '.') + s = 0; + else { - int segchar; + /* Reject leading zeroes, just like we do for RelFileNumber. */ + if (name[0] < '1' || name[0] > '9') + return false; - for (segchar = 1; isdigit((unsigned char) name[pos + segchar]); ++segchar) - ; - if (segchar <= 1) + errno = 0; + s = strtoul(name + 1, &endp, 10); + if (errno || name + 1 == endp || s <= 0 || s > PG_UINT32_MAX) return false; - pos += segchar; + name = endp; } /* Now we should be at the end. */ - if (name[pos] != '\0') + if (*name != '\0') return false; + + /* Set out parameters and return. */ + *relnumber = (RelFileNumber) n; + *fork = f; + *segno = (unsigned) s; return true; } diff --git a/src/include/storage/reinit.h b/src/include/storage/reinit.h index e2bbb5abe9..f8eb7ce234 100644 --- a/src/include/storage/reinit.h +++ b/src/include/storage/reinit.h @@ -20,8 +20,9 @@ extern void ResetUnloggedRelations(int op); extern bool parse_filename_for_nontemp_relation(const char *name, - int *relnumchars, - ForkNumber *fork); + RelFileNumber *relnumber, + ForkNumber *fork, + unsigned *segno); #define UNLOGGED_RELATION_CLEANUP 0x0001 #define UNLOGGED_RELATION_INIT 0x0002 -- 2.37.1 (Apple Git-137.1) ^ permalink raw reply [nested|flat] 22+ messages in thread
* Re: trying again to get incremental backup @ 2023-10-19 19:18 David Steele <[email protected]> parent: Robert Haas <[email protected]> 0 siblings, 1 reply; 22+ messages in thread From: David Steele @ 2023-10-19 19:18 UTC (permalink / raw) To: Robert Haas <[email protected]>; Dilip Kumar <[email protected]>; +Cc: Andres Freund <[email protected]>; pgsql-hackers On 10/19/23 12:05, Robert Haas wrote: > On Wed, Oct 4, 2023 at 4:08 PM Robert Haas <[email protected]> wrote: >> Clearly there's a good amount of stuff to sort out here, but we've >> still got quite a bit of time left before feature freeze so I'd like >> to have a go at it. Please let me know your thoughts, if you have any. > > Apparently, nobody has any thoughts, but here's an updated patch set > anyway. The main change, other than rebasing, is that I did a bunch > more documentation work on the main patch (0005). I'm much happier > with it now, although I expect it may need more adjustments here and > there as outstanding design questions get settled. > > After some thought, I think that it should be fine to commit 0001 and > 0002 as independent refactoring patches, and I plan to go ahead and do > that pretty soon unless somebody objects. 0001 looks pretty good to me. The only thing I find a little troublesome is the repeated construction of file names with/without segment numbers in ResetUnloggedRelationsInDbspaceDir(), .e.g.: + if (segno == 0) + snprintf(dstpath, sizeof(dstpath), "%s/%u", + dbspacedirname, relNumber); + else + snprintf(dstpath, sizeof(dstpath), "%s/%u.%u", + dbspacedirname, relNumber, segno); If this happened three times I'd definitely want a helper function, but even with two I think it would be a bit nicer. 0002 is definitely a good idea. FWIW pgBackRest does this conversion but also errors if it does not succeed. We have never seen a report of this error happening in the wild, so I think it must be pretty rare if it does happen. Regards, -David ^ permalink raw reply [nested|flat] 22+ messages in thread
* Re: trying again to get incremental backup @ 2023-10-19 20:00 Robert Haas <[email protected]> parent: David Steele <[email protected]> 0 siblings, 1 reply; 22+ messages in thread From: Robert Haas @ 2023-10-19 20:00 UTC (permalink / raw) To: David Steele <[email protected]>; +Cc: Dilip Kumar <[email protected]>; Andres Freund <[email protected]>; pgsql-hackers On Thu, Oct 19, 2023 at 3:18 PM David Steele <[email protected]> wrote: > 0001 looks pretty good to me. The only thing I find a little troublesome > is the repeated construction of file names with/without segment numbers > in ResetUnloggedRelationsInDbspaceDir(), .e.g.: > > + if (segno == 0) > + snprintf(dstpath, sizeof(dstpath), "%s/%u", > + dbspacedirname, relNumber); > + else > + snprintf(dstpath, sizeof(dstpath), "%s/%u.%u", > + dbspacedirname, relNumber, segno); > > > If this happened three times I'd definitely want a helper function, but > even with two I think it would be a bit nicer. Personally I think that would make the code harder to read rather than easier. I agree that repeating code isn't great, but this is a relatively brief idiom and pretty self-explanatory. If other people agree with you I can change it, but to me it's not an improvement. > 0002 is definitely a good idea. FWIW pgBackRest does this conversion but > also errors if it does not succeed. We have never seen a report of this > error happening in the wild, so I think it must be pretty rare if it > does happen. Cool, but ... how about the main patch set? It's nice to get some of these refactoring bits and pieces out of the way, but if I spend the effort to work out what I think are the right answers to the remaining design questions for the main patch set and then find out after I've done all that that you have massive objections, I'm going to be annoyed. I've been trying to get this feature into PostgreSQL for years, and if I don't succeed this time, I want the reason to be something better than "well, I didn't find out that David disliked X until five minutes before I was planning to type 'git push'." I'm not really concerned about detailed bug-hunting in the main patches just yet. The time for that will come. But if you have views on how to resolve the design questions that I mentioned in a couple of emails back, or intend to advocate vigorously against the whole concept for some reason, let's try to sort that out sooner rather than later. -- Robert Haas EDB: http://www.enterprisedb.com ^ permalink raw reply [nested|flat] 22+ messages in thread
* Re: trying again to get incremental backup @ 2023-10-20 13:20 Jakub Wartak <[email protected]> parent: Robert Haas <[email protected]> 2 siblings, 1 reply; 22+ messages in thread From: Jakub Wartak @ 2023-10-20 13:20 UTC (permalink / raw) To: Robert Haas <[email protected]>; +Cc: Dilip Kumar <[email protected]>; Andres Freund <[email protected]>; pgsql-hackers Hi Robert, On Wed, Oct 4, 2023 at 10:09 PM Robert Haas <[email protected]> wrote: > > On Tue, Oct 3, 2023 at 2:21 PM Robert Haas <[email protected]> wrote: > > Here's a new patch set, also addressing Jakub's observation that > > MINIMUM_VERSION_FOR_WAL_SUMMARIES needed updating. > > Here's yet another new version.[..] Okay, so another good news - related to the patch version #4. Not-so-tiny stress test consisting of pgbench run for 24h straight (with incremental backups every 2h, with base of initial full backup), followed by two PITRs (one not using incremental backup and one using to to illustrate the performance point - and potentially spot any errors in between). In both cases it worked fine. Pgbench has this behaviour that it doesn't cause space growth over time - it produces lots of WAL instead. Some stats: START DBSIZE: ~3.3GB (pgbench -i -s 200 --partitions=8) END DBSIZE: ~4.3GB RUN DURATION: 24h (pgbench -P 1 -R 100 -T 86400) WALARCHIVES-24h: 77GB FULL-DB-BACKUP-SIZE: 3.4GB INCREMENTAL-BACKUP-11-SIZE: 3.5GB Env: Azure VM D4s (4VCPU), Debian 11, gcc 10.2, normal build (asserts and debug disabled) The increments were taken every 2h just to see if they would fail for any reason - they did not. PITR RTO RESULTS (copy/pg_combinebackup time + recovery time): 1. time to restore from fullbackup (+ recovery of 24h WAL[77GB]): 53s + 4640s =~ 78min 2. time to restore from fullbackup+incremental backup from 2h ago (+ recovery of 2h WAL [5.4GB]): 68s + 190s =~ 4min18s I could probably pre populate the DB with 1TB cold data (not touched to be touched pgbench at all), just for the sake of argument, and that would probably could be demonstrated how space efficient the incremental backup can be, but most of time would be time wasted on copying the 1TB here... > - I would like some feedback on the generation of WAL summary files. > Right now, I have it enabled by default, and summaries are kept for a > week. That means that, with no additional setup, you can take an > incremental backup as long as the reference backup was taken in the > last week. I've just noticed one thing when recovery is progress: is summarization working during recovery - in the background - an expected behaviour? I'm wondering about that, because after freshly restored and recovered DB, one would need to create a *new* full backup and only from that point new summaries would have any use? Sample log: 2023-10-20 11:10:02.288 UTC [64434] LOG: restored log file "000000010000000200000022" from archive 2023-10-20 11:10:02.599 UTC [64434] LOG: restored log file "000000010000000200000023" from archive 2023-10-20 11:10:02.769 UTC [64446] LOG: summarized WAL on TLI 1 from 2/139B1130 to 2/239B1518 2023-10-20 11:10:02.923 UTC [64434] LOG: restored log file "000000010000000200000024" from archive 2023-10-20 11:10:03.193 UTC [64434] LOG: restored log file "000000010000000200000025" from archive 2023-10-20 11:10:03.345 UTC [64432] LOG: restartpoint starting: wal 2023-10-20 11:10:03.407 UTC [64446] LOG: summarized WAL on TLI 1 from 2/239B1518 to 2/25B609D0 2023-10-20 11:10:03.521 UTC [64434] LOG: restored log file "000000010000000200000026" from archive 2023-10-20 11:10:04.429 UTC [64434] LOG: restored log file "000000010000000200000027" from archive > > - On a related note, I haven't yet tested this on a standby, which is > a thing that I definitely need to do. I don't know of a reason why it > shouldn't be possible for all of this machinery to work on a standby > just as it does on a primary, but then we need the WAL summarizer to > run there too, which could end up being a waste if nobody ever tries > to take an incremental backup. I wonder how that should be reflected > in the configuration. We could do something like what we've done for > archive_mode, where on means "only on if this is a primary" and you > have to say always if you want it to run on standbys as well ... but > I'm not sure if that's a design pattern that we really want to > replicate into more places. I'd be somewhat inclined to just make > whatever configuration parameters we need to configure this thing on > the primary also work on standbys, and you can set each server up as > you please. But I'm open to other suggestions. I'll try to play with some standby restores in future, stay tuned. Regards, -J. ^ permalink raw reply [nested|flat] 22+ messages in thread
* Re: trying again to get incremental backup @ 2023-10-20 15:30 David Steele <[email protected]> parent: Robert Haas <[email protected]> 0 siblings, 1 reply; 22+ messages in thread From: David Steele @ 2023-10-20 15:30 UTC (permalink / raw) To: Robert Haas <[email protected]>; +Cc: Dilip Kumar <[email protected]>; Andres Freund <[email protected]>; pgsql-hackers On 10/19/23 16:00, Robert Haas wrote: > On Thu, Oct 19, 2023 at 3:18 PM David Steele <[email protected]> wrote: >> 0001 looks pretty good to me. The only thing I find a little troublesome >> is the repeated construction of file names with/without segment numbers >> in ResetUnloggedRelationsInDbspaceDir(), .e.g.: >> >> + if (segno == 0) >> + snprintf(dstpath, sizeof(dstpath), "%s/%u", >> + dbspacedirname, relNumber); >> + else >> + snprintf(dstpath, sizeof(dstpath), "%s/%u.%u", >> + dbspacedirname, relNumber, segno); >> >> >> If this happened three times I'd definitely want a helper function, but >> even with two I think it would be a bit nicer. > > Personally I think that would make the code harder to read rather than > easier. I agree that repeating code isn't great, but this is a > relatively brief idiom and pretty self-explanatory. If other people > agree with you I can change it, but to me it's not an improvement. Then I'm fine with it as is. >> 0002 is definitely a good idea. FWIW pgBackRest does this conversion but >> also errors if it does not succeed. We have never seen a report of this >> error happening in the wild, so I think it must be pretty rare if it >> does happen. > > Cool, but ... how about the main patch set? It's nice to get some of > these refactoring bits and pieces out of the way, but if I spend the > effort to work out what I think are the right answers to the remaining > design questions for the main patch set and then find out after I've > done all that that you have massive objections, I'm going to be > annoyed. I've been trying to get this feature into PostgreSQL for > years, and if I don't succeed this time, I want the reason to be > something better than "well, I didn't find out that David disliked X > until five minutes before I was planning to type 'git push'." I simply have not had time to look at the main patch set in any detail. > I'm not really concerned about detailed bug-hunting in the main > patches just yet. The time for that will come. But if you have views > on how to resolve the design questions that I mentioned in a couple of > emails back, or intend to advocate vigorously against the whole > concept for some reason, let's try to sort that out sooner rather than > later. In my view this feature puts the cart way before the horse. I'd think higher priority features might be parallelism, a backup repository, expiration management, archiving, or maybe even a restore command. It seems the only goal here is to make pg_basebackup a tool for external backup software to use, which might be OK, but I don't believe this feature really advances pg_basebackup as a usable piece of stand-alone software. If people really think that start/stop backup is too complicated an interface how are they supposed to track page incrementals and get them to a place where pg_combinebackup can put them backup together? If automation is required to use this feature, shouldn't pg_basebackup implement that automation? I have plenty of thoughts about the implementation as well, but I have a lot on my plate right now and I don't have time to get into it. I don't plan to stand in your way on this feature. I'm reviewing what patches I can out of courtesy and to be sure that nothing adjacent to your work is being affected. My apologies if my reviews are not meeting your expectations, but I am contributing as my time constraints allow. Regards, -David ^ permalink raw reply [nested|flat] 22+ messages in thread
* Re: trying again to get incremental backup @ 2023-10-23 15:44 Robert Haas <[email protected]> parent: David Steele <[email protected]> 0 siblings, 0 replies; 22+ messages in thread From: Robert Haas @ 2023-10-23 15:44 UTC (permalink / raw) To: David Steele <[email protected]>; +Cc: Dilip Kumar <[email protected]>; Andres Freund <[email protected]>; pgsql-hackers On Fri, Oct 20, 2023 at 11:30 AM David Steele <[email protected]> wrote: > Then I'm fine with it as is. OK, thanks. > In my view this feature puts the cart way before the horse. I'd think > higher priority features might be parallelism, a backup repository, > expiration management, archiving, or maybe even a restore command. > > It seems the only goal here is to make pg_basebackup a tool for external > backup software to use, which might be OK, but I don't believe this > feature really advances pg_basebackup as a usable piece of stand-alone > software. If people really think that start/stop backup is too > complicated an interface how are they supposed to track page > incrementals and get them to a place where pg_combinebackup can put them > backup together? If automation is required to use this feature, > shouldn't pg_basebackup implement that automation? > > I have plenty of thoughts about the implementation as well, but I have a > lot on my plate right now and I don't have time to get into it. > > I don't plan to stand in your way on this feature. I'm reviewing what > patches I can out of courtesy and to be sure that nothing adjacent to > your work is being affected. My apologies if my reviews are not meeting > your expectations, but I am contributing as my time constraints allow. Sorry, I realize reading this response that I probably didn't do a very good job writing that email and came across sounding like a jerk. Possibly, I actually am a jerk. Whether it just sounded like it or I actually am, I apologize. But your last paragraph here gets at my real question, which is whether you were going to try to block the feature. I recognize that we have different priorities when it comes to what would make most sense to implement in PostgreSQL, and that's OK, or at least, it's OK with me. I also don't have any particular expectation about how much you should review the patch or in what level of detail, and I have sincerely appreciated your feedback thus far. If you are able to continue to provide more, that's great, and if that's not, well, you're not obligated. What I was concerned about was whether that review was a precursor to a vigorous attempt to keep the main patch from getting committed, because if that was going to be the case, then I'd like to surface that conflict sooner rather than later. It sounds like that's not an issue, which is great. At the risk of drifting into the fraught question of what I *should* be implementing rather than the hopefully-less-fraught question of whether what I am actually implementing is any good, I see incremental backup as a way of removing some of the use cases for the low-level backup API. If you said "but people still will have lots of reasons to use it," I would agree; and if you said "people can still screw things up with pg_basebackup," I would also agree. Nonetheless, most of the disasters I've personally seen have stemmed from the use of the low-level API rather than from the use of pg_basebackup, though there are exceptions. I also think a lot of the use of the low-level API is driven by it being just too darn slow to copy the whole database, and incremental backup can help with that in some circumstances. Also, I have worked fairly hard to try to make sure that if you misuse pg_combinebackup, or fail to use it altogether, you'll get an error rather than silent data corruption. I would be interested to hear about scenarios where the checks that I've implemented can be defeated by something that is plausibly described as stupidity rather than malice. I'm not sure we can fix all such cases, but I'm very alert to the horror that will befall me if user error looks exactly like a bug in the code. For my own sanity, we have to be able to distinguish those cases. Moreover, we also need to be able to distinguish backup-time bugs from reassembly-time bugs, which is why I've got the pg_walsummary tool, and why pg_combinebackup has the ability to emit fairly detailed debugging output. I anticipate those things being useful in investigating bug reports when they show up. I won't be too surprised if it turns out that more work on sanity-checking and/or debugging tools is needed, but I think your concern about people misusing stuff is bang on target and I really want to do whatever we can to avoid that when possible and detect it when it happens. -- Robert Haas EDB: http://www.enterprisedb.com ^ permalink raw reply [nested|flat] 22+ messages in thread
* Re: trying again to get incremental backup @ 2023-10-23 19:34 Robert Haas <[email protected]> parent: Jakub Wartak <[email protected]> 0 siblings, 0 replies; 22+ messages in thread From: Robert Haas @ 2023-10-23 19:34 UTC (permalink / raw) To: Jakub Wartak <[email protected]>; +Cc: Dilip Kumar <[email protected]>; Andres Freund <[email protected]>; pgsql-hackers On Fri, Oct 20, 2023 at 9:20 AM Jakub Wartak <[email protected]> wrote: > Okay, so another good news - related to the patch version #4. > Not-so-tiny stress test consisting of pgbench run for 24h straight > (with incremental backups every 2h, with base of initial full backup), > followed by two PITRs (one not using incremental backup and one using > to to illustrate the performance point - and potentially spot any > errors in between). In both cases it worked fine. This is great testing, thanks. What might be even better is to test whether the resulting backups are correct, somehow. > I've just noticed one thing when recovery is progress: is > summarization working during recovery - in the background - an > expected behaviour? I'm wondering about that, because after freshly > restored and recovered DB, one would need to create a *new* full > backup and only from that point new summaries would have any use? Actually, I think you could take an incremental backup relative to a full backup from a previous timeline. But the question of what summarization ought to do (or not do) during recovery, and whether it ought to be enabled by default, and what the retention policy ought to be are very much live ones. Right now, it's enabled by default and keeps summaries for a week, assuming you don't reset your local clock and that it advances at the same speed as the universe's own clock. But that's all debatable. Any views? Meanwhile, here's a new patch set. I went ahead and committed the first two preparatory patches, as I said earlier that I intended to do. And here I've adjusted the main patch, which is now 0003, for the addition of XLOG_CHECKPOINT_REDO, which permitted me to simplify a few things. wal_summarize_mb now feels like a bit of a silly GUC -- presumably you'd never care, unless you had an absolutely gigantic inter-checkpoint WAL distance. And if you have that, maybe you should also have enough memory to summarize all that WAL. Or maybe not: perhaps it's better to write WAL summaries more than once per checkpoint when checkpoints are really big. But I'm worried that the GUC will become a source of needless confusion for users. For most people, it seems like emitting one summary per checkpoint should be totally fine, and they might prefer a simple Boolean GUC, summarize_wal = true | false, over this. I'm just not quite sure about the corner cases. -- Robert Haas EDB: http://www.enterprisedb.com Attachments: [application/octet-stream] v6-0004-Add-new-pg_walsummary-tool.patch (11.4K, ../../CA+Tgmoad7igbt46K+JHG6UEZ56Y5SVBYq5e7OjcKkgD3SStNmg@mail.gmail.com/2-v6-0004-Add-new-pg_walsummary-tool.patch) download | inline diff: From c66a2ab3cbee191f1ba0d97994b8a7a8e0086c68 Mon Sep 17 00:00:00 2001 From: Robert Haas <[email protected]> Date: Wed, 14 Jun 2023 11:31:39 -0400 Subject: [PATCH v6 4/4] Add new pg_walsummary tool. This can dump the contents of WAL summary files, either those in pg_wal/summaries, or the INCREMENTAL_BACKUP files that are part of an incremental backup proper. XXX. Needs documentation and tests. --- src/bin/Makefile | 1 + src/bin/meson.build | 1 + src/bin/pg_walsummary/.gitignore | 1 + src/bin/pg_walsummary/Makefile | 42 ++++ src/bin/pg_walsummary/meson.build | 24 +++ src/bin/pg_walsummary/pg_walsummary.c | 278 ++++++++++++++++++++++++++ 6 files changed, 347 insertions(+) create mode 100644 src/bin/pg_walsummary/.gitignore create mode 100644 src/bin/pg_walsummary/Makefile create mode 100644 src/bin/pg_walsummary/meson.build create mode 100644 src/bin/pg_walsummary/pg_walsummary.c diff --git a/src/bin/Makefile b/src/bin/Makefile index aa2210925e..f98f58d39e 100644 --- a/src/bin/Makefile +++ b/src/bin/Makefile @@ -31,6 +31,7 @@ SUBDIRS = \ pg_upgrade \ pg_verifybackup \ pg_waldump \ + pg_walsummary \ pgbench \ psql \ scripts diff --git a/src/bin/meson.build b/src/bin/meson.build index 4cb6fd59bb..d1e9ef4409 100644 --- a/src/bin/meson.build +++ b/src/bin/meson.build @@ -17,6 +17,7 @@ subdir('pg_test_timing') subdir('pg_upgrade') subdir('pg_verifybackup') subdir('pg_waldump') +subdir('pg_walsummary') subdir('pgbench') subdir('pgevent') subdir('psql') diff --git a/src/bin/pg_walsummary/.gitignore b/src/bin/pg_walsummary/.gitignore new file mode 100644 index 0000000000..d71ec192fa --- /dev/null +++ b/src/bin/pg_walsummary/.gitignore @@ -0,0 +1 @@ +pg_walsummary diff --git a/src/bin/pg_walsummary/Makefile b/src/bin/pg_walsummary/Makefile new file mode 100644 index 0000000000..852f7208f6 --- /dev/null +++ b/src/bin/pg_walsummary/Makefile @@ -0,0 +1,42 @@ +#------------------------------------------------------------------------- +# +# Makefile for src/bin/pg_walsummary +# +# Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group +# Portions Copyright (c) 1994, Regents of the University of California +# +# src/bin/pg_walsummary/Makefile +# +#------------------------------------------------------------------------- + +PGFILEDESC = "pg_walsummary - print contents of WAL summary files" +PGAPPICON=win32 + +subdir = src/bin/pg_walsummary +top_builddir = ../../.. +include $(top_builddir)/src/Makefile.global + +override CPPFLAGS := -I$(libpq_srcdir) $(CPPFLAGS) +LDFLAGS_INTERNAL += -L$(top_builddir)/src/fe_utils -lpgfeutils + +OBJS = \ + $(WIN32RES) \ + pg_walsummary.o + +all: pg_walsummary + +pg_walsummary: $(OBJS) | submake-libpgport submake-libpgfeutils + $(CC) $(CFLAGS) $^ $(LDFLAGS) $(LDFLAGS_EX) $(LIBS) -o $@$(X) + + +install: all installdirs + $(INSTALL_PROGRAM) pg_walsummary$(X) '$(DESTDIR)$(bindir)/pg_walsummary$(X)' + +installdirs: + $(MKDIR_P) '$(DESTDIR)$(bindir)' + +uninstall: + rm -f '$(DESTDIR)$(bindir)/pg_walsummary$(X)' + +clean distclean maintainer-clean: + rm -f pg_walsummary$(X) $(OBJS) diff --git a/src/bin/pg_walsummary/meson.build b/src/bin/pg_walsummary/meson.build new file mode 100644 index 0000000000..c2092960c6 --- /dev/null +++ b/src/bin/pg_walsummary/meson.build @@ -0,0 +1,24 @@ +# Copyright (c) 2022-2023, PostgreSQL Global Development Group + +pg_walsummary_sources = files( + 'pg_walsummary.c', +) + +if host_system == 'windows' + pg_walsummary_sources += rc_bin_gen.process(win32ver_rc, extra_args: [ + '--NAME', 'pg_walsummary', + '--FILEDESC', 'pg_walsummary - print contents of WAL summary files',]) +endif + +pg_walsummary = executable('pg_walsummary', + pg_walsummary_sources, + dependencies: [frontend_code], + kwargs: default_bin_args, +) +bin_targets += pg_walsummary + +tests += { + 'name': 'pg_walsummary', + 'sd': meson.current_source_dir(), + 'bd': meson.current_build_dir() +} diff --git a/src/bin/pg_walsummary/pg_walsummary.c b/src/bin/pg_walsummary/pg_walsummary.c new file mode 100644 index 0000000000..0304a42026 --- /dev/null +++ b/src/bin/pg_walsummary/pg_walsummary.c @@ -0,0 +1,278 @@ +/*------------------------------------------------------------------------- + * + * pg_walsummary.c + * Prints the contents of WAL summary files. + * + * Copyright (c) 2017-2023, PostgreSQL Global Development Group + * + * IDENTIFICATION + * src/bin/pg_walsummary/pg_walsummary.c + * + *------------------------------------------------------------------------- + */ +#include "postgres_fe.h" + +#include <fcntl.h> +#include <limits.h> + +#include "common/blkreftable.h" +#include "common/logging.h" +#include "fe_utils/option_utils.h" +#include "lib/stringinfo.h" +#include "getopt_long.h" + +typedef struct ws_options +{ + bool individual; + bool quiet; +} ws_options; + +typedef struct ws_file_info +{ + int fd; + char *filename; +} ws_file_info; + +static BlockNumber *block_buffer = NULL; +static unsigned block_buffer_size = 512; /* Initial size. */ + +static void dump_one_relation(ws_options *opt, RelFileLocator *rlocator, + ForkNumber forknum, BlockNumber limit_block, + BlockRefTableReader *reader); +static void help(const char *progname); +static int compare_block_numbers(const void *a, const void *b); +static int walsummary_read_callback(void *callback_arg, void *data, + int length); +static void walsummary_error_callback(void *callback_arg, char *fmt,...); + +/* + * Main program. + */ +int +main(int argc, char *argv[]) +{ + static struct option long_options[] = { + {"individual", no_argument, NULL, 'i'}, + {"quiet", no_argument, NULL, 'q'}, + {NULL, 0, NULL, 0} + }; + + const char *progname; + int optindex; + int c; + ws_options opt; + + pg_logging_init(argv[0]); + progname = get_progname(argv[0]); + handle_help_version_opts(argc, argv, progname, help); + + /* process command-line options */ + while ((c = getopt_long(argc, argv, "f:iqw:", + long_options, &optindex)) != -1) + { + switch (c) + { + case 'i': + opt.individual = true; + break; + case 'q': + opt.quiet = true; + break; + default: + /* getopt_long already emitted a complaint */ + pg_log_error_hint("Try \"%s --help\" for more information.", progname); + exit(1); + } + } + + if (optind >= argc) + { + pg_log_error("%s: no input files specified", progname); + pg_log_error_hint("Try \"%s --help\" for more information.", progname); + exit(1); + } + + while (optind < argc) + { + ws_file_info ws; + BlockRefTableReader *reader; + RelFileLocator rlocator; + ForkNumber forknum; + BlockNumber limit_block; + + ws.filename = argv[optind++]; + if ((ws.fd = open(ws.filename, O_RDONLY | PG_BINARY, 0)) < 0) + pg_fatal("could not open file \"%s\": %m", ws.filename); + + reader = CreateBlockRefTableReader(walsummary_read_callback, &ws, + ws.filename, + walsummary_error_callback, NULL); + while (BlockRefTableReaderNextRelation(reader, &rlocator, &forknum, + &limit_block)) + dump_one_relation(&opt, &rlocator, forknum, limit_block, reader); + + DestroyBlockRefTableReader(reader); + close(ws.fd); + } + + exit(0); +} + +/* + * Dump details for one relation. + */ +static void +dump_one_relation(ws_options *opt, RelFileLocator *rlocator, + ForkNumber forknum, BlockNumber limit_block, + BlockRefTableReader *reader) +{ + unsigned i = 0; + unsigned nblocks; + BlockNumber startblock = InvalidBlockNumber; + BlockNumber endblock = InvalidBlockNumber; + + /* Dump limit block, if any. */ + if (limit_block != InvalidBlockNumber) + printf("TS %u, DB %u, REL %u, FORK %s: limit %u\n", + rlocator->spcOid, rlocator->dbOid, rlocator->relNumber, + forkNames[forknum], limit_block); + + /* If we haven't allocated a block buffer yet, do that now. */ + if (block_buffer == NULL) + block_buffer = palloc_array(BlockNumber, block_buffer_size); + + /* Try to fill the block buffer. */ + nblocks = BlockRefTableReaderGetBlocks(reader, + block_buffer, + block_buffer_size); + + /* If we filled the block buffer completely, we must enlarge it. */ + while (nblocks >= block_buffer_size) + { + unsigned new_size; + + /* Double the size, being careful about overflow. */ + new_size = block_buffer_size * 2; + if (new_size < block_buffer_size) + new_size = PG_UINT32_MAX; + block_buffer = repalloc_array(block_buffer, BlockNumber, new_size); + + /* Try to fill the newly-allocated space. */ + nblocks += + BlockRefTableReaderGetBlocks(reader, + block_buffer + block_buffer_size, + new_size - block_buffer_size); + + /* Save the new size for later calls. */ + block_buffer_size = new_size; + } + + /* If we don't need to produce any output, skip the rest of this. */ + if (opt->quiet) + return; + + /* + * Sort the returned block numbers. If the block reference table was using + * the bitmap representation for a given chunk, the block numbers in that + * chunk will already be sorted, but when the array-of-offsets + * representation is used, we can receive block numbers here out of order. + */ + qsort(block_buffer, nblocks, sizeof(BlockNumber), compare_block_numbers); + + /* Dump block references. */ + while (i < nblocks) + { + /* + * Find the next range of blocks to print, but if --individual was + * specified, then consider each block a separate range. + */ + startblock = endblock = block_buffer[i++]; + if (!opt->individual) + { + while (i < nblocks && block_buffer[i] == endblock + 1) + { + endblock++; + i++; + } + } + + /* Format this range of block numbers as a string. */ + if (startblock == endblock) + printf("TS %u, DB %u, REL %u, FORK %s: block %u\n", + rlocator->spcOid, rlocator->dbOid, rlocator->relNumber, + forkNames[forknum], startblock); + else + printf("TS %u, DB %u, REL %u, FORK %s: blocks %u..%u\n", + rlocator->spcOid, rlocator->dbOid, rlocator->relNumber, + forkNames[forknum], startblock, endblock); + } +} + +/* + * Quicksort comparator for block numbers. + */ +static int +compare_block_numbers(const void *a, const void *b) +{ + BlockNumber aa = *(BlockNumber *) a; + BlockNumber bb = *(BlockNumber *) b; + + if (aa > bb) + return 1; + else if (aa == bb) + return 0; + else + return -1; +} + +/* + * Error callback. + */ +void +walsummary_error_callback(void *callback_arg, char *fmt,...) +{ + va_list ap; + + va_start(ap, fmt); + pg_log_generic_v(PG_LOG_ERROR, PG_LOG_PRIMARY, fmt, ap); + va_end(ap); + + exit(1); +} + +/* + * Read callback. + */ +int +walsummary_read_callback(void *callback_arg, void *data, int length) +{ + ws_file_info *ws = callback_arg; + int rc; + + if ((rc = read(ws->fd, data, length)) < 0) + pg_fatal("could not read file \"%s\": %m", ws->filename); + + return rc; +} + +/* + * help + * + * Prints help page for the program + * + * progname: the name of the executed program, such as "pg_walsummary" + */ +static void +help(const char *progname) +{ + printf(_("%s prints the contents of a WAL summary file.\n\n"), progname); + printf(_("Usage:\n")); + printf(_(" %s [OPTION]... FILE...\n"), progname); + printf(_("\nOptions:\n")); + printf(_(" -i, --individual list block numbers individually, not as ranges\n")); + printf(_(" -q, --quiet don't print anything, just parse the files\n")); + printf(_(" -?, --help show this help, then exit\n")); + + printf(_("\nReport bugs to <%s>.\n"), PACKAGE_BUGREPORT); + printf(_("%s home page: <%s>\n"), PACKAGE_NAME, PACKAGE_URL); +} -- 2.37.1 (Apple Git-137.1) [application/octet-stream] v6-0002-Move-src-bin-pg_verifybackup-parse_manifest.c-int.patch (4.3K, ../../CA+Tgmoad7igbt46K+JHG6UEZ56Y5SVBYq5e7OjcKkgD3SStNmg@mail.gmail.com/3-v6-0002-Move-src-bin-pg_verifybackup-parse_manifest.c-int.patch) download | inline diff: From fac0a392b62254066300c051b077ef78a9d4cbcb Mon Sep 17 00:00:00 2001 From: Robert Haas <[email protected]> Date: Tue, 3 Oct 2023 13:32:45 -0400 Subject: [PATCH v6 2/4] Move src/bin/pg_verifybackup/parse_manifest.c into src/common. This makes it possible for the code to be easily reused by other client-side tools, and/or by the server. --- src/bin/pg_verifybackup/Makefile | 1 - src/bin/pg_verifybackup/meson.build | 1 - src/bin/pg_verifybackup/pg_verifybackup.c | 2 +- src/common/Makefile | 1 + src/common/meson.build | 1 + src/{bin/pg_verifybackup => common}/parse_manifest.c | 4 ++-- src/{bin/pg_verifybackup => include/common}/parse_manifest.h | 2 +- 7 files changed, 6 insertions(+), 6 deletions(-) rename src/{bin/pg_verifybackup => common}/parse_manifest.c (99%) rename src/{bin/pg_verifybackup => include/common}/parse_manifest.h (97%) diff --git a/src/bin/pg_verifybackup/Makefile b/src/bin/pg_verifybackup/Makefile index 596df15118..8f04fa662c 100644 --- a/src/bin/pg_verifybackup/Makefile +++ b/src/bin/pg_verifybackup/Makefile @@ -21,7 +21,6 @@ LDFLAGS_INTERNAL += -L$(top_builddir)/src/fe_utils -lpgfeutils $(libpq_pgport) OBJS = \ $(WIN32RES) \ - parse_manifest.o \ pg_verifybackup.o all: pg_verifybackup diff --git a/src/bin/pg_verifybackup/meson.build b/src/bin/pg_verifybackup/meson.build index 9369da1bc6..58f780d1a6 100644 --- a/src/bin/pg_verifybackup/meson.build +++ b/src/bin/pg_verifybackup/meson.build @@ -1,7 +1,6 @@ # Copyright (c) 2022-2023, PostgreSQL Global Development Group pg_verifybackup_sources = files( - 'parse_manifest.c', 'pg_verifybackup.c' ) diff --git a/src/bin/pg_verifybackup/pg_verifybackup.c b/src/bin/pg_verifybackup/pg_verifybackup.c index 059836f0e6..ce423a03d4 100644 --- a/src/bin/pg_verifybackup/pg_verifybackup.c +++ b/src/bin/pg_verifybackup/pg_verifybackup.c @@ -20,9 +20,9 @@ #include "common/hashfn.h" #include "common/logging.h" +#include "common/parse_manifest.h" #include "fe_utils/simple_list.h" #include "getopt_long.h" -#include "parse_manifest.h" #include "pgtime.h" /* diff --git a/src/common/Makefile b/src/common/Makefile index 70884be00c..3c8effc533 100644 --- a/src/common/Makefile +++ b/src/common/Makefile @@ -66,6 +66,7 @@ OBJS_COMMON = \ kwlookup.o \ link-canary.o \ md5_common.o \ + parse_manifest.o \ percentrepl.o \ pg_get_line.o \ pg_lzcompress.o \ diff --git a/src/common/meson.build b/src/common/meson.build index ae05ac63cf..aa646f96a3 100644 --- a/src/common/meson.build +++ b/src/common/meson.build @@ -18,6 +18,7 @@ common_sources = files( 'kwlookup.c', 'link-canary.c', 'md5_common.c', + 'parse_manifest.c', 'percentrepl.c', 'pg_get_line.c', 'pg_lzcompress.c', diff --git a/src/bin/pg_verifybackup/parse_manifest.c b/src/common/parse_manifest.c similarity index 99% rename from src/bin/pg_verifybackup/parse_manifest.c rename to src/common/parse_manifest.c index f0acd9f1e7..9895f2f17d 100644 --- a/src/bin/pg_verifybackup/parse_manifest.c +++ b/src/common/parse_manifest.c @@ -6,15 +6,15 @@ * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group * Portions Copyright (c) 1994, Regents of the University of California * - * src/bin/pg_verifybackup/parse_manifest.c + * src/common/parse_manifest.c * *------------------------------------------------------------------------- */ #include "postgres_fe.h" -#include "parse_manifest.h" #include "common/jsonapi.h" +#include "common/parse_manifest.h" /* * Semantic states for JSON manifest parsing. diff --git a/src/bin/pg_verifybackup/parse_manifest.h b/src/include/common/parse_manifest.h similarity index 97% rename from src/bin/pg_verifybackup/parse_manifest.h rename to src/include/common/parse_manifest.h index 7387a917a2..7b24c5d785 100644 --- a/src/bin/pg_verifybackup/parse_manifest.h +++ b/src/include/common/parse_manifest.h @@ -6,7 +6,7 @@ * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group * Portions Copyright (c) 1994, Regents of the University of California * - * src/bin/pg_verifybackup/parse_manifest.h + * src/include/common/parse_manifest.h * *------------------------------------------------------------------------- */ -- 2.37.1 (Apple Git-137.1) [application/octet-stream] v6-0001-Change-how-a-base-backup-decides-which-files-have.patch (11.4K, ../../CA+Tgmoad7igbt46K+JHG6UEZ56Y5SVBYq5e7OjcKkgD3SStNmg@mail.gmail.com/4-v6-0001-Change-how-a-base-backup-decides-which-files-have.patch) download | inline diff: From 621ea9af483466cbf08cbcca10a4650c2518f235 Mon Sep 17 00:00:00 2001 From: Robert Haas <[email protected]> Date: Tue, 3 Oct 2023 13:32:28 -0400 Subject: [PATCH v6 1/4] Change how a base backup decides which files have checksums. Previously, it thought that any plain file located under global, base, or a tablespace directory had checksums unless it was in a short list of excluded files. Now, it thinks that files in those directories have checksums if parse_filename_for_nontemp_relation says that they are relation files. (Temporary relation files don't matter because they're excluded from the backup anyway.) This changes the behavior if you have stray files not managed by PostgreSQL in the relevant directories. Previously, you'd get some kind of checksum-related complaint if such files existed, assuming that the cluster had checksums enabled and that the base backup wasn't run with NOVERIFY_CHECKSUMS. Now, you won't get those complaints any more. That seems like an improvement to me, because those files were presumably not created by PostgreSQL and so there is no reason to think that they would be checksummed like a PostgreSQL relation file. (If we want to complain about such files, we should complain about them existing at all, not just about their checksums.) The point of this change is to make the code more consistent. sendDir() was already calling parse_filename_for_nontemp_relation() as part of an effort to determine which files to include in the backup. So, it already had the information about whether a certain file was a relation file. sendFile() then used a separate method, embodied in is_checksummed_file(), to make what is essentially the same determination. It's better not to make the same decision using two different methods, especially in closely-related code. --- src/backend/backup/basebackup.c | 172 ++++++++++---------------------- 1 file changed, 55 insertions(+), 117 deletions(-) diff --git a/src/backend/backup/basebackup.c b/src/backend/backup/basebackup.c index b537f46219..4ba63ad8a6 100644 --- a/src/backend/backup/basebackup.c +++ b/src/backend/backup/basebackup.c @@ -82,7 +82,8 @@ static int64 sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeo backup_manifest_info *manifest, Oid spcoid); static bool sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, struct stat *statbuf, bool missing_ok, - Oid dboid, Oid spcoid, + Oid dboid, Oid spcoid, RelFileNumber relfilenumber, + unsigned segno, backup_manifest_info *manifest); static off_t read_file_data_into_buffer(bbsink *sink, const char *readfilename, int fd, @@ -104,7 +105,6 @@ static void convert_link_to_directory(const char *pathbuf, struct stat *statbuf) static void perform_base_backup(basebackup_options *opt, bbsink *sink); static void parse_basebackup_options(List *options, basebackup_options *opt); static int compareWalFileNames(const ListCell *a, const ListCell *b); -static bool is_checksummed_file(const char *fullpath, const char *filename); static int basebackup_read_file(int fd, char *buf, size_t nbytes, off_t offset, const char *filename, bool partial_read_ok); @@ -213,23 +213,6 @@ static const struct exclude_list_item excludeFiles[] = {NULL, false} }; -/* - * List of files excluded from checksum validation. - * - * Note: this list should be kept in sync with what pg_checksums.c - * includes. - */ -static const struct exclude_list_item noChecksumFiles[] = { - {"pg_control", false}, - {"pg_filenode.map", false}, - {"pg_internal.init", true}, - {"PG_VERSION", false}, -#ifdef EXEC_BACKEND - {"config_exec_params", true}, -#endif - {NULL, false} -}; - /* * Actually do a base backup for the specified tablespaces. * @@ -356,7 +339,8 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) errmsg("could not stat file \"%s\": %m", XLOG_CONTROL_FILE))); sendFile(sink, XLOG_CONTROL_FILE, XLOG_CONTROL_FILE, &statbuf, - false, InvalidOid, InvalidOid, &manifest); + false, InvalidOid, InvalidOid, + InvalidRelFileNumber, 0, &manifest); } else { @@ -625,7 +609,8 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) errmsg("could not stat file \"%s\": %m", pathbuf))); sendFile(sink, pathbuf, pathbuf, &statbuf, false, - InvalidOid, InvalidOid, &manifest); + InvalidOid, InvalidOid, InvalidRelFileNumber, 0, + &manifest); /* unconditionally mark file as archived */ StatusFilePath(pathbuf, fname, ".done"); @@ -1163,7 +1148,8 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, struct stat statbuf; int64 size = 0; const char *lastDir; /* Split last dir from parent path. */ - bool isDbDir = false; /* Does this directory contain relations? */ + bool isRelationDir = false; /* Does directory contain relations? */ + Oid dboid = InvalidOid; /* * Determine if the current path is a database directory that can contain @@ -1190,17 +1176,23 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, strncmp(lastDir - (sizeof(TABLESPACE_VERSION_DIRECTORY) - 1), TABLESPACE_VERSION_DIRECTORY, sizeof(TABLESPACE_VERSION_DIRECTORY) - 1) == 0)) - isDbDir = true; + { + isRelationDir = true; + dboid = atooid(lastDir + 1); + } } + else if (strcmp(path, "./global") == 0) + isRelationDir = true; dir = AllocateDir(path); while ((de = ReadDir(dir, path)) != NULL) { int excludeIdx; bool excludeFound; - RelFileNumber relNumber; - ForkNumber relForkNum; - unsigned segno; + RelFileNumber relfilenumber = InvalidRelFileNumber; + ForkNumber relForkNum = InvalidForkNumber; + unsigned segno = 0; + bool isRelationFile = false; /* Skip special stuff */ if (strcmp(de->d_name, ".") == 0 || strcmp(de->d_name, "..") == 0) @@ -1248,37 +1240,40 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, if (excludeFound) continue; + /* + * If there could be non-temporary relation files in this directory, + * try to parse the filename. + */ + if (isRelationDir) + isRelationFile = + parse_filename_for_nontemp_relation(de->d_name, + &relfilenumber, + &relForkNum, &segno); + /* Exclude all forks for unlogged tables except the init fork */ - if (isDbDir && - parse_filename_for_nontemp_relation(de->d_name, &relNumber, - &relForkNum, &segno)) + if (isRelationFile && relForkNum != INIT_FORKNUM) { - /* Never exclude init forks */ - if (relForkNum != INIT_FORKNUM) - { - char initForkFile[MAXPGPATH]; + char initForkFile[MAXPGPATH]; - /* - * If any other type of fork, check if there is an init fork - * with the same RelFileNumber. If so, the file can be - * excluded. - */ - snprintf(initForkFile, sizeof(initForkFile), "%s/%u_init", - path, relNumber); + /* + * If any other type of fork, check if there is an init fork with + * the same RelFileNumber. If so, the file can be excluded. + */ + snprintf(initForkFile, sizeof(initForkFile), "%s/%u_init", + path, relfilenumber); - if (lstat(initForkFile, &statbuf) == 0) - { - elog(DEBUG2, - "unlogged relation file \"%s\" excluded from backup", - de->d_name); + if (lstat(initForkFile, &statbuf) == 0) + { + elog(DEBUG2, + "unlogged relation file \"%s\" excluded from backup", + de->d_name); - continue; - } + continue; } } /* Exclude temporary relations */ - if (isDbDir && looks_like_temp_rel_name(de->d_name)) + if (OidIsValid(dboid) && looks_like_temp_rel_name(de->d_name)) { elog(DEBUG2, "temporary relation file \"%s\" excluded from backup", @@ -1417,8 +1412,8 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, if (!sizeonly) sent = sendFile(sink, pathbuf, pathbuf + basepathlen + 1, &statbuf, - true, isDbDir ? atooid(lastDir + 1) : InvalidOid, spcoid, - manifest); + true, dboid, spcoid, + relfilenumber, segno, manifest); if (sent || sizeonly) { @@ -1440,40 +1435,6 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, return size; } -/* - * Check if a file should have its checksum validated. - * We validate checksums on files in regular tablespaces - * (including global and default) only, and in those there - * are some files that are explicitly excluded. - */ -static bool -is_checksummed_file(const char *fullpath, const char *filename) -{ - /* Check that the file is in a tablespace */ - if (strncmp(fullpath, "./global/", 9) == 0 || - strncmp(fullpath, "./base/", 7) == 0 || - strncmp(fullpath, "/", 1) == 0) - { - int excludeIdx; - - /* Compare file against noChecksumFiles skip list */ - for (excludeIdx = 0; noChecksumFiles[excludeIdx].name != NULL; excludeIdx++) - { - int cmplen = strlen(noChecksumFiles[excludeIdx].name); - - if (!noChecksumFiles[excludeIdx].match_prefix) - cmplen++; - if (strncmp(filename, noChecksumFiles[excludeIdx].name, - cmplen) == 0) - return false; - } - - return true; - } - else - return false; -} - /* * Given the member, write the TAR header & send the file. * @@ -1488,6 +1449,7 @@ is_checksummed_file(const char *fullpath, const char *filename) static bool sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, struct stat *statbuf, bool missing_ok, Oid dboid, Oid spcoid, + RelFileNumber relfilenumber, unsigned segno, backup_manifest_info *manifest) { int fd; @@ -1495,8 +1457,6 @@ sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, int checksum_failures = 0; off_t cnt; pgoff_t bytes_done = 0; - int segmentno = 0; - char *segmentpath; bool verify_checksum = false; pg_checksum_context checksum_ctx; @@ -1522,36 +1482,14 @@ sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, */ Assert((sink->bbs_buffer_length % BLCKSZ) == 0); - if (!noverify_checksums && DataChecksumsEnabled()) - { - char *filename; - - /* - * Get the filename (excluding path). As last_dir_separator() - * includes the last directory separator, we chop that off by - * incrementing the pointer. - */ - filename = last_dir_separator(readfilename) + 1; - - if (is_checksummed_file(readfilename, filename)) - { - verify_checksum = true; - - /* - * Cut off at the segment boundary (".") to get the segment number - * in order to mix it into the checksum. - */ - segmentpath = strstr(filename, "."); - if (segmentpath != NULL) - { - segmentno = atoi(segmentpath + 1); - if (segmentno == 0) - ereport(ERROR, - (errmsg("invalid segment number %d in file \"%s\"", - segmentno, filename))); - } - } - } + /* + * If we weren't told not to verify checksums, and if checksums are + * enabled for this cluster, and if this is a relation file, then verify + * the checksum. + */ + if (!noverify_checksums && DataChecksumsEnabled() && + RelFileNumberIsValid(relfilenumber)) + verify_checksum = true; /* * Loop until we read the amount of data the caller told us to expect. The @@ -1566,7 +1504,7 @@ sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, /* Try to read some more data. */ cnt = read_file_data_into_buffer(sink, readfilename, fd, bytes_done, remaining, - blkno + segmentno * RELSEG_SIZE, + blkno + segno * RELSEG_SIZE, verify_checksum, &checksum_failures); -- 2.37.1 (Apple Git-137.1) [application/octet-stream] v6-0003-Prototype-patch-for-incremental-backup.patch (333.5K, ../../CA+Tgmoad7igbt46K+JHG6UEZ56Y5SVBYq5e7OjcKkgD3SStNmg@mail.gmail.com/5-v6-0003-Prototype-patch-for-incremental-backup.patch) download | inline diff: From a381fdbb31ba2752f89b64dd46506fb530cc0355 Mon Sep 17 00:00:00 2001 From: Robert Haas <[email protected]> Date: Wed, 14 Jun 2023 11:31:29 -0400 Subject: [PATCH v6 3/4] Prototype patch for incremental backup. We don't differentiate between incremental and differential backups; an incremental backup can be based either on a full backup or on a previous incremental backup. This adds a new background process, the WAL summarizer, whose behavor is governed by new GUCs wal_summarize_mb and wal_summarize_keep_time. This writes out WAL summary files to $PGDATA/pg_wal/summaries. Each summary file contains information for a certain range of LSNs on a certain TLI. For each relation, it stores a "limit block" which is 0 if a relation is created or destroyed within a certain range of WAL records, or otherwise the shortest length to which the relation was truncated during that range of WAL records, or otherwise InvalidBlockNumber. In addition, it stores any blocks which have been modified during that range of WAL records, but excluding blocks which were removed by truncation after they were modified and which were never modified thereafter. In other words, it tells us which blocks need to copied in case of an incremental backup covering that range of WAL records. To take an incremental backup, you use the new replication command UPLOAD_MANIFEST to upload the manifest for the prior backup. This prior backup could either be a full backup or another incremental backup. You then use BASE_BACKUP with the INCREMENTAL option to take the backup. pg_basebackup now has an --incremental=PATH_TO_MANIFEST option to trigger this behavior. An incremental backup is like a regular full backup except that some relation files are replaced with files with names like INCREMENTAL.${ORIGINAL_NAME}, and the backup_label file contains additional lines identifying it as an incremental backup. The new pg_combinebackup tool can be used to reconstruct a data directory from a full backup and a series of incremental backups. Open issues: - Should we remove wal_summarize_mb, or replace it with a Boolean on/off switch? - How should we control generation and retention of summary files? What should be the defaults? - Needs to be tested on a standby. - Should we send the whole backup manifest to the server or, say, just an LSN? - Should the timeout when waiting for WAL summaries be configurable? If it is, then the maximum sleep time for the WAL summarizer needs to vary accordingly. - It would be nice (but not essential) to do something about incremental JSON parsing. - Might need more tests. Patch by me. Thanks to Dilip Kumar and Andres Freund for some helpful design discussions. Reviewed by Dilip Kumar and Jakub Wartak. --- doc/src/sgml/backup.sgml | 89 +- doc/src/sgml/ref/allfiles.sgml | 1 + doc/src/sgml/ref/pg_basebackup.sgml | 37 +- doc/src/sgml/ref/pg_combinebackup.sgml | 228 +++ doc/src/sgml/reference.sgml | 1 + src/backend/access/transam/xlog.c | 100 +- src/backend/access/transam/xlogbackup.c | 10 + src/backend/access/transam/xlogrecovery.c | 6 + src/backend/backup/Makefile | 5 +- src/backend/backup/basebackup.c | 334 +++- src/backend/backup/basebackup_incremental.c | 873 +++++++++++ src/backend/backup/meson.build | 3 + src/backend/backup/walsummary.c | 356 +++++ src/backend/backup/walsummaryfuncs.c | 169 ++ src/backend/postmaster/Makefile | 1 + src/backend/postmaster/auxprocess.c | 8 + src/backend/postmaster/meson.build | 1 + src/backend/postmaster/postmaster.c | 53 + src/backend/postmaster/walsummarizer.c | 1363 +++++++++++++++++ src/backend/replication/repl_gram.y | 14 +- src/backend/replication/repl_scanner.l | 2 + src/backend/replication/walsender.c | 162 +- src/backend/storage/ipc/ipci.c | 3 + src/backend/storage/lmgr/lwlocknames.txt | 1 + src/backend/utils/activity/pgstat_io.c | 4 +- .../utils/activity/wait_event_names.txt | 5 + src/backend/utils/init/miscinit.c | 3 + src/backend/utils/misc/guc_tables.c | 29 + src/backend/utils/misc/postgresql.conf.sample | 5 + src/bin/Makefile | 1 + src/bin/initdb/initdb.c | 1 + src/bin/meson.build | 1 + src/bin/pg_basebackup/bbstreamer_file.c | 1 + src/bin/pg_basebackup/pg_basebackup.c | 110 +- src/bin/pg_basebackup/t/010_pg_basebackup.pl | 4 +- src/bin/pg_combinebackup/.gitignore | 1 + src/bin/pg_combinebackup/Makefile | 52 + src/bin/pg_combinebackup/backup_label.c | 281 ++++ src/bin/pg_combinebackup/backup_label.h | 29 + src/bin/pg_combinebackup/copy_file.c | 169 ++ src/bin/pg_combinebackup/copy_file.h | 19 + src/bin/pg_combinebackup/load_manifest.c | 245 +++ src/bin/pg_combinebackup/load_manifest.h | 67 + src/bin/pg_combinebackup/meson.build | 35 + src/bin/pg_combinebackup/pg_combinebackup.c | 1270 +++++++++++++++ src/bin/pg_combinebackup/reconstruct.c | 618 ++++++++ src/bin/pg_combinebackup/reconstruct.h | 32 + src/bin/pg_combinebackup/t/001_basic.pl | 23 + .../pg_combinebackup/t/002_compare_backups.pl | 154 ++ src/bin/pg_combinebackup/write_manifest.c | 293 ++++ src/bin/pg_combinebackup/write_manifest.h | 33 + src/bin/pg_resetwal/pg_resetwal.c | 36 + src/common/Makefile | 1 + src/common/blkreftable.c | 1309 ++++++++++++++++ src/common/meson.build | 1 + src/include/access/xlog.h | 1 + src/include/access/xlogbackup.h | 2 + src/include/backup/basebackup.h | 5 +- src/include/backup/basebackup_incremental.h | 56 + src/include/backup/walsummary.h | 49 + src/include/catalog/pg_proc.dat | 19 + src/include/common/blkreftable.h | 120 ++ src/include/miscadmin.h | 3 + src/include/nodes/replnodes.h | 9 + src/include/postmaster/walsummarizer.h | 31 + src/include/storage/proc.h | 9 +- src/include/utils/guc_tables.h | 1 + src/test/perl/PostgreSQL/Test/Cluster.pm | 21 +- src/test/recovery/t/001_stream_rep.pl | 2 + src/test/recovery/t/019_replslot_limit.pl | 3 + .../t/035_standby_logical_decoding.pl | 1 + src/tools/pgindent/typedefs.list | 23 + 72 files changed, 8937 insertions(+), 70 deletions(-) create mode 100644 doc/src/sgml/ref/pg_combinebackup.sgml create mode 100644 src/backend/backup/basebackup_incremental.c create mode 100644 src/backend/backup/walsummary.c create mode 100644 src/backend/backup/walsummaryfuncs.c create mode 100644 src/backend/postmaster/walsummarizer.c create mode 100644 src/bin/pg_combinebackup/.gitignore create mode 100644 src/bin/pg_combinebackup/Makefile create mode 100644 src/bin/pg_combinebackup/backup_label.c create mode 100644 src/bin/pg_combinebackup/backup_label.h create mode 100644 src/bin/pg_combinebackup/copy_file.c create mode 100644 src/bin/pg_combinebackup/copy_file.h create mode 100644 src/bin/pg_combinebackup/load_manifest.c create mode 100644 src/bin/pg_combinebackup/load_manifest.h create mode 100644 src/bin/pg_combinebackup/meson.build create mode 100644 src/bin/pg_combinebackup/pg_combinebackup.c create mode 100644 src/bin/pg_combinebackup/reconstruct.c create mode 100644 src/bin/pg_combinebackup/reconstruct.h create mode 100644 src/bin/pg_combinebackup/t/001_basic.pl create mode 100644 src/bin/pg_combinebackup/t/002_compare_backups.pl create mode 100644 src/bin/pg_combinebackup/write_manifest.c create mode 100644 src/bin/pg_combinebackup/write_manifest.h create mode 100644 src/common/blkreftable.c create mode 100644 src/include/backup/basebackup_incremental.h create mode 100644 src/include/backup/walsummary.h create mode 100644 src/include/common/blkreftable.h create mode 100644 src/include/postmaster/walsummarizer.h diff --git a/doc/src/sgml/backup.sgml b/doc/src/sgml/backup.sgml index 8cb24d6ae5..b3468eea3c 100644 --- a/doc/src/sgml/backup.sgml +++ b/doc/src/sgml/backup.sgml @@ -857,12 +857,79 @@ test ! -f /mnt/server/archivedir/00000001000000A900000065 && cp pg_wal/0 </para> </sect2> + <sect2 id="backup-incremental-backup"> + <title>Making an Incremental Backup</title> + + <para> + You can use <xref linkend="app-pgbasebackup"/> to take an incremental + backup by specifying the <literal>--incremental</literal> option. You must + supply, as an argument to <literal>--incremental</literal>, the backup + manifest to an earlier backup from the same server. In the resulting + backup, non-relation files will be included in their entirety, but some + relation files may be replaced by smaller incremental files which contain + only the blocks which have been changed since the earlier backup and enough + metadata to reconstruct the current version of the file. + </para> + + <para> + To figure out which blocks need to be backed up, the server uses WAL + summaries, which are stored in the data directory, inside the directory + <literal>pg_wal/summaries</literal>. If the required summary files are not + present, an attempt to take an incremental backup will fail. The summaries + present in this directory must cover all LSNs from the start LSN of the + prior backup to the start LSN of the current backup. Since the server looks + for WAL summaries just after establishing the start LSN of the current + backup, the necessary summary files probably won't be instantly present + on disk, but the server will wait for any missing files to show up. + This also helps if the WAL summarization process has fallen behind. + However, if the necessary files have already been removed, or if the WAL + summarizer doesn't catch up quickly enough, the incremental backup will + fail. + </para> + + <para> + When restoring an incremental backup, it will be necessary to have not + only the incremental backup itself but also all earlier backups that + are required to supply the blocks omitted from the incremental backup. + See <xref linkend="app-pgcombinebackup"/> for further information about + this requirement. + </para> + + <para> + Note that all of the requirements for making use of a full backup also + apply to an incremental backup. For instance, you still need all of the + WAL segment files generated during and after the file system backup, and + any relevant WAL history files. And you still need to create a + <literal>recovery.signal</literal> (or <literal>standby.signal</literal>) + and perform recovery, as described in + <xref linkend="backup-pitr-recovery" />. The requirement to have earlier + backups available at restore time and to use + <literal>pg_combinebackup</literal> is an additional requirement on top of + everything else. Keep in mind that <application>PostgreSQL</application> + has no built-in mechanism to figure out which backups are still needed as + a basis for restoring later incremental backups. You must keep track of + the relationships between your full and incremental backups on your own, + and be certain not to remove earlier backups if they might be needed when + restoring later incremental backups. + </para> + + <para> + Incremental backups typically only make sense for relatively large + databases where a significant portion of the data does not change, or only + changes slowly. For a small database, it's simpler to ignore the existence + of incremental backups and simply take full backups, which are simpler + to manage. For a large database all of which is heavily modified, + incremental backups won't be much smaller than full backups. + </para> + </sect2> + <sect2 id="backup-lowlevel-base-backup"> <title>Making a Base Backup Using the Low Level API</title> <para> - The procedure for making a base backup using the low level - APIs contains a few more steps than - the <xref linkend="app-pgbasebackup"/> method, but is relatively + Instead of taking a full or incremental base backup using + <xref linkend="app-pgbasebackup"/>, you can take a base backup using the + low-level API. This procedure contains a few more steps than + the <application>pg_basebackup</application> method, but is relatively simple. It is very important that these steps are executed in sequence, and that the success of a step is verified before proceeding to the next step. @@ -1118,7 +1185,8 @@ SELECT * FROM pg_backup_stop(wait_for_archive => true); </listitem> <listitem> <para> - Restore the database files from your file system backup. Be sure that they + If you're restoring a full backup, you can restore the database files + directly into the target directories. Be sure that they are restored with the right ownership (the database system user, not <literal>root</literal>!) and with the right permissions. If you are using tablespaces, @@ -1126,6 +1194,19 @@ SELECT * FROM pg_backup_stop(wait_for_archive => true); were correctly restored. </para> </listitem> + <listitem> + <para> + If you're restoring an incremental backup, you'll need to restore the + incremental backup and all earlier backups upon which it directly or + indirectly depends to the machine where you are performing the restore. + These backups will need to be placed in separate directories, not the + target directories where you want the running server to end up. + Once this is done, use <xref linkend="app-pgcombinebackup"/> to pull + data from the full backup and all of the subsequent incremental backups + and write out a synthetic full backup to the target directories. As above, + verify that permissions and tablespace links are correct. + </para> + </listitem> <listitem> <para> Remove any files present in <filename>pg_wal/</filename>; these came from the diff --git a/doc/src/sgml/ref/allfiles.sgml b/doc/src/sgml/ref/allfiles.sgml index 54b5f22d6e..fda4690eab 100644 --- a/doc/src/sgml/ref/allfiles.sgml +++ b/doc/src/sgml/ref/allfiles.sgml @@ -202,6 +202,7 @@ Complete list of usable sgml source files in this directory. <!ENTITY pgBasebackup SYSTEM "pg_basebackup.sgml"> <!ENTITY pgbench SYSTEM "pgbench.sgml"> <!ENTITY pgChecksums SYSTEM "pg_checksums.sgml"> +<!ENTITY pgCombinebackup SYSTEM "pg_combinebackup.sgml"> <!ENTITY pgConfig SYSTEM "pg_config-ref.sgml"> <!ENTITY pgControldata SYSTEM "pg_controldata.sgml"> <!ENTITY pgCtl SYSTEM "pg_ctl-ref.sgml"> diff --git a/doc/src/sgml/ref/pg_basebackup.sgml b/doc/src/sgml/ref/pg_basebackup.sgml index 712568a62d..50536d0521 100644 --- a/doc/src/sgml/ref/pg_basebackup.sgml +++ b/doc/src/sgml/ref/pg_basebackup.sgml @@ -38,11 +38,25 @@ PostgreSQL documentation </para> <para> - <application>pg_basebackup</application> makes an exact copy of the database - cluster's files, while making sure the server is put into and - out of backup mode automatically. Backups are always taken of the entire - database cluster; it is not possible to back up individual databases or - database objects. For selective backups, another tool such as + <application>pg_basebackup</application> can take a full or incremental + base backup of the database. When used to take a full backup, it makes an + exact copy of the database cluster's files. When used to take an incremental + backup, some files that would have been part of a full backup may be + replaced with incremental versions of the same files, containing only those + blocks that have been modified since the reference backup. An incremental + backup cannot be used directly; instead, + <xref linkend="app-pgcombinebackup"/> must first + be used to combine it with the previous backups upon which it depends. + See <xref linkend="backup-incremental-backup" /> for more information + about incremental backups, and <xref linkend="backup-pitr-recovery" /> + for steps to recover from a backup. + </para> + + <para> + In any mode, <application>pg_basebackup</application> makes sure the server + is put into and out of backup mode automatically. Backups are always taken of + the entire database cluster; it is not possible to back up individual + databases or database objects. For selective backups, another tool such as <xref linkend="app-pgdump"/> must be used. </para> @@ -197,6 +211,19 @@ PostgreSQL documentation </listitem> </varlistentry> + <varlistentry> + <term><option>-i <replaceable class="parameter">old_manifest_file</replaceable></option></term> + <term><option>--incremental=<replaceable class="parameter">old_meanifest_file</replaceable></option></term> + <listitem> + <para> + Performs an <link linkend="backup-incremental-backup">incremental + backup</link>. The backup manifest for the reference + backup must be provided, and will be uploaded to the server, which will + respond by sending the requested incremental backup. + </para> + </listitem> + </varlistentry> + <varlistentry> <term><option>-R</option></term> <term><option>--write-recovery-conf</option></term> diff --git a/doc/src/sgml/ref/pg_combinebackup.sgml b/doc/src/sgml/ref/pg_combinebackup.sgml new file mode 100644 index 0000000000..6cac73573f --- /dev/null +++ b/doc/src/sgml/ref/pg_combinebackup.sgml @@ -0,0 +1,228 @@ +<!-- +doc/src/sgml/ref/pg_combinebackup.sgml +PostgreSQL documentation +--> + +<refentry id="app-pgcombinebackup"> + <indexterm zone="app-pgcombinebackup"> + <primary>pg_combinebackup</primary> + </indexterm> + + <refmeta> + <refentrytitle><application>pg_combinebackup</application></refentrytitle> + <manvolnum>1</manvolnum> + <refmiscinfo>Application</refmiscinfo> + </refmeta> + + <refnamediv> + <refname>pg_combinebackup</refname> + <refpurpose>reconstruct a full backup from an incremental backup and dependent backups</refpurpose> + </refnamediv> + + <refsynopsisdiv> + <cmdsynopsis> + <command>pg_combinebackup</command> + <arg rep="repeat"><replaceable>option</replaceable></arg> + <arg rep="repeat"><replaceable>backup_directory</replaceable></arg> + </cmdsynopsis> + </refsynopsisdiv> + + <refsect1> + <title>Description</title> + <para> + <application>pg_combinebackup</application> is used to reconstruct a + synthetic full backup from an + <link linkend="backup-incremental-backup">incremental backup</link> and the + earlier backups upon which it depends. + </para> + + <para> + Specify all of the required backups on the command line from oldest to newest. + That is, the first backup directory should be the path to the full backup, and + the last should be the path to the final incremental backup + that you wish to restore. The reconstructed backup will be written to the + output directory specified by the <option>-o</option> option. + </para> + + <para> + Although <application>pg_combinebackup</application> will attempt to verify + that the backups you specify form a legal backup chain from which a correct + full backup can be reconstructed, it is not designed to help you keep track + of which backups depend on which other backups. If you remove the one or + more of the previous backups upon which your incremental + backup relies, you will not be able to restore it. + </para> + + <para> + Since the output of <application>pg_combinebackup</application> is a + synthetic full backup, it can be used as an input to a future invocation of + <application>pg_combinebackup</application>. The synthetic full backup would + be specified on the command line in lieu of the chain of backups from which + it was reconstructed. + </para> + </refsect1> + + <refsect1> + <title>Options</title> + + <para> + <variablelist> + <varlistentry> + <term><option>-d</option></term> + <term><option>--debug</option></term> + <listitem> + <para> + Print lots of debug logging output on <filename>stderr</filename>. + </para> + </listitem> + </varlistentry> + + <varlistentry> + <term><option>-T <replaceable class="parameter">olddir</replaceable>=<replaceable class="parameter">newdir</replaceable></option></term> + <term><option>--tablespace-mapping=<replaceable class="parameter">olddir</replaceable>=<replaceable class="parameter">newdir</replaceable></option></term> + <listitem> + <para> + Relocates the tablespace in directory <replaceable>olddir</replaceable> + to <replaceable>newdir</replaceable> during the backup. + <replaceable>olddir</replaceable> is the absolute path of the tablespace + as it exists in the first backup specified on the command line, + and <replaceable>newdir</replaceable> is the absolute path to use for the + tablespace in the reconstructed backup. If either path needs to contain + an equal sign (<literal>=</literal>), precede that with a backslash. + This option can be specified multiple times for multiple tablespaces. + </para> + </listitem> + </varlistentry> + + <varlistentry> + <term><option>-N</option></term> + <term><option>--no-sync</option></term> + <listitem> + <para> + By default, <command>pg_combinebackup</command> will wait for all files + to be written safely to disk. This option causes + <command>pg_combinebackup</command> to return without waiting, which is + faster, but means that a subsequent operating system crash can leave + the output backup corrupt. Generally, this option is useful for testing + but should not be used when creating a production installation. + </para> + </listitem> + </varlistentry> + + <varlistentry> + <term><option>-o <replaceable class="parameter">outputdir</replaceable></option></term> + <term><option>--output=<replaceable class="parameter">outputdir</replaceable></option></term> + <listitem> + <para> + Specifies the output directory to which the synthetic full backup + should be written. Currently, this argument is required. + </para> + </listitem> + </varlistentry> + + <varlistentry> + <term><option>--sync-method</option></term> + <listitem> + <para> + When set to <literal>fsync</literal>, which is the default, + <command>pg_combinebackup</command> will recursively open and synchronize + all files in the backup directory. When the plain format is used, the + search for files will follow symbolic links for the WAL directory and + each configured tablespace. + </para> + <para> + On Linux, <literal>syncfs</literal> may be used instead to ask the + operating system to synchronize the whole file system that contains the + backup directory. When the plain format is used, + <command>pg_combinebackup</command> will also synchronize the file systems + that contain the WAL files and each tablespace. See + <xref linkend="syncfs"/> for more information about using + <function>syncfs()</function>. + </para> + <para> + This option has no effect when <option>--no-sync</option> is used. + </para> + </listitem> + </varlistentry> + + <varlistentry> + <term><option>--manifest-checksums=<replaceable class="parameter">algorithm</replaceable></option></term> + <listitem> + <para> + Like <xref linkend="app-pgbasebackup"/>, + <application>pg_combinebackup</application> writes a backup manifest + in the output directory. This option specifies the checksum algorithm + that should be applied to each file included in the backup manifest. + Currently, the available algorithms are <literal>NONE</literal>, + <literal>CRC32C</literal>, <literal>SHA224</literal>, + <literal>SHA256</literal>, <literal>SHA384</literal>, + and <literal>SHA512</literal>. The default is <literal>CRC32C</literal>. + </para> + </listitem> + </varlistentry> + + <varlistentry> + <term><option>--no-manifest</option></term> + <listitem> + <para> + Disables generation of a backup manifest. If this option is not + specified, a backup manifest for the reconstructed backup will be + written to the output directory. + </para> + </listitem> + </varlistentry> + </variablelist> + + <variablelist> + <varlistentry> + <term><option>-V</option></term> + <term><option>--version</option></term> + <listitem> + <para> + Prints the <application>pg_combinebackup</application> version and exits. + </para> + </listitem> + </varlistentry> + + <varlistentry> + <term><option>-?</option></term> + <term><option>--help</option></term> + <listitem> + <para> + Shows help about <application>pg_combinebackup</application> command line + arguments, and exits. + </para> + </listitem> + </varlistentry> + + </variablelist> + </para> + + </refsect1> + + <refsect1> + <title>Environment</title> + + <para> + This utility, like most other <productname>PostgreSQL</productname> utilities, + uses the environment variables supported by <application>libpq</application> + (see <xref linkend="libpq-envars"/>). + </para> + + <para> + The environment variable <envar>PG_COLOR</envar> specifies whether to use + color in diagnostic messages. Possible values are + <literal>always</literal>, <literal>auto</literal> and + <literal>never</literal>. + </para> + </refsect1> + + <refsect1> + <title>See Also</title> + + <simplelist type="inline"> + <member><xref linkend="app-pgbasebackup"/></member> + </simplelist> + </refsect1> + +</refentry> diff --git a/doc/src/sgml/reference.sgml b/doc/src/sgml/reference.sgml index e11b4b6130..a07d2b5e01 100644 --- a/doc/src/sgml/reference.sgml +++ b/doc/src/sgml/reference.sgml @@ -250,6 +250,7 @@ &pgamcheck; &pgBasebackup; &pgbench; + &pgCombinebackup; &pgConfig; &pgDump; &pgDumpall; diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 40461923ea..9ddad7864f 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -77,6 +77,7 @@ #include "port/pg_iovec.h" #include "postmaster/bgwriter.h" #include "postmaster/startup.h" +#include "postmaster/walsummarizer.h" #include "postmaster/walwriter.h" #include "replication/logical.h" #include "replication/origin.h" @@ -3555,6 +3556,43 @@ XLogGetLastRemovedSegno(void) return lastRemovedSegNo; } +/* + * Return the oldest WAL segment on the given TLI that still exists in + * XLOGDIR, or 0 if none. + */ +XLogSegNo +XLogGetOldestSegno(TimeLineID tli) +{ + DIR *xldir; + struct dirent *xlde; + XLogSegNo oldest_segno = 0; + + xldir = AllocateDir(XLOGDIR); + while ((xlde = ReadDir(xldir, XLOGDIR)) != NULL) + { + TimeLineID file_tli; + XLogSegNo file_segno; + + /* Ignore files that are not XLOG segments */ + if (!IsXLogFileName(xlde->d_name)) + continue; + + /* Parse filename to get TLI and segno. */ + XLogFromFileName(xlde->d_name, &file_tli, &file_segno, + wal_segment_size); + + /* Ignore anything that's not from the TLI of interest. */ + if (tli != file_tli) + continue; + + /* If it's the oldest so far, update oldest_segno. */ + if (oldest_segno == 0 || file_segno < oldest_segno) + oldest_segno = file_segno; + } + + FreeDir(xldir); + return oldest_segno; +} /* * Update the last removed segno pointer in shared memory, to reflect that the @@ -3834,8 +3872,8 @@ RemoveXlogFile(const struct dirent *segment_de, } /* - * Verify whether pg_wal and pg_wal/archive_status exist. - * If the latter does not exist, recreate it. + * Verify whether pg_wal, pg_wal/archive_status, and pg_wal/summaries exist. + * If the latter do not exist, recreate them. * * It is not the goal of this function to verify the contents of these * directories, but to help in cases where someone has performed a cluster @@ -3878,6 +3916,26 @@ ValidateXLOGDirectoryStructure(void) (errmsg("could not create missing directory \"%s\": %m", path))); } + + /* Check for summaries */ + snprintf(path, MAXPGPATH, XLOGDIR "/summaries"); + if (stat(path, &stat_buf) == 0) + { + /* Check for weird cases where it exists but isn't a directory */ + if (!S_ISDIR(stat_buf.st_mode)) + ereport(FATAL, + (errmsg("required WAL directory \"%s\" does not exist", + path))); + } + else + { + ereport(LOG, + (errmsg("creating missing WAL directory \"%s\"", path))); + if (MakePGDirectory(path) < 0) + ereport(FATAL, + (errmsg("could not create missing directory \"%s\": %m", + path))); + } } /* @@ -5202,9 +5260,9 @@ StartupXLOG(void) #endif /* - * Verify that pg_wal and pg_wal/archive_status exist. In cases where - * someone has performed a copy for PITR, these directories may have been - * excluded and need to be re-created. + * Verify that pg_wal, pg_wal/archive_status, and pg_wal/summaries exist. + * In cases where someone has performed a copy for PITR, these directories + * may have been excluded and need to be re-created. */ ValidateXLOGDirectoryStructure(); @@ -6921,6 +6979,24 @@ CreateCheckPoint(int flags) */ END_CRIT_SECTION(); + /* + * WAL summaries end when the next XLOG_CHECKPOINT_REDO or + * XLOG_CHECKPOINT_SHUTDOWN record is reached. This is the first point + * where (a) we're not inside of a critical section and (b) we can be + * certain that the relevant record has been flushed to disk, which must + * happen before it can be summarized. + * + * If this is a shutdown checkpoint, then this happens reasonably promptly: + * we've only just inserted and flushed the XLOG_CHECKPOINT_SHUTDOWN + * record. If this is not a shutdown checkpoint, then this might not be + * very prompt at all: the XLOG_CHECKPOINT_REDO record was written before + * we began flushing data to disk, and that could be many minutes ago at + * this point. However, we don't XLogFlush() after inserting that record, + * so we're not guaranteed that it's on disk until after the above call + * that flushes the XLOG_CHECKPOINT_ONLINE record. + */ + SetWalSummarizerLatch(); + /* * Let smgr do post-checkpoint cleanup (eg, deleting old files). */ @@ -7595,6 +7671,20 @@ KeepLogSeg(XLogRecPtr recptr, XLogSegNo *logSegNo) } } + /* + * If WAL summarization is in use, don't remove WAL that has yet to be + * summarized. + */ + keep = GetOldestUnsummarizedLSN(NULL, NULL); + if (keep != InvalidXLogRecPtr) + { + XLogSegNo unsummarized_segno; + + XLByteToSeg(keep, unsummarized_segno, wal_segment_size); + if (unsummarized_segno < segno) + segno = unsummarized_segno; + } + /* but, keep at least wal_keep_size if that's set */ if (wal_keep_size_mb > 0) { diff --git a/src/backend/access/transam/xlogbackup.c b/src/backend/access/transam/xlogbackup.c index 21d68133ae..f51d4282bb 100644 --- a/src/backend/access/transam/xlogbackup.c +++ b/src/backend/access/transam/xlogbackup.c @@ -77,6 +77,16 @@ build_backup_content(BackupState *state, bool ishistoryfile) appendStringInfo(result, "STOP TIMELINE: %u\n", state->stoptli); } + /* either both istartpoint and istarttli should be set, or neither */ + Assert(XLogRecPtrIsInvalid(state->istartpoint) == (state->istarttli == 0)); + if (!XLogRecPtrIsInvalid(state->istartpoint)) + { + appendStringInfo(result, "INCREMENTAL FROM LSN: %X/%X\n", + LSN_FORMAT_ARGS(state->istartpoint)); + appendStringInfo(result, "INCREMENTAL FROM TLI: %u\n", + state->istarttli); + } + data = result->data; pfree(result); diff --git a/src/backend/access/transam/xlogrecovery.c b/src/backend/access/transam/xlogrecovery.c index 315e4b27cb..6cde31ee23 100644 --- a/src/backend/access/transam/xlogrecovery.c +++ b/src/backend/access/transam/xlogrecovery.c @@ -1284,6 +1284,12 @@ read_backup_label(XLogRecPtr *checkPointLoc, TimeLineID *backupLabelTLI, tli_from_file, BACKUP_LABEL_FILE))); } + if (fscanf(lfp, "INCREMENTAL FROM LSN: %X/%X\n", &hi, &lo) > 0) + ereport(FATAL, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("this is an incremental backup, not a data directory"), + errhint("Use pg_combinebackup to reconstruct a valid data directory."))); + if (ferror(lfp) || FreeFile(lfp)) ereport(FATAL, (errcode_for_file_access(), diff --git a/src/backend/backup/Makefile b/src/backend/backup/Makefile index b21bd8ff43..751e6d3d5e 100644 --- a/src/backend/backup/Makefile +++ b/src/backend/backup/Makefile @@ -19,12 +19,15 @@ OBJS = \ basebackup.o \ basebackup_copy.o \ basebackup_gzip.o \ + basebackup_incremental.o \ basebackup_lz4.o \ basebackup_zstd.o \ basebackup_progress.o \ basebackup_server.o \ basebackup_sink.o \ basebackup_target.o \ - basebackup_throttle.o + basebackup_throttle.o \ + walsummary.o \ + walsummaryfuncs.o include $(top_srcdir)/src/backend/common.mk diff --git a/src/backend/backup/basebackup.c b/src/backend/backup/basebackup.c index 4ba63ad8a6..8a70a9ae41 100644 --- a/src/backend/backup/basebackup.c +++ b/src/backend/backup/basebackup.c @@ -20,8 +20,10 @@ #include "access/xlogbackup.h" #include "backup/backup_manifest.h" #include "backup/basebackup.h" +#include "backup/basebackup_incremental.h" #include "backup/basebackup_sink.h" #include "backup/basebackup_target.h" +#include "catalog/pg_tablespace_d.h" #include "commands/defrem.h" #include "common/compression.h" #include "common/file_perm.h" @@ -64,6 +66,7 @@ typedef struct bool fastcheckpoint; bool nowait; bool includewal; + bool incremental; uint32 maxrate; bool sendtblspcmapfile; bool send_to_client; @@ -76,21 +79,28 @@ typedef struct } basebackup_options; static int64 sendTablespace(bbsink *sink, char *path, Oid spcoid, bool sizeonly, - struct backup_manifest_info *manifest); + struct backup_manifest_info *manifest, + IncrementalBackupInfo *ib); static int64 sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, List *tablespaces, bool sendtblspclinks, - backup_manifest_info *manifest, Oid spcoid); + backup_manifest_info *manifest, Oid spcoid, + IncrementalBackupInfo *ib); static bool sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, struct stat *statbuf, bool missing_ok, Oid dboid, Oid spcoid, RelFileNumber relfilenumber, unsigned segno, - backup_manifest_info *manifest); + backup_manifest_info *manifest, + unsigned num_incremental_blocks, + BlockNumber *incremental_blocks, + unsigned truncation_block_length); static off_t read_file_data_into_buffer(bbsink *sink, const char *readfilename, int fd, off_t offset, size_t length, BlockNumber blkno, bool verify_checksum, int *checksum_failures); +static void push_to_sink(bbsink *sink, pg_checksum_context *checksum_ctx, + size_t *bytes_done, void *data, size_t length); static bool verify_page_checksum(Page page, XLogRecPtr start_lsn, BlockNumber blkno, uint16 *expected_checksum); @@ -102,7 +112,8 @@ static int64 _tarWriteHeader(bbsink *sink, const char *filename, bool sizeonly); static void _tarWritePadding(bbsink *sink, int len); static void convert_link_to_directory(const char *pathbuf, struct stat *statbuf); -static void perform_base_backup(basebackup_options *opt, bbsink *sink); +static void perform_base_backup(basebackup_options *opt, bbsink *sink, + IncrementalBackupInfo *ib); static void parse_basebackup_options(List *options, basebackup_options *opt); static int compareWalFileNames(const ListCell *a, const ListCell *b); static int basebackup_read_file(int fd, char *buf, size_t nbytes, off_t offset, @@ -220,7 +231,8 @@ static const struct exclude_list_item excludeFiles[] = * clobbered by longjmp" from stupider versions of gcc. */ static void -perform_base_backup(basebackup_options *opt, bbsink *sink) +perform_base_backup(basebackup_options *opt, bbsink *sink, + IncrementalBackupInfo *ib) { bbsink_state state; XLogRecPtr endptr; @@ -270,6 +282,10 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) ListCell *lc; tablespaceinfo *newti; + /* If this is an incremental backup, execute preparatory steps. */ + if (ib != NULL) + PrepareForIncrementalBackup(ib, backup_state); + /* Add a node for the base directory at the end */ newti = palloc0(sizeof(tablespaceinfo)); newti->size = -1; @@ -289,10 +305,10 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) if (tmp->path == NULL) tmp->size = sendDir(sink, ".", 1, true, state.tablespaces, - true, NULL, InvalidOid); + true, NULL, InvalidOid, NULL); else tmp->size = sendTablespace(sink, tmp->path, tmp->oid, true, - NULL); + NULL, NULL); state.bytes_total += tmp->size; } state.bytes_total_is_valid = true; @@ -330,7 +346,7 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) /* Then the bulk of the files... */ sendDir(sink, ".", 1, false, state.tablespaces, - sendtblspclinks, &manifest, InvalidOid); + sendtblspclinks, &manifest, InvalidOid, ib); /* ... and pg_control after everything else. */ if (lstat(XLOG_CONTROL_FILE, &statbuf) != 0) @@ -340,7 +356,7 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) XLOG_CONTROL_FILE))); sendFile(sink, XLOG_CONTROL_FILE, XLOG_CONTROL_FILE, &statbuf, false, InvalidOid, InvalidOid, - InvalidRelFileNumber, 0, &manifest); + InvalidRelFileNumber, 0, &manifest, 0, NULL, 0); } else { @@ -348,7 +364,7 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) bbsink_begin_archive(sink, archive_name); - sendTablespace(sink, ti->path, ti->oid, false, &manifest); + sendTablespace(sink, ti->path, ti->oid, false, &manifest, ib); } /* @@ -610,7 +626,7 @@ perform_base_backup(basebackup_options *opt, bbsink *sink) sendFile(sink, pathbuf, pathbuf, &statbuf, false, InvalidOid, InvalidOid, InvalidRelFileNumber, 0, - &manifest); + &manifest, 0, NULL, 0); /* unconditionally mark file as archived */ StatusFilePath(pathbuf, fname, ".done"); @@ -686,6 +702,7 @@ parse_basebackup_options(List *options, basebackup_options *opt) bool o_checkpoint = false; bool o_nowait = false; bool o_wal = false; + bool o_incremental = false; bool o_maxrate = false; bool o_tablespace_map = false; bool o_noverify_checksums = false; @@ -764,6 +781,15 @@ parse_basebackup_options(List *options, basebackup_options *opt) opt->includewal = defGetBoolean(defel); o_wal = true; } + else if (strcmp(defel->defname, "incremental") == 0) + { + if (o_incremental) + ereport(ERROR, + (errcode(ERRCODE_SYNTAX_ERROR), + errmsg("duplicate option \"%s\"", defel->defname))); + opt->incremental = defGetBoolean(defel); + o_incremental = true; + } else if (strcmp(defel->defname, "max_rate") == 0) { int64 maxrate; @@ -956,7 +982,7 @@ parse_basebackup_options(List *options, basebackup_options *opt) * the filesystem, bypassing the buffer cache. */ void -SendBaseBackup(BaseBackupCmd *cmd) +SendBaseBackup(BaseBackupCmd *cmd, IncrementalBackupInfo *ib) { basebackup_options opt; bbsink *sink; @@ -980,6 +1006,20 @@ SendBaseBackup(BaseBackupCmd *cmd) set_ps_display(activitymsg); } + /* + * If we're asked to perform an incremental backup and the user has not + * supplied a manifest, that's an ERROR. + * + * If we're asked to perform a full backup and the user did supply a + * manifest, just ignore it. + */ + if (!opt.incremental) + ib = NULL; + else if (ib == NULL) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("must UPLOAD_MANIFEST before performing an incremental BASE_BACKUP"))); + /* * If the target is specifically 'client' then set up to stream the backup * to the client; otherwise, it's being sent someplace else and should not @@ -1011,7 +1051,7 @@ SendBaseBackup(BaseBackupCmd *cmd) */ PG_TRY(); { - perform_base_backup(&opt, sink); + perform_base_backup(&opt, sink, ib); } PG_FINALLY(); { @@ -1086,7 +1126,7 @@ sendFileWithContent(bbsink *sink, const char *filename, const char *content, */ static int64 sendTablespace(bbsink *sink, char *path, Oid spcoid, bool sizeonly, - backup_manifest_info *manifest) + backup_manifest_info *manifest, IncrementalBackupInfo *ib) { int64 size; char pathbuf[MAXPGPATH]; @@ -1120,7 +1160,7 @@ sendTablespace(bbsink *sink, char *path, Oid spcoid, bool sizeonly, /* Send all the files in the tablespace version directory */ size += sendDir(sink, pathbuf, strlen(path), sizeonly, NIL, true, manifest, - spcoid); + spcoid, ib); return size; } @@ -1140,7 +1180,7 @@ sendTablespace(bbsink *sink, char *path, Oid spcoid, bool sizeonly, static int64 sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, List *tablespaces, bool sendtblspclinks, backup_manifest_info *manifest, - Oid spcoid) + Oid spcoid, IncrementalBackupInfo *ib) { DIR *dir; struct dirent *de; @@ -1149,7 +1189,16 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, int64 size = 0; const char *lastDir; /* Split last dir from parent path. */ bool isRelationDir = false; /* Does directory contain relations? */ + bool isGlobalDir = false; Oid dboid = InvalidOid; + BlockNumber *relative_block_numbers = NULL; + + /* + * Since this array is relatively large, avoid putting it on the stack. + * But we don't need it at all if this is not an incremental backup. + */ + if (ib != NULL) + relative_block_numbers = palloc(sizeof(BlockNumber) * RELSEG_SIZE); /* * Determine if the current path is a database directory that can contain @@ -1182,7 +1231,10 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, } } else if (strcmp(path, "./global") == 0) + { isRelationDir = true; + isGlobalDir = true; + } dir = AllocateDir(path); while ((de = ReadDir(dir, path)) != NULL) @@ -1331,11 +1383,13 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, &statbuf, sizeonly); /* - * Also send archive_status directory (by hackishly reusing - * statbuf from above ...). + * Also send archive_status and summaries directories (by + * hackishly reusing statbuf from above ...). */ size += _tarWriteHeader(sink, "./pg_wal/archive_status", NULL, &statbuf, sizeonly); + size += _tarWriteHeader(sink, "./pg_wal/summaries", NULL, + &statbuf, sizeonly); continue; /* don't recurse into pg_wal */ } @@ -1404,33 +1458,88 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, if (!skip_this_dir) size += sendDir(sink, pathbuf, basepathlen, sizeonly, tablespaces, - sendtblspclinks, manifest, spcoid); + sendtblspclinks, manifest, spcoid, ib); } else if (S_ISREG(statbuf.st_mode)) { bool sent = false; + unsigned num_blocks_required = 0; + unsigned truncation_block_length = 0; + char tarfilenamebuf[MAXPGPATH * 2]; + char *tarfilename = pathbuf + basepathlen + 1; + FileBackupMethod method = BACK_UP_FILE_FULLY; - if (!sizeonly) - sent = sendFile(sink, pathbuf, pathbuf + basepathlen + 1, &statbuf, - true, dboid, spcoid, - relfilenumber, segno, manifest); + if (ib != NULL && isRelationFile) + { + Oid relspcoid; + char *lookup_path; - if (sent || sizeonly) + if (OidIsValid(spcoid)) + { + relspcoid = spcoid; + lookup_path = psprintf("pg_tblspc/%u/%s", spcoid, + pathbuf + basepathlen + 1); + } + else + { + if (isGlobalDir) + relspcoid = GLOBALTABLESPACE_OID; + else + relspcoid = DEFAULTTABLESPACE_OID; + lookup_path = pstrdup(pathbuf + basepathlen + 1); + } + + method = GetFileBackupMethod(ib, lookup_path, dboid, relspcoid, + relfilenumber, relForkNum, + segno, statbuf.st_size, + &num_blocks_required, + relative_block_numbers, + &truncation_block_length); + if (method == BACK_UP_FILE_INCREMENTALLY) + { + statbuf.st_size = + GetIncrementalFileSize(num_blocks_required); + snprintf(tarfilenamebuf, sizeof(tarfilenamebuf), + "%s/INCREMENTAL.%s", + path + basepathlen + 1, + de->d_name); + tarfilename = tarfilenamebuf; + } + + pfree(lookup_path); + } + + if (method != DO_NOT_BACK_UP_FILE) { - /* Add size. */ - size += statbuf.st_size; + if (!sizeonly) + sent = sendFile(sink, pathbuf, tarfilename, &statbuf, + true, dboid, spcoid, + relfilenumber, segno, manifest, + num_blocks_required, + method == BACK_UP_FILE_INCREMENTALLY ? relative_block_numbers : NULL, + truncation_block_length); + + if (sent || sizeonly) + { + /* Add size. */ + size += statbuf.st_size; - /* Pad to a multiple of the tar block size. */ - size += tarPaddingBytesRequired(statbuf.st_size); + /* Pad to a multiple of the tar block size. */ + size += tarPaddingBytesRequired(statbuf.st_size); - /* Size of the header for the file. */ - size += TAR_BLOCK_SIZE; + /* Size of the header for the file. */ + size += TAR_BLOCK_SIZE; + } } } else ereport(WARNING, (errmsg("skipping special file \"%s\"", pathbuf))); } + + if (relative_block_numbers != NULL) + pfree(relative_block_numbers); + FreeDir(dir); return size; } @@ -1443,6 +1552,12 @@ sendDir(bbsink *sink, const char *path, int basepathlen, bool sizeonly, * If dboid is anything other than InvalidOid then any checksum failures * detected will get reported to the cumulative stats system. * + * If the file is to be sent incrementally, then num_incremental_blocks + * should be the number of blocks to be sent, and incremental_blocks + * an array of block numbers relative to the start of the current segment. + * If the whole file is to be sent, then incremental_blocks should be NULL, + * and num_incremental_blocks can have any value, as it will be ignored. + * * Returns true if the file was successfully sent, false if 'missing_ok', * and the file did not exist. */ @@ -1450,7 +1565,8 @@ static bool sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, struct stat *statbuf, bool missing_ok, Oid dboid, Oid spcoid, RelFileNumber relfilenumber, unsigned segno, - backup_manifest_info *manifest) + backup_manifest_info *manifest, unsigned num_incremental_blocks, + BlockNumber *incremental_blocks, unsigned truncation_block_length) { int fd; BlockNumber blkno = 0; @@ -1459,6 +1575,7 @@ sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, pgoff_t bytes_done = 0; bool verify_checksum = false; pg_checksum_context checksum_ctx; + int ibindex = 0; if (pg_checksum_init(&checksum_ctx, manifest->checksum_type) < 0) elog(ERROR, "could not initialize checksum of file \"%s\"", @@ -1491,22 +1608,111 @@ sendFile(bbsink *sink, const char *readfilename, const char *tarfilename, RelFileNumberIsValid(relfilenumber)) verify_checksum = true; + /* + * If we're sending an incremental file, write the file header. + */ + if (incremental_blocks != NULL) + { + unsigned magic = INCREMENTAL_MAGIC; + size_t header_bytes_done = 0; + + /* Emit header data. */ + push_to_sink(sink, &checksum_ctx, &header_bytes_done, + &magic, sizeof(magic)); + push_to_sink(sink, &checksum_ctx, &header_bytes_done, + &num_incremental_blocks, sizeof(num_incremental_blocks)); + push_to_sink(sink, &checksum_ctx, &header_bytes_done, + &truncation_block_length, sizeof(truncation_block_length)); + push_to_sink(sink, &checksum_ctx, &header_bytes_done, + incremental_blocks, + sizeof(BlockNumber) * num_incremental_blocks); + + /* Flush out any data still in the buffer so it's again empty. */ + if (header_bytes_done > 0) + { + bbsink_archive_contents(sink, header_bytes_done); + if (pg_checksum_update(&checksum_ctx, + (uint8 *) sink->bbs_buffer, + header_bytes_done) < 0) + elog(ERROR, "could not update checksum of base backup"); + } + + /* Update our notion of file position. */ + bytes_done += sizeof(magic); + bytes_done += sizeof(num_incremental_blocks); + bytes_done += sizeof(truncation_block_length); + bytes_done += sizeof(BlockNumber) * num_incremental_blocks; + } + /* * Loop until we read the amount of data the caller told us to expect. The * file could be longer, if it was extended while we were sending it, but * for a base backup we can ignore such extended data. It will be restored * from WAL. */ - while (bytes_done < statbuf->st_size) + while (1) { - size_t remaining = statbuf->st_size - bytes_done; + /* + * Determine whether we've read all the data that we need, and if not, + * read some more. + */ + if (incremental_blocks == NULL) + { + size_t remaining = statbuf->st_size - bytes_done; + + /* + * If we've read the required number of bytes, then it's time to + * stop. + */ + if (bytes_done >= statbuf->st_size) + break; + + /* + * Read as many bytes as will fit in the buffer, or however many + * are left to read, whichever is less. + */ + cnt = read_file_data_into_buffer(sink, readfilename, fd, + bytes_done, remaining, + blkno + segno * RELSEG_SIZE, + verify_checksum, + &checksum_failures); + } + else + { + BlockNumber relative_blkno; - /* Try to read some more data. */ - cnt = read_file_data_into_buffer(sink, readfilename, fd, bytes_done, - remaining, - blkno + segno * RELSEG_SIZE, - verify_checksum, - &checksum_failures); + /* + * If we've read all the blocks, then it's time to stop. + */ + if (ibindex >= num_incremental_blocks) + break; + + /* + * Read just one block, whichever one is the next that we're + * supposed to include. + */ + relative_blkno = incremental_blocks[ibindex++]; + cnt = read_file_data_into_buffer(sink, readfilename, fd, + relative_blkno * BLCKSZ, + BLCKSZ, + relative_blkno + segno * RELSEG_SIZE, + verify_checksum, + &checksum_failures); + + /* + * If we get a partial read, that must mean that the relation is + * being truncated. Ultimately, it should be truncated to a + * multiple of BLCKSZ, since this path should only be reached for + * relation files, but we might transiently observe an + * intermediate value. + * + * It should be fine to treat this just as if the entire block had + * been truncated away - i.e. fill this and all later blocks with + * zeroes. WAL replay will fix things up. + */ + if (cnt < BLCKSZ) + break; + } /* * If the amount of data we were able to read was not a multiple of @@ -1689,6 +1895,56 @@ read_file_data_into_buffer(bbsink *sink, const char *readfilename, int fd, return cnt; } +/* + * Push data into a bbsink. + * + * It's better, when possible, to read data directly into the bbsink's buffer, + * rather than using this function to copy it into the buffer; this function is + * for cases where that approach is not practical. + * + * bytes_done should point to a count of the number of bytes that are + * currently used in the bbsink's buffer. Upon return, the bytes identified by + * data and length will have been copied into the bbsink's buffer, flushing + * as required, and *bytes_done will have been updated accordingly. If the + * buffer was flushed, the previous contents will also have been fed to + * checksum_ctx. + * + * Note that after one or more calls to this function it is the caller's + * responsibility to perform any required final flush. + */ +static void +push_to_sink(bbsink *sink, pg_checksum_context *checksum_ctx, + size_t *bytes_done, void *data, size_t length) +{ + while (length > 0) + { + size_t bytes_to_copy; + + /* + * We use < here rather than <= so that if the data exactly fills the + * remaining buffer space, we trigger a flush now. + */ + if (length < sink->bbs_buffer_length - *bytes_done) + { + /* Append remaining data to buffer. */ + memcpy(sink->bbs_buffer + *bytes_done, data, length); + *bytes_done += length; + return; + } + + /* Copy until buffer is full and flush it. */ + bytes_to_copy = sink->bbs_buffer_length - *bytes_done; + memcpy(sink->bbs_buffer + *bytes_done, data, bytes_to_copy); + data = ((char *) data) + bytes_to_copy; + length -= bytes_to_copy; + bbsink_archive_contents(sink, sink->bbs_buffer_length); + if (pg_checksum_update(checksum_ctx, (uint8 *) sink->bbs_buffer, + sink->bbs_buffer_length) < 0) + elog(ERROR, "could not update checksum"); + *bytes_done = 0; + } +} + /* * Try to verify the checksum for the provided page, if it seems appropriate * to do so. diff --git a/src/backend/backup/basebackup_incremental.c b/src/backend/backup/basebackup_incremental.c new file mode 100644 index 0000000000..20cc00bded --- /dev/null +++ b/src/backend/backup/basebackup_incremental.c @@ -0,0 +1,873 @@ +/*------------------------------------------------------------------------- + * + * basebackup_incremental.c + * code for incremental backup support + * + * This code isn't actually in charge of taking an incremental backup; + * the actual construction of the incremental backup happens in + * basebackup.c. Here, we're concerned with providing the necessary + * supports for that operation. In particular, we need to parse the + * backup manifest supplied by the user taking the incremental backup + * and extract the required information from it. + * + * Portions Copyright (c) 2010-2022, PostgreSQL Global Development Group + * + * IDENTIFICATION + * src/backend/backup/basebackup_incremental.c + * + *------------------------------------------------------------------------- + */ +#include "postgres.h" + +#include "access/timeline.h" +#include "access/xlog.h" +#include "access/xlogrecovery.h" +#include "backup/basebackup_incremental.h" +#include "backup/walsummary.h" +#include "common/blkreftable.h" +#include "common/parse_manifest.h" +#include "common/hashfn.h" +#include "postmaster/walsummarizer.h" + +#define BLOCKS_PER_READ 512 + +/* + * Details extracted from the WAL ranges present in the supplied backup manifest. + */ +typedef struct +{ + TimeLineID tli; + XLogRecPtr start_lsn; + XLogRecPtr end_lsn; +} backup_wal_range; + +/* + * Details extracted from the file list present in the supplied backup manifest. + */ +typedef struct +{ + uint32 status; + const char *path; + size_t size; +} backup_file_entry; + +static uint32 hash_string_pointer(const char *s); +#define SH_PREFIX backup_file +#define SH_ELEMENT_TYPE backup_file_entry +#define SH_KEY_TYPE const char * +#define SH_KEY path +#define SH_HASH_KEY(tb, key) hash_string_pointer(key) +#define SH_EQUAL(tb, a, b) (strcmp(a, b) == 0) +#define SH_SCOPE static inline +#define SH_DECLARE +#define SH_DEFINE +#include "lib/simplehash.h" + +struct IncrementalBackupInfo +{ + /* Memory context for this object and its subsidiary objects. */ + MemoryContext mcxt; + + /* Temporary buffer for storing the manifest while parsing it. */ + StringInfoData buf; + + /* WAL ranges extracted from the backup manifest. */ + List *manifest_wal_ranges; + + /* + * Files extracted from the backup manifest. + * + * We don't really need this information, because we use WAL summaries to + * figure what's changed. It would be unsafe to just rely on the list of + * files that existed before, because it's possible for a file to be + * removed and a new one created with the same name and different + * contents. In such cases, the whole file must still be sent. We can tell + * from the WAL summaries whether that happened, but not from the file + * list. + * + * Nonetheless, this data is useful for sanity checking. If a file that we + * think we shouldn't need to send is not present in the manifest for the + * prior backup, something has gone terribly wrong. We retain the file + * names and sizes, but not the checksums or last modified times, for + * which we have no use. + * + * One significant downside of storing this data is that it consumes + * memory. If that turns out to be a problem, we might have to decide not + * to retain this information, or to make it optional. + */ + backup_file_hash *manifest_files; + + /* + * Block-reference table for the incremental backup. + * + * It's possible that storing the entire block-reference table in memory + * will be a problem for some users. The in-memory format that we're using + * here is pretty efficient, converging to little more than 1 bit per + * block for relation forks with large numbers of modified blocks. It's + * possible, however, that if you try to perform an incremental backup of + * a database with a sufficiently large number of relations on a + * sufficiently small machine, you could run out of memory here. If that + * turns out to be a problem in practice, we'll need to be more clever. + */ + BlockRefTable *brtab; +}; + +static void manifest_process_file(JsonManifestParseContext *, + char *pathname, + size_t size, + pg_checksum_type checksum_type, + int checksum_length, + uint8 *checksum_payload); +static void manifest_process_wal_range(JsonManifestParseContext *, + TimeLineID tli, + XLogRecPtr start_lsn, + XLogRecPtr end_lsn); +static void manifest_report_error(JsonManifestParseContext *ib, + const char *fmt,...) + pg_attribute_printf(2, 3) pg_attribute_noreturn(); + +/* + * Create a new object for storing information extracted from the manifest + * supplied when creating an incremental backup. + */ +IncrementalBackupInfo * +CreateIncrementalBackupInfo(MemoryContext mcxt) +{ + IncrementalBackupInfo *ib; + MemoryContext oldcontext; + + oldcontext = MemoryContextSwitchTo(mcxt); + + ib = palloc0(sizeof(IncrementalBackupInfo)); + ib->mcxt = mcxt; + initStringInfo(&ib->buf); + + /* + * It's hard to guess how many files a "typical" installation will have in + * the data directory, but a fresh initdb creates almost 1000 files as of + * this writing, so it seems to make sense for our estimate to + * substantially higher. + */ + ib->manifest_files = backup_file_create(mcxt, 10000, NULL); + + MemoryContextSwitchTo(oldcontext); + + return ib; +} + +/* + * Before taking an incremental backup, the caller must supply the backup + * manifest from a prior backup. Each chunk of manifest data recieved + * from the client should be passed to this function. + */ +void +AppendIncrementalManifestData(IncrementalBackupInfo *ib, const char *data, + int len) +{ + MemoryContext oldcontext; + + /* Switch to our memory context. */ + oldcontext = MemoryContextSwitchTo(ib->mcxt); + + /* + * XXX. Our json parser is at present incapable of parsing json blobs + * incrementally, so we have to accumulate the entire backup manifest + * before we can do anything with it. This should really be fixed, since + * some users might have very large numbers of files in the data + * directory. + */ + appendBinaryStringInfo(&ib->buf, data, len); + + /* Switch back to previous memory context. */ + MemoryContextSwitchTo(oldcontext); +} + +/* + * Finalize an IncrementalBackupInfo object after all manifest data has + * been supplied via calls to AppendIncrementalManifestData. + */ +void +FinalizeIncrementalManifest(IncrementalBackupInfo *ib) +{ + JsonManifestParseContext context; + MemoryContext oldcontext; + + /* Switch to our memory context. */ + oldcontext = MemoryContextSwitchTo(ib->mcxt); + + /* Parse the manifest. */ + context.private_data = ib; + context.perfile_cb = manifest_process_file; + context.perwalrange_cb = manifest_process_wal_range; + context.error_cb = manifest_report_error; + json_parse_manifest(&context, ib->buf.data, ib->buf.len); + + /* Done with the buffer, so release memory. */ + pfree(ib->buf.data); + ib->buf.data = NULL; + + /* Switch back to previous memory context. */ + MemoryContextSwitchTo(oldcontext); +} + +/* + * Prepare to take an incremental backup. + * + * Before this function is called, AppendIncrementalManifestData and + * FinalizeIncrementalManifest should have already been called to pass all + * the manifest data to this object. + * + * This function performs sanity checks on the data extracted from the + * manifest and figures out for which WAL ranges we need summaries, and + * whether those summaries are available. Then, it reads and combines the + * data from those summary files. It also updates the backup_state with the + * reference TLI and LSN for the prior backup. + */ +void +PrepareForIncrementalBackup(IncrementalBackupInfo *ib, + BackupState *backup_state) +{ + MemoryContext oldcontext; + List *expectedTLEs; + List *all_wslist, + *required_wslist = NIL; + ListCell *lc; + TimeLineHistoryEntry **tlep; + int num_wal_ranges; + int i; + bool found_backup_start_tli = false; + TimeLineID earliest_wal_range_tli = 0; + XLogRecPtr earliest_wal_range_start_lsn; + TimeLineID latest_wal_range_tli = 0; + XLogRecPtr summarized_lsn; + + Assert(ib->buf.data == NULL); + + /* Switch to our memory context. */ + oldcontext = MemoryContextSwitchTo(ib->mcxt); + + /* + * Match up the TLIs that appear in the WAL ranges of the backup manifest + * with those that appear in this server's timeline history. We expect + * every backup_wal_range to match to a TimeLineHistoryEntry; if it does + * not, that's an error. + * + * This loop also decides which of the WAL ranges is the manifest is most + * ancient and which one is the newest, according to the timeline history + * of this server, and stores TLIs of those WAL ranges into + * earliest_wal_range_tli and latest_wal_range_tli. It also updates + * earliest_wal_range_start_lsn to the start LSN of the WAL range for + * earliest_wal_range_tli. + * + * Note that the return value of readTimeLineHistory puts the latest + * timeline at the beginning of the list, not the end. Hence, the earliest + * TLI is the one that occurs nearest the end of the list returned by + * readTimeLineHistory, and the latest TLI is the one that occurs closest + * to the beginning. + */ + expectedTLEs = readTimeLineHistory(backup_state->starttli); + num_wal_ranges = list_length(ib->manifest_wal_ranges); + tlep = palloc0(num_wal_ranges * sizeof(TimeLineHistoryEntry *)); + for (i = 0; i < num_wal_ranges; ++i) + { + backup_wal_range *range = list_nth(ib->manifest_wal_ranges, i); + bool saw_earliest_wal_range_tli = false; + bool saw_latest_wal_range_tli = false; + + /* Search this server's history for this WAL range's TLI. */ + foreach(lc, expectedTLEs) + { + TimeLineHistoryEntry *tle = lfirst(lc); + + if (tle->tli == range->tli) + { + tlep[i] = tle; + break; + } + + if (tle->tli == earliest_wal_range_tli) + saw_earliest_wal_range_tli = true; + if (tle->tli == latest_wal_range_tli) + saw_latest_wal_range_tli = true; + } + + /* + * An incremental backup can only be taken relative to a backup that + * represents a previous state of this server. If the backup requires + * WAL from a timeline that's not in our history, that definitely + * isn't the case. + */ + if (tlep[i] == NULL) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("timeline %u found in manifest, but not in this server's history", + range->tli))); + + /* + * If we found this TLI in the server's history before encountering + * the latest TLI seen so far in the server's history, then this TLI + * is the latest one seen so far. + * + * If on the other hand we saw the earliest TLI seen so far before + * finding this TLI, this TLI is earlier than the earliest one seen so + * far. And if this is the first TLI for which we've searched, it's + * also the earliest one seen so far. + * + * On the first loop iteration, both things should necessarily be + * true. + */ + if (!saw_latest_wal_range_tli) + latest_wal_range_tli = range->tli; + if (earliest_wal_range_tli == 0 || saw_earliest_wal_range_tli) + { + earliest_wal_range_tli = range->tli; + earliest_wal_range_start_lsn = range->start_lsn; + } + } + + /* + * Propagate information about the prior backup into the backup_label that + * will be generated for this backup. + */ + backup_state->istartpoint = earliest_wal_range_start_lsn; + backup_state->istarttli = earliest_wal_range_tli; + + /* + * Sanity check start and end LSNs for the WAL ranges in the manifest. + * + * Commonly, there won't be any timeline switches during the prior backup + * at all, but if there are, they should happen at the same LSNs that this + * server switched timelines. + * + * Whether there are any timeline switches during the prior backup or not, + * the prior backup shouldn't require any WAL from a timeline prior to the + * start of that timeline. It also shouldn't require any WAL from later + * than the start of this backup. + * + * If any of these sanity checks fail, one possible explanation is that + * the user has generated WAL on the same timeline with the same LSNs more + * than once. For instance, if two standbys running on timeline 1 were + * both promoted and (due to a broken archiving setup) both selected new + * timeline ID 2, then it's possible that one of these checks might trip. + * + * Note that there are lots of ways for the user to do something very bad + * without tripping any of these checks, and they are not intended to be + * comprehensive. It's pretty hard to see how we could be certain of + * anything here. However, if there's a problem staring us right in the + * face, it's best to report it, so we do. + */ + for (i = 0; i < num_wal_ranges; ++i) + { + backup_wal_range *range = list_nth(ib->manifest_wal_ranges, i); + + if (range->tli == earliest_wal_range_tli) + { + if (range->start_lsn < tlep[i]->begin) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("manifest requires WAL from initial timeline %u starting at %X/%X, but that timeline begins at %X/%X", + range->tli, + LSN_FORMAT_ARGS(range->start_lsn), + LSN_FORMAT_ARGS(tlep[i]->begin)))); + } + else + { + if (range->start_lsn != tlep[i]->begin) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("manifest requires WAL from continuation timeline %u starting at %X/%X, but that timeline begins at %X/%X", + range->tli, + LSN_FORMAT_ARGS(range->start_lsn), + LSN_FORMAT_ARGS(tlep[i]->begin)))); + } + + if (range->tli == latest_wal_range_tli) + { + if (range->end_lsn > backup_state->startpoint) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("manifest requires WAL from final timeline %u ending at %X/%X, but this backup starts at %X/%X", + range->tli, + LSN_FORMAT_ARGS(range->end_lsn), + LSN_FORMAT_ARGS(backup_state->startpoint)))); + } + else + { + if (range->end_lsn != tlep[i]->end) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("manifest requires WAL from non-final timeline %u ending at %X/%X, but this server switched timelines at %X/%X", + range->tli, + LSN_FORMAT_ARGS(range->end_lsn), + LSN_FORMAT_ARGS(tlep[i]->end)))); + } + + } + + /* + * Wait for WAL summarization to catch up to the backup start LSN (but + * time out if it doesn't do so quickly enough). + */ + /* XXX make timeout configurable */ + summarized_lsn = WaitForWalSummarization(backup_state->startpoint, 60000); + if (summarized_lsn < backup_state->startpoint) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("timeout waiting for WAL summarization"), + errdetail("This backup requires WAL to be summarized up to %X/%X, but summarizer has only reached %X/%X.", + LSN_FORMAT_ARGS(backup_state->startpoint), + LSN_FORMAT_ARGS(summarized_lsn)))); + + /* + * Retrieve a list of all WAL summaries on any timeline that overlap with + * the LSN range of interest. We could instead call GetWalSummaries() once + * per timeline in the loop that follows, but that would involve reading + * the directory multiple times. It should be mildly faster - and perhaps + * a bit safer - to do it just once. + */ + all_wslist = GetWalSummaries(0, earliest_wal_range_start_lsn, + backup_state->startpoint); + + /* + * We need WAL summaries for everything that happened during the prior + * backup and everything that happened afterward up until the point where + * the current backup started. + */ + foreach(lc, expectedTLEs) + { + TimeLineHistoryEntry *tle = lfirst(lc); + XLogRecPtr tli_start_lsn = tle->begin; + XLogRecPtr tli_end_lsn = tle->end; + XLogRecPtr tli_missing_lsn = InvalidXLogRecPtr; + List *tli_wslist; + + /* + * Working through the history of this server from the current + * timeline backwards, we skip everything until we find the timeline + * where this backup started. Most of the time, this means we won't + * skip anything at all, as it's unlikely that the timeline has + * changed since the beginning of the backup moments ago. + */ + if (tle->tli == backup_state->starttli) + { + found_backup_start_tli = true; + tli_end_lsn = backup_state->startpoint; + } + else if (!found_backup_start_tli) + continue; + + /* + * Find the summaries that overlap the LSN range of interest for this + * timeline. If this is the earliest timeline involved, the range of + * interest begins with the start LSN of the prior backup; otherwise, + * it begins at the LSN at which this timeline came into existence. If + * this is the latest TLI involved, the range of interest ends at the + * start LSN of the current backup; otherwise, it ends at the point + * where we switched from this timeline to the next one. + */ + if (tle->tli == earliest_wal_range_tli) + tli_start_lsn = earliest_wal_range_start_lsn; + tli_wslist = FilterWalSummaries(all_wslist, tle->tli, + tli_start_lsn, tli_end_lsn); + + /* + * There is no guarantee that the WAL summaries we found cover the + * entire range of LSNs for which summaries are required, or indeed + * that we found any WAL summaries at all. Check whether we have a + * problem of that sort. + */ + if (!WalSummariesAreComplete(tli_wslist, tli_start_lsn, tli_end_lsn, + &tli_missing_lsn)) + { + if (XLogRecPtrIsInvalid(tli_missing_lsn)) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("WAL summaries are required on timeline %u from %X/%X to %X/%X, but no summaries for that timeline and LSN range exist", + tle->tli, + LSN_FORMAT_ARGS(tli_start_lsn), + LSN_FORMAT_ARGS(tli_end_lsn)))); + else + ereport(ERROR, + (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("WAL summaries are required on timeline %u from %X/%X to %X/%X, but the summaries for that timeline and LSN range are incomplete", + tle->tli, + LSN_FORMAT_ARGS(tli_start_lsn), + LSN_FORMAT_ARGS(tli_end_lsn)), + errdetail("The first unsummarized LSN is this range is %X/%X.", + LSN_FORMAT_ARGS(tli_missing_lsn)))); + } + + /* + * Remember that we need to read these summaries. + * + * Technically, it's possible that this could read more files than + * required, since tli_wslist in theory could contain redundant + * summaries. For instance, if we have a summary from 0/10000000 to + * 0/20000000 and also one from 0/00000000 to 0/30000000, then the + * latter subsumes the former and the former could be ignored. + * + * We ignore this possibility because the WAL summarizer only tries to + * generate summaries that do not overlap. If somehow they exist, + * we'll do a bit of extra work but the results should still be + * correct. + */ + required_wslist = list_concat(required_wslist, tli_wslist); + + /* + * Timelines earlier than the one in which the prior backup began are + * not relevant. + */ + if (tle->tli == earliest_wal_range_tli) + break; + } + + /* + * Read all of the required block reference table files and merge all of + * the data into a single in-memory block reference table. + * + * See the comments for struct IncrementalBackupInfo for some thoughts on + * memory usage. + */ + ib->brtab = CreateEmptyBlockRefTable(); + foreach(lc, required_wslist) + { + WalSummaryFile *ws = lfirst(lc); + WalSummaryIO wsio; + BlockRefTableReader *reader; + RelFileLocator rlocator; + ForkNumber forknum; + BlockNumber limit_block; + BlockNumber blocks[BLOCKS_PER_READ]; + + wsio.file = OpenWalSummaryFile(ws, false); + wsio.filepos = 0; + ereport(DEBUG1, + (errmsg_internal("reading WAL summary file \"%s\"", + FilePathName(wsio.file)))); + reader = CreateBlockRefTableReader(ReadWalSummary, &wsio, + FilePathName(wsio.file), + ReportWalSummaryError, NULL); + while (BlockRefTableReaderNextRelation(reader, &rlocator, &forknum, + &limit_block)) + { + BlockRefTableSetLimitBlock(ib->brtab, &rlocator, + forknum, limit_block); + + while (1) + { + unsigned nblocks; + unsigned i; + + nblocks = BlockRefTableReaderGetBlocks(reader, blocks, + BLOCKS_PER_READ); + if (nblocks == 0) + break; + + for (i = 0; i < nblocks; ++i) + BlockRefTableMarkBlockModified(ib->brtab, &rlocator, + forknum, blocks[i]); + } + } + DestroyBlockRefTableReader(reader); + FileClose(wsio.file); + } + + /* Switch back to previous memory context. */ + MemoryContextSwitchTo(oldcontext); +} + +/* + * Get the pathname that should be used when a file is sent incrementally. + * + * The result is a palloc'd string. + */ +char * +GetIncrementalFilePath(Oid dboid, Oid spcoid, RelFileNumber relfilenumber, + ForkNumber forknum, unsigned segno) +{ + char *path; + char *lastslash; + char *ipath; + + path = GetRelationPath(dboid, spcoid, relfilenumber, InvalidBackendId, + forknum); + + lastslash = strrchr(path, '/'); + Assert(lastslash != NULL); + *lastslash = '\0'; + + if (segno > 0) + ipath = psprintf("%s/INCREMENTAL.%s.%u", path, lastslash + 1, segno); + else + ipath = psprintf("%s/INCREMENTAL.%s", path, lastslash + 1); + + pfree(path); + + return ipath; +} + +/* + * How should we back up a particular file as part of an incremental backup? + * + * If the return value is BACK_UP_FILE_FULLY, caller should back up the whole + * file just as if this were not an incremental backup. + * + * If the return value is BACK_UP_FILE_INCREMENTALLY, caller should include + * an incremental file in the backup instead of the entire file. On return, + * *num_blocks_required will be set to the number of blocks that need to be + * sent, and the actual block numbers will have been stored in + * relative_block_numbers, which should be an array of at least RELSEG_SIZE. + * In addition, *truncation_block_length will be set to the value that should + * be included in the incremental file. + * + * If the return value is DO_NOT_BACK_UP_FILE, the caller should not include + * the file in the backup at all. + */ +FileBackupMethod +GetFileBackupMethod(IncrementalBackupInfo *ib, char *path, + Oid dboid, Oid spcoid, + RelFileNumber relfilenumber, ForkNumber forknum, + unsigned segno, size_t size, + unsigned *num_blocks_required, + BlockNumber *relative_block_numbers, + unsigned *truncation_block_length) +{ + BlockNumber absolute_block_numbers[RELSEG_SIZE]; + BlockNumber limit_block; + BlockNumber start_blkno; + BlockNumber stop_blkno; + RelFileLocator rlocator; + BlockRefTableEntry *brtentry; + unsigned i; + unsigned nblocks; + + /* Should only be called after PrepareForIncrementalBackup. */ + Assert(ib->buf.data == NULL); + + /* + * dboid could be InvalidOid if shared rel, but spcoid and relfilenumber + * should have legal values. + */ + Assert(OidIsValid(spcoid)); + Assert(RelFileNumberIsValid(relfilenumber)); + + /* + * If the file size is too large or not a multiple of BLCKSZ, then + * something weird is happening, so give up and send the whole file. + */ + if ((size % BLCKSZ) != 0 || size / BLCKSZ > RELSEG_SIZE) + return BACK_UP_FILE_FULLY; + + /* + * The free-space map fork is not properly WAL-logged, so we need to + * backup the entire file every time. + */ + if (forknum == FSM_FORKNUM) + return BACK_UP_FILE_FULLY; + + /* + * Check whether this file is part of the prior backup. If it isn't, back + * up the whole file. + */ + if (backup_file_lookup(ib->manifest_files, path) == NULL) + { + char *ipath; + + ipath = GetIncrementalFilePath(dboid, spcoid, relfilenumber, + forknum, segno); + if (backup_file_lookup(ib->manifest_files, ipath) == NULL) + return BACK_UP_FILE_FULLY; + } + + /* Look up the block reference table entry. */ + rlocator.spcOid = spcoid; + rlocator.dbOid = dboid; + rlocator.relNumber = relfilenumber; + brtentry = BlockRefTableGetEntry(ib->brtab, &rlocator, forknum, + &limit_block); + + /* + * If there is no entry, then there have been no WAL-logged changes to the + * relation since the predecessor backup was taken, so we can back it up + * incrementally and need not include any modified blocks. + * + * However, if the file is zero-length, we should do a full backup, + * because an incremental file is always more than zero length, and it's + * silly to take an incremental backup when a full backup would be + * smaller. + */ + if (brtentry == NULL) + { + *num_blocks_required = 0; + *truncation_block_length = size / BLCKSZ; + if (size == 0) + return BACK_UP_FILE_FULLY; + return BACK_UP_FILE_INCREMENTALLY; + } + + /* + * If the limit_block is less than or equal to the point where this + * segment starts, send the whole file. + */ + if (limit_block <= segno * RELSEG_SIZE) + return BACK_UP_FILE_FULLY; + + /* + * Get relevant entries from the block reference table entry. + * + * We shouldn't overflow computing the start or stop block numbers, but if + * it manages to happen somehow, detect it and throw an error. + */ + start_blkno = segno * RELSEG_SIZE; + stop_blkno = start_blkno + (size / BLCKSZ); + if (start_blkno / RELSEG_SIZE != segno || stop_blkno < start_blkno) + ereport(ERROR, + errcode(ERRCODE_INTERNAL_ERROR), + errmsg_internal("overflow computing block number bounds for segment %u with size %lu", + segno, size)); + nblocks = BlockRefTableEntryGetBlocks(brtentry, start_blkno, stop_blkno, + absolute_block_numbers, RELSEG_SIZE); + Assert(nblocks <= RELSEG_SIZE); + + /* + * If we're going to have to send nearly all of the blocks, then just send + * the whole file, because that won't require much extra storage or + * transfer and will speed up and simplify backup restoration. It's not + * clear what threshold is most appropriate here and perhaps it ought to + * be configurable, but for now we're just going to say that if we'd need + * to send 90% of the blocks anyway, give up and send the whole file. + * + * NB: If you change the threshold here, at least make sure to back up the + * file fully when every single block must be sent, because there's + * nothing good about sending an incremental file in that case. + */ + if (nblocks * BLCKSZ > size * 0.9) + return BACK_UP_FILE_FULLY; + + /* + * Looks like we can send an incremental file. + * + * Return the relevant details to the caller, transposing absolute block + * numbers to relative block numbers. + * + * The truncation block length is the minimum length of the reconstructed + * file. Any block numbers below this threshold that are not present in + * the backup need to be fetched from the prior backup. At or above this + * threshold, blocks should only be included in the result if they are + * present in the backup. (This may require inserting zero blocks if the + * blocks included in the backup are non-consecutive.) + */ + for (i = 0; i < nblocks; ++i) + relative_block_numbers[i] = absolute_block_numbers[i] - start_blkno; + *num_blocks_required = nblocks; + *truncation_block_length = + Min(size / BLCKSZ, limit_block - segno * RELSEG_SIZE); + return BACK_UP_FILE_INCREMENTALLY; +} + +/* + * Compute the size for an incremental file containing a given number of blocks. + */ +extern size_t +GetIncrementalFileSize(unsigned num_blocks_required) +{ + size_t result; + + /* Make sure we're not going to overflow. */ + Assert(num_blocks_required <= RELSEG_SIZE); + + /* + * Three four byte quantities (magic number, truncation block length, + * block count) followed by block numbers followed by block contents. + */ + result = 3 * sizeof(uint32); + result += (BLCKSZ + sizeof(BlockNumber)) * num_blocks_required; + + return result; +} + +/* + * Helper function for filemap hash table. + */ +static uint32 +hash_string_pointer(const char *s) +{ + unsigned char *ss = (unsigned char *) s; + + return hash_bytes(ss, strlen(s)); +} + +/* + * This callback is invoked for each file mentioned in the backup manifest. + * + * We store the path to each file and the size of each file for sanity-checking + * purposes. For further details, see comments for IncrementalBackupInfo. + */ +static void +manifest_process_file(JsonManifestParseContext *context, + char *pathname, size_t size, + pg_checksum_type checksum_type, + int checksum_length, + uint8 *checksum_payload) +{ + IncrementalBackupInfo *ib = context->private_data; + backup_file_entry *entry; + bool found; + + entry = backup_file_insert(ib->manifest_files, pathname, &found); + if (!found) + { + entry->path = MemoryContextStrdup(ib->manifest_files->ctx, + pathname); + entry->size = size; + } +} + +/* + * This callback is invoked for each WAL range mentioned in the backup + * manifest. + * + * We're just interested in learning the oldest LSN and the corresponding TLI + * that appear in any WAL range. + */ +static void +manifest_process_wal_range(JsonManifestParseContext *context, + TimeLineID tli, XLogRecPtr start_lsn, + XLogRecPtr end_lsn) +{ + IncrementalBackupInfo *ib = context->private_data; + backup_wal_range *range = palloc(sizeof(backup_wal_range)); + + range->tli = tli; + range->start_lsn = start_lsn; + range->end_lsn = end_lsn; + ib->manifest_wal_ranges = lappend(ib->manifest_wal_ranges, range); +} + +/* + * This callback is invoked if an error occurs while parsing the backup + * manifest. + */ +static void +manifest_report_error(JsonManifestParseContext *context, const char *fmt,...) +{ + StringInfoData errbuf; + + initStringInfo(&errbuf); + + for (;;) + { + va_list ap; + int needed; + + va_start(ap, fmt); + needed = appendStringInfoVA(&errbuf, fmt, ap); + va_end(ap); + if (needed == 0) + break; + enlargeStringInfo(&errbuf, needed); + } + + ereport(ERROR, + errmsg_internal("%s", errbuf.data)); +} diff --git a/src/backend/backup/meson.build b/src/backend/backup/meson.build index 11a79bbf80..19c355ceca 100644 --- a/src/backend/backup/meson.build +++ b/src/backend/backup/meson.build @@ -5,6 +5,7 @@ backend_sources += files( 'basebackup.c', 'basebackup_copy.c', 'basebackup_gzip.c', + 'basebackup_incremental.c', 'basebackup_lz4.c', 'basebackup_progress.c', 'basebackup_server.c', @@ -12,4 +13,6 @@ backend_sources += files( 'basebackup_target.c', 'basebackup_throttle.c', 'basebackup_zstd.c', + 'walsummary.c', + 'walsummaryfuncs.c' ) diff --git a/src/backend/backup/walsummary.c b/src/backend/backup/walsummary.c new file mode 100644 index 0000000000..ebf4ea038d --- /dev/null +++ b/src/backend/backup/walsummary.c @@ -0,0 +1,356 @@ +/*------------------------------------------------------------------------- + * + * walsummary.c + * Functions for accessing and managing WAL summary data. + * + * Portions Copyright (c) 2010-2022, PostgreSQL Global Development Group + * + * src/backend/backup/walsummary.c + * + *------------------------------------------------------------------------- + */ + +#include "postgres.h" + +#include <sys/stat.h> +#include <unistd.h> + +#include "access/xlog_internal.h" +#include "backup/walsummary.h" +#include "utils/wait_event.h" + +static bool IsWalSummaryFilename(char *filename); +static int ListComparatorForWalSummaryFiles(const ListCell *a, + const ListCell *b); + +/* + * Get a list of WAL summaries. + * + * If tli != 0, only WAL summaries with the indicated TLI will be included. + * + * If start_lsn != InvalidXLogRecPtr, only summaries that end before the + * indicated LSN will be included. + * + * If end_lsn != InvalidXLogRecPtr, only summaries that start before the + * indicated LSN will be included. + * + * The intent is that you can call GetWalSummaries(tli, start_lsn, end_lsn) + * to get all WAL summaries on the indicated timeline that overlap the + * specified LSN range. + */ +List * +GetWalSummaries(TimeLineID tli, XLogRecPtr start_lsn, XLogRecPtr end_lsn) +{ + DIR *sdir; + struct dirent *dent; + List *result = NIL; + + sdir = AllocateDir(XLOGDIR "/summaries"); + while ((dent = ReadDir(sdir, XLOGDIR "/summaries")) != NULL) + { + WalSummaryFile *ws; + uint32 tmp[5]; + TimeLineID file_tli; + XLogRecPtr file_start_lsn; + XLogRecPtr file_end_lsn; + + /* Decode filename, or skip if it's not in the expected format. */ + if (!IsWalSummaryFilename(dent->d_name)) + continue; + sscanf(dent->d_name, "%08X%08X%08X%08X%08X", + &tmp[0], &tmp[1], &tmp[2], &tmp[3], &tmp[4]); + file_tli = tmp[0]; + file_start_lsn = ((uint64) tmp[1]) << 32 | tmp[2]; + file_end_lsn = ((uint64) tmp[3]) << 32 | tmp[4]; + + /* Skip if it doesn't match the filter criteria. */ + if (tli != 0 && tli != file_tli) + continue; + if (!XLogRecPtrIsInvalid(start_lsn) && start_lsn > file_end_lsn) + continue; + if (!XLogRecPtrIsInvalid(end_lsn) && end_lsn < file_start_lsn) + continue; + + /* Add it to the list. */ + ws = palloc(sizeof(WalSummaryFile)); + ws->tli = file_tli; + ws->start_lsn = file_start_lsn; + ws->end_lsn = file_end_lsn; + result = lappend(result, ws); + } + FreeDir(sdir); + + return result; +} + +/* + * Build a new list of WAL summaries based on an existing list, but filtering + * out summaries that don't match the search parameters. + * + * If tli != 0, only WAL summaries with the indicated TLI will be included. + * + * If start_lsn != InvalidXLogRecPtr, only summaries that end before the + * indicated LSN will be included. + * + * If end_lsn != InvalidXLogRecPtr, only summaries that start before the + * indicated LSN will be included. + */ +List * +FilterWalSummaries(List *wslist, TimeLineID tli, + XLogRecPtr start_lsn, XLogRecPtr end_lsn) +{ + List *result = NIL; + ListCell *lc; + + /* Loop over input. */ + foreach(lc, wslist) + { + WalSummaryFile *ws = lfirst(lc); + + /* Skip if it doesn't match the filter criteria. */ + if (tli != 0 && tli != ws->tli) + continue; + if (!XLogRecPtrIsInvalid(start_lsn) && start_lsn > ws->end_lsn) + continue; + if (!XLogRecPtrIsInvalid(end_lsn) && end_lsn < ws->start_lsn) + continue; + + /* Add it to the result list. */ + result = lappend(result, ws); + } + + return result; +} + +/* + * Check whether the supplied list of WalSummaryFile objects covers the + * whole range of LSNs from start_lsn to end_lsn. This function ignores + * timelines, so the caller should probably filter using the appropriate + * timeline before calling this. + * + * If the whole range of LSNs is covered, returns true, otherwise false. + * If false is returned, *missing_lsn is set either to InvalidXLogRecPtr + * if there are no WAL summary files in the input list, or to the first LSN + * in the range that is not covered by a WAL summary file in the input list. + */ +bool +WalSummariesAreComplete(List *wslist, XLogRecPtr start_lsn, + XLogRecPtr end_lsn, XLogRecPtr *missing_lsn) +{ + XLogRecPtr current_lsn = start_lsn; + ListCell *lc; + + /* Special case for empty list. */ + if (wslist == NIL) + { + *missing_lsn = InvalidXLogRecPtr; + return false; + } + + /* Make a private copy of the list and sort it by start LSN. */ + wslist = list_copy(wslist); + list_sort(wslist, ListComparatorForWalSummaryFiles); + + /* + * Consider summary files in order of increasing start_lsn, advancing the + * known-summarized range from start_lsn toward end_lsn. + * + * Normally, the summary files should cover non-overlapping WAL ranges, + * but this algorithm is intended to be correct even in case of overlap. + */ + foreach(lc, wslist) + { + WalSummaryFile *ws = lfirst(lc); + + if (ws->start_lsn > current_lsn) + { + /* We found a gap. */ + break; + } + if (ws->end_lsn > current_lsn) + { + /* + * Next summary extends beyond end of previous summary, so extend + * the end of the range known to be summarized. + */ + current_lsn = ws->end_lsn; + + /* + * If the range we know to be summarized has reached the required + * end LSN, we have proved completeness. + */ + if (current_lsn >= end_lsn) + return true; + } + } + + /* + * We either ran out of summary files without reaching the end LSN, or we + * hit a gap in the sequence that resulted in us bailing out of the loop + * above. + */ + *missing_lsn = current_lsn; + return false; +} + +/* + * Open a WAL summary file. + * + * This will throw an error in case of trouble. As an exception, if + * missing_ok = true and the trouble is specifically that the file does + * not exist, it will not throw an error and will return a value less than 0. + */ +File +OpenWalSummaryFile(WalSummaryFile *ws, bool missing_ok) +{ + char path[MAXPGPATH]; + File file; + + snprintf(path, MAXPGPATH, + XLOGDIR "/summaries/%08X%08X%08X%08X%08X.summary", + ws->tli, + LSN_FORMAT_ARGS(ws->start_lsn), + LSN_FORMAT_ARGS(ws->end_lsn)); + + file = PathNameOpenFile(path, O_RDONLY); + if (file < 0 && (errno != EEXIST || !missing_ok)) + ereport(ERROR, + (errcode_for_file_access(), + errmsg("could not open file \"%s\": %m", path))); + + return file; +} + +/* + * Remove a WAL summary file if the last modification time precedes the + * cutoff time. + */ +void +RemoveWalSummaryIfOlderThan(WalSummaryFile *ws, time_t cutoff_time) +{ + char path[MAXPGPATH]; + struct stat statbuf; + + snprintf(path, MAXPGPATH, + XLOGDIR "/summaries/%08X%08X%08X%08X%08X.summary", + ws->tli, + LSN_FORMAT_ARGS(ws->start_lsn), + LSN_FORMAT_ARGS(ws->end_lsn)); + + if (lstat(path, &statbuf) != 0) + { + if (errno == ENOENT) + return; + ereport(ERROR, + (errcode_for_file_access(), + errmsg("could not stat file \"%s\": %m", path))); + } + if (statbuf.st_mtime >= cutoff_time) + return; + if (unlink(path) != 0) + ereport(ERROR, + (errcode_for_file_access(), + errmsg("could not stat file \"%s\": %m", path))); + ereport(DEBUG2, + (errmsg_internal("removing file \"%s\"", path))); +} + +/* + * Test whether a filename looks like a WAL summary file. + */ +static bool +IsWalSummaryFilename(char *filename) +{ + return strspn(filename, "0123456789ABCDEF") == 40 && + strcmp(filename + 40, ".summary") == 0; +} + +/* + * Data read callback for use with CreateBlockRefTableReader. + */ +int +ReadWalSummary(void *wal_summary_io, void *data, int length) +{ + WalSummaryIO *io = wal_summary_io; + int nbytes; + + nbytes = FileRead(io->file, data, length, io->filepos, + WAIT_EVENT_WAL_SUMMARY_READ); + if (nbytes < 0) + ereport(ERROR, + (errcode_for_file_access(), + errmsg("could not write file \"%s\": %m", + FilePathName(io->file)))); + + io->filepos += nbytes; + return nbytes; +} + +/* + * Data write callback for use with WriteBlockRefTable. + */ +int +WriteWalSummary(void *wal_summary_io, void *data, int length) +{ + WalSummaryIO *io = wal_summary_io; + int nbytes; + + nbytes = FileWrite(io->file, data, length, io->filepos, + WAIT_EVENT_WAL_SUMMARY_WRITE); + if (nbytes < 0) + ereport(ERROR, + (errcode_for_file_access(), + errmsg("could not write file \"%s\": %m", + FilePathName(io->file)))); + if (nbytes != length) + ereport(ERROR, + (errcode_for_file_access(), + errmsg("could not write file \"%s\": wrote only %d of %d bytes at offset %u", + FilePathName(io->file), nbytes, + length, (unsigned) io->filepos), + errhint("Check free disk space."))); + + io->filepos += nbytes; + return nbytes; +} + +/* + * Error-reporting callback for use with CreateBlockRefTableReader. + */ +void +ReportWalSummaryError(void *callback_arg, char *fmt,...) +{ + StringInfoData buf; + va_list ap; + int needed; + + initStringInfo(&buf); + for (;;) + { + va_start(ap, fmt); + needed = appendStringInfoVA(&buf, fmt, ap); + va_end(ap); + if (needed == 0) + break; + enlargeStringInfo(&buf, needed); + } + ereport(ERROR, + errcode(ERRCODE_DATA_CORRUPTED), + errmsg_internal("%s", buf.data)); +} + +/* + * Comparator to sort a List of WalSummaryFile objects by start_lsn. + */ +static int +ListComparatorForWalSummaryFiles(const ListCell *a, const ListCell *b) +{ + WalSummaryFile *ws1 = lfirst(a); + WalSummaryFile *ws2 = lfirst(b); + + if (ws1->start_lsn < ws2->start_lsn) + return -1; + if (ws1->start_lsn > ws2->start_lsn) + return 1; + return 0; +} diff --git a/src/backend/backup/walsummaryfuncs.c b/src/backend/backup/walsummaryfuncs.c new file mode 100644 index 0000000000..2e77d38b4a --- /dev/null +++ b/src/backend/backup/walsummaryfuncs.c @@ -0,0 +1,169 @@ +/*------------------------------------------------------------------------- + * + * walsummaryfuncs.c + * SQL-callable functions for accessing WAL summary data. + * + * Portions Copyright (c) 2010-2022, PostgreSQL Global Development Group + * + * src/backend/backup/walsummaryfuncs.c + * + *------------------------------------------------------------------------- + */ + +#include "postgres.h" + +#include "backup/walsummary.h" +#include "common/blkreftable.h" +#include "funcapi.h" +#include "miscadmin.h" +#include "utils/fmgrprotos.h" +#include "utils/pg_lsn.h" + +#define NUM_WS_ATTS 3 +#define NUM_SUMMARY_ATTS 6 +#define MAX_BLOCKS_PER_CALL 256 + +/* + * List the WAL summary files available in pg_wal/summaries. + */ +Datum +pg_available_wal_summaries(PG_FUNCTION_ARGS) +{ + ReturnSetInfo *rsi; + List *wslist; + ListCell *lc; + Datum values[NUM_WS_ATTS]; + bool nulls[NUM_WS_ATTS]; + + InitMaterializedSRF(fcinfo, 0); + rsi = (ReturnSetInfo *) fcinfo->resultinfo; + + memset(nulls, 0, sizeof(nulls)); + + wslist = GetWalSummaries(0, InvalidXLogRecPtr, InvalidXLogRecPtr); + foreach(lc, wslist) + { + WalSummaryFile *ws = (WalSummaryFile *) lfirst(lc); + HeapTuple tuple; + + CHECK_FOR_INTERRUPTS(); + + values[0] = Int64GetDatum((int64) ws->tli); + values[1] = LSNGetDatum(ws->start_lsn); + values[2] = LSNGetDatum(ws->end_lsn); + + tuple = heap_form_tuple(rsi->setDesc, values, nulls); + tuplestore_puttuple(rsi->setResult, tuple); + } + + return (Datum) 0; +} + +/* + * List the contents of a WAL summary file identified by TLI, start LSN, + * and end LSN. + */ +Datum +pg_wal_summary_contents(PG_FUNCTION_ARGS) +{ + ReturnSetInfo *rsi; + Datum values[NUM_SUMMARY_ATTS]; + bool nulls[NUM_SUMMARY_ATTS]; + WalSummaryFile ws; + WalSummaryIO io; + BlockRefTableReader *reader; + int64 raw_tli; + RelFileLocator rlocator; + ForkNumber forknum; + BlockNumber limit_block; + + InitMaterializedSRF(fcinfo, 0); + rsi = (ReturnSetInfo *) fcinfo->resultinfo; + memset(nulls, 0, sizeof(nulls)); + + /* + * Since the timeline could at least in theory be more than 2^31, and + * since we don't have unsigned types at the SQL level, it is passed as a + * 64-bit integer. Test whether it's out of range. + */ + raw_tli = PG_GETARG_INT64(0); + if (raw_tli < 1 || raw_tli > PG_INT32_MAX) + ereport(ERROR, + errcode(ERRCODE_INVALID_PARAMETER_VALUE), + errmsg("invalid timeline %lld", (long long) raw_tli)); + + /* Prepare to read the specified WAL summry file. */ + ws.tli = (TimeLineID) raw_tli; + ws.start_lsn = PG_GETARG_LSN(1); + ws.end_lsn = PG_GETARG_LSN(2); + io.filepos = 0; + io.file = OpenWalSummaryFile(&ws, false); + reader = CreateBlockRefTableReader(ReadWalSummary, &io, + FilePathName(io.file), + ReportWalSummaryError, NULL); + + /* Loop over relation forks. */ + while (BlockRefTableReaderNextRelation(reader, &rlocator, &forknum, + &limit_block)) + { + BlockNumber blocks[MAX_BLOCKS_PER_CALL]; + HeapTuple tuple; + + CHECK_FOR_INTERRUPTS(); + + values[0] = ObjectIdGetDatum(rlocator.relNumber); + values[1] = ObjectIdGetDatum(rlocator.spcOid); + values[2] = ObjectIdGetDatum(rlocator.dbOid); + values[3] = Int16GetDatum((int16) forknum); + + /* Loop over blocks within the current relation fork. */ + while (true) + { + unsigned nblocks; + unsigned i; + + CHECK_FOR_INTERRUPTS(); + + nblocks = BlockRefTableReaderGetBlocks(reader, blocks, + MAX_BLOCKS_PER_CALL); + if (nblocks == 0) + break; + + /* + * For each block that we specifically know to have been modified, + * emit a row with that block number and limit_block = false. + */ + values[5] = BoolGetDatum(false); + for (i = 0; i < nblocks; ++i) + { + values[4] = Int64GetDatum((int64) blocks[i]); + + tuple = heap_form_tuple(rsi->setDesc, values, nulls); + tuplestore_puttuple(rsi->setResult, tuple); + } + + /* + * If the limit block is not InvalidBlockNumber, emit an exta row + * with that block number and limit_block = true. + * + * There is no point in doing this when the limit_block is + * InvalidBlockNumber, because no block with that number or any + * higher number can ever exist. + */ + if (BlockNumberIsValid(limit_block)) + { + values[4] = Int64GetDatum((int64) limit_block); + values[5] = BoolGetDatum(true); + + tuple = heap_form_tuple(rsi->setDesc, values, nulls); + tuplestore_puttuple(rsi->setResult, tuple); + } + } + } + + /* Cleanup */ + DestroyBlockRefTableReader(reader); + FileClose(io.file); + + return (Datum) 0; +} diff --git a/src/backend/postmaster/Makefile b/src/backend/postmaster/Makefile index 047448b34e..367a46c617 100644 --- a/src/backend/postmaster/Makefile +++ b/src/backend/postmaster/Makefile @@ -24,6 +24,7 @@ OBJS = \ postmaster.o \ startup.o \ syslogger.o \ + walsummarizer.o \ walwriter.o include $(top_srcdir)/src/backend/common.mk diff --git a/src/backend/postmaster/auxprocess.c b/src/backend/postmaster/auxprocess.c index cae6feb356..0c15c1777d 100644 --- a/src/backend/postmaster/auxprocess.c +++ b/src/backend/postmaster/auxprocess.c @@ -21,6 +21,7 @@ #include "postmaster/auxprocess.h" #include "postmaster/bgwriter.h" #include "postmaster/startup.h" +#include "postmaster/walsummarizer.h" #include "postmaster/walwriter.h" #include "replication/walreceiver.h" #include "storage/bufmgr.h" @@ -80,6 +81,9 @@ AuxiliaryProcessMain(AuxProcType auxtype) case WalReceiverProcess: MyBackendType = B_WAL_RECEIVER; break; + case WalSummarizerProcess: + MyBackendType = B_WAL_SUMMARIZER; + break; default: elog(PANIC, "unrecognized process type: %d", (int) MyAuxProcType); MyBackendType = B_INVALID; @@ -161,6 +165,10 @@ AuxiliaryProcessMain(AuxProcType auxtype) WalReceiverMain(); proc_exit(1); + case WalSummarizerProcess: + WalSummarizerMain(); + proc_exit(1); + default: elog(PANIC, "unrecognized process type: %d", (int) MyAuxProcType); proc_exit(1); diff --git a/src/backend/postmaster/meson.build b/src/backend/postmaster/meson.build index cda921fd10..a30eb6692f 100644 --- a/src/backend/postmaster/meson.build +++ b/src/backend/postmaster/meson.build @@ -12,5 +12,6 @@ backend_sources += files( 'postmaster.c', 'startup.c', 'syslogger.c', + 'walsummarizer.c', 'walwriter.c', ) diff --git a/src/backend/postmaster/postmaster.c b/src/backend/postmaster/postmaster.c index 9cb624eab8..86f6cf2feb 100644 --- a/src/backend/postmaster/postmaster.c +++ b/src/backend/postmaster/postmaster.c @@ -115,6 +115,7 @@ #include "postmaster/pgarch.h" #include "postmaster/postmaster.h" #include "postmaster/syslogger.h" +#include "postmaster/walsummarizer.h" #include "replication/logicallauncher.h" #include "replication/walsender.h" #include "storage/fd.h" @@ -252,6 +253,7 @@ static pid_t StartupPID = 0, CheckpointerPID = 0, WalWriterPID = 0, WalReceiverPID = 0, + WalSummarizerPID = 0, AutoVacPID = 0, PgArchPID = 0, SysLoggerPID = 0; @@ -443,6 +445,7 @@ static bool CreateOptsFile(int argc, char *argv[], char *fullprogname); static pid_t StartChildProcess(AuxProcType type); static void StartAutovacuumWorker(void); static void MaybeStartWalReceiver(void); +static void MaybeStartWalSummarizer(void); static void InitPostmasterDeathWatchHandle(void); /* @@ -562,6 +565,7 @@ static void ShmemBackendArrayRemove(Backend *bn); #define StartCheckpointer() StartChildProcess(CheckpointerProcess) #define StartWalWriter() StartChildProcess(WalWriterProcess) #define StartWalReceiver() StartChildProcess(WalReceiverProcess) +#define StartWalSummarizer() StartChildProcess(WalSummarizerProcess) /* Macros to check exit status of a child process */ #define EXIT_STATUS_0(st) ((st) == 0) @@ -1833,6 +1837,9 @@ ServerLoop(void) if (WalReceiverRequested) MaybeStartWalReceiver(); + /* If we need to start a WAL summarizer, try to do that now */ + MaybeStartWalSummarizer(); + /* Get other worker processes running, if needed */ if (StartWorkerNeeded || HaveCrashedWorker) maybe_start_bgworkers(); @@ -2657,6 +2664,8 @@ process_pm_reload_request(void) signal_child(WalWriterPID, SIGHUP); if (WalReceiverPID != 0) signal_child(WalReceiverPID, SIGHUP); + if (WalSummarizerPID != 0) + signal_child(WalSummarizerPID, SIGHUP); if (AutoVacPID != 0) signal_child(AutoVacPID, SIGHUP); if (PgArchPID != 0) @@ -3010,6 +3019,7 @@ process_pm_child_exit(void) BgWriterPID = StartBackgroundWriter(); if (WalWriterPID == 0) WalWriterPID = StartWalWriter(); + MaybeStartWalSummarizer(); /* * Likewise, start other special children as needed. In a restart @@ -3128,6 +3138,20 @@ process_pm_child_exit(void) continue; } + /* + * Was it the wal summarizer? Normal exit can be ignored; we'll start + * a new one at the next iteration of the postmaster's main loop, if + * necessary. Any other exit condition is treated as a crash. + */ + if (pid == WalSummarizerPID) + { + WalSummarizerPID = 0; + if (!EXIT_STATUS_0(exitstatus)) + HandleChildCrash(pid, exitstatus, + _("WAL summarizer process")); + continue; + } + /* * Was it the autovacuum launcher? Normal exit can be ignored; we'll * start a new one at the next iteration of the postmaster's main @@ -3523,6 +3547,12 @@ HandleChildCrash(int pid, int exitstatus, const char *procname) else if (WalReceiverPID != 0 && take_action) sigquit_child(WalReceiverPID); + /* Take care of the walsummarizer too */ + if (pid == WalSummarizerPID) + WalSummarizerPID = 0; + else if (WalSummarizerPID != 0 && take_action) + sigquit_child(WalSummarizerPID); + /* Take care of the autovacuum launcher too */ if (pid == AutoVacPID) AutoVacPID = 0; @@ -3673,6 +3703,8 @@ PostmasterStateMachine(void) signal_child(StartupPID, SIGTERM); if (WalReceiverPID != 0) signal_child(WalReceiverPID, SIGTERM); + if (WalSummarizerPID != 0) + signal_child(WalSummarizerPID, SIGTERM); /* checkpointer, archiver, stats, and syslogger may continue for now */ /* Now transition to PM_WAIT_BACKENDS state to wait for them to die */ @@ -3699,6 +3731,7 @@ PostmasterStateMachine(void) if (CountChildren(BACKEND_TYPE_ALL - BACKEND_TYPE_WALSND) == 0 && StartupPID == 0 && WalReceiverPID == 0 && + WalSummarizerPID == 0 && BgWriterPID == 0 && (CheckpointerPID == 0 || (!FatalError && Shutdown < ImmediateShutdown)) && @@ -3796,6 +3829,7 @@ PostmasterStateMachine(void) /* These other guys should be dead already */ Assert(StartupPID == 0); Assert(WalReceiverPID == 0); + Assert(WalSummarizerPID == 0); Assert(BgWriterPID == 0); Assert(CheckpointerPID == 0); Assert(WalWriterPID == 0); @@ -4017,6 +4051,8 @@ TerminateChildren(int signal) signal_child(WalWriterPID, signal); if (WalReceiverPID != 0) signal_child(WalReceiverPID, signal); + if (WalSummarizerPID != 0) + signal_child(WalSummarizerPID, signal); if (AutoVacPID != 0) signal_child(AutoVacPID, signal); if (PgArchPID != 0) @@ -5364,6 +5400,10 @@ StartChildProcess(AuxProcType type) ereport(LOG, (errmsg("could not fork WAL receiver process: %m"))); break; + case WalSummarizerProcess: + ereport(LOG, + (errmsg("could not fork WAL summarizer process: %m"))); + break; default: ereport(LOG, (errmsg("could not fork process: %m"))); @@ -5500,6 +5540,19 @@ MaybeStartWalReceiver(void) } } +/* + * MaybeStartWalSummarizer + * Start the WAL summarizer process, if not running and our state allows. + */ +static void +MaybeStartWalSummarizer(void) +{ + if (wal_summarize_mb != 0 && WalSummarizerPID == 0 && + (pmState == PM_RUN || pmState == PM_HOT_STANDBY) && + Shutdown <= SmartShutdown) + WalSummarizerPID = StartWalSummarizer(); +} + /* * Create the opts file diff --git a/src/backend/postmaster/walsummarizer.c b/src/backend/postmaster/walsummarizer.c new file mode 100644 index 0000000000..4ded951119 --- /dev/null +++ b/src/backend/postmaster/walsummarizer.c @@ -0,0 +1,1363 @@ +/*------------------------------------------------------------------------- + * + * walsummarizer.c + * + * Background process to perform WAL summarization, if it is enabled. + * It continuously scans the write-ahead log and periodically emits a + * summary file which indicates which blocks in which relation forks + * were modified by WAL records in the LSN range covered by the summary + * file. See walsummary.c and blkreftable.c for more details on the + * naming and contents of WAL summary files. + * + * If configured to do, this background process will also remove WAL + * summary files when the file timestamp is older than a configurable + * threshold (but only if the WAL has been removed first). + * + * Portions Copyright (c) 1996-2022, PostgreSQL Global Development Group + * + * IDENTIFICATION + * src/backend/postmaster/walsummarizer.c + * + *------------------------------------------------------------------------- + */ +#include "postgres.h" + +#include "access/timeline.h" +#include "access/xlog.h" +#include "access/xlog_internal.h" +#include "access/xlogrecovery.h" +#include "access/xlogutils.h" +#include "backup/walsummary.h" +#include "catalog/storage_xlog.h" +#include "common/blkreftable.h" +#include "libpq/pqsignal.h" +#include "miscadmin.h" +#include "postmaster/bgwriter.h" +#include "postmaster/interrupt.h" +#include "postmaster/walsummarizer.h" +#include "replication/walreceiver.h" +#include "storage/fd.h" +#include "storage/ipc.h" +#include "storage/lwlock.h" +#include "storage/latch.h" +#include "storage/proc.h" +#include "storage/procsignal.h" +#include "storage/shmem.h" +#include "storage/spin.h" +#include "utils/guc.h" +#include "utils/memutils.h" +#include "utils/wait_event.h" + +/* + * Data in shared memory related to WAL summarization. + */ +typedef struct +{ + /* + * These fields are protected by WALSummarizerLock. + * + * Until we've discovered what summary files already exist on disk and + * stored that information in shared memory, initialized is false and the + * other fields here contain no meaningful information. After that has + * been done, initialized is true. + * + * summarized_tli and summarized_lsn indicate the last LSN and TLI at + * which the next summary file will start. Normally, these are the LSN and + * TLI at which the last file ended; in such case, lsn_is_exact is true. + * If, however, the LSN is just an approximation, then lsn_is_exact is + * false. This can happen if, for example, there are no existing WAL + * summary files at startup. In that case, we have to derive the position + * at which to start summarizing from the WAL files that exist on disk, + * and so the LSN might point to the start of the next file even though + * that might happen to be in the middle of a WAL record. + * + * summarizer_pgprocno is the pgprocno value for the summarizer process, + * if one is running, or else INVALID_PGPROCNO. + * + * pending_lsn is used by the summarizer to advertise the ending LSN of a + * record it has recently read. It shouldn't ever be less than + * summarized_lsn, but might be greater, because the summarizer buffers + * data for a range of LSNs in memory before writing out a new file. + */ + bool initialized; + TimeLineID summarized_tli; + XLogRecPtr summarized_lsn; + bool lsn_is_exact; + int summarizer_pgprocno; + XLogRecPtr pending_lsn; + + /* + * This field handles its own synchronizaton. + */ + ConditionVariable summary_file_cv; +} WalSummarizerData; + +/* + * Private data for our xlogreader's page read callback. + */ +typedef struct +{ + TimeLineID tli; + bool historic; + XLogRecPtr read_upto; + bool end_of_wal; + bool waited; +} SummarizerReadLocalXLogPrivate; + +/* Pointer to shared memory state. */ +static WalSummarizerData *WalSummarizerCtl; + +/* + * When we reach end of WAL and need to read more, we sleep for a number of + * milliseconds that is a integer multiple of MS_PER_SLEEP_QUANTUM. This is + * the multiplier. It should vary between 1 and MAX_SLEEP_QUANTA, depending + * on system activity. See summarizer_wait_for_wal() for how we adjust this. + */ +static long sleep_quanta = 1; + +/* + * The sleep time will always be a multiple of 200ms and will not exceed + * thirty seconds (150 * 200 = 30 * 1000). Note that the timeout here needs + * to be substntially less than the maximum amount of time for which an + * incremental backup will wait for this process to catch up. Otherwise, an + * incremental backup might time out on an idle system just because we sleep + * for too long. + */ +#define MAX_SLEEP_QUANTA 150 +#define MS_PER_SLEEP_QUANTUM 200 + +/* + * This is a count of the number of pages of WAL that we've read since the + * last time we waited for more WAL to appear. + */ +static long pages_read_since_last_sleep = 0; + +/* + * Most recent RedoRecPtr value observed by MaybeRemoveOldWalSummaries. + */ +static XLogRecPtr redo_pointer_at_last_summary_removal = InvalidXLogRecPtr; + +/* + * GUC parameters + */ +int wal_summarize_mb = 256; +int wal_summarize_keep_time = 7 * 24 * 60; + +static XLogRecPtr GetLatestLSN(TimeLineID *tli); +static void HandleWalSummarizerInterrupts(void); +static XLogRecPtr SummarizeWAL(TimeLineID tli, bool historic, + XLogRecPtr start_lsn, bool exact, + XLogRecPtr cutoff_lsn, XLogRecPtr maximum_lsn); +static void SummarizeSmgrRecord(XLogReaderState *xlogreader, + BlockRefTable *brtab); +static void SummarizeXactRecord(XLogReaderState *xlogreader, + BlockRefTable *brtab); +static bool SummarizeXlogRecord(XLogReaderState *xlogreader); +static int summarizer_read_local_xlog_page(XLogReaderState *state, + XLogRecPtr targetPagePtr, + int reqLen, + XLogRecPtr targetRecPtr, + char *cur_page); +static void summarizer_wait_for_wal(void); +static void MaybeRemoveOldWalSummaries(void); + +/* + * Amount of shared memory required for this module. + */ +Size +WalSummarizerShmemSize(void) +{ + return sizeof(WalSummarizerData); +} + +/* + * Create or attach to shared memory segment for this module. + */ +void +WalSummarizerShmemInit(void) +{ + bool found; + + WalSummarizerCtl = (WalSummarizerData *) + ShmemInitStruct("Wal Summarizer Ctl", WalSummarizerShmemSize(), + &found); + + if (!found) + { + /* + * First time through, so initialize. + * + * We're just filling in dummy values here -- the real initialization + * will happen when GetOldestUnsummarizedLSN() is called for the first + * time. + */ + WalSummarizerCtl->initialized = false; + WalSummarizerCtl->summarized_tli = 0; + WalSummarizerCtl->summarized_lsn = InvalidXLogRecPtr; + WalSummarizerCtl->lsn_is_exact = false; + WalSummarizerCtl->summarizer_pgprocno = INVALID_PGPROCNO; + WalSummarizerCtl->pending_lsn = InvalidXLogRecPtr; + ConditionVariableInit(&WalSummarizerCtl->summary_file_cv); + } +} + +/* + * Entry point for walsummarizer process. + */ +void +WalSummarizerMain(void) +{ + sigjmp_buf local_sigjmp_buf; + MemoryContext context; + + /* + * Within this function, 'current_lsn' and 'current_tli' refer to the + * point from which the next WAL summary file should start. 'exact' is + * true if 'current_lsn' is known to be the start of a WAL recod or WAL + * segment, and false if it might be in the middle of a record someplace. + * + * 'switch_lsn' and 'switch_tli', if set, are the LSN at which we need to + * switch to a new timeline and the timeline to which we need to switch. + * If not set, we either haven't figured out the answers yet or we're + * already on the latest timeline. + */ + XLogRecPtr current_lsn; + TimeLineID current_tli; + bool exact; + XLogRecPtr switch_lsn = InvalidXLogRecPtr; + TimeLineID switch_tli = 0; + + ereport(DEBUG1, + (errmsg_internal("WAL summarizer started"))); + + /* + * Properly accept or ignore signals the postmaster might send us + * + * We have no particular use for SIGINT at the moment, but seems + * reasonable to treat like SIGTERM. + */ + pqsignal(SIGHUP, SignalHandlerForConfigReload); + pqsignal(SIGINT, SignalHandlerForShutdownRequest); + pqsignal(SIGTERM, SignalHandlerForShutdownRequest); + /* SIGQUIT handler was already set up by InitPostmasterChild */ + pqsignal(SIGALRM, SIG_IGN); + pqsignal(SIGPIPE, SIG_IGN); + pqsignal(SIGUSR1, procsignal_sigusr1_handler); + pqsignal(SIGUSR2, SIG_IGN); /* not used */ + + /* Advertise ourselves. */ + LWLockAcquire(WALSummarizerLock, LW_EXCLUSIVE); + WalSummarizerCtl->summarizer_pgprocno = MyProc->pgprocno; + LWLockRelease(WALSummarizerLock); + + /* Create and switch to a memory context that we can reset on error. */ + context = AllocSetContextCreate(TopMemoryContext, + "Wal Summarizer", + ALLOCSET_DEFAULT_SIZES); + MemoryContextSwitchTo(context); + + /* + * Reset some signals that are accepted by postmaster but not here + */ + pqsignal(SIGCHLD, SIG_DFL); + + /* + * If an exception is encountered, processing resumes here. + */ + if (sigsetjmp(local_sigjmp_buf, 1) != 0) + { + /* Since not using PG_TRY, must reset error stack by hand */ + error_context_stack = NULL; + + /* Prevent interrupts while cleaning up */ + HOLD_INTERRUPTS(); + + /* Report the error to the server log */ + EmitErrorReport(); + + /* Release resources we might have acquired. */ + LWLockReleaseAll(); + ConditionVariableCancelSleep(); + pgstat_report_wait_end(); + ReleaseAuxProcessResources(false); + AtEOXact_Files(false); + AtEOXact_HashTables(false); + + /* + * Now return to normal top-level context and clear ErrorContext for + * next time. + */ + MemoryContextSwitchTo(context); + FlushErrorState(); + + /* Flush any leaked data in the top-level context */ + MemoryContextResetAndDeleteChildren(context); + + /* Now we can allow interrupts again */ + RESUME_INTERRUPTS(); + + /* + * Sleep for 10 seconds before attempting to resume operations in + * order to avoid excessing logging. + * + * Many of the likely error conditions are things that will repeat + * every time. For example, if the WAL can't be read or the summary + * can't be written, only administrator action will cure the problem. + * So a really fast retry time doesn't seem to be especially + * beneficial, and it will clutter the logs. + */ + (void) WaitLatch(MyLatch, + WL_TIMEOUT | WL_EXIT_ON_PM_DEATH, + 10000, + WAIT_EVENT_WAL_SUMMARIZER_ERROR); + } + + /* We can now handle ereport(ERROR) */ + PG_exception_stack = &local_sigjmp_buf; + + /* + * Unblock signals (they were blocked when the postmaster forked us) + */ + sigprocmask(SIG_SETMASK, &UnBlockSig, NULL); + + /* + * Fetch information about previous progress from shared memory. + * + * If we discover that WAL summarization is not enabled, just exit. + */ + current_lsn = GetOldestUnsummarizedLSN(¤t_tli, &exact); + if (XLogRecPtrIsInvalid(current_lsn)) + proc_exit(0); + + /* + * Loop forever + */ + for (;;) + { + XLogRecPtr latest_lsn; + TimeLineID latest_tli; + XLogRecPtr cutoff_lsn; + XLogRecPtr end_of_summary_lsn; + + /* Flush any leaked data in the top-level context */ + MemoryContextResetAndDeleteChildren(context); + + /* Process any signals received recently. */ + HandleWalSummarizerInterrupts(); + + /* If it's time to remove any old WAL summaries, do that now. */ + MaybeRemoveOldWalSummaries(); + + /* Find the LSN and TLI up to which we can safely summarize. */ + latest_lsn = GetLatestLSN(&latest_tli); + + /* + * If we're summarizing a historic timeline and we haven't yet + * computed the point at which to switch to the next timeline, do that + * now. + * + * Note that if this is a standby, what was previously the current + * timeline could become historic at any time. + * + * We could try to make this more efficient by caching the results of + * readTimeLineHistory when latest_tli has not changed, but since we + * only have to do this once per timeline switch, we probably wouldn't + * save any significant amount of work in practice. + */ + if (current_tli != latest_tli && XLogRecPtrIsInvalid(switch_lsn)) + { + List *tles = readTimeLineHistory(latest_tli); + + switch_lsn = tliSwitchPoint(current_tli, tles, &switch_tli); + elog(DEBUG2, + "switch point from TLI %u to TLI %u is at %X/%X", + current_tli, switch_tli, LSN_FORMAT_ARGS(switch_lsn)); + } + + /* + * wal_summarize_mb sets a soft limit on the amont of WAL covered by a + * single summary file. If we read a WAL record that ends after the + * cutoff LSN computed here, we'll stop the summary. In most cases, it + * will actually stop earlier than that, but this is here as a + * backstop. + */ + cutoff_lsn = current_lsn + wal_summarize_mb * 1024 * 1024; + if (!XLogRecPtrIsInvalid(switch_lsn) && cutoff_lsn > switch_lsn) + cutoff_lsn = switch_lsn; + elog(DEBUG2, + "WAL summarization cutoff is TLI %d @ %X/%X, flush position is %X/%X", + current_tli, LSN_FORMAT_ARGS(cutoff_lsn), LSN_FORMAT_ARGS(latest_lsn)); + + /* Summarize WAL. */ + end_of_summary_lsn = SummarizeWAL(current_tli, + current_tli != latest_tli, + current_lsn, exact, + cutoff_lsn, latest_lsn); + Assert(!XLogRecPtrIsInvalid(end_of_summary_lsn)); + Assert(end_of_summary_lsn >= current_lsn); + + /* + * Update state for next loop iteration. + * + * Next summary file should start from exactly where this one ended. + * Timeline remains unchanged unless a switch LSN was computed and we + * have reached it. + */ + current_lsn = end_of_summary_lsn; + exact = true; + if (!XLogRecPtrIsInvalid(switch_lsn) && cutoff_lsn >= switch_lsn) + { + current_tli = switch_tli; + switch_lsn = InvalidXLogRecPtr; + switch_tli = 0; + } + + /* Update state in shared memory. */ + LWLockAcquire(WALSummarizerLock, LW_EXCLUSIVE); + Assert(WalSummarizerCtl->pending_lsn <= end_of_summary_lsn); + WalSummarizerCtl->summarized_lsn = end_of_summary_lsn; + WalSummarizerCtl->summarized_tli = current_tli; + WalSummarizerCtl->lsn_is_exact = true; + WalSummarizerCtl->pending_lsn = end_of_summary_lsn; + LWLockRelease(WALSummarizerLock); + + /* Wake up anyone waiting for more summary files to be written. */ + ConditionVariableBroadcast(&WalSummarizerCtl->summary_file_cv); + } +} + +/* + * Get the oldest LSN in this server's timeline history that has not yet been + * summarized. + * + * If *tli != NULL, it will be set to the TLI for the LSN that is returned. + * + * If *lsn_is_exact != NULL, it will be set to true if the returned LSN is + * necessarily the start of a WAL record and false if it's just the beginning + * of a WAL segment. + */ +XLogRecPtr +GetOldestUnsummarizedLSN(TimeLineID *tli, bool *lsn_is_exact) +{ + TimeLineID latest_tli; + LWLockMode mode = LW_SHARED; + int n; + List *tles; + XLogRecPtr unsummarized_lsn; + TimeLineID unsummarized_tli = 0; + bool should_make_exact = false; + List *existing_summaries; + ListCell *lc; + + /* If not summarizing WAL, do nothing. */ + if (wal_summarize_mb == 0) + return InvalidXLogRecPtr; + + /* + * Initially, we acquire the lock in shared mode and try to fetch the + * required information. If the data structure hasn't been initialized, we + * reacquire the lock in shared mode so that we can initialize it. + * However, if someone else does that first before we get the lock, then + * we can just return the requested information after all. + */ + while (true) + { + LWLockAcquire(WALSummarizerLock, mode); + + if (WalSummarizerCtl->initialized) + { + unsummarized_lsn = WalSummarizerCtl->summarized_lsn; + if (tli != NULL) + *tli = WalSummarizerCtl->summarized_tli; + if (lsn_is_exact != NULL) + *lsn_is_exact = WalSummarizerCtl->lsn_is_exact; + LWLockRelease(WALSummarizerLock); + return unsummarized_lsn; + } + + if (mode == LW_EXCLUSIVE) + break; + + LWLockRelease(WALSummarizerLock); + mode = LW_EXCLUSIVE; + } + + /* + * The data structure needs to be initialized, and we are the first to + * obtain the lock in exclusive mode, so it's our job to do that + * initialization. + * + * So, find the oldest timeline on which WAL still exists, and the + * earliest segment for which it exists. + */ + (void) GetLatestLSN(&latest_tli); + tles = readTimeLineHistory(latest_tli); + for (n = list_length(tles) - 1; n >= 0; --n) + { + TimeLineHistoryEntry *tle = list_nth(tles, n); + XLogSegNo oldest_segno; + + oldest_segno = XLogGetOldestSegno(tle->tli); + if (oldest_segno != 0) + { + /* Compute oldest LSN that still exists on disk. */ + XLogSegNoOffsetToRecPtr(oldest_segno, 0, wal_segment_size, + unsummarized_lsn); + + unsummarized_tli = tle->tli; + break; + } + } + + /* It really should not be possible for us to find no WAL. */ + if (unsummarized_tli == 0) + ereport(ERROR, + errcode(ERRCODE_INTERNAL_ERROR), + errmsg_internal("no WAL found on timeline %d", latest_tli)); + + /* + * Don't try to summarize anything older than the end LSN of the newest + * summary file that exists for this timeline. + */ + existing_summaries = + GetWalSummaries(unsummarized_tli, + InvalidXLogRecPtr, InvalidXLogRecPtr); + foreach(lc, existing_summaries) + { + WalSummaryFile *ws = lfirst(lc); + + if (ws->end_lsn > unsummarized_lsn) + { + unsummarized_lsn = ws->end_lsn; + should_make_exact = true; + } + } + + /* Update shared memory with the discovered values. */ + WalSummarizerCtl->initialized = true; + WalSummarizerCtl->summarized_lsn = unsummarized_lsn; + WalSummarizerCtl->summarized_tli = unsummarized_tli; + WalSummarizerCtl->lsn_is_exact = should_make_exact; + WalSummarizerCtl->pending_lsn = unsummarized_lsn; + + /* Also return the to the caller as required. */ + if (tli != NULL) + *tli = WalSummarizerCtl->summarized_tli; + if (lsn_is_exact != NULL) + *lsn_is_exact = WalSummarizerCtl->lsn_is_exact; + LWLockRelease(WALSummarizerLock); + + return unsummarized_lsn; +} + +/* + * Attempt to set the WAL summarizer's latch. + * + * This might not work, because there's no guarantee that the WAL summarizer + * process was successfully started, and it also might have started but + * subsequently terminated. So, under normal circumstances, this will get the + * latch set, but there's no guarantee. + */ +void +SetWalSummarizerLatch(void) +{ + int pgprocno; + + LWLockAcquire(WALSummarizerLock, LW_EXCLUSIVE); + pgprocno = WalSummarizerCtl->summarizer_pgprocno; + LWLockRelease(WALSummarizerLock); + + if (pgprocno != INVALID_PGPROCNO) + SetLatch(&ProcGlobal->allProcs[pgprocno].procLatch); +} + +/* + * Wait until WAL summarization reaches the given LSN, but not longer than + * the given timeout. + * + * The return value is the first still-unsummarized LSN. If it's greater than + * or equal to the passed LSN, then that LSN was reached. If not, we timed out. + */ +XLogRecPtr +WaitForWalSummarization(XLogRecPtr lsn, long timeout) +{ + TimestampTz start_time = GetCurrentTimestamp(); + TimestampTz deadline = TimestampTzPlusMilliseconds(start_time, timeout); + XLogRecPtr summarized_lsn; + + Assert(!XLogRecPtrIsInvalid(lsn)); + Assert(timeout > 0); + + while (1) + { + TimestampTz now; + long remaining_timeout; + + /* + * If the LSN summarized on disk has reached the target value, stop. + */ + LWLockAcquire(WALSummarizerLock, LW_EXCLUSIVE); + summarized_lsn = WalSummarizerCtl->summarized_lsn; + LWLockRelease(WALSummarizerLock); + if (summarized_lsn >= lsn) + break; + + /* Timeout reached? If yes, stop. */ + now = GetCurrentTimestamp(); + remaining_timeout = TimestampDifferenceMilliseconds(now, deadline); + if (remaining_timeout <= 0) + break; + + /* Wait and see. */ + ConditionVariableTimedSleep(&WalSummarizerCtl->summary_file_cv, + remaining_timeout, + WAIT_EVENT_WAL_SUMMARY_READY); + } + + return summarized_lsn; +} + +/* + * Get the latest LSN that is eligible to be summarized, and set *tli to the + * corresponding timeline. + */ +static XLogRecPtr +GetLatestLSN(TimeLineID *tli) +{ + if (!RecoveryInProgress()) + { + /* Don't summarize WAL before it's flushed. */ + return GetFlushRecPtr(tli); + } + else + { + XLogRecPtr flush_lsn; + TimeLineID flush_tli; + XLogRecPtr replay_lsn; + TimeLineID replay_tli; + + /* + * What we really want to know is how much WAL has been flushed to + * disk, but the only flush position available is the one provided by + * the walreceiver, which may not be running, because this could be + * crash recovery or recovery via restore_command. So use either the + * WAL receiver's flush position or the replay position, whichever is + * further ahead, on the theory that if the WAL has been replayed then + * it must also have been flushed to disk. + */ + flush_lsn = GetWalRcvFlushRecPtr(NULL, &flush_tli); + replay_lsn = GetXLogReplayRecPtr(&replay_tli); + if (flush_lsn > replay_lsn) + { + *tli = flush_tli; + return flush_lsn; + } + else + { + *tli = replay_tli; + return replay_lsn; + } + } +} + +/* + * Interrupt handler for main loop of WAL summarizer process. + */ +static void +HandleWalSummarizerInterrupts(void) +{ + if (ProcSignalBarrierPending) + ProcessProcSignalBarrier(); + + if (ConfigReloadPending) + { + ConfigReloadPending = false; + ProcessConfigFile(PGC_SIGHUP); + } + + if (ShutdownRequestPending || wal_summarize_mb == 0) + { + ereport(DEBUG1, + errmsg_internal("WAL summarizer shutting down")); + proc_exit(0); + } + + /* Perform logging of memory contexts of this process */ + if (LogMemoryContextPending) + ProcessLogMemoryContextInterrupt(); +} + +/* + * Summarize a range of WAL records on a single timeline. + * + * 'tli' is the timeline to be summarized. 'historic' should be false if the + * timeline in question is the latest one and true otherwise. + * + * 'start_lsn' is the point at which we should start summarizing. If this + * value comes from the end LSN of the previous record as returned by the + * xlograder machinery, 'exact' should be true; otherwise, 'exact' should + * be false, and this function will search forward for the start of a valid + * WAL record. + * + * 'cutoff_lsn' is the point at which we should stop summarizing. The first + * record that ends at or after cutoff_lsn will be the last one included + * in the summary. + * + * 'maximum_lsn' identifies the point beyond which we can't count on being + * able to read any more WAL. It should be the switch point when reading a + * historic timeline, or the most-recently-measured end of WAL when reading + * the current timeline. + * + * The return value is the LSN at which the WAL summary actually ends. Most + * often, a summary file ends because we notice that a checkpoint has + * occurred and reach the redo pointer of that checkpoint, but sometimes + * we stop for other reasons, such as a timeline switch, or reading a record + * that ends after the cutoff_lsn. + */ +static XLogRecPtr +SummarizeWAL(TimeLineID tli, bool historic, + XLogRecPtr start_lsn, bool exact, + XLogRecPtr cutoff_lsn, XLogRecPtr maximum_lsn) +{ + SummarizerReadLocalXLogPrivate *private_data; + XLogReaderState *xlogreader; + XLogRecPtr summary_start_lsn; + XLogRecPtr summary_end_lsn = cutoff_lsn; + char temp_path[MAXPGPATH]; + char final_path[MAXPGPATH]; + WalSummaryIO io; + BlockRefTable *brtab = CreateEmptyBlockRefTable(); + + /* Initialize private data for xlogreader. */ + private_data = (SummarizerReadLocalXLogPrivate *) + palloc0(sizeof(SummarizerReadLocalXLogPrivate)); + private_data->tli = tli; + private_data->historic = historic; + private_data->read_upto = maximum_lsn; + + /* Create xlogreader. */ + xlogreader = XLogReaderAllocate(wal_segment_size, NULL, + XL_ROUTINE(.page_read = &summarizer_read_local_xlog_page, + .segment_open = &wal_segment_open, + .segment_close = &wal_segment_close), + private_data); + if (xlogreader == NULL) + ereport(ERROR, + (errcode(ERRCODE_OUT_OF_MEMORY), + errmsg("out of memory"), + errdetail("Failed while allocating a WAL reading processor."))); + + /* + * When exact = false, we're starting from an arbitrary point in the WAL + * and must search forward for the start of the next record. + * + * When exact = true, start_lsn should be either the LSN where a record + * begins, or the LSN of a page where the page header is immediately + * followed by the start of a new record. XLogBeginRead should tolerate + * either case. + * + * We need to allow for both cases because the behavior of xlogreader + * varies. When a record spans two or more xlog pages, the ending LSN + * reported by xlogreader will be the starting LSN of the following + * record, but when an xlog page boundary falls between two records, the + * end LSN for the first will be reported as the first byte of the + * following page. We can't know until we read that page how large the + * header will be, but we'll have to skip over it to find the next record. + */ + if (exact) + { + /* + * Even if start_lsn is the beginning of a page rather than the + * beginning of the first record on that page, we should still use it + * as the start LSN for the summary file. That's because we detect + * missing summary files by looking for cases where the end LSN of one + * file is less than the start LSN of the next file. When only a page + * header is skipped, nothing has been missed. + */ + XLogBeginRead(xlogreader, start_lsn); + summary_start_lsn = start_lsn; + } + else + { + summary_start_lsn = XLogFindNextRecord(xlogreader, start_lsn); + if (XLogRecPtrIsInvalid(summary_start_lsn)) + { + /* + * If we hit end-of-WAL while trying to find the next valid + * record, we must be on a historic timeline that has no valid + * records that begin after start_lsn and before end of WAL. + */ + if (private_data->end_of_wal) + { + ereport(DEBUG1, + errmsg_internal("could not read WAL from timeline %d at %X/%X: end of WAL at %X/%X", + tli, + LSN_FORMAT_ARGS(start_lsn), + LSN_FORMAT_ARGS(private_data->read_upto))); + + /* + * The timeline ends at or after start_lsn, without containing + * any records. Thus, we must make sure the main loop does not + * iterate. If start_lsn is the end of the timeline, then we + * won't actually emit an empty summary file, but otherwise, + * we must, to capture the fact that the LSN range in question + * contains no interesting WAL records. + */ + summary_start_lsn = start_lsn; + summary_end_lsn = private_data->read_upto; + cutoff_lsn = xlogreader->EndRecPtr; + } + else + ereport(ERROR, + (errmsg("could not find a valid record after %X/%X", + LSN_FORMAT_ARGS(start_lsn)))); + } + + /* We shouldn't go backward. */ + Assert(summary_start_lsn >= start_lsn); + } + + /* + * Main loop: read xlog records one by one. + */ + while (xlogreader->EndRecPtr < cutoff_lsn) + { + int block_id; + char *errormsg; + XLogRecord *record; + bool stop_requested = false; + + /* We shouldn't go backward. */ + Assert(summary_start_lsn <= xlogreader->EndRecPtr); + + /* + * This flag tracks whether the read of a particular record had to + * wait for more WAL to arrive, so reset it before reading the next + * record. + */ + private_data->waited = false; + + /* Now read the next record. */ + record = XLogReadRecord(xlogreader, &errormsg); + if (record == NULL) + { + SummarizerReadLocalXLogPrivate *private_data; + + private_data = (SummarizerReadLocalXLogPrivate *) + xlogreader->private_data; + if (private_data->end_of_wal) + { + /* + * This timeline must be historic and must end before we were + * able to read a complete record. + */ + ereport(DEBUG1, + errmsg_internal("could not read WAL from timeline %d at %X/%X: end of WAL at %X/%X", + tli, + LSN_FORMAT_ARGS(xlogreader->EndRecPtr), + LSN_FORMAT_ARGS(private_data->read_upto))); + /* Summary ends at end of WAL. */ + summary_end_lsn = private_data->read_upto; + break; + } + if (errormsg) + ereport(ERROR, + (errcode_for_file_access(), + errmsg("could not read WAL at %X/%X: %s", + LSN_FORMAT_ARGS(xlogreader->EndRecPtr), errormsg))); + else + ereport(ERROR, + (errcode_for_file_access(), + errmsg("could not read WAL at %X/%X", + LSN_FORMAT_ARGS(xlogreader->EndRecPtr)))); + } + + /* We shouldn't go backward. */ + Assert(summary_start_lsn <= xlogreader->EndRecPtr); + + if (xlogreader->ReadRecPtr >= cutoff_lsn) + { + /* + * Woops! We've read a record that *starts* after the cutoff LSN, + * contrary to our goal of reading only until we hit the first + * record that ends at or after the cutoff LSN. Pretend we didn't + * read it after all by bailing out of this loop right here, + * before we do anything with this record. + * + * This can happen because the last record before the cutoff LSN + * might be continued across multiple pages, and then we might + * come to a page with XLP_FIRST_IS_OVERWRITE_CONTRECORD set. In + * that case, the record that was continued across multiple pages + * is incomplete and will be disregarded, and the read will + * restart from the beginning of the page that is flagged + * XLP_FIRST_IS_OVERWRITE_CONTRECORD. + * + * If this case occurs, we can fairly say that the current summary + * file ends at the cutoff LSN exactly. The first record on the + * page marked XLP_FIRST_IS_OVERWRITE_CONTRECORD will be + * discovered when generating the next summary file. + */ + summary_end_lsn = cutoff_lsn; + break; + } + + /* Special handling for particular types of WAL records. */ + switch (XLogRecGetRmid(xlogreader)) + { + case RM_SMGR_ID: + SummarizeSmgrRecord(xlogreader, brtab); + break; + case RM_XACT_ID: + SummarizeXactRecord(xlogreader, brtab); + break; + case RM_XLOG_ID: + stop_requested = SummarizeXlogRecord(xlogreader); + break; + default: + break; + } + + /* + * If we've been told that it's time to end this WAL summary file, + * do so. As an exception, if there's nothing included in this WAL + * summary file yet, then stoppng doesn't make any sense, and we + * should wait until the next stop point instead. + */ + if (stop_requested && xlogreader->ReadRecPtr > summary_start_lsn) + { + summary_end_lsn = xlogreader->ReadRecPtr; + break; + } + + /* Feed block references from xlog record to block reference table. */ + for (block_id = 0; block_id <= XLogRecMaxBlockId(xlogreader); + block_id++) + { + RelFileLocator rlocator; + ForkNumber forknum; + BlockNumber blocknum; + + if (!XLogRecGetBlockTagExtended(xlogreader, block_id, &rlocator, + &forknum, &blocknum, NULL)) + continue; + + BlockRefTableMarkBlockModified(brtab, &rlocator, forknum, + blocknum); + } + + /* Update our notion of where this summary file ends. */ + summary_end_lsn = xlogreader->EndRecPtr; + + /* Also update shared memory. */ + LWLockAcquire(WALSummarizerLock, LW_EXCLUSIVE); + Assert(summary_end_lsn >= WalSummarizerCtl->pending_lsn); + Assert(summary_end_lsn >= WalSummarizerCtl->summarized_lsn); + WalSummarizerCtl->pending_lsn = summary_end_lsn; + LWLockRelease(WALSummarizerLock); + } + + /* Destroy xlogreader. */ + pfree(xlogreader->private_data); + XLogReaderFree(xlogreader); + + /* + * If a timeline switch occurs, we may fail to make any progress at all + * before exiting the loop above. If that happens, we don't write a WAL + * summary file at all. + */ + if (summary_end_lsn > summary_start_lsn) + { + /* Generate temporary and final path name. */ + snprintf(temp_path, MAXPGPATH, + XLOGDIR "/summaries/temp.summary"); + snprintf(final_path, MAXPGPATH, + XLOGDIR "/summaries/%08X%08X%08X%08X%08X.summary", + tli, + LSN_FORMAT_ARGS(summary_start_lsn), + LSN_FORMAT_ARGS(summary_end_lsn)); + + /* Open the temporary file for writing. */ + io.filepos = 0; + io.file = PathNameOpenFile(temp_path, O_WRONLY | O_CREAT | O_TRUNC); + if (io.file < 0) + ereport(ERROR, + (errcode_for_file_access(), + errmsg("could not create file \"%s\": %m", temp_path))); + + /* Write the data. */ + WriteBlockRefTable(brtab, WriteWalSummary, &io); + + /* Close temporary file and shut down xlogreader. */ + FileClose(io.file); + + /* Tell the user what we did. */ + ereport(LOG, + errmsg("summarized WAL on TLI %d from %X/%X to %X/%X", + tli, + LSN_FORMAT_ARGS(summary_start_lsn), + LSN_FORMAT_ARGS(summary_end_lsn))); + + /* Durably rename the new summary into place. */ + durable_rename(temp_path, final_path, ERROR); + } + + return summary_end_lsn; +} + +/* + * Special handling for WAL records with RM_SMGR_ID. + */ +static void +SummarizeSmgrRecord(XLogReaderState *xlogreader, BlockRefTable *brtab) +{ + uint8 info = XLogRecGetInfo(xlogreader) & ~XLR_INFO_MASK; + + if (info == XLOG_SMGR_CREATE) + { + xl_smgr_create *xlrec; + + /* + * If a new relation fork is created on disk, there is no point + * tracking anything about which blocks have been modified, because + * the whole thing will be new. Hence, set the limit block for this + * fork to 0. + * + * Ignore the FSM fork, which is not fully WAL-logged. + */ + xlrec = (xl_smgr_create *) XLogRecGetData(xlogreader); + + if (xlrec->forkNum != FSM_FORKNUM) + BlockRefTableSetLimitBlock(brtab, &xlrec->rlocator, + xlrec->forkNum, 0); + } + else if (info == XLOG_SMGR_TRUNCATE) + { + xl_smgr_truncate *xlrec; + + xlrec = (xl_smgr_truncate *) XLogRecGetData(xlogreader); + + /* + * If a relation fork is truncated on disk, there is in point in + * tracking anything about block modifications beyond the truncation + * point. + * + * We ignore SMGR_TRUNCATE_FSM here because the FSM isn't fully + * WAL-logged and thus we can't track modified blocks for it anyway. + */ + if ((xlrec->flags & SMGR_TRUNCATE_HEAP) != 0) + BlockRefTableSetLimitBlock(brtab, &xlrec->rlocator, + MAIN_FORKNUM, xlrec->blkno); + if ((xlrec->flags & SMGR_TRUNCATE_VM) != 0) + BlockRefTableSetLimitBlock(brtab, &xlrec->rlocator, + VISIBILITYMAP_FORKNUM, xlrec->blkno); + } +} + +/* + * Special handling for WAL recods with RM_XACT_ID. + */ +static void +SummarizeXactRecord(XLogReaderState *xlogreader, BlockRefTable *brtab) +{ + uint8 info = XLogRecGetInfo(xlogreader) & ~XLR_INFO_MASK; + uint8 xact_info = info & XLOG_XACT_OPMASK; + + if (xact_info == XLOG_XACT_COMMIT || + xact_info == XLOG_XACT_COMMIT_PREPARED) + { + xl_xact_commit *xlrec = (xl_xact_commit *) XLogRecGetData(xlogreader); + xl_xact_parsed_commit parsed; + int i; + + ParseCommitRecord(XLogRecGetInfo(xlogreader), xlrec, &parsed); + for (i = 0; i < parsed.nrels; ++i) + { + ForkNumber forknum; + + for (forknum = 0; forknum <= MAX_FORKNUM; ++forknum) + if (forknum != FSM_FORKNUM) + BlockRefTableSetLimitBlock(brtab, &parsed.xlocators[i], + forknum, 0); + } + } + else if (xact_info == XLOG_XACT_ABORT || + xact_info == XLOG_XACT_ABORT_PREPARED) + { + xl_xact_abort *xlrec = (xl_xact_abort *) XLogRecGetData(xlogreader); + xl_xact_parsed_abort parsed; + int i; + + ParseAbortRecord(XLogRecGetInfo(xlogreader), xlrec, &parsed); + for (i = 0; i < parsed.nrels; ++i) + { + ForkNumber forknum; + + for (forknum = 0; forknum <= MAX_FORKNUM; ++forknum) + if (forknum != FSM_FORKNUM) + BlockRefTableSetLimitBlock(brtab, &parsed.xlocators[i], + forknum, 0); + } + } +} + +/* + * Special handling for WAL recods with RM_XLOG_ID. + */ +static bool +SummarizeXlogRecord(XLogReaderState *xlogreader) +{ + uint8 info = XLogRecGetInfo(xlogreader) & ~XLR_INFO_MASK; + + if (info == XLOG_CHECKPOINT_REDO || info == XLOG_CHECKPOINT_SHUTDOWN) + { + /* + * This is an LSN at which redo might begin, so we'd like summarization + * to stop just before this WAL record. + */ + return true; + } + + return false; +} + +/* + * Similar to read_local_xlog_page, but limited to read from one particular + * timeline. If the end of WAL is reached, it will wait for more if reading + * from the current timeline, or give up if reading from a historic timeline. + * In the latter case, it will also set private_data->end_of_wal = true. + * + * Caller must set private_data->tli to the TLI of interest, + * private_data->read_upto to the lowest LSN that is not known to be safe + * to read on that timeline, and private_data->historic to true if and only + * if the timeline is not the current timeline. This function will update + * private_data->read_upto and private_data->historic if more WAL appears + * on the current timeline or if the current timeline becomes historic. + */ +static int +summarizer_read_local_xlog_page(XLogReaderState *state, + XLogRecPtr targetPagePtr, int reqLen, + XLogRecPtr targetRecPtr, char *cur_page) +{ + int count; + WALReadError errinfo; + SummarizerReadLocalXLogPrivate *private_data; + + private_data = (SummarizerReadLocalXLogPrivate *) + state->private_data; + + while (true) + { + if (targetPagePtr + XLOG_BLCKSZ <= private_data->read_upto) + { + /* + * more than one block available; read only that block, have + * caller come back if they need more. + */ + count = XLOG_BLCKSZ; + break; + } + else if (targetPagePtr + reqLen > private_data->read_upto) + { + /* We don't seem to have enough data. */ + if (private_data->historic) + { + /* + * This is a historic timeline, so there will never be any + * more data than we have currently. + */ + private_data->end_of_wal = true; + return -1; + } + else + { + XLogRecPtr latest_lsn; + TimeLineID latest_tli; + + /* + * This is - or at least was up until very recently - the + * current timeline, so more data might show up. Delay here + * so we don't tight-loop. + */ + HandleWalSummarizerInterrupts(); + summarizer_wait_for_wal(); + private_data->waited = true; + + /* Recheck end-of-WAL. */ + latest_lsn = GetLatestLSN(&latest_tli); + if (private_data->tli == latest_tli) + { + /* Still the current timeline, update max LSN. */ + Assert(latest_lsn >= private_data->read_upto); + private_data->read_upto = latest_lsn; + } + else + { + List *tles = readTimeLineHistory(latest_tli); + XLogRecPtr switchpoint; + + /* + * The timeline we're scanning is no longer the latest + * one. Figure out when it ended and allow reads up to + * exactly that point. + */ + private_data->historic = true; + switchpoint = tliSwitchPoint(private_data->tli, tles, + NULL); + Assert(switchpoint >= private_data->read_upto); + private_data->read_upto = switchpoint; + } + + /* Go around and try again. */ + } + } + else + { + /* enough bytes available to satisfy the request */ + count = private_data->read_upto - targetPagePtr; + break; + } + } + + /* + * Even though we just determined how much of the page can be validly read + * as 'count', read the whole page anyway. It's guaranteed to be + * zero-padded up to the page boundary if it's incomplete. + */ + if (!WALRead(state, cur_page, targetPagePtr, XLOG_BLCKSZ, + private_data->tli, &errinfo)) + WALReadRaiseError(&errinfo); + + /* Track that we read a page, for sleep time calculation. */ + ++pages_read_since_last_sleep; + + /* number of valid bytes in the buffer */ + return count; +} + +/* + * Sleep for long enough that we believe it's likely that more WAL will + * be available afterwards. + */ +static void +summarizer_wait_for_wal(void) +{ + if (pages_read_since_last_sleep == 0) + { + /* + * No pages were read since the last sleep, so double the sleep time, + * but not beyond the maximum allowable value. + */ + sleep_quanta = Min(sleep_quanta * 2, MAX_SLEEP_QUANTA); + } + else if (pages_read_since_last_sleep > 1) + { + /* + * Multiple pages were read since the last sleep, so reduce the sleep + * time. + * + * A large burst of activity should be able to quickly reduce the + * sleep time to the minimum, but we don't want a handful of extra WAL + * records to provoke a strong reaction. We choose to reduce the sleep + * time by 1 quantum for each page read beyond the first, which is a + * fairly arbitrary way of trying to be reactive without + * overrreacting. + */ + if (pages_read_since_last_sleep > sleep_quanta - 1) + sleep_quanta = 1; + else + sleep_quanta -= pages_read_since_last_sleep; + } + + /* OK, now sleep. */ + (void) WaitLatch(MyLatch, + WL_LATCH_SET | WL_TIMEOUT | WL_EXIT_ON_PM_DEATH, + sleep_quanta * MS_PER_SLEEP_QUANTUM, + WAIT_EVENT_WAL_SUMMARIZER_WAL); + ResetLatch(MyLatch); + + /* Reset count of pages read. */ + pages_read_since_last_sleep = 0; +} + +/* + * Most recent RedoRecPtr value observed by RemoveOldWalSummaries. + */ +static void +MaybeRemoveOldWalSummaries(void) +{ + XLogRecPtr redo_pointer = GetRedoRecPtr(); + List *wslist; + time_t cutoff_time; + + /* If WAL summary removal is disabled, don't do anything. */ + if (wal_summarize_keep_time == 0) + return; + + /* + * If the redo pointer has not advanced, don't do anything. + * + * This has the effect that we only try to remove old WAL summary files + * once per checkpoint cycle. + */ + if (redo_pointer == redo_pointer_at_last_summary_removal) + return; + redo_pointer_at_last_summary_removal = redo_pointer; + + /* + * Files should only be removed if the last modification time precedes the + * cutoff time we compute here. + */ + cutoff_time = time(NULL) - 60 * wal_summarize_keep_time; + + /* Get all the summaries that currently exist. */ + wslist = GetWalSummaries(0, InvalidXLogRecPtr, InvalidXLogRecPtr); + + /* Loop until all summaries have been considered for removal. */ + while (wslist != NIL) + { + ListCell *lc; + XLogSegNo oldest_segno; + XLogRecPtr oldest_lsn = InvalidXLogRecPtr; + TimeLineID selected_tli; + + CHECK_FOR_INTERRUPTS(); + + /* + * Pick a timeline for which some summary files still exist on disk, + * and find the oldest LSN that still exists on disk for that + * timeline. + */ + selected_tli = ((WalSummaryFile *) linitial(wslist))->tli; + oldest_segno = XLogGetOldestSegno(selected_tli); + if (oldest_segno != 0) + XLogSegNoOffsetToRecPtr(oldest_segno, 0, wal_segment_size, + oldest_lsn); + + + /* Consider each WAL file on the selected timeline in turn. */ + foreach(lc, wslist) + { + WalSummaryFile *ws = lfirst(lc); + + CHECK_FOR_INTERRUPTS(); + + /* If it's not on this timeline, it's not time to consider it. */ + if (selected_tli != ws->tli) + continue; + + /* + * If the WAL doesn't exist any more, we can remove it if the file + * modification time is old enough. + */ + if (XLogRecPtrIsInvalid(oldest_lsn) || ws->end_lsn <= oldest_lsn) + RemoveWalSummaryIfOlderThan(ws, cutoff_time); + + /* + * Whether we we removed the file or not, we need not consider it + * again. + */ + wslist = foreach_delete_current(wslist, lc); + pfree(ws); + } + } +} diff --git a/src/backend/replication/repl_gram.y b/src/backend/replication/repl_gram.y index 0c874e33cf..a5d118ed68 100644 --- a/src/backend/replication/repl_gram.y +++ b/src/backend/replication/repl_gram.y @@ -76,11 +76,12 @@ Node *replication_parse_result; %token K_EXPORT_SNAPSHOT %token K_NOEXPORT_SNAPSHOT %token K_USE_SNAPSHOT +%token K_UPLOAD_MANIFEST %type <node> command %type <node> base_backup start_replication start_logical_replication create_replication_slot drop_replication_slot identify_system - read_replication_slot timeline_history show + read_replication_slot timeline_history show upload_manifest %type <list> generic_option_list %type <defelt> generic_option %type <uintval> opt_timeline @@ -114,6 +115,7 @@ command: | read_replication_slot | timeline_history | show + | upload_manifest ; /* @@ -307,6 +309,15 @@ timeline_history: } ; +/* UPLOAD_MANIFEST doesn't currently accept any arguments */ +upload_manifest: + K_UPLOAD_MANIFEST + { + UploadManifestCmd *cmd = makeNode(UploadManifestCmd); + + $$ = (Node *) cmd; + } + opt_physical: K_PHYSICAL | /* EMPTY */ @@ -411,6 +422,7 @@ ident_or_keyword: | K_EXPORT_SNAPSHOT { $$ = "export_snapshot"; } | K_NOEXPORT_SNAPSHOT { $$ = "noexport_snapshot"; } | K_USE_SNAPSHOT { $$ = "use_snapshot"; } + | K_UPLOAD_MANIFEST { $$ = "upload_manifest"; } ; %% diff --git a/src/backend/replication/repl_scanner.l b/src/backend/replication/repl_scanner.l index 1cc7fb858c..4805da08ee 100644 --- a/src/backend/replication/repl_scanner.l +++ b/src/backend/replication/repl_scanner.l @@ -136,6 +136,7 @@ EXPORT_SNAPSHOT { return K_EXPORT_SNAPSHOT; } NOEXPORT_SNAPSHOT { return K_NOEXPORT_SNAPSHOT; } USE_SNAPSHOT { return K_USE_SNAPSHOT; } WAIT { return K_WAIT; } +UPLOAD_MANIFEST { return K_UPLOAD_MANIFEST; } {space}+ { /* do nothing */ } @@ -303,6 +304,7 @@ replication_scanner_is_replication_command(void) case K_DROP_REPLICATION_SLOT: case K_READ_REPLICATION_SLOT: case K_TIMELINE_HISTORY: + case K_UPLOAD_MANIFEST: case K_SHOW: /* Yes; push back the first token so we can parse later. */ repl_pushed_back_token = first_token; diff --git a/src/backend/replication/walsender.c b/src/backend/replication/walsender.c index e250b0567e..b33b86671b 100644 --- a/src/backend/replication/walsender.c +++ b/src/backend/replication/walsender.c @@ -58,6 +58,7 @@ #include "access/xlogrecovery.h" #include "access/xlogutils.h" #include "backup/basebackup.h" +#include "backup/basebackup_incremental.h" #include "catalog/pg_authid.h" #include "catalog/pg_type.h" #include "commands/dbcommands.h" @@ -137,6 +138,17 @@ bool wake_wal_senders = false; */ static XLogReaderState *xlogreader = NULL; +/* + * If the UPLOAD_MANIFEST command is used to provide a backup manifest in + * preparation for an incremental backup, uploaded_manifest will be point + * to an object containing information about its contexts, and + * uploaded_manifest_mcxt will point to the memory context that contains + * that object and all of its subordinate data. Otherwise, both values will + * be NULL. + */ +static IncrementalBackupInfo *uploaded_manifest = NULL; +static MemoryContext uploaded_manifest_mcxt = NULL; + /* * These variables keep track of the state of the timeline we're currently * sending. sendTimeLine identifies the timeline. If sendTimeLineIsHistoric, @@ -233,6 +245,9 @@ static void XLogSendLogical(void); static void WalSndDone(WalSndSendDataCallback send_data); static XLogRecPtr GetStandbyFlushRecPtr(TimeLineID *tli); static void IdentifySystem(void); +static void UploadManifest(void); +static bool HandleUploadManifestPacket(StringInfo buf, off_t *offset, + IncrementalBackupInfo *ib); static void ReadReplicationSlot(ReadReplicationSlotCmd *cmd); static void CreateReplicationSlot(CreateReplicationSlotCmd *cmd); static void DropReplicationSlot(DropReplicationSlotCmd *cmd); @@ -660,6 +675,143 @@ SendTimeLineHistory(TimeLineHistoryCmd *cmd) pq_endmessage(&buf); } +/* + * Handle UPLOAD_MANIFEST command. + */ +static void +UploadManifest(void) +{ + MemoryContext mcxt; + IncrementalBackupInfo *ib; + off_t offset = 0; + StringInfoData buf; + + /* + * parsing the manifest will use the cryptohash stuff, which requires a + * resource owner + */ + Assert(CurrentResourceOwner == NULL); + CurrentResourceOwner = ResourceOwnerCreate(NULL, "base backup"); + + /* Prepare to read manifest data into a temporary context. */ + mcxt = AllocSetContextCreate(CurrentMemoryContext, + "incremental backup information", + ALLOCSET_DEFAULT_SIZES); + ib = CreateIncrementalBackupInfo(mcxt); + + /* Send a CopyInResponse message */ + pq_beginmessage(&buf, 'G'); + pq_sendbyte(&buf, 0); + pq_sendint16(&buf, 0); + pq_endmessage_reuse(&buf); + pq_flush(); + + /* Recieve packets from client until done. */ + while (HandleUploadManifestPacket(&buf, &offset, ib)) + ; + + /* Finish up manifest processing. */ + FinalizeIncrementalManifest(ib); + + /* + * Discard any old manifest information and arrange to preserve the new + * information we just got. + * + * We assume that MemoryContextDelete and MemoryContextSetParent won't + * fail, and thus we shouldn't end up bailing out of here in such a way as + * to leave dangling pointrs. + */ + if (uploaded_manifest_mcxt != NULL) + MemoryContextDelete(uploaded_manifest_mcxt); + MemoryContextSetParent(mcxt, CacheMemoryContext); + uploaded_manifest = ib; + uploaded_manifest_mcxt = mcxt; + + /* clean up the resource owner we created */ + WalSndResourceCleanup(true); +} + +/* + * Process one packet received during the handling of an UPLOAD_MANIFEST + * operation. + * + * 'buf' is scratch space. This function expects it to be initialized, doesn't + * care what the current contents are, and may override them with completely + * new contents. + * + * The return value is true if the caller should continue processing + * additional packets and false if the UPLOAD_MANIFEST operation is complete. + */ +static bool +HandleUploadManifestPacket(StringInfo buf, off_t *offset, + IncrementalBackupInfo *ib) +{ + int mtype; + int maxmsglen; + + HOLD_CANCEL_INTERRUPTS(); + + pq_startmsgread(); + mtype = pq_getbyte(); + if (mtype == EOF) + ereport(ERROR, + (errcode(ERRCODE_CONNECTION_FAILURE), + errmsg("unexpected EOF on client connection with an open transaction"))); + + switch (mtype) + { + case 'd': /* CopyData */ + maxmsglen = PQ_LARGE_MESSAGE_LIMIT; + break; + case 'c': /* CopyDone */ + case 'f': /* CopyFail */ + case 'H': /* Flush */ + case 'S': /* Sync */ + maxmsglen = PQ_SMALL_MESSAGE_LIMIT; + break; + default: + ereport(ERROR, + (errcode(ERRCODE_PROTOCOL_VIOLATION), + errmsg("unexpected message type 0x%02X during COPY from stdin", + mtype))); + maxmsglen = 0; /* keep compiler quiet */ + break; + } + + /* Now collect the message body */ + if (pq_getmessage(buf, maxmsglen)) + ereport(ERROR, + (errcode(ERRCODE_CONNECTION_FAILURE), + errmsg("unexpected EOF on client connection with an open transaction"))); + RESUME_CANCEL_INTERRUPTS(); + + /* Process the message */ + switch (mtype) + { + case 'd': /* CopyData */ + AppendIncrementalManifestData(ib, buf->data, buf->len); + return true; + + case 'c': /* CopyDone */ + return false; + + case 'H': /* Sync */ + case 'S': /* Flush */ + /* Ignore these while in CopyOut mode as we do elsewhere. */ + return true; + + case 'f': + ereport(ERROR, + (errcode(ERRCODE_QUERY_CANCELED), + errmsg("COPY from stdin failed: %s", + pq_getmsgstring(buf)))); + } + + /* Not reached. */ + Assert(false); + return false; +} + /* * Handle START_REPLICATION command. * @@ -1801,7 +1953,7 @@ exec_replication_command(const char *cmd_string) cmdtag = "BASE_BACKUP"; set_ps_display(cmdtag); PreventInTransactionBlock(true, cmdtag); - SendBaseBackup((BaseBackupCmd *) cmd_node); + SendBaseBackup((BaseBackupCmd *) cmd_node, uploaded_manifest); EndReplicationCommand(cmdtag); break; @@ -1863,6 +2015,14 @@ exec_replication_command(const char *cmd_string) } break; + case T_UploadManifestCmd: + cmdtag = "UPLOAD_MANIFEST"; + set_ps_display(cmdtag); + PreventInTransactionBlock(true, cmdtag); + UploadManifest(); + EndReplicationCommand(cmdtag); + break; + default: elog(ERROR, "unrecognized replication command node tag: %u", cmd_node->type); diff --git a/src/backend/storage/ipc/ipci.c b/src/backend/storage/ipc/ipci.c index a3d8eacb8d..3a6729003a 100644 --- a/src/backend/storage/ipc/ipci.c +++ b/src/backend/storage/ipc/ipci.c @@ -31,6 +31,7 @@ #include "postmaster/bgworker_internals.h" #include "postmaster/bgwriter.h" #include "postmaster/postmaster.h" +#include "postmaster/walsummarizer.h" #include "replication/logicallauncher.h" #include "replication/origin.h" #include "replication/slot.h" @@ -136,6 +137,7 @@ CalculateShmemSize(int *num_semaphores) size = add_size(size, ReplicationOriginShmemSize()); size = add_size(size, WalSndShmemSize()); size = add_size(size, WalRcvShmemSize()); + size = add_size(size, WalSummarizerShmemSize()); size = add_size(size, PgArchShmemSize()); size = add_size(size, ApplyLauncherShmemSize()); size = add_size(size, BTreeShmemSize()); @@ -291,6 +293,7 @@ CreateSharedMemoryAndSemaphores(void) ReplicationOriginShmemInit(); WalSndShmemInit(); WalRcvShmemInit(); + WalSummarizerShmemInit(); PgArchShmemInit(); ApplyLauncherShmemInit(); diff --git a/src/backend/storage/lmgr/lwlocknames.txt b/src/backend/storage/lmgr/lwlocknames.txt index f72f2906ce..d621f5507f 100644 --- a/src/backend/storage/lmgr/lwlocknames.txt +++ b/src/backend/storage/lmgr/lwlocknames.txt @@ -54,3 +54,4 @@ XactTruncationLock 44 WrapLimitsVacuumLock 46 NotifyQueueTailLock 47 WaitEventExtensionLock 48 +WALSummarizerLock 49 diff --git a/src/backend/utils/activity/pgstat_io.c b/src/backend/utils/activity/pgstat_io.c index 490d5a9ab7..8109aee6f0 100644 --- a/src/backend/utils/activity/pgstat_io.c +++ b/src/backend/utils/activity/pgstat_io.c @@ -296,7 +296,8 @@ pgstat_io_snapshot_cb(void) * - Syslogger because it is not connected to shared memory * - Archiver because most relevant archiving IO is delegated to a * specialized command or module -* - WAL Receiver and WAL Writer IO is not tracked in pg_stat_io for now +* - WAL Receiver, WAL Writer, and WAL Summarizer IO are not tracked in +* pg_stat_io for now * * Function returns true if BackendType participates in the cumulative stats * subsystem for IO and false if it does not. @@ -318,6 +319,7 @@ pgstat_tracks_io_bktype(BackendType bktype) case B_LOGGER: case B_WAL_RECEIVER: case B_WAL_WRITER: + case B_WAL_SUMMARIZER: return false; case B_AUTOVAC_LAUNCHER: diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index d7995931bd..7e79163466 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -56,6 +56,7 @@ RECOVERY_WAL_STREAM "Waiting in main loop of startup process for WAL to arrive, SYSLOGGER_MAIN "Waiting in main loop of syslogger process." WAL_RECEIVER_MAIN "Waiting in main loop of WAL receiver process." WAL_SENDER_MAIN "Waiting in main loop of WAL sender process." +WAL_SUMMARIZER_WAL "Waiting in WAL summarizer for more WAL to be generated." WAL_WRITER_MAIN "Waiting in main loop of WAL writer process." @@ -142,6 +143,7 @@ SAFE_SNAPSHOT "Waiting to obtain a valid snapshot for a <literal>READ ONLY DEFER SYNC_REP "Waiting for confirmation from a remote server during synchronous replication." WAL_RECEIVER_EXIT "Waiting for the WAL receiver to exit." WAL_RECEIVER_WAIT_START "Waiting for startup process to send initial data for streaming replication." +WAL_SUMMARY_READY "Waiting for a new WAL summary to be generated." XACT_GROUP_UPDATE "Waiting for the group leader to update transaction status at end of a parallel operation." @@ -162,6 +164,7 @@ REGISTER_SYNC_REQUEST "Waiting while sending synchronization requests to the che SPIN_DELAY "Waiting while acquiring a contended spinlock." VACUUM_DELAY "Waiting in a cost-based vacuum delay point." VACUUM_TRUNCATE "Waiting to acquire an exclusive lock to truncate off any empty pages at the end of a table vacuumed." +WAL_SUMMARIZER_ERROR "Waiting after a WAL summarizer error." # @@ -243,6 +246,8 @@ WAL_COPY_WRITE "Waiting for a write when creating a new WAL segment by copying a WAL_INIT_SYNC "Waiting for a newly initialized WAL file to reach durable storage." WAL_INIT_WRITE "Waiting for a write while initializing a new WAL file." WAL_READ "Waiting for a read from a WAL file." +WAL_SUMMARY_READ "Waiting for a read from a WAL summary file." +WAL_SUMMARY_WRITE "Waiting for a write to a WAL summary file." WAL_SYNC "Waiting for a WAL file to reach durable storage." WAL_SYNC_METHOD_ASSIGN "Waiting for data to reach durable storage while assigning a new WAL sync method." WAL_WRITE "Waiting for a write to a WAL file." diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c index 182d666852..94e7944748 100644 --- a/src/backend/utils/init/miscinit.c +++ b/src/backend/utils/init/miscinit.c @@ -306,6 +306,9 @@ GetBackendTypeDesc(BackendType backendType) case B_WAL_SENDER: backendDesc = "walsender"; break; + case B_WAL_SUMMARIZER: + backendDesc = "walsummarizer"; + break; case B_WAL_WRITER: backendDesc = "walwriter"; break; diff --git a/src/backend/utils/misc/guc_tables.c b/src/backend/utils/misc/guc_tables.c index 4c58574166..faf42bdbfb 100644 --- a/src/backend/utils/misc/guc_tables.c +++ b/src/backend/utils/misc/guc_tables.c @@ -63,6 +63,7 @@ #include "postmaster/postmaster.h" #include "postmaster/startup.h" #include "postmaster/syslogger.h" +#include "postmaster/walsummarizer.h" #include "postmaster/walwriter.h" #include "replication/logicallauncher.h" #include "replication/slot.h" @@ -704,6 +705,8 @@ const char *const config_group_names[] = gettext_noop("Write-Ahead Log / Archive Recovery"), /* WAL_RECOVERY_TARGET */ gettext_noop("Write-Ahead Log / Recovery Target"), + /* WAL_SUMMARIZATION */ + gettext_noop("Write-Ahead Log / Summarization"), /* REPLICATION_SENDING */ gettext_noop("Replication / Sending Servers"), /* REPLICATION_PRIMARY */ @@ -3181,6 +3184,32 @@ struct config_int ConfigureNamesInt[] = check_wal_segment_size, NULL, NULL }, + { + {"wal_summarize_mb", PGC_SIGHUP, WAL_SUMMARIZATION, + gettext_noop("Number of bytes of WAL per summary file."), + gettext_noop("Smaller values minimize extra work performed by incremental backup, but increase the number of files on disk."), + GUC_UNIT_MB, + }, + &wal_summarize_mb, + 256, + 0, + INT_MAX, + NULL, NULL, NULL + }, + + { + {"wal_summarize_keep_time", PGC_SIGHUP, WAL_SUMMARIZATION, + gettext_noop("Time for which WAL summary files should be kept."), + NULL, + GUC_UNIT_MIN, + }, + &wal_summarize_keep_time, + 7 * 24 * 60, /* 1 week */ + 0, + INT_MAX, + NULL, NULL, NULL + }, + { {"autovacuum_naptime", PGC_SIGHUP, AUTOVACUUM, gettext_noop("Time to sleep between autovacuum runs."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index d08d55c3fe..4736606ac1 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -299,6 +299,11 @@ #recovery_target_action = 'pause' # 'pause', 'promote', 'shutdown' # (change requires restart) +# - WAL Summarization - + +#wal_summarize_mb = 256 # MB of WAL per summary file, 0 disables +#wal_summarize_keep_time = '7d' # when to remove old summary files, 0 = never + #------------------------------------------------------------------------------ # REPLICATION diff --git a/src/bin/Makefile b/src/bin/Makefile index 373077bf52..aa2210925e 100644 --- a/src/bin/Makefile +++ b/src/bin/Makefile @@ -19,6 +19,7 @@ SUBDIRS = \ pg_archivecleanup \ pg_basebackup \ pg_checksums \ + pg_combinebackup \ pg_config \ pg_controldata \ pg_ctl \ diff --git a/src/bin/initdb/initdb.c b/src/bin/initdb/initdb.c index 0c6f5ceb0a..e68b40d2b5 100644 --- a/src/bin/initdb/initdb.c +++ b/src/bin/initdb/initdb.c @@ -227,6 +227,7 @@ static char *extra_options = ""; static const char *const subdirs[] = { "global", "pg_wal/archive_status", + "pg_wal/summaries", "pg_commit_ts", "pg_dynshmem", "pg_notify", diff --git a/src/bin/meson.build b/src/bin/meson.build index 67cb50630c..4cb6fd59bb 100644 --- a/src/bin/meson.build +++ b/src/bin/meson.build @@ -5,6 +5,7 @@ subdir('pg_amcheck') subdir('pg_archivecleanup') subdir('pg_basebackup') subdir('pg_checksums') +subdir('pg_combinebackup') subdir('pg_config') subdir('pg_controldata') subdir('pg_ctl') diff --git a/src/bin/pg_basebackup/bbstreamer_file.c b/src/bin/pg_basebackup/bbstreamer_file.c index 45f32974ff..6b78ee283d 100644 --- a/src/bin/pg_basebackup/bbstreamer_file.c +++ b/src/bin/pg_basebackup/bbstreamer_file.c @@ -296,6 +296,7 @@ should_allow_existing_directory(const char *pathname) if (strcmp(filename, "pg_wal") == 0 || strcmp(filename, "pg_xlog") == 0 || strcmp(filename, "archive_status") == 0 || + strcmp(filename, "summaries") == 0 || strcmp(filename, "pg_tblspc") == 0) return true; diff --git a/src/bin/pg_basebackup/pg_basebackup.c b/src/bin/pg_basebackup/pg_basebackup.c index 1a8cef345d..33416b11cf 100644 --- a/src/bin/pg_basebackup/pg_basebackup.c +++ b/src/bin/pg_basebackup/pg_basebackup.c @@ -101,6 +101,11 @@ typedef void (*WriteDataCallback) (size_t nbytes, char *buf, */ #define MINIMUM_VERSION_FOR_TERMINATED_TARFILE 150000 +/* + * pg_wal/summaries exists beginning with version 17. + */ +#define MINIMUM_VERSION_FOR_WAL_SUMMARIES 170000 + /* * Different ways to include WAL */ @@ -217,7 +222,8 @@ static void ReceiveBackupManifestInMemoryChunk(size_t r, char *copybuf, void *callback_data); static void BaseBackup(char *compression_algorithm, char *compression_detail, CompressionLocation compressloc, - pg_compress_specification *client_compress); + pg_compress_specification *client_compress, + char *incremental_manifest); static bool reached_end_position(XLogRecPtr segendpos, uint32 timeline, bool segment_finished); @@ -390,6 +396,8 @@ usage(void) printf(_("\nOptions controlling the output:\n")); printf(_(" -D, --pgdata=DIRECTORY receive base backup into directory\n")); printf(_(" -F, --format=p|t output format (plain (default), tar)\n")); + printf(_(" -i, --incremental=OLDMANIFEST\n")); + printf(_(" take incremental or differential backup\n")); printf(_(" -r, --max-rate=RATE maximum transfer rate to transfer data directory\n" " (in kB/s, or use suffix \"k\" or \"M\")\n")); printf(_(" -R, --write-recovery-conf\n" @@ -688,6 +696,23 @@ StartLogStreamer(char *startpos, uint32 timeline, char *sysidentifier, if (pg_mkdir_p(statusdir, pg_dir_create_mode) != 0 && errno != EEXIST) pg_fatal("could not create directory \"%s\": %m", statusdir); + + /* + * For newer server versions, likewise create pg_wal/summaries + */ + if (PQserverVersion(conn) < MINIMUM_VERSION_FOR_WAL_SUMMARIES) + { + char summarydir[MAXPGPATH]; + + snprintf(summarydir, sizeof(summarydir), "%s/%s/summaries", + basedir, + PQserverVersion(conn) < MINIMUM_VERSION_FOR_PG_WAL ? + "pg_xlog" : "pg_wal"); + + if (pg_mkdir_p(statusdir, pg_dir_create_mode) != 0 && + errno != EEXIST) + pg_fatal("could not create directory \"%s\": %m", summarydir); + } } /* @@ -1728,7 +1753,9 @@ ReceiveBackupManifestInMemoryChunk(size_t r, char *copybuf, static void BaseBackup(char *compression_algorithm, char *compression_detail, - CompressionLocation compressloc, pg_compress_specification *client_compress) + CompressionLocation compressloc, + pg_compress_specification *client_compress, + char *incremental_manifest) { PGresult *res; char *sysidentifier; @@ -1794,7 +1821,74 @@ BaseBackup(char *compression_algorithm, char *compression_detail, exit(1); /* - * Start the actual backup + * If the user wants an incremental backup, we must upload the manifest + * for the previous backup upon which it is to be based. + */ + if (incremental_manifest != NULL) + { + int fd; + char mbuf[65536]; + int nbytes; + + /* XXX add a server version check here */ + + /* Open the file. */ + fd = open(incremental_manifest, O_RDONLY | PG_BINARY, 0); + if (fd < 0) + pg_fatal("could not open file \"%s\": %m", incremental_manifest); + + /* Tell the server what we want to do. */ + if (PQsendQuery(conn, "UPLOAD_MANIFEST") == 0) + pg_fatal("could not send replication command \"%s\": %s", + "UPLOAD_MANIFEST", PQerrorMessage(conn)); + res = PQgetResult(conn); + if (PQresultStatus(res) != PGRES_COPY_IN) + { + if (PQresultStatus(res) == PGRES_FATAL_ERROR) + pg_fatal("could not upload manifest: %s", + PQerrorMessage(conn)); + else + pg_fatal("could not upload manifest: unexpected status %s", + PQresStatus(PQresultStatus(res))); + } + + /* Loop, reading from the file and sending the data to the server. */ + while ((nbytes = read(fd, mbuf, sizeof mbuf)) > 0) + { + if (PQputCopyData(conn, mbuf, nbytes) < 0) + pg_fatal("could not send COPY data: %s", + PQerrorMessage(conn)); + } + + /* Bail out if we exited the loop due to an error. */ + if (nbytes < 0) + pg_fatal("could not read file \"%s\": %m", incremental_manifest); + + /* End the COPY operation. */ + if (PQputCopyEnd(conn, NULL) < 0) + pg_fatal("could not send end-of-COPY: %s", + PQerrorMessage(conn)); + + /* See whether the server is happy with what we sent. */ + res = PQgetResult(conn); + if (PQresultStatus(res) == PGRES_FATAL_ERROR) + pg_fatal("could not upload manifest: %s", + PQerrorMessage(conn)); + else if (PQresultStatus(res) != PGRES_COMMAND_OK) + pg_fatal("could not upload manifest: unexpected status %s", + PQresStatus(PQresultStatus(res))); + + /* Consume ReadyForQuery message from server. */ + res = PQgetResult(conn); + if (res != NULL) + pg_fatal("unexpected extra result while sending manifest"); + + /* Add INCREMENTAL option to BASE_BACKUP command. */ + AppendPlainCommandOption(&buf, use_new_option_syntax, "INCREMENTAL"); + } + + /* + * Continue building up the options list for the BASE_BACKUP command. */ AppendStringCommandOption(&buf, use_new_option_syntax, "LABEL", label); if (estimatesize) @@ -1901,6 +1995,7 @@ BaseBackup(char *compression_algorithm, char *compression_detail, else basebkp = psprintf("BASE_BACKUP %s", buf.data); + /* OK, try to start the backup. */ if (PQsendQuery(conn, basebkp) == 0) pg_fatal("could not send replication command \"%s\": %s", "BASE_BACKUP", PQerrorMessage(conn)); @@ -2256,6 +2351,7 @@ main(int argc, char **argv) {"version", no_argument, NULL, 'V'}, {"pgdata", required_argument, NULL, 'D'}, {"format", required_argument, NULL, 'F'}, + {"incremental", required_argument, NULL, 'i'}, {"checkpoint", required_argument, NULL, 'c'}, {"create-slot", no_argument, NULL, 'C'}, {"max-rate", required_argument, NULL, 'r'}, @@ -2293,6 +2389,7 @@ main(int argc, char **argv) int option_index; char *compression_algorithm = "none"; char *compression_detail = NULL; + char *incremental_manifest = NULL; CompressionLocation compressloc = COMPRESS_LOCATION_UNSPECIFIED; pg_compress_specification client_compress; @@ -2317,7 +2414,7 @@ main(int argc, char **argv) atexit(cleanup_directories_atexit); - while ((c = getopt_long(argc, argv, "c:Cd:D:F:h:l:nNp:Pr:Rs:S:t:T:U:vwWX:zZ:", + while ((c = getopt_long(argc, argv, "c:Cd:D:F:h:i:l:nNp:Pr:Rs:S:t:T:U:vwWX:zZ:", long_options, &option_index)) != -1) { switch (c) @@ -2352,6 +2449,9 @@ main(int argc, char **argv) case 'h': dbhost = pg_strdup(optarg); break; + case 'i': + incremental_manifest = pg_strdup(optarg); + break; case 'l': label = pg_strdup(optarg); break; @@ -2765,7 +2865,7 @@ main(int argc, char **argv) } BaseBackup(compression_algorithm, compression_detail, compressloc, - &client_compress); + &client_compress, incremental_manifest); success = true; return 0; diff --git a/src/bin/pg_basebackup/t/010_pg_basebackup.pl b/src/bin/pg_basebackup/t/010_pg_basebackup.pl index b9f5e1266b..bf765291e7 100644 --- a/src/bin/pg_basebackup/t/010_pg_basebackup.pl +++ b/src/bin/pg_basebackup/t/010_pg_basebackup.pl @@ -223,10 +223,10 @@ SKIP: "check backup dir permissions"); } -# Only archive_status directory should be copied in pg_wal/. +# Only archive_status and summaries directories should be copied in pg_wal/. is_deeply( [ sort(slurp_dir("$tempdir/backup/pg_wal/")) ], - [ sort qw(. .. archive_status) ], + [ sort qw(. .. archive_status summaries) ], 'no WAL files copied'); # Contents of these directories should not be copied. diff --git a/src/bin/pg_combinebackup/.gitignore b/src/bin/pg_combinebackup/.gitignore new file mode 100644 index 0000000000..d7e617438c --- /dev/null +++ b/src/bin/pg_combinebackup/.gitignore @@ -0,0 +1 @@ +pg_combinebackup diff --git a/src/bin/pg_combinebackup/Makefile b/src/bin/pg_combinebackup/Makefile new file mode 100644 index 0000000000..78ba05e624 --- /dev/null +++ b/src/bin/pg_combinebackup/Makefile @@ -0,0 +1,52 @@ +#------------------------------------------------------------------------- +# +# Makefile for src/bin/pg_combinebackup +# +# Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group +# Portions Copyright (c) 1994, Regents of the University of California +# +# src/bin/pg_combinebackup/Makefile +# +#------------------------------------------------------------------------- + +PGFILEDESC = "pg_combinebackup - combine incremental backups" +PGAPPICON=win32 + +subdir = src/bin/pg_combinebackup +top_builddir = ../../.. +include $(top_builddir)/src/Makefile.global + +override CPPFLAGS := -I$(libpq_srcdir) $(CPPFLAGS) +LDFLAGS_INTERNAL += -L$(top_builddir)/src/fe_utils -lpgfeutils + +OBJS = \ + $(WIN32RES) \ + pg_combinebackup.o \ + backup_label.o \ + copy_file.o \ + load_manifest.o \ + reconstruct.o \ + write_manifest.o + +all: pg_combinebackup + +pg_combinebackup: $(OBJS) | submake-libpgport submake-libpgfeutils + $(CC) $(CFLAGS) $^ $(LDFLAGS) $(LDFLAGS_EX) $(LIBS) -o $@$(X) + +install: all installdirs + $(INSTALL_PROGRAM) pg_combinebackup$(X) '$(DESTDIR)$(bindir)/pg_combinebackup$(X)' + +installdirs: + $(MKDIR_P) '$(DESTDIR)$(bindir)' + +uninstall: + rm -f '$(DESTDIR)$(bindir)/pg_combinebackup$(X)' + +clean distclean maintainer-clean: + rm -f pg_combinebackup$(X) $(OBJS) + +check: + $(prove_check) + +installcheck: + $(prove_installcheck) diff --git a/src/bin/pg_combinebackup/backup_label.c b/src/bin/pg_combinebackup/backup_label.c new file mode 100644 index 0000000000..2a62aa6fad --- /dev/null +++ b/src/bin/pg_combinebackup/backup_label.c @@ -0,0 +1,281 @@ +/*------------------------------------------------------------------------- + * + * Read and manipulate backup label files + * + * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group + * Portions Copyright (c) 1994, Regents of the University of California + * + * src/bin/pg_combinebackup/backup_label.c + * + *------------------------------------------------------------------------- + */ +#include "postgres_fe.h" + +#include <unistd.h> + +#include "access/xlogdefs.h" +#include "backup_label.h" +#include "common/logging.h" +#include "common/file_perm.h" +#include "write_manifest.h" + +static int get_eol_offset(StringInfo buf); +static bool line_starts_with(char *s, char *e, char *match, char **sout); +static bool parse_lsn(char *s, char *e, XLogRecPtr *lsn, char **c); +static bool parse_tli(char *s, char *e, TimeLineID *tli); + +/* + * Parse a backup label file, starting at buf->cursor. + * + * We expect to find a START WAL LOCATION line, followed by a LSN, followed + * by a space; the resulting LSN is stored into *start_lsn. + * + * We expect to find a START TIMELINE line, followed by a TLI, followed by + * a newline; the resulting TLI is stored into *start_tli. + * + * We expect to find either both INCREMENTAL FROM LSN and INCREMENTAL FROM TLI + * or neither. If these are found, they should be followed by an LSN or TLI + * respectively and then by a newline, and the values will be stored into + * *previous_lsn and *previous_tli, respectively. + * + * Other lines in the provided backup_label data are ignored. filename is used + * for error reporting; errors are fatal. + */ +void +parse_backup_label(char *filename, StringInfo buf, + TimeLineID *start_tli, XLogRecPtr *start_lsn, + TimeLineID *previous_tli, XLogRecPtr *previous_lsn) +{ + int found = 0; + + *start_tli = 0; + *start_lsn = InvalidXLogRecPtr; + *previous_tli = 0; + *previous_lsn = InvalidXLogRecPtr; + + while (buf->cursor < buf->len) + { + char *s = &buf->data[buf->cursor]; + int eo = get_eol_offset(buf); + char *e = &buf->data[eo]; + char *c; + + if (line_starts_with(s, e, "START WAL LOCATION: ", &s)) + { + if (!parse_lsn(s, e, start_lsn, &c)) + pg_fatal("%s: could not parse START WAL LOCATION", + filename); + if (c >= e || *c != ' ') + pg_fatal("%s: improper terminator for START WAL LOCATION", + filename); + found |= 1; + } + else if (line_starts_with(s, e, "START TIMELINE: ", &s)) + { + if (!parse_tli(s, e, start_tli)) + pg_fatal("%s: could not parse TLI for START TIMELINE", + filename); + if (*start_tli == 0) + pg_fatal("%s: invalid TLI", filename); + found |= 2; + } + else if (line_starts_with(s, e, "INCREMENTAL FROM LSN: ", &s)) + { + if (!parse_lsn(s, e, previous_lsn, &c)) + pg_fatal("%s: could not parse INCREMENTAL FROM LSN", + filename); + if (c >= e || *c != '\n') + pg_fatal("%s: improper terminator for INCREMENTAL FROM LSN", + filename); + found |= 4; + } + else if (line_starts_with(s, e, "INCREMENTAL FROM TLI: ", &s)) + { + if (!parse_tli(s, e, previous_tli)) + pg_fatal("%s: could not parse INCREMENTAL FROM TLI", + filename); + if (*previous_tli == 0) + pg_fatal("%s: invalid TLI", filename); + found |= 8; + } + + buf->cursor = eo; + } + + if ((found & 1) == 0) + pg_fatal("%s: could not find START WAL LOCATION", filename); + if ((found & 2) == 0) + pg_fatal("%s: could not find START TIMELINE", filename); + if ((found & 4) != 0 && (found & 8) == 0) + pg_fatal("%s: INCREMENTAL FROM LSN requires INCREMENTAL FROM TLI", filename); + if ((found & 8) != 0 && (found & 4) == 0) + pg_fatal("%s: INCREMENTAL FROM TLI requires INCREMENTAL FROM LSN", filename); +} + +/* + * Write a backup label file to the output directory. + * + * This will be identical to the provided backup_label file, except that the + * INCREMENTAL FROM LSN and INCREMENTAL FROM TLI lines will be omitted. + * + * The new file will be checksummed using the specified algorithm. If + * mwriter != NULL, it will be added to the manifest. + */ +void +write_backup_label(char *output_directory, StringInfo buf, + pg_checksum_type checksum_type, manifest_writer *mwriter) +{ + char output_filename[MAXPGPATH]; + int output_fd; + pg_checksum_context checksum_ctx; + uint8 checksum_payload[PG_CHECKSUM_MAX_LENGTH]; + int checksum_length; + + pg_checksum_init(&checksum_ctx, checksum_type); + + snprintf(output_filename, MAXPGPATH, "%s/backup_label", output_directory); + + if ((output_fd = open(output_filename, + O_WRONLY | O_CREAT | O_EXCL | PG_BINARY, + pg_file_create_mode)) < 0) + pg_fatal("could not open file \"%s\": %m", output_filename); + + while (buf->cursor < buf->len) + { + char *s = &buf->data[buf->cursor]; + int eo = get_eol_offset(buf); + char *e = &buf->data[eo]; + + if (!line_starts_with(s, e, "INCREMENTAL FROM LSN: ", NULL) && + !line_starts_with(s, e, "INCREMENTAL FROM TLI: ", NULL)) + { + ssize_t wb; + + wb = write(output_fd, s, e - s); + if (wb != e - s) + { + if (wb < 0) + pg_fatal("could not write file \"%s\": %m", output_filename); + else + pg_fatal("could not write file \"%s\": wrote only %d of %d bytes", + output_filename, (int) wb, (int) (e - s)); + } + if (pg_checksum_update(&checksum_ctx, (uint8 *) s, e - s) < 0) + pg_fatal("could not update checksum of file \"%s\"", + output_filename); + } + + buf->cursor = eo; + } + + if (close(output_fd) != 0) + pg_fatal("could not close \"%s\": %m", output_filename); + + checksum_length = pg_checksum_final(&checksum_ctx, checksum_payload); + + if (mwriter != NULL) + { + struct stat sb; + + /* + * We could track the length ourselves, but must stat() to get the + * mtime. + */ + if (stat(output_filename, &sb) < 0) + pg_fatal("could not stat file \"%s\": %m", output_filename); + add_file_to_manifest(mwriter, "backup_label", sb.st_size, + sb.st_mtime, checksum_type, + checksum_length, checksum_payload); + } +} + +/* + * Return the offset at which the next line in the buffer starts, or there + * is none, the offset at which the buffer ends. + * + * The search begins at buf->cursor. + */ +static int +get_eol_offset(StringInfo buf) +{ + int eo = buf->cursor; + + while (eo < buf->len) + { + if (buf->data[eo] == '\n') + return eo + 1; + ++eo; + } + + return eo; +} + +/* + * Test whether the line that runs from s to e (inclusive of *s, but not + * inclusive of *e) starts with the match string provided, and return true + * or false according to whether or not this is the case. + * + * If the function returns true and if *sout != NULL, stores a pointer to the + * byte following the match into *sout. + */ +static bool +line_starts_with(char *s, char *e, char *match, char **sout) +{ + while (s < e && *match != '\0' && *s == *match) + ++s, ++match; + + if (*match == '\0' && sout != NULL) + *sout = s; + + return (*match == '\0'); +} + +/* + * Parse an LSN starting at s and not stopping at or before e. The return value + * is true on success and otherwise false. On success, stores the result into + * *lsn and sets *c to the first character that is not part of the LSN. + */ +static bool +parse_lsn(char *s, char *e, XLogRecPtr *lsn, char **c) +{ + char save = *e; + int nchars; + bool success; + unsigned hi; + unsigned lo; + + *e = '\0'; + success = (sscanf(s, "%X/%X%n", &hi, &lo, &nchars) == 2); + *e = save; + + if (success) + { + *lsn = ((XLogRecPtr) hi) << 32 | (XLogRecPtr) lo; + *c = s + nchars; + } + + return success; +} + +/* + * Parse a TLI starting at s and stopping at or before e. The return value is + * true on success and otherwise false. On success, stores the result into + * *tli. If the first character that is not part of the TLI is anything other + * than a newline, that is deemed a failure. + */ +static bool +parse_tli(char *s, char *e, TimeLineID *tli) +{ + char save = *e; + int nchars; + bool success; + + *e = '\0'; + success = (sscanf(s, "%u%n", tli, &nchars) == 1); + *e = save; + + if (success && s[nchars] != '\n') + success = false; + + return success; +} diff --git a/src/bin/pg_combinebackup/backup_label.h b/src/bin/pg_combinebackup/backup_label.h new file mode 100644 index 0000000000..08d6ed67a9 --- /dev/null +++ b/src/bin/pg_combinebackup/backup_label.h @@ -0,0 +1,29 @@ +/*------------------------------------------------------------------------- + * + * Read and manipulate backup label files + * + * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group + * Portions Copyright (c) 1994, Regents of the University of California + * + * src/bin/pg_combinebackup/backup_label.h + * + *------------------------------------------------------------------------- + */ +#ifndef BACKUP_LABEL_H +#define BACKUP_LABEL_H + +#include "common/checksum_helper.h" +#include "lib/stringinfo.h" + +struct manifest_writer; + +extern void parse_backup_label(char *filename, StringInfo buf, + TimeLineID *start_tli, + XLogRecPtr *start_lsn, + TimeLineID *previous_tli, + XLogRecPtr *previous_lsn); +extern void write_backup_label(char *output_directory, StringInfo buf, + pg_checksum_type checksum_type, + struct manifest_writer *mwriter); + +#endif /* BACKUP_LABEL_H */ diff --git a/src/bin/pg_combinebackup/copy_file.c b/src/bin/pg_combinebackup/copy_file.c new file mode 100644 index 0000000000..8ba6cc09e4 --- /dev/null +++ b/src/bin/pg_combinebackup/copy_file.c @@ -0,0 +1,169 @@ +/* + * Copy entire files. + * + * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group + * Portions Copyright (c) 1994, Regents of the University of California + * + * src/bin/pg_combinebackup/copy_file.h + * + *------------------------------------------------------------------------- + */ +#include "postgres_fe.h" + +#ifdef HAVE_COPYFILE_H +#include <copyfile.h> +#endif +#include <fcntl.h> +#include <sys/stat.h> +#include <unistd.h> + +#include "common/file_perm.h" +#include "common/logging.h" +#include "copy_file.h" + +static void copy_file_blocks(const char *src, const char *dst, + pg_checksum_context *checksum_ctx); + +#ifdef WIN32 +static void copy_file_copyfile(const char *src, const char *dst); +#endif + +/* + * Copy a regular file, optionally computing a checksum, and emitting + * appropriate debug messages. But if we're in dry-run mode, then just emit + * the messages and don't copy anything. + */ +void +copy_file(const char *src, const char *dst, + pg_checksum_context *checksum_ctx, bool dry_run) +{ + /* + * In dry-run mode, we don't actually copy anything, nor do we read any + * data from the source file, but we do verify that we can open it. + */ + if (dry_run) + { + int fd; + + if ((fd = open(src, O_RDONLY | PG_BINARY)) < 0) + pg_fatal("could not open \"%s\": %m", src); + if (close(fd) < 0) + pg_fatal("could not close \"%s\": %m", src); + } + + /* + * If we don't need to compute a checksum, then we can use any special + * operating system primitives that we know about to copy the file; this + * may be quicker than a naive block copy. + */ + if (checksum_ctx->type != CHECKSUM_TYPE_NONE) + { + char *strategy_name = NULL; + void (*strategy_implementation) (const char *, const char *) = NULL; + +#ifdef WIN32 + strategy_name = "CopyFile"; + strategy_implementation = copy_file_copyfile; +#endif + + if (strategy_name != NULL) + { + if (dry_run) + pg_log_debug("would copy \"%s\" to \"%s\" using strategy %s", + src, dst, strategy_name); + else + { + pg_log_debug("copying \"%s\" to \"%s\" using strategy %s", + src, dst, strategy_name); + (*strategy_implementation) (src, dst); + } + return; + } + } + + /* + * Fall back to the simple approach of reading and writing all the blocks, + * feeding them into the checksum context as we go. + */ + if (dry_run) + { + if (checksum_ctx->type == CHECKSUM_TYPE_NONE) + pg_log_debug("would copy \"%s\" to \"%s\"", + src, dst); + else + pg_log_debug("would copy \"%s\" to \"%s\" and checksum with %s", + src, dst, pg_checksum_type_name(checksum_ctx->type)); + } + else + { + if (checksum_ctx->type == CHECKSUM_TYPE_NONE) + pg_log_debug("copying \"%s\" to \"%s\"", + src, dst); + else + pg_log_debug("copying \"%s\" to \"%s\" and checksumming with %s", + src, dst, pg_checksum_type_name(checksum_ctx->type)); + copy_file_blocks(src, dst, checksum_ctx); + } +} + +/* + * Copy a file block by block, and optionally compute a checksum as we go. + */ +static void +copy_file_blocks(const char *src, const char *dst, + pg_checksum_context *checksum_ctx) +{ + int src_fd; + int dest_fd; + uint8 *buffer; + const int buffer_size = 50 * BLCKSZ; + ssize_t rb; + unsigned offset = 0; + + if ((src_fd = open(src, O_RDONLY | PG_BINARY, 0)) < 0) + pg_fatal("could not open file \"%s\": %m", src); + + if ((dest_fd = open(dst, O_WRONLY | O_CREAT | O_EXCL | PG_BINARY, + pg_file_create_mode)) < 0) + pg_fatal("could not open file \"%s\": %m", dst); + + buffer = pg_malloc(buffer_size); + + while ((rb = read(src_fd, buffer, buffer_size)) > 0) + { + ssize_t wb; + + if ((wb = write(dest_fd, buffer, rb)) != rb) + { + if (wb < 0) + pg_fatal("could not write file \"%s\": %m", dst); + else + pg_fatal("could not write file \"%s\": wrote only %d of %d bytes at offset %u", + dst, (int) wb, (int) rb, offset); + } + + if (pg_checksum_update(checksum_ctx, buffer, rb) < 0) + pg_fatal("could not update checksum of file \"%s\"", dst); + + offset += rb; + } + + if (rb < 0) + pg_fatal("could not read file \"%s\": %m", dst); + + pg_free(buffer); + close(src_fd); + close(dest_fd); +} + +#ifdef WIN32 +static void +copy_file_copyfile(const char *src, const char *dst) +{ + if (CopyFile(src, dst, true) == 0) + { + _dosmaperr(GetLastError()); + pg_fatal("could not copy \"%s\" to \"%s\": %m", src, dst); + } +} +#endif /* WIN32 */ diff --git a/src/bin/pg_combinebackup/copy_file.h b/src/bin/pg_combinebackup/copy_file.h new file mode 100644 index 0000000000..031030bacb --- /dev/null +++ b/src/bin/pg_combinebackup/copy_file.h @@ -0,0 +1,19 @@ +/* + * Copy entire files. + * + * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group + * Portions Copyright (c) 1994, Regents of the University of California + * + * src/bin/pg_combinebackup/copy_file.h + * + *------------------------------------------------------------------------- + */ +#ifndef COPY_FILE_H +#define COPY_FILE_H + +#include "common/checksum_helper.h" + +extern void copy_file(const char *src, const char *dst, + pg_checksum_context *checksum_ctx, bool dry_run); + +#endif /* COPY_FILE_H */ diff --git a/src/bin/pg_combinebackup/load_manifest.c b/src/bin/pg_combinebackup/load_manifest.c new file mode 100644 index 0000000000..d0b8de7912 --- /dev/null +++ b/src/bin/pg_combinebackup/load_manifest.c @@ -0,0 +1,245 @@ +/*------------------------------------------------------------------------- + * + * Load data from a backup manifest into memory. + * + * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group + * Portions Copyright (c) 1994, Regents of the University of California + * + * src/bin/pg_combinebackup/load_manifest.c + * + *------------------------------------------------------------------------- + */ + +#include "postgres_fe.h" + +#include <sys/stat.h> +#include <unistd.h> + +#include "common/hashfn.h" +#include "common/logging.h" +#include "common/parse_manifest.h" +#include "load_manifest.h" + +/* + * For efficiency, we'd like our hash table containing information about the + * manifest to start out with approximately the correct number of entries. + * There's no way to know the exact number of entries without reading the whole + * file, but we can get an estimate by dividing the file size by the estimated + * number of bytes per line. + * + * This could be off by about a factor of two in either direction, because the + * checksum algorithm has a big impact on the line lengths; e.g. a SHA512 + * checksum is 128 hex bytes, whereas a CRC-32C value is only 8, and there + * might be no checksum at all. + */ +#define ESTIMATED_BYTES_PER_MANIFEST_LINE 100 + +/* + * Define a hash table which we can use to store information about the files + * mentioned in the backup manifest. + */ +static uint32 hash_string_pointer(char *s); +#define SH_PREFIX manifest_files +#define SH_ELEMENT_TYPE manifest_file +#define SH_KEY_TYPE char * +#define SH_KEY pathname +#define SH_HASH_KEY(tb, key) hash_string_pointer(key) +#define SH_EQUAL(tb, a, b) (strcmp(a, b) == 0) +#define SH_SCOPE extern +#define SH_RAW_ALLOCATOR pg_malloc0 +#define SH_DEFINE +#include "lib/simplehash.h" + +static void record_manifest_details_for_file(JsonManifestParseContext *context, + char *pathname, size_t size, + pg_checksum_type checksum_type, + int checksum_length, + uint8 *checksum_payload); +static void record_manifest_details_for_wal_range(JsonManifestParseContext *context, + TimeLineID tli, + XLogRecPtr start_lsn, + XLogRecPtr end_lsn); +static void report_manifest_error(JsonManifestParseContext *context, + const char *fmt,...) + pg_attribute_printf(2, 3) pg_attribute_noreturn(); + +/* + * Load backup_manifest files from an array of backups and produces an array + * of manifest_data objects. + * + * NB: Since load_backup_manifest() can return NULL, the resulting array could + * contain NULL entries. + */ +manifest_data ** +load_backup_manifests(int n_backups, char **backup_directories) +{ + manifest_data **result; + int i; + + result = pg_malloc(sizeof(manifest_data *) * n_backups); + for (i = 0; i < n_backups; ++i) + result[i] = load_backup_manifest(backup_directories[i]); + + return result; +} + +/* + * Parse the backup_manifest file in the named backup directory. Construct a + * hash table with information about all the files it mentions, and a linked + * list of all the WAL ranges it mentions. + * + * If the backup_manifest file simply doesn't exist, logs a warning and returns + * NULL. Any other error, or any error parsing the contents of the file, is + * fatal. + */ +manifest_data * +load_backup_manifest(char *backup_directory) +{ + char pathname[MAXPGPATH]; + int fd; + struct stat statbuf; + off_t estimate; + uint32 initial_size; + manifest_files_hash *ht; + char *buffer; + int rc; + JsonManifestParseContext context; + manifest_data *result; + + /* Open the manifest file. */ + snprintf(pathname, MAXPGPATH, "%s/backup_manifest", backup_directory); + if ((fd = open(pathname, O_RDONLY | PG_BINARY, 0)) < 0) + { + if (errno == EEXIST) + { + pg_log_warning("\"%s\" does not exist", pathname); + return NULL; + } + pg_fatal("could not open file \"%s\": %m", pathname); + } + + /* Figure out how big the manifest is. */ + if (fstat(fd, &statbuf) != 0) + pg_fatal("could not stat file \"%s\": %m", pathname); + + /* Guess how large to make the hash table based on the manifest size. */ + estimate = statbuf.st_size / ESTIMATED_BYTES_PER_MANIFEST_LINE; + initial_size = Min(PG_UINT32_MAX, Max(estimate, 256)); + + /* Create the hash table. */ + ht = manifest_files_create(initial_size, NULL); + + /* + * Slurp in the whole file. + * + * This is not ideal, but there's currently no way to get pg_parse_json() + * to perform incremental parsing. + */ + buffer = pg_malloc(statbuf.st_size); + rc = read(fd, buffer, statbuf.st_size); + if (rc != statbuf.st_size) + { + if (rc < 0) + pg_fatal("could not read file \"%s\": %m", pathname); + else + pg_fatal("could not read file \"%s\": read %d of %lld", + pathname, rc, (long long int) statbuf.st_size); + } + + /* Close the manifest file. */ + close(fd); + + /* Parse the manifest. */ + result = pg_malloc0(sizeof(manifest_data)); + result->files = ht; + context.private_data = result; + context.perfile_cb = record_manifest_details_for_file; + context.perwalrange_cb = record_manifest_details_for_wal_range; + context.error_cb = report_manifest_error; + json_parse_manifest(&context, buffer, statbuf.st_size); + + /* All done. */ + pfree(buffer); + return result; +} + +/* + * Report an error while parsing the manifest. + * + * We consider all such errors to be fatal errors. The manifest parser + * expects this function not to return. + */ +static void +report_manifest_error(JsonManifestParseContext *context, const char *fmt,...) +{ + va_list ap; + + va_start(ap, fmt); + pg_log_generic_v(PG_LOG_ERROR, PG_LOG_PRIMARY, gettext(fmt), ap); + va_end(ap); + + exit(1); +} + +/* + * Record details extracted from the backup manifest for one file. + */ +static void +record_manifest_details_for_file(JsonManifestParseContext *context, + char *pathname, size_t size, + pg_checksum_type checksum_type, + int checksum_length, uint8 *checksum_payload) +{ + manifest_data *manifest = context->private_data; + manifest_file *m; + bool found; + + /* Make a new entry in the hash table for this file. */ + m = manifest_files_insert(manifest->files, pathname, &found); + if (found) + pg_fatal("duplicate path name in backup manifest: \"%s\"", pathname); + + /* Initialize the entry. */ + m->size = size; + m->checksum_type = checksum_type; + m->checksum_length = checksum_length; + m->checksum_payload = checksum_payload; +} + +/* + * Record details extracted from the backup manifest for one WAL range. + */ +static void +record_manifest_details_for_wal_range(JsonManifestParseContext *context, + TimeLineID tli, + XLogRecPtr start_lsn, XLogRecPtr end_lsn) +{ + manifest_data *manifest = context->private_data; + manifest_wal_range *range; + + /* Allocate and initialize a struct describing this WAL range. */ + range = palloc(sizeof(manifest_wal_range)); + range->tli = tli; + range->start_lsn = start_lsn; + range->end_lsn = end_lsn; + range->prev = manifest->last_wal_range; + range->next = NULL; + + /* Add it to the end of the list. */ + if (manifest->first_wal_range == NULL) + manifest->first_wal_range = range; + else + manifest->last_wal_range->next = range; + manifest->last_wal_range = range; +} + +/* + * Helper function for manifest_files hash table. + */ +static uint32 +hash_string_pointer(char *s) +{ + unsigned char *ss = (unsigned char *) s; + + return hash_bytes(ss, strlen(s)); +} diff --git a/src/bin/pg_combinebackup/load_manifest.h b/src/bin/pg_combinebackup/load_manifest.h new file mode 100644 index 0000000000..2bfeeff156 --- /dev/null +++ b/src/bin/pg_combinebackup/load_manifest.h @@ -0,0 +1,67 @@ +/*------------------------------------------------------------------------- + * + * Load data from a backup manifest into memory. + * + * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group + * Portions Copyright (c) 1994, Regents of the University of California + * + * src/bin/pg_combinebackup/load_manifest.h + * + *------------------------------------------------------------------------- + */ +#ifndef LOAD_MANIFEST_H +#define LOAD_MANIFEST_H + +#include "access/xlogdefs.h" +#include "common/checksum_helper.h" + +/* + * Each file described by the manifest file is parsed to produce an object + * like this. + */ +typedef struct manifest_file +{ + uint32 status; /* hash status */ + char *pathname; + size_t size; + pg_checksum_type checksum_type; + int checksum_length; + uint8 *checksum_payload; +} manifest_file; + +#define SH_PREFIX manifest_files +#define SH_ELEMENT_TYPE manifest_file +#define SH_KEY_TYPE char * +#define SH_SCOPE extern +#define SH_RAW_ALLOCATOR pg_malloc0 +#define SH_DECLARE +#include "lib/simplehash.h" + +/* + * Each WAL range described by the manifest file is parsed to produce an + * object like this. + */ +typedef struct manifest_wal_range +{ + TimeLineID tli; + XLogRecPtr start_lsn; + XLogRecPtr end_lsn; + struct manifest_wal_range *next; + struct manifest_wal_range *prev; +} manifest_wal_range; + +/* + * All the data parsed from a backup_manifest file. + */ +typedef struct manifest_data +{ + manifest_files_hash *files; + manifest_wal_range *first_wal_range; + manifest_wal_range *last_wal_range; +} manifest_data; + +extern manifest_data *load_backup_manifest(char *backup_directory); +extern manifest_data **load_backup_manifests(int n_backups, + char **backup_directories); + +#endif /* LOAD_MANIFEST_H */ diff --git a/src/bin/pg_combinebackup/meson.build b/src/bin/pg_combinebackup/meson.build new file mode 100644 index 0000000000..a6036dea74 --- /dev/null +++ b/src/bin/pg_combinebackup/meson.build @@ -0,0 +1,35 @@ +# Copyright (c) 2022-2023, PostgreSQL Global Development Group + +pg_combinebackup_sources = files( + 'pg_combinebackup.c', + 'backup_label.c', + 'copy_file.c', + 'load_manifest.c', + 'reconstruct.c', + 'write_manifest.c', +) + +if host_system == 'windows' + pg_combinebackup_sources += rc_bin_gen.process(win32ver_rc, extra_args: [ + '--NAME', 'pg_combinebackup', + '--FILEDESC', 'pg_combinebackup - combine incremental backups',]) +endif + +pg_combinebackup = executable('pg_combinebackup', + pg_combinebackup_sources, + dependencies: [frontend_code], + kwargs: default_bin_args, +) +bin_targets += pg_combinebackup + +tests += { + 'name': 'pg_combinebackup', + 'sd': meson.current_source_dir(), + 'bd': meson.current_build_dir(), + 'tap': { + 'tests': [ + 't/001_basic.pl', + 't/002_compare_backups.pl', + ], + } +} diff --git a/src/bin/pg_combinebackup/pg_combinebackup.c b/src/bin/pg_combinebackup/pg_combinebackup.c new file mode 100644 index 0000000000..32d2846433 --- /dev/null +++ b/src/bin/pg_combinebackup/pg_combinebackup.c @@ -0,0 +1,1270 @@ +/*------------------------------------------------------------------------- + * + * pg_combinebackup.c + * Combine incremental backups with prior backups. + * + * Copyright (c) 2017-2023, PostgreSQL Global Development Group + * + * IDENTIFICATION + * src/bin/pg_combinebackup/pg_combinebackup.c + * + *------------------------------------------------------------------------- + */ +#include "postgres_fe.h" + +#include <dirent.h> +#include <fcntl.h> +#include <limits.h> + +#include "access/xlogdefs.h" +#include "backup_label.h" +#include "common/blkreftable.h" +#include "common/checksum_helper.h" +#include "common/controldata_utils.h" +#include "common/file_perm.h" +#include "common/file_utils.h" +#include "common/logging.h" +#include "copy_file.h" +#include "fe_utils/option_utils.h" +#include "lib/stringinfo.h" +#include "load_manifest.h" +#include "getopt_long.h" +#include "reconstruct.h" +#include "write_manifest.h" + +/* Incremental file naming convention. */ +#define INCREMENTAL_PREFIX "INCREMENTAL." +#define INCREMENTAL_PREFIX_LENGTH 12 + +/* + * Tracking for directories that need to be removed, or have their contents + * removed, if the operation fails. + */ +typedef struct cb_cleanup_dir +{ + char *target_path; + bool rmtopdir; + struct cb_cleanup_dir *next; +} cb_cleanup_dir; + +/* + * Stores a tablespace mapping provided using -T, --tablespace-mapping. + */ +typedef struct cb_tablespace_mapping +{ + char old_dir[MAXPGPATH]; + char new_dir[MAXPGPATH]; + struct cb_tablespace_mapping *next; +} cb_tablespace_mapping; + +/* + * Stores data parsed from all command-line options. + */ +typedef struct cb_options +{ + bool debug; + char *output; + bool dry_run; + bool no_sync; + cb_tablespace_mapping *tsmappings; + pg_checksum_type manifest_checksums; + bool no_manifest; + DataDirSyncMethod sync_method; +} cb_options; + +/* + * Data about a tablespace. + * + * Every normal tablespace needs a tablespace mapping, but in-place tablespaces + * don't, so the list of tablespaces can contain more entries than the list of + * tablespace mappings. + */ +typedef struct cb_tablespace +{ + Oid oid; + bool in_place; + char old_dir[MAXPGPATH]; + char new_dir[MAXPGPATH]; + struct cb_tablespace *next; +} cb_tablespace; + +/* Directories to be removed if we exit uncleanly. */ +cb_cleanup_dir *cleanup_dir_list = NULL; + +static void add_tablespace_mapping(cb_options *opt, char *arg); +static StringInfo check_backup_label_files(int n_backups, char **backup_dirs); +static void check_control_files(int n_backups, char **backup_dirs); +static void check_input_dir_permissions(char *dir); +static void cleanup_directories_atexit(void); +static void create_output_directory(char *dirname, cb_options *opt); +static void help(const char *progname); +static bool parse_oid(char *s, Oid *result); +static void process_directory_recursively(Oid tsoid, + char *input_directory, + char *output_directory, + char *relative_path, + int n_prior_backups, + char **prior_backup_dirs, + manifest_data **manifests, + manifest_writer *mwriter, + cb_options *opt); +static int read_pg_version_file(char *directory); +static void remember_to_cleanup_directory(char *target_path, bool rmtopdir); +static void reset_directory_cleanup_list(void); +static cb_tablespace *scan_for_existing_tablespaces(char *pathname, + cb_options *opt); +static void slurp_file(int fd, char *filename, StringInfo buf, int maxlen); + +/* + * Main program. + */ +int +main(int argc, char *argv[]) +{ + static struct option long_options[] = { + {"debug", no_argument, NULL, 'd'}, + {"dry-run", no_argument, NULL, 'n'}, + {"no-sync", no_argument, NULL, 'N'}, + {"output", required_argument, NULL, 'o'}, + {"tablespace-mapping", no_argument, NULL, 'T'}, + {"manifest-checksums", required_argument, NULL, 1}, + {"no-manifest", no_argument, NULL, 2}, + {"sync-method", required_argument, NULL, 3}, + {NULL, 0, NULL, 0} + }; + + const char *progname; + char *last_input_dir; + int optindex; + int c; + int n_backups; + int n_prior_backups; + int version; + char **prior_backup_dirs; + cb_options opt; + cb_tablespace *tablespaces; + cb_tablespace *ts; + StringInfo last_backup_label; + manifest_data **manifests; + manifest_writer *mwriter; + + pg_logging_init(argv[0]); + progname = get_progname(argv[0]); + handle_help_version_opts(argc, argv, progname, help); + + memset(&opt, 0, sizeof(opt)); + opt.manifest_checksums = CHECKSUM_TYPE_CRC32C; + opt.sync_method = DATA_DIR_SYNC_METHOD_FSYNC; + + /* process command-line options */ + while ((c = getopt_long(argc, argv, "do:nNPT:", + long_options, &optindex)) != -1) + { + switch (c) + { + case 'd': + opt.debug = true; + pg_logging_increase_verbosity(); + break; + case 'o': + opt.output = optarg; + break; + case 'n': + opt.dry_run = true; + break; + case 'N': + opt.no_sync = true; + break; + case 'T': + add_tablespace_mapping(&opt, optarg); + break; + case 1: + if (!pg_checksum_parse_type(optarg, + &opt.manifest_checksums)) + pg_fatal("unrecognized checksum algorithm: \"%s\"", + optarg); + break; + case 2: + opt.no_manifest = true; + break; + case 3: + if (!parse_sync_method(optarg, &opt.sync_method)) + exit(1); + break; + default: + /* getopt_long already emitted a complaint */ + pg_log_error_hint("Try \"%s --help\" for more information.", progname); + exit(1); + } + } + + if (optind >= argc) + { + pg_log_error("%s: no input directories specified", progname); + pg_log_error_hint("Try \"%s --help\" for more information.", progname); + exit(1); + } + + if (opt.output == NULL) + pg_fatal("no output directory specified"); + + /* If no manifest is needed, no checksums are needed, either. */ + if (opt.no_manifest) + opt.manifest_checksums = CHECKSUM_TYPE_NONE; + + /* Read the server version from the final backup. */ + version = read_pg_version_file(argv[argc - 1]); + + /* Sanity-check control files. */ + n_backups = argc - optind; + check_control_files(n_backups, argv + optind); + + /* Sanity-check backup_label files, and get the contents of the last one. */ + last_backup_label = check_backup_label_files(n_backups, argv + optind); + + /* Load backup manifests. */ + manifests = load_backup_manifests(n_backups, argv + optind); + + /* Figure out which tablespaces are going to be included in the output. */ + last_input_dir = argv[argc - 1]; + check_input_dir_permissions(last_input_dir); + tablespaces = scan_for_existing_tablespaces(last_input_dir, &opt); + + /* + * Create output directories. + * + * We create one output directory for the main data directory plus one for + * each non-in-place tablespace. create_output_directory() will arrange + * for those directories to be cleaned up on failure. In-place tablespaces + * aren't handled at this stage because they're located beneath the main + * output directory, and thus the cleanup of that directory will get rid + * of them. Plus, the pg_tblspc directory that needs to contain them + * doesn't exist yet. + */ + atexit(cleanup_directories_atexit); + create_output_directory(opt.output, &opt); + for (ts = tablespaces; ts != NULL; ts = ts->next) + if (!ts->in_place) + create_output_directory(ts->new_dir, &opt); + + /* If we need to write a backup_manifest, prepare to do so. */ + if (!opt.dry_run && !opt.no_manifest) + mwriter = create_manifest_writer(opt.output); + else + mwriter = NULL; + + /* Write backup label into output directory. */ + if (opt.dry_run) + pg_log_debug("would generate \"%s/backup_label\"", opt.output); + else + { + pg_log_debug("generating \"%s/backup_label\"", opt.output); + last_backup_label->cursor = 0; + write_backup_label(opt.output, last_backup_label, + opt.manifest_checksums, mwriter); + } + + /* + * We'll need the pathnames to the prior backups. By "prior" we mean all + * but the last one listed on the command line. + */ + n_prior_backups = argc - optind - 1; + prior_backup_dirs = argv + optind; + + /* Process everything that's not part of a user-defined tablespace. */ + pg_log_debug("processing backup directory \"%s\"", last_input_dir); + process_directory_recursively(InvalidOid, last_input_dir, opt.output, + NULL, n_prior_backups, prior_backup_dirs, + manifests, mwriter, &opt); + + /* Process user-defined tablespaces. */ + for (ts = tablespaces; ts != NULL; ts = ts->next) + { + pg_log_debug("processing tablespace directory \"%s\"", ts->old_dir); + + /* + * If it's a normal tablespace, we need to set up a symbolic link from + * pg_tblspc/${OID} to the target directory; if it's an in-place + * tablespace, we need to create a directory at pg_tblspc/${OID}. + */ + if (!ts->in_place) + { + char linkpath[MAXPGPATH]; + + snprintf(linkpath, MAXPGPATH, "%s/pg_tblspc/%u", opt.output, + ts->oid); + + if (opt.dry_run) + pg_log_debug("would create symbolic link from \"%s\" to \"%s\"", + linkpath, ts->new_dir); + else + { + pg_log_debug("creating symbolic link from \"%s\" to \"%s\"", + linkpath, ts->new_dir); + if (symlink(ts->new_dir, linkpath) != 0) + pg_fatal("could not create symbolic link from \"%s\" to \"%s\": %m", + linkpath, ts->new_dir); + } + } + else + { + if (opt.dry_run) + pg_log_debug("would create directory \"%s\"", ts->new_dir); + else + { + pg_log_debug("creating directory \"%s\"", ts->new_dir); + if (pg_mkdir_p(ts->new_dir, pg_dir_create_mode) == -1) + pg_fatal("could not create directory \"%s\": %m", + ts->new_dir); + } + } + + /* OK, now handle the directory contents. */ + process_directory_recursively(ts->oid, ts->old_dir, ts->new_dir, + NULL, n_prior_backups, prior_backup_dirs, + manifests, mwriter, &opt); + } + + /* Finalize the backup_manifest, if we're generating one. */ + if (mwriter != NULL) + finalize_manifest(mwriter, + manifests[n_prior_backups]->first_wal_range); + + /* fsync that output directory unless we've been told not to do so */ + if (!opt.no_sync) + { + if (opt.dry_run) + pg_log_debug("would recursively fsync \"%s\"", opt.output); + else + { + pg_log_debug("recursively fsyncing \"%s\"", opt.output); + sync_pgdata(opt.output, version * 10000, opt.sync_method); + } + } + + /* It's a success, so don't remove the output directories. */ + reset_directory_cleanup_list(); + exit(0); +} + +/* + * Process the option argument for the -T, --tablespace-mapping switch. + */ +static void +add_tablespace_mapping(cb_options *opt, char *arg) +{ + cb_tablespace_mapping *tsmap = pg_malloc0(sizeof(cb_tablespace_mapping)); + char *dst; + char *dst_ptr; + char *arg_ptr; + + /* + * Basically, we just want to copy everything before the equals sign to + * tsmap->old_dir and everything afterwards to tsmap->new_dir, but if + * there's more or less than one equals sign, that's an error, and if + * there's an equals sign preceded by a backslash, don't treat it as a + * field separator but instead copy a literal equals sign. + */ + dst_ptr = dst = tsmap->old_dir; + for (arg_ptr = arg; *arg_ptr != '\0'; arg_ptr++) + { + if (dst_ptr - dst >= MAXPGPATH) + pg_fatal("directory name too long"); + + if (*arg_ptr == '\\' && *(arg_ptr + 1) == '=') + ; /* skip backslash escaping = */ + else if (*arg_ptr == '=' && (arg_ptr == arg || *(arg_ptr - 1) != '\\')) + { + if (tsmap->new_dir[0] != '\0') + pg_fatal("multiple \"=\" signs in tablespace mapping"); + else + dst = dst_ptr = tsmap->new_dir; + } + else + *dst_ptr++ = *arg_ptr; + } + if (!tsmap->old_dir[0] || !tsmap->new_dir[0]) + pg_fatal("invalid tablespace mapping format \"%s\", must be \"OLDDIR=NEWDIR\"", arg); + + /* + * All tablespaces are created with absolute directories, so specifying a + * non-absolute path here would never match, possibly confusing users. + * + * In contrast to pg_basebackup, both the old and new directories are on + * the local machine, so the local machine's definition of an absolute + * path is the only relevant one. + */ + if (!is_absolute_path(tsmap->old_dir)) + pg_fatal("old directory is not an absolute path in tablespace mapping: %s", + tsmap->old_dir); + + if (!is_absolute_path(tsmap->new_dir)) + pg_fatal("old directory is not an absolute path in tablespace mapping: %s", + tsmap->new_dir); + + /* Canonicalize paths to avoid spurious failures when comparing. */ + canonicalize_path(tsmap->old_dir); + canonicalize_path(tsmap->new_dir); + + /* Add it to the list. */ + tsmap->next = opt->tsmappings; + opt->tsmappings = tsmap; +} + +/* + * Check that the backup_label files form a coherent backup chain, and return + * the contents of the backup_label file from the latest backup. + */ +static StringInfo +check_backup_label_files(int n_backups, char **backup_dirs) +{ + StringInfo buf = makeStringInfo(); + StringInfo lastbuf = buf; + int i; + TimeLineID check_tli = 0; + XLogRecPtr check_lsn = InvalidXLogRecPtr; + + /* Try to read each backup_label file in turn, last to first. */ + for (i = n_backups - 1; i >= 0; --i) + { + char pathbuf[MAXPGPATH]; + int fd; + TimeLineID start_tli; + TimeLineID previous_tli; + XLogRecPtr start_lsn; + XLogRecPtr previous_lsn; + + /* Open the backup_label file. */ + snprintf(pathbuf, MAXPGPATH, "%s/backup_label", backup_dirs[i]); + pg_log_debug("reading \"%s\"", pathbuf); + if ((fd = open(pathbuf, O_RDONLY, 0)) < 0) + pg_fatal("could not open file \"%s\": %m", pathbuf); + + /* + * Slurp the whole file into memory. + * + * The exact size limit that we impose here doesn't really matter -- + * most of what's supposed to be in the file is fixed size and quite + * short. However, the length of the backup_label is limited (at least + * by some parts of the code) to MAXGPATH, so include that value in + * the maximum length that we tolerate. + */ + slurp_file(fd, pathbuf, buf, 10000 + MAXPGPATH); + + /* Close the file. */ + if (close(fd) != 0) + pg_fatal("could not close \"%s\": %m", pathbuf); + + /* Parse the file contents. */ + parse_backup_label(pathbuf, buf, &start_tli, &start_lsn, + &previous_tli, &previous_lsn); + + /* + * Sanity checks. + * + * XXX. It's actually not required that start_lsn == check_lsn. It + * would be OK if start_lsn > check_lsn provided that start_lsn is + * less than or equal to the relevant switchpoint. But at the moment + * we don't have that information. + */ + if (i > 0 && previous_tli == 0) + pg_fatal("backup at \"%s\" is a full backup, but only the first backup should be a full backup", + backup_dirs[i]); + if (i == 0 && previous_tli != 0) + pg_fatal("backup at \"%s\" is an incremental backup, but the first backup should be a full backup", + backup_dirs[i]); + if (i < n_backups - 1 && start_tli != check_tli) + pg_fatal("backup at \"%s\" starts on timeline %u, but expected %u", + backup_dirs[i], start_tli, check_tli); + if (i < n_backups - 1 && start_lsn != check_lsn) + pg_fatal("backup at \"%s\" starts at LSN %X/%X, but expected %X/%X", + backup_dirs[i], + LSN_FORMAT_ARGS(start_lsn), + LSN_FORMAT_ARGS(check_lsn)); + check_tli = previous_tli; + check_lsn = previous_lsn; + + /* + * The last backup label in the chain needs to be saved for later use, + * while the others are only needed within this loop. + */ + if (lastbuf == buf) + buf = makeStringInfo(); + else + resetStringInfo(buf); + } + + /* Free memory that we don't need any more. */ + if (lastbuf != buf) + { + pfree(buf->data); + pfree(buf); + } + + /* + * Return the data from the first backup_info that we read (which is the + * backup_label from the last directory specified on the command line). + */ + return lastbuf; +} + +/* + * Sanity check control files. + */ +static void +check_control_files(int n_backups, char **backup_dirs) +{ + int i; + uint64 system_identifier; + + /* Try to read each control file in turn, last to first. */ + for (i = n_backups - 1; i >= 0; --i) + { + ControlFileData *control_file; + bool crc_ok; + + pg_log_debug("reading \"%s/global/pg_control\"", backup_dirs[i]); + control_file = get_controlfile(backup_dirs[i], &crc_ok); + + /* Control file contents not meaningful if CRC is bad. */ + if (!crc_ok) + pg_fatal("%s/global/pg_control: crc is incorrect", backup_dirs[i]); + + /* Can't interpret control file if not current version. */ + if (control_file->pg_control_version != PG_CONTROL_VERSION) + pg_fatal("%s/global/pg_control: unexpected control file version", + backup_dirs[i]); + + /* System identifiers should all match. */ + if (i == n_backups - 1) + system_identifier = control_file->system_identifier; + else if (system_identifier != control_file->system_identifier) + pg_fatal("%s/global/pg_control: expected system identifier %llu, but found %llu", + backup_dirs[i], (unsigned long long) system_identifier, + (unsigned long long) control_file->system_identifier); + + /* Release memory. */ + pfree(control_file); + } + + /* + * If debug output is enabled, make a note of the system identifier that + * we found in all of the relevant control files. + */ + pg_log_debug("system identifier is %llu", + (unsigned long long) system_identifier); +} + +/* + * Set default permissions for new files and directories based on the + * permissions of the given directory. The intent here is that the output + * directory should use the same permissions scheme as the final input + * directory. + */ +static void +check_input_dir_permissions(char *dir) +{ + struct stat st; + + if (stat(dir, &st) != 0) + pg_fatal("could not stat \"%s\": %m", dir); + + SetDataDirectoryCreatePerm(st.st_mode); +} + +/* + * Clean up output directories before exiting. + */ +static void +cleanup_directories_atexit(void) +{ + while (cleanup_dir_list != NULL) + { + cb_cleanup_dir *dir = cleanup_dir_list; + + if (dir->rmtopdir) + { + pg_log_info("removing output directory \"%s\"", dir->target_path); + if (!rmtree(dir->target_path, dir->rmtopdir)) + pg_log_error("failed to remove output directory"); + } + else + { + pg_log_info("removing contents of output directory \"%s\"", + dir->target_path); + if (!rmtree(dir->target_path, dir->rmtopdir)) + pg_log_error("failed to remove contents of output directory"); + } + + cleanup_dir_list = cleanup_dir_list->next; + pfree(dir); + } +} + +/* + * Create the named output directory, unless it already exists or we're in + * dry-run mode. If it already exists but is not empty, that's a fatal error. + * + * Adds the created directory to the list of directories to be cleaned up + * at process exit. + */ +static void +create_output_directory(char *dirname, cb_options *opt) +{ + switch (pg_check_dir(dirname)) + { + case 0: + if (opt->dry_run) + { + pg_log_debug("would create directory \"%s\"", dirname); + return; + } + pg_log_debug("creating directory \"%s\"", dirname); + if (pg_mkdir_p(dirname, pg_dir_create_mode) == -1) + pg_fatal("could not create directory \"%s\": %m", dirname); + remember_to_cleanup_directory(dirname, true); + break; + + case 1: + pg_log_debug("using existing directory \"%s\"", dirname); + remember_to_cleanup_directory(dirname, false); + break; + + case 2: + case 3: + case 4: + pg_fatal("directory \"%s\" exists but is not empty", dirname); + + case -1: + pg_fatal("could not access directory \"%s\": %m", dirname); + } +} + +/* + * help + * + * Prints help page for the program + * + * progname: the name of the executed program, such as "pg_combinebackup" + */ +static void +help(const char *progname) +{ + printf(_("%s combines incremental backups.\n\n"), progname); + printf(_("Usage:\n")); + printf(_(" %s [OPTION]... DIRECTORY...\n"), progname); + printf(_("\nOptions:\n")); + printf(_(" -d, --debug generate lots of debugging output\n")); + printf(_(" -n, --dry-run don't actually do anything\n")); + printf(_(" -N, --no-sync do not wait for changes to be written safely to disk\n")); + printf(_(" -o, --output output directory\n")); + printf(_(" -T, --tablespace-mapping=OLDDIR=NEWDIR\n")); + printf(_(" relocate tablespace in OLDDIR to NEWDIR\n")); + printf(_(" --manifest-checksums=SHA{224,256,384,512}|CRC32C|NONE\n" + " use algorithm for manifest checksums\n")); + printf(_(" --no-manifest suppress generation of backup manifest\n")); + printf(_(" --sync-method=METHOD set method for syncing files to disk\n")); + printf(_(" -?, --help show this help, then exit\n")); + + printf(_("\nReport bugs to <%s>.\n"), PACKAGE_BUGREPORT); + printf(_("%s home page: <%s>\n"), PACKAGE_NAME, PACKAGE_URL); +} + +/* + * Try to parse a string as a non-zero OID without leading zeroes. + * + * If it works, return true and set *result to the answer, else return false. + */ +static bool +parse_oid(char *s, Oid *result) +{ + Oid oid; + char *ep; + + errno = 0; + oid = strtoul(s, &ep, 10); + if (errno != 0 || *ep != '\0' || oid < 1 || oid > PG_UINT32_MAX) + return false; + + *result = oid; + return true; +} + +/* + * Copy files from the input directory to the output directory, reconstructing + * full files from incremental files as required. + * + * If processing is a user-defined tablespace, the tsoid should be the OID + * of that tablespace and input_directory and output_directory should be the + * toplevel input and output directories for that tablespace. Otherwise, + * tsoid should be InvalidOid and input_directory and output_directory should + * be the main input and output directories. + * + * relative_path is the path beneath the given input and output directories + * that we are currently processing. If NULL, it indicates that we're + * processing the input and output directories themselves. + * + * n_prior_backups is the number of prior backups that we have available. + * This doesn't count the very last backup, which is referenced by + * output_directory, just the older ones. prior_backup_dirs is an array of + * the locations of those previous backups. + */ +static void +process_directory_recursively(Oid tsoid, + char *input_directory, + char *output_directory, + char *relative_path, + int n_prior_backups, + char **prior_backup_dirs, + manifest_data **manifests, + manifest_writer *mwriter, + cb_options *opt) +{ + char ifulldir[MAXPGPATH]; + char ofulldir[MAXPGPATH]; + char manifest_prefix[MAXPGPATH]; + DIR *dir; + struct dirent *de; + bool is_pg_tblspc; + bool is_pg_wal; + manifest_data *latest_manifest = manifests[n_prior_backups]; + pg_checksum_type checksum_type; + + StaticAssertStmt(strlen(INCREMENTAL_PREFIX) == INCREMENTAL_PREFIX_LENGTH, + "INCREMENTAL_PREFIX_LENGTH is incorrect"); + + /* + * pg_tblspc and pg_wal are special cases, so detect those here. + * + * pg_tblspc is only special at the top level, but subdirectories of + * pg_wal are just as special as the top level directory. + * + * Since incremental backup does not exist in pre-v10 versions, we don't + * have to worry about the old pg_xlog naming. + */ + is_pg_tblspc = !OidIsValid(tsoid) && relative_path != NULL && + strcmp(relative_path, "pg_tblspc") == 0; + is_pg_wal = !OidIsValid(tsoid) && relative_path != NULL && + (strcmp(relative_path, "pg_wal") == 0 || + strncmp(relative_path, "pg_wal/", 7) == 0); + + /* + * If we're under pg_wal, then we don't need checksums, because these + * files aren't included in the backup manifest. Otherwise use whatever + * type of checksum is configured. + */ + if (!is_pg_wal) + checksum_type = opt->manifest_checksums; + else + checksum_type = CHECKSUM_TYPE_NONE; + + /* + * Append the relative path to the input and output directories, and + * figure out the appropriate prefix to add to files in this directory + * when looking them up in a backup manifest. + */ + if (relative_path == NULL) + { + strncpy(ifulldir, input_directory, MAXPGPATH); + strncpy(ofulldir, output_directory, MAXPGPATH); + if (OidIsValid(tsoid)) + snprintf(manifest_prefix, MAXPGPATH, "pg_tblspc/%u/", tsoid); + else + manifest_prefix[0] = '\0'; + } + else + { + snprintf(ifulldir, MAXPGPATH, "%s/%s", input_directory, + relative_path); + snprintf(ofulldir, MAXPGPATH, "%s/%s", output_directory, + relative_path); + if (OidIsValid(tsoid)) + snprintf(manifest_prefix, MAXPGPATH, "pg_tblspc/%u/%s/", + tsoid, relative_path); + else + snprintf(manifest_prefix, MAXPGPATH, "%s/", relative_path); + } + + /* + * Toplevel output directories have already been created by the time this + * function is called, but any subdirectories are our responsibility. + */ + if (relative_path != NULL) + { + if (opt->dry_run) + pg_log_debug("would create directory \"%s\"", ofulldir); + else + { + pg_log_debug("creating directory \"%s\"", ofulldir); + if (mkdir(ofulldir, pg_dir_create_mode) == -1) + pg_fatal("could not create directory \"%s\": %m", ofulldir); + } + } + + /* It's time to scan the directory. */ + if ((dir = opendir(ifulldir)) == NULL) + pg_fatal("could not open directory \"%s\": %m", ifulldir); + while (errno = 0, (de = readdir(dir)) != NULL) + { + PGFileType type; + char ifullpath[MAXPGPATH]; + char ofullpath[MAXPGPATH]; + char manifest_path[MAXPGPATH]; + Oid oid = InvalidOid; + int checksum_length = 0; + uint8 *checksum_payload = NULL; + pg_checksum_context checksum_ctx; + + /* Ignore "." and ".." entries. */ + if (strcmp(de->d_name, ".") == 0 || + strcmp(de->d_name, "..") == 0) + continue; + + /* Construct input path. */ + snprintf(ifullpath, MAXPGPATH, "%s/%s", ifulldir, de->d_name); + + /* Figure out what kind of directory entry this is. */ + type = get_dirent_type(ifullpath, de, false, PG_LOG_ERROR); + if (type == PGFILETYPE_ERROR) + exit(1); + + /* + * If we're processing pg_tblspc, then check whether the filename + * looks like it could be a tablespace OID. If so, and if the + * directory entry is a symbolic link or a directory, skip it. + * + * Our goal here is to ignore anything that would have been considered + * by scan_for_existing_tablespaces to be a tablespace. + */ + if (is_pg_tblspc && parse_oid(de->d_name, &oid) && + (type == PGFILETYPE_LNK || type == PGFILETYPE_DIR)) + continue; + + /* If it's a directory, recurse. */ + if (type == PGFILETYPE_DIR) + { + char new_relative_path[MAXPGPATH]; + + /* Append new pathname component to relative path. */ + if (relative_path == NULL) + strncpy(new_relative_path, de->d_name, MAXPGPATH); + else + snprintf(new_relative_path, MAXPGPATH, "%s/%s", relative_path, + de->d_name); + + /* And recurse. */ + process_directory_recursively(tsoid, + input_directory, output_directory, + new_relative_path, + n_prior_backups, prior_backup_dirs, + manifests, mwriter, opt); + continue; + } + + /* Skip anything that's not a regular file. */ + if (type != PGFILETYPE_REG) + { + if (type == PGFILETYPE_LNK) + pg_log_warning("skipping symbolic link \"%s\"", ifullpath); + else + pg_log_warning("skipping special file \"%s\"", ifullpath); + continue; + } + + /* + * Skip the backup_label and backup_manifest files; they require + * special handling and are handled elsewhere. + */ + if (relative_path == NULL && + (strcmp(de->d_name, "backup_label") == 0 || + strcmp(de->d_name, "backup_manifest") == 0)) + continue; + + /* + * If it's an incremental file, hand it off to the reconstruction + * code, which will figure out what to do. + */ + if (strncmp(de->d_name, INCREMENTAL_PREFIX, + INCREMENTAL_PREFIX_LENGTH) == 0) + { + /* Output path should not include "INCREMENTAL." prefix. */ + snprintf(ofullpath, MAXPGPATH, "%s/%s", ofulldir, + de->d_name + INCREMENTAL_PREFIX_LENGTH); + + + /* Manifest path likewise omits incremental prefix. */ + snprintf(manifest_path, MAXPGPATH, "%s%s", manifest_prefix, + de->d_name + INCREMENTAL_PREFIX_LENGTH); + + /* Reconstruction logic will do the rest. */ + reconstruct_from_incremental_file(ifullpath, ofullpath, + relative_path, + de->d_name + INCREMENTAL_PREFIX_LENGTH, + n_prior_backups, + prior_backup_dirs, + manifests, + manifest_path, + checksum_type, + &checksum_length, + &checksum_payload, + opt->dry_run); + } + else + { + /* Construct the path that the backup_manifest will use. */ + snprintf(manifest_path, MAXPGPATH, "%s%s", manifest_prefix, + de->d_name); + + /* + * It's not an incremental file, so we need to copy the entire + * file to the output directory. + * + * If a checksum of the required type already exists in the + * backup_manifest for the final input directory, we can save some + * work by reusing that checksum instead of computing a new one. + */ + if (checksum_type != CHECKSUM_TYPE_NONE && + latest_manifest != NULL) + { + manifest_file *mfile; + + mfile = manifest_files_lookup(latest_manifest->files, + manifest_path); + if (mfile == NULL) + { + /* + * The directory is out of sync with the backup_manifest, + * so emit a warning. + */ + pg_log_warning("\"%s/backup_manifest\" contains no entry for \"%s\"", + input_directory, manifest_path); + } + else if (mfile->checksum_type == checksum_type) + { + checksum_length = mfile->checksum_length; + checksum_payload = mfile->checksum_payload; + } + } + + /* + * If we're reusing a checksum, then we don't need copy_file() to + * compute one for us, but otherwise, it needs to compute whatever + * type of checksum we need. + */ + if (checksum_length != 0) + pg_checksum_init(&checksum_ctx, CHECKSUM_TYPE_NONE); + else + pg_checksum_init(&checksum_ctx, checksum_type); + + /* Actually copy the file. */ + snprintf(ofullpath, MAXPGPATH, "%s/%s", ofulldir, de->d_name); + copy_file(ifullpath, ofullpath, &checksum_ctx, opt->dry_run); + + /* + * If copy_file() performed a checksum calculation for us, then + * save the results (except in dry-run mode, when there's no + * point). + */ + if (checksum_ctx.type != CHECKSUM_TYPE_NONE && !opt->dry_run) + { + checksum_payload = pg_malloc(PG_CHECKSUM_MAX_LENGTH); + checksum_length = pg_checksum_final(&checksum_ctx, + checksum_payload); + } + } + + /* Generate manifest entry, if needed. */ + if (mwriter != NULL) + { + struct stat sb; + + /* + * In order to generate a manifest entry, we need the file size + * and mtime. We have no way to know the correct mtime except to + * stat() the file, so just do that and get the size as well. + * + * If we didn't need the mtime here, we could try to obtain the + * file size from the reconstruction or file copy process above, + * although that is actually not convenient in all cases. If we + * write the file ourselves then clearly we can keep a count of + * bytes, but if we use something like CopyFile() then it's + * trickier. Since we have to stat() anyway to get the mtime, + * there's no point in worrying about it. + */ + if (stat(ofullpath, &sb) < 0) + pg_fatal("could not stat file \"%s\": %m", ofullpath); + + /* OK, now do the work. */ + add_file_to_manifest(mwriter, manifest_path, + sb.st_size, sb.st_mtime, + checksum_type, checksum_length, + checksum_payload); + } + + /* Avoid leaking memory. */ + if (checksum_payload != NULL) + pfree(checksum_payload); + } + + closedir(dir); +} + +/* + * Read the version number from PG_VERSION and convert it to the usual server + * version number format. (e.g. If PG_VERSION contains "14\n" this function + * will return 140000) + */ +static int +read_pg_version_file(char *directory) +{ + char filename[MAXPGPATH]; + StringInfoData buf; + int fd; + int version; + char *ep; + + /* Construct pathname. */ + snprintf(filename, MAXPGPATH, "%s/PG_VERSION", directory); + + /* Open file. */ + if ((fd = open(filename, O_RDONLY, 0)) < 0) + pg_fatal("could not open file \"%s\": %m", filename); + + /* Read into memory. Length limit of 128 should be more than generous. */ + initStringInfo(&buf); + slurp_file(fd, filename, &buf, 128); + + /* Close the file. */ + if (close(fd) != 0) + pg_fatal("could not close \"%s\": %m", filename); + + /* Convert to integer. */ + errno = 0; + version = strtoul(buf.data, &ep, 10); + if (errno != 0 || *ep != '\n') + { + /* + * Incremental backup is not relevant to very old server versions that + * used multi-part version number (e.g. 9.6, or 8.4). So if we see + * what looks like the beginning of such a version number, just bail + * out. + */ + if (version < 10 && *ep == '.') + pg_fatal("%s: server version too old\n", filename); + pg_fatal("%s: could not parse version number\n", filename); + } + + /* Debugging output. */ + pg_log_debug("read server version %d from \"%s\"", version, filename); + + /* Release memory and return result. */ + pfree(buf.data); + return version * 10000; +} + +/* + * Add a directory to the list of output directories to clean up. + */ +static void +remember_to_cleanup_directory(char *target_path, bool rmtopdir) +{ + cb_cleanup_dir *dir = pg_malloc(sizeof(cb_cleanup_dir)); + + dir->target_path = target_path; + dir->rmtopdir = rmtopdir; + dir->next = cleanup_dir_list; + cleanup_dir_list = dir; +} + +/* + * Empty out the list of directories scheduled for cleanup a exit. + * + * We want to remove the output directories only on a failure, so call this + * function when we know that the operation has succeeded. + * + * Since we only expect this to be called when we're about to exit, we could + * just set cleanup_dir_list to NULL and be done with it, but we free the + * memory to be tidy. + */ +static void +reset_directory_cleanup_list(void) +{ + while (cleanup_dir_list != NULL) + { + cb_cleanup_dir *dir = cleanup_dir_list; + + cleanup_dir_list = cleanup_dir_list->next; + pfree(dir); + } +} + +/* + * Scan the pg_tblspc directory of the final input backup to get a canonical + * list of what tablespaces are part of the backup. + * + * 'pathname' should be the path to the toplevel backup directory for the + * final backup in the backup chain. + */ +static cb_tablespace * +scan_for_existing_tablespaces(char *pathname, cb_options *opt) +{ + char pg_tblspc[MAXPGPATH]; + DIR *dir; + struct dirent *de; + cb_tablespace *tslist = NULL; + + snprintf(pg_tblspc, MAXPGPATH, "%s/pg_tblspc", pathname); + pg_log_debug("scanning \"%s\"", pg_tblspc); + + if ((dir = opendir(pg_tblspc)) == NULL) + pg_fatal("could not open directory \"%s\": %m", pathname); + + while (errno = 0, (de = readdir(dir)) != NULL) + { + Oid oid; + char tblspcdir[MAXPGPATH]; + char link_target[MAXPGPATH]; + int link_length; + cb_tablespace *ts; + cb_tablespace *otherts; + PGFileType type; + + /* Silently ignore "." and ".." entries. */ + if (strcmp(de->d_name, ".") == 0 || strcmp(de->d_name, "..") == 0) + continue; + + /* Construct full pathname. */ + snprintf(tblspcdir, MAXPGPATH, "%s/%s", pg_tblspc, de->d_name); + + /* Ignore any file name that doesn't look like a proper OID. */ + if (!parse_oid(de->d_name, &oid)) + { + pg_log_debug("skipping \"%s\" because the filename is not a legal tablespace OID", + tblspcdir); + continue; + } + + /* Only symbolic links and directories are tablespaces. */ + type = get_dirent_type(tblspcdir, de, false, PG_LOG_ERROR); + if (type == PGFILETYPE_ERROR) + exit(1); + if (type != PGFILETYPE_LNK && type != PGFILETYPE_DIR) + { + pg_log_debug("skipping \"%s\" because it is neither a symbolic link nor a directory", + tblspcdir); + continue; + } + + /* Create a new tablespace object. */ + ts = pg_malloc0(sizeof(cb_tablespace)); + ts->oid = oid; + + /* + * If it's a link, it's not an in-place tablespace. Otherwise, it must + * be a directory, and thus an in-place tablespace. + */ + if (type == PGFILETYPE_LNK) + { + cb_tablespace_mapping *tsmap; + + /* Read the link target. */ + link_length = readlink(tblspcdir, link_target, sizeof(link_target)); + if (link_length < 0) + pg_fatal("could not read symbolic link \"%s\": %m", + tblspcdir); + if (link_length >= sizeof(link_target)) + pg_fatal("symbolic link \"%s\" is too long", tblspcdir); + link_target[link_length] = '\0'; + if (!is_absolute_path(link_target)) + pg_fatal("symbolic link \"%s\" is relative", tblspcdir); + + /* Caonicalize the link target. */ + canonicalize_path(link_target); + + /* + * Find the corresponding tablespace mapping and copy the relevant + * details into the new tablespace entry. + */ + for (tsmap = opt->tsmappings; tsmap != NULL; tsmap = tsmap->next) + { + if (strcmp(tsmap->old_dir, link_target) == 0) + { + strncpy(ts->old_dir, tsmap->old_dir, MAXPGPATH); + strncpy(ts->new_dir, tsmap->new_dir, MAXPGPATH); + ts->in_place = false; + break; + } + } + + /* Every non-in-place tablespace must be mapped. */ + if (tsmap == NULL) + pg_fatal("tablespace at \"%s\" has no tablespace mapping", + link_target); + } + else + { + /* + * For an in-place tablespace, there's no separate directory, so + * we just record the paths within the data directories. + */ + snprintf(ts->old_dir, MAXPGPATH, "%s/%s", pg_tblspc, de->d_name); + snprintf(ts->new_dir, MAXPGPATH, "%s/pg_tblpc/%s", opt->output, + de->d_name); + ts->in_place = true; + } + + /* Tablespaces should not share a directory. */ + for (otherts = tslist; otherts != NULL; otherts = otherts->next) + if (strcmp(ts->new_dir, otherts->new_dir) == 0) + pg_fatal("tablespaces with OIDs %u and %u both point at \"%s\"", + otherts->oid, oid, ts->new_dir); + + /* Add this tablespace to the list. */ + ts->next = tslist; + tslist = ts; + } + + return tslist; +} + +/* + * Read a file into a StringInfo. + * + * fd is used for the actual file I/O, filename for error reporting purposes. + * A file longer than maxlen is a fatal error. + */ +static void +slurp_file(int fd, char *filename, StringInfo buf, int maxlen) +{ + struct stat st; + ssize_t rb; + + /* Check file size, and complain if it's too large. */ + if (fstat(fd, &st) != 0) + pg_fatal("could not stat \"%s\": %m", filename); + if (st.st_size > maxlen) + pg_fatal("file \"%s\" is too large", filename); + + /* Make sure we have enough space. */ + enlargeStringInfo(buf, st.st_size); + + /* Read the data. */ + rb = read(fd, &buf->data[buf->len], st.st_size); + + /* + * We don't expect any concurrent changes, so we should read exactly the + * expected number of bytes. + */ + if (rb != st.st_size) + { + if (rb < 0) + pg_fatal("could not read file \"%s\": %m", filename); + else + pg_fatal("could not read file \"%s\": read only %d of %d bytes", + filename, (int) rb, (int) st.st_size); + } + + /* Adjust buffer length for new data and restore trailing-\0 invariant */ + buf->len += rb; + buf->data[buf->len] = '\0'; +} diff --git a/src/bin/pg_combinebackup/reconstruct.c b/src/bin/pg_combinebackup/reconstruct.c new file mode 100644 index 0000000000..c774bf1842 --- /dev/null +++ b/src/bin/pg_combinebackup/reconstruct.c @@ -0,0 +1,618 @@ +/*------------------------------------------------------------------------- + * + * reconstruct.c + * Reconstruct full file from incremental file and backup chain. + * + * Copyright (c) 2017-2023, PostgreSQL Global Development Group + * + * IDENTIFICATION + * src/bin/pg_combinebackup/reconstruct.c + * + *------------------------------------------------------------------------- + */ +#include "postgres_fe.h" + +#include <unistd.h> + +#include "backup/basebackup_incremental.h" +#include "common/logging.h" +#include "common/file_perm.h" +#include "copy_file.h" +#include "reconstruct.h" +#include "storage/block.h" + +/* + * An rfile stores the data that we need in order to be able to use some file + * on disk for reconstruction. For any given output file, we create one rfile + * per backup that we need to consult when we constructing that output file. + * + * If we find a full version of the file in the backup chain, then only + * filename and fd are initialized; the remaining fields are 0 or NULL. + * For an incremental file, header_length, num_blocks, relative_block_numbers, + * and truncation_block_length are also set. + * + * num_blocks_read and highest_offset_read always start out as 0. + */ +typedef struct rfile +{ + char *filename; + int fd; + size_t header_length; + unsigned num_blocks; + BlockNumber *relative_block_numbers; + unsigned truncation_block_length; + unsigned num_blocks_read; + off_t highest_offset_read; +} rfile; + +static void debug_reconstruction(int n_source, + rfile **sources, + bool dry_run); +static unsigned find_reconstructed_block_length(rfile *s); +static rfile *make_incremental_rfile(char *filename); +static rfile *make_rfile(char *filename, bool missing_ok); +static void write_reconstructed_file(char *input_filename, + char *output_filename, + unsigned block_length, + rfile **sourcemap, + off_t *offsetmap, + pg_checksum_context *checksum_ctx, + bool dry_run); +static void read_bytes(rfile *rf, void *buffer, unsigned length); + +/* + * Reconstruct a full file from an incremental file and a chain of prior + * backups. + * + * input_filename should be the path to the incremental file, and + * output_filename should be the path where the reconstructed file is to be + * written. + * + * relative_path should be the relative path to the directory containing this + * file. bare_file_name should be the name of the file within that directory, + * without "INCREMENTAL.". + * + * n_prior_backups is the number of prior backups, and prior_backup_dirs is + * an array of pathnames where those backups can be found. + */ +void +reconstruct_from_incremental_file(char *input_filename, + char *output_filename, + char *relative_path, + char *bare_file_name, + int n_prior_backups, + char **prior_backup_dirs, + manifest_data **manifests, + char *manifest_path, + pg_checksum_type checksum_type, + int *checksum_length, + uint8 **checksum_payload, + bool dry_run) +{ + rfile **source; + rfile *latest_source = NULL; + rfile **sourcemap; + off_t *offsetmap; + unsigned block_length; + unsigned num_missing_blocks; + unsigned i; + unsigned sidx = n_prior_backups; + bool full_copy_possible = true; + int copy_source_index = -1; + rfile *copy_source = NULL; + pg_checksum_context checksum_ctx; + + /* + * Every block must come either from the latest version of the file or + * from one of the prior backups. + */ + source = pg_malloc0(sizeof(rfile *) * (1 + n_prior_backups)); + + /* + * Use the information from the latest incremental file to figure out how + * long the reconstructed file should be. + */ + latest_source = make_incremental_rfile(input_filename); + source[n_prior_backups] = latest_source; + block_length = find_reconstructed_block_length(latest_source); + + /* + * For each block in the output file, we need to know from which file we + * need to obtain it and at what offset in that file it's stored. + * sourcemap gives us the first of these things, and offsetmap the latter. + */ + sourcemap = pg_malloc0(sizeof(rfile *) * block_length); + offsetmap = pg_malloc0(sizeof(off_t) * block_length); + + /* + * Blocks prior to the truncation_block_length threshold must be obtained + * from some prior backup, while those after that threshold are left as + * zeroes if not present in the newest incremental file. + * num_missing_blocks counts the number of blocks that we must be found + * somewhere in the backup chain, and is thus initially equal to + * truncation_block_length. + */ + num_missing_blocks = latest_source->truncation_block_length; + + /* + * Every block that is present in the newest incremental file should be + * sourced from that file. If it precedes the truncation_block_length, + * it's a block that we would otherwise have had to find in an older + * backup and thus reduces the number of blocks remaining to be found by + * one; otherwise, it's an extra block that needs to be included in the + * output but would not have needed to be found in an older backup if it + * had not been present. + */ + for (i = 0; i < latest_source->num_blocks; ++i) + { + BlockNumber b = latest_source->relative_block_numbers[i]; + + Assert(b < block_length); + sourcemap[b] = latest_source; + offsetmap[b] = latest_source->header_length + (i * BLCKSZ); + if (b < latest_source->truncation_block_length) + num_missing_blocks--; + + /* + * A full copy of a file from an earlier backup is only possible if no + * blocks are needed from any later incremental file. + */ + full_copy_possible = false; + } + + while (num_missing_blocks > 0) + { + char source_filename[MAXPGPATH]; + rfile *s; + + /* + * Move to the next backup in the chain. If there are no more, then + * something has gone wrong and reconstruction has failed. + */ + if (sidx == 0) + pg_fatal("reconstruction for file \"%s\" failed to find %u required blocks", + output_filename, num_missing_blocks); + --sidx; + + /* + * Look for the full file in the previous backup. If not found, then + * look for an incremental file instead. + */ + snprintf(source_filename, MAXPGPATH, "%s/%s/%s", + prior_backup_dirs[sidx], relative_path, bare_file_name); + if ((s = make_rfile(source_filename, true)) == NULL) + { + snprintf(source_filename, MAXPGPATH, "%s/%s/INCREMENTAL.%s", + prior_backup_dirs[sidx], relative_path, bare_file_name); + s = make_incremental_rfile(source_filename); + } + source[sidx] = s; + + /* + * If s->header_length == 0, then this is a full file; otherwise, it's + * an incremental file. + */ + if (s->header_length != 0) + { + /* + * Since we found another incremental file, source all blocks from + * it that we need but don't yet have. + */ + for (i = 0; i < s->num_blocks; ++i) + { + BlockNumber b = s->relative_block_numbers[i]; + + if (b < latest_source->truncation_block_length && + sourcemap[b] == NULL) + { + sourcemap[b] = s; + offsetmap[b] = s->header_length + (i * BLCKSZ); + + Assert(num_missing_blocks > 0); + --num_missing_blocks; + + /* + * A full copy of a file from an earlier backup is only + * possible if no blocks are needed from any later + * incremental file. + */ + full_copy_possible = false; + } + } + } + else + { + BlockNumber b; + + /* + * Since we found a full file, source all remaining required + * blocks from it. + */ + for (b = 0; b < latest_source->truncation_block_length; ++b) + { + if (sourcemap[b] == NULL) + { + sourcemap[b] = s; + offsetmap[b] = b * BLCKSZ; + + Assert(num_missing_blocks > 0); + --num_missing_blocks; + } + } + Assert(num_missing_blocks == 0); + + /* + * If a full copy looks possible, check whether the resulting file + * should be exactly as long as the source file is. If so, a full + * copy is acceptable, otherwise not. + */ + if (full_copy_possible) + { + struct stat sb; + uint64 expected_length; + + if (fstat(s->fd, &sb) < 0) + pg_fatal("could not stat \"%s\": %m", s->filename); + expected_length = + (uint64) latest_source->truncation_block_length; + expected_length *= BLCKSZ; + if (expected_length == sb.st_size) + { + copy_source = s; + copy_source_index = sidx; + } + } + } + } + + /* + * If a checksum of the required type already exists in the + * backup_manifest for the relevant input directory, we can save some work + * by reusing that checksum instead of computing a new one. + */ + if (copy_source_index >= 0 && manifests[copy_source_index] != NULL && + checksum_type != CHECKSUM_TYPE_NONE) + { + manifest_file *mfile; + + mfile = manifest_files_lookup(manifests[copy_source_index]->files, + manifest_path); + if (mfile == NULL) + { + /* + * The directory is out of sync with the backup_manifest, so emit + * a warning. + */ + pg_log_warning("\"%s/backup_manifest\" contains no entry for \"%s\"", + prior_backup_dirs[copy_source_index], + manifest_path); + } + else if (mfile->checksum_type == checksum_type) + { + *checksum_length = mfile->checksum_length; + *checksum_payload = pg_malloc(*checksum_length); + memcpy(*checksum_payload, mfile->checksum_payload, + *checksum_length); + checksum_type = CHECKSUM_TYPE_NONE; + } + } + + /* Prepare for checksum calculation, if required. */ + pg_checksum_init(&checksum_ctx, checksum_type); + + /* + * If the full file can be created by copying a file from an older backup + * in the chain without needing to overwrite any blocks or truncate the + * result, then forget about performing reconstruction and just copy that + * file in its entirety. + * + * Otherwise, reconstruct. + */ + if (copy_source != NULL) + copy_file(copy_source->filename, output_filename, + &checksum_ctx, dry_run); + else + { + write_reconstructed_file(input_filename, output_filename, + block_length, sourcemap, offsetmap, + &checksum_ctx, dry_run); + debug_reconstruction(n_prior_backups + 1, source, dry_run); + } + + /* Save results of checksum calculation. */ + if (checksum_type != CHECKSUM_TYPE_NONE) + { + *checksum_payload = pg_malloc(PG_CHECKSUM_MAX_LENGTH); + *checksum_length = pg_checksum_final(&checksum_ctx, + *checksum_payload); + } + + /* + * Close files and release memory. + */ + for (i = 0; i <= n_prior_backups; ++i) + { + rfile *s = source[i]; + + if (s == NULL) + continue; + if (close(s->fd) != 0) + pg_fatal("could not close \"%s\": %m", s->filename); + if (s->relative_block_numbers != NULL) + pfree(s->relative_block_numbers); + pg_free(s->filename); + } + pfree(sourcemap); + pfree(offsetmap); + pfree(source); +} + +/* + * Perform post-reconstruction logging and sanity checks. + */ +static void +debug_reconstruction(int n_source, rfile **sources, bool dry_run) +{ + unsigned i; + + for (i = 0; i < n_source; ++i) + { + rfile *s = sources[i]; + + /* Ignore source if not used. */ + if (s == NULL) + continue; + + /* If no data is needed from this file, we can ignore it. */ + if (s->num_blocks_read == 0) + continue; + + /* Debug logging. */ + if (dry_run) + pg_log_debug("would have read %u blocks from \"%s\"", + s->num_blocks_read, s->filename); + else + pg_log_debug("read %u blocks from \"%s\"", + s->num_blocks_read, s->filename); + + /* + * In dry-run mode, we don't actually try to read data from the file, + * but we do try to verify that the file is long enough that we could + * have read the data if we'd tried. + * + * If this fails, then it means that a non-dry-run attempt would fail, + * complaining of not being able to read the required bytes from the + * file. + */ + if (dry_run) + { + struct stat sb; + + if (fstat(s->fd, &sb) < 0) + pg_fatal("could not stat \"%s\": %m", s->filename); + if (sb.st_size < s->highest_offset_read) + pg_fatal("file \"%s\" is too short: expected %llu, found %llu", + s->filename, + (unsigned long long) s->highest_offset_read, + (unsigned long long) sb.st_size); + } + } +} + +/* + * When we perform reconstruction using an incremental file, the output file + * should be at least as long as the truncation_block_length. Any blocks + * present in the incremental file increase the output length as far as is + * necessary to include those blocks. + */ +static unsigned +find_reconstructed_block_length(rfile *s) +{ + unsigned block_length = s->truncation_block_length; + unsigned i; + + for (i = 0; i < s->num_blocks; ++i) + if (s->relative_block_numbers[i] >= block_length) + block_length = s->relative_block_numbers[i] + 1; + + return block_length; +} + +/* + * Initialize an incremental rfile, reading the header so that we know which + * blocks it contains. + */ +static rfile * +make_incremental_rfile(char *filename) +{ + rfile *rf; + unsigned magic; + + rf = make_rfile(filename, false); + + /* Read and validate magic number. */ + read_bytes(rf, &magic, sizeof(magic)); + if (magic != INCREMENTAL_MAGIC) + pg_fatal("file \"%s\" has bad incremental magic number (0x%x not 0x%x)", + filename, magic, INCREMENTAL_MAGIC); + + /* Read block count. */ + read_bytes(rf, &rf->num_blocks, sizeof(rf->num_blocks)); + if (rf->num_blocks > RELSEG_SIZE) + pg_fatal("file \"%s\" has block count %u in excess of segment size %u", + filename, rf->num_blocks, RELSEG_SIZE); + + /* Read truncation block length. */ + read_bytes(rf, &rf->truncation_block_length, + sizeof(rf->truncation_block_length)); + if (rf->truncation_block_length > RELSEG_SIZE) + pg_fatal("file \"%s\" has truncation block length %u in excess of segment size %u", + filename, rf->truncation_block_length, RELSEG_SIZE); + + /* Read block numbers if there are any. */ + if (rf->num_blocks > 0) + { + rf->relative_block_numbers = + pg_malloc0(sizeof(BlockNumber) * rf->num_blocks); + read_bytes(rf, rf->relative_block_numbers, + sizeof(BlockNumber) * rf->num_blocks); + } + + /* Remember length of header. */ + rf->header_length = sizeof(magic) + sizeof(rf->num_blocks) + + sizeof(rf->truncation_block_length) + + sizeof(BlockNumber) * rf->num_blocks; + + return rf; +} + +/* + * Allocate and perform basic initialization of an rfile. + */ +static rfile * +make_rfile(char *filename, bool missing_ok) +{ + rfile *rf; + + rf = pg_malloc0(sizeof(rfile)); + rf->filename = pstrdup(filename); + if ((rf->fd = open(filename, O_RDONLY | PG_BINARY, 0)) < 0) + { + if (missing_ok && errno == ENOENT) + { + pg_free(rf); + return NULL; + } + pg_fatal("could not open file \"%s\": %m", filename); + } + + return rf; +} + +/* + * Read the indicated number of bytes from an rfile into the buffer. + */ +static void +read_bytes(rfile *rf, void *buffer, unsigned length) +{ + unsigned rb = read(rf->fd, buffer, length); + + if (rb != length) + { + if (rb < 0) + pg_fatal("could not read file \"%s\": %m", rf->filename); + else + pg_fatal("could not read file \"%s\": read only %d of %d bytes", + rf->filename, (int) rb, length); + } +} + +/* + * Write out a reconstructed file. + */ +static void +write_reconstructed_file(char *input_filename, + char *output_filename, + unsigned block_length, + rfile **sourcemap, + off_t *offsetmap, + pg_checksum_context *checksum_ctx, + bool dry_run) +{ + int wfd = -1; + unsigned i; + unsigned zero_blocks = 0; + + /* Debugging output. */ + if (dry_run) + pg_log_debug("would reconstruct \"%s\" (%u blocks, checksum %s)", + output_filename, block_length, + pg_checksum_type_name(checksum_ctx->type)); + else + pg_log_debug("reconstructing \"%s\" (%u blocks, checksum %s)", + output_filename, block_length, + pg_checksum_type_name(checksum_ctx->type)); + + /* Open the output file, except in dry_run mode. */ + if (!dry_run && + (wfd = open(output_filename, + O_RDWR | PG_BINARY | O_CREAT | O_EXCL, + pg_file_create_mode)) < 0) + pg_fatal("could not open file \"%s\": %m", output_filename); + + /* Read and write the blocks as required. */ + for (i = 0; i < block_length; ++i) + { + uint8 buffer[BLCKSZ]; + rfile *s = sourcemap[i]; + unsigned wb; + + /* Update accounting information. */ + if (s == NULL) + ++zero_blocks; + else + { + s->num_blocks_read++; + s->highest_offset_read = Max(s->highest_offset_read, + offsetmap[i] + BLCKSZ); + } + + /* Skip the rest of this in dry-run mode. */ + if (dry_run) + continue; + + /* Read or zero-fill the block as appropriate. */ + if (s == NULL) + { + /* + * New block not mentioned in the WAL summary. Should have been an + * uninitialized block, so just zero-fill it. + */ + memset(buffer, 0, BLCKSZ); + } + else + { + unsigned rb; + + /* Read the block from the correct source, except if dry-run. */ + rb = pg_pread(s->fd, buffer, BLCKSZ, offsetmap[i]); + if (rb != BLCKSZ) + { + if (rb < 0) + pg_fatal("could not read file \"%s\": %m", s->filename); + else + pg_fatal("could not read file \"%s\": read only %d of %d bytes at offset %u", + s->filename, (int) rb, BLCKSZ, + (unsigned) offsetmap[i]); + } + } + + /* Write out the block. */ + if ((wb = write(wfd, buffer, BLCKSZ)) != BLCKSZ) + { + if (wb < 0) + pg_fatal("could not write file \"%s\": %m", output_filename); + else + pg_fatal("could not write file \"%s\": wrote only %d of %d bytes", + output_filename, (int) wb, BLCKSZ); + } + + /* Update the checksum computation. */ + if (pg_checksum_update(checksum_ctx, buffer, BLCKSZ) < 0) + pg_fatal("could not update checksum of file \"%s\"", + output_filename); + } + + /* Debugging output. */ + if (zero_blocks > 0) + { + if (dry_run) + pg_log_debug("would have zero-filled %u blocks", zero_blocks); + else + pg_log_debug("zero-filled %u blocks", zero_blocks); + } + + /* Close the output file. */ + if (wfd >= 0 && close(wfd) != 0) + pg_fatal("could not close \"%s\": %m", output_filename); +} diff --git a/src/bin/pg_combinebackup/reconstruct.h b/src/bin/pg_combinebackup/reconstruct.h new file mode 100644 index 0000000000..c599a70d42 --- /dev/null +++ b/src/bin/pg_combinebackup/reconstruct.h @@ -0,0 +1,32 @@ +/*------------------------------------------------------------------------- + * + * reconstruct.h + * Reconstruct full file from incremental file and backup chain. + * + * Copyright (c) 2017-2023, PostgreSQL Global Development Group + * + * IDENTIFICATION + * src/bin/pg_combinebackup/reconstruct.h + * + *------------------------------------------------------------------------- + */ +#ifndef RECONSTRUCT_H +#define RECONSTRUCT_H + +#include "common/checksum_helper.h" +#include "load_manifest.h" + +extern void reconstruct_from_incremental_file(char *input_filename, + char *output_filename, + char *relative_path, + char *bare_file_name, + int n_prior_backups, + char **prior_backup_dirs, + manifest_data **manifests, + char *manifest_path, + pg_checksum_type checksum_type, + int *checksum_length, + uint8 **checksum_payload, + bool dry_run); + +#endif diff --git a/src/bin/pg_combinebackup/t/001_basic.pl b/src/bin/pg_combinebackup/t/001_basic.pl new file mode 100644 index 0000000000..fb66075d1a --- /dev/null +++ b/src/bin/pg_combinebackup/t/001_basic.pl @@ -0,0 +1,23 @@ +# Copyright (c) 2021-2023, PostgreSQL Global Development Group + +use strict; +use warnings; +use PostgreSQL::Test::Utils; +use Test::More; + +my $tempdir = PostgreSQL::Test::Utils::tempdir; + +program_help_ok('pg_combinebackup'); +program_version_ok('pg_combinebackup'); +program_options_handling_ok('pg_combinebackup'); + +command_fails_like( + ['pg_combinebackup'], + qr/no input directories specified/, + 'input directories must be specified'); +command_fails_like( + [ 'pg_combinebackup', $tempdir ], + qr/no output directory specified/, + 'output directory must be specified'); + +done_testing(); diff --git a/src/bin/pg_combinebackup/t/002_compare_backups.pl b/src/bin/pg_combinebackup/t/002_compare_backups.pl new file mode 100644 index 0000000000..3d9238f366 --- /dev/null +++ b/src/bin/pg_combinebackup/t/002_compare_backups.pl @@ -0,0 +1,154 @@ +# Copyright (c) 2021-2023, PostgreSQL Global Development Group + +use strict; +use warnings; +use File::Compare; +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +# Set up a new database instance. +my $primary = PostgreSQL::Test::Cluster->new('primary'); +$primary->init(has_archiving => 1, allows_streaming => 1); +$primary->append_conf('postgresql.conf', 'autovacuum = off'); +$primary->start; + +# Create some test tables, each containing one row of data, plus a whole +# extra database. +$primary->safe_psql('postgres', <<EOM); +CREATE TABLE will_change (a int, b text); +INSERT INTO will_change VALUES (1, 'initial test row'); +CREATE TABLE will_grow (a int, b text); +INSERT INTO will_grow VALUES (1, 'initial test row'); +CREATE TABLE will_shrink (a int, b text); +INSERT INTO will_shrink VALUES (1, 'initial test row'); +CREATE TABLE will_get_vacuumed (a int, b text); +INSERT INTO will_get_vacuumed VALUES (1, 'initial test row'); +CREATE TABLE will_get_dropped (a int, b text); +INSERT INTO will_get_dropped VALUES (1, 'initial test row'); +CREATE TABLE will_get_rewritten (a int, b text); +INSERT INTO will_get_rewritten VALUES (1, 'initial test row'); +CREATE DATABASE db_will_get_dropped; +EOM + +# Take a full backup. +my $backup1path = $primary->backup_dir . '/backup1'; +$primary->command_ok( + [ 'pg_basebackup', '-D', $backup1path, '--no-sync', '-cfast' ], + "full backup"); + +# Now make some database changes. +$primary->safe_psql('postgres', <<EOM); +UPDATE will_change SET b = 'modified value' WHERE a = 1; +INSERT INTO will_grow + SELECT g, 'additional row' FROM generate_series(2, 5000) g; +TRUNCATE will_shrink; +VACUUM will_get_vacuumed; +DROP TABLE will_get_dropped; +CREATE TABLE newly_created (a int, b text); +INSERT INTO newly_created VALUES (1, 'row for new table'); +VACUUM FULL will_get_rewritten; +DROP DATABASE db_will_get_dropped; +CREATE DATABASE db_newly_created; +EOM + +# Take an incremental backup. +my $backup2path = $primary->backup_dir . '/backup2'; +$primary->command_ok( + [ 'pg_basebackup', '-D', $backup2path, '--no-sync', '-cfast', + '--incremental', $backup1path . '/backup_manifest' ], + "incremental backup"); + +# Find an LSN to which either backup can be recovered. +my $lsn = $primary->safe_psql('postgres', "SELECT pg_current_wal_lsn();"); + +# Make sure that the WAL segment containing that LSN has been archived. +# PostgreSQL won't issue two consecutive XLOG_SWITCH records, and the backup +# just issued one, so call txid_current() to generate some WAL activity +# before calling pg_switch_wal(). +$primary->safe_psql('postgres', 'SELECT txid_current();'); +$primary->safe_psql('postgres', 'SELECT pg_switch_wal()'); + +# Now wait for the LSN we chose above to be archived. +my $archive_wait_query = + "SELECT pg_walfile_name('$lsn') <= last_archived_wal FROM pg_stat_archiver;"; +$primary->poll_query_until('postgres', $archive_wait_query) + or die "Timed out while waiting for WAL segment to be archived"; + +# Perform PITR from the full backup. Disable archive_mode so that the archive +# doesn't find out about the new timeline; that way, the later PITR below will +# choose the same timeline. +my $pitr1 = PostgreSQL::Test::Cluster->new('pitr1'); +$pitr1->init_from_backup($primary, 'backup1', + standby => 1, has_restoring => 1); +$pitr1->append_conf('postgresql.conf', qq{ +recovery_target_lsn = '$lsn' +recovery_target_action = 'promote' +archive_mode = 'off' +}); +$pitr1->start(); + +# Perform PITR to the same LSN from the incremental backup. Use the same +# basic configuration as before. +my $pitr2 = PostgreSQL::Test::Cluster->new('pitr2'); +$pitr2->init_from_backup($primary, 'backup2', + standby => 1, has_restoring => 1, + combine_with_prior => [ 'backup1' ]); +$pitr2->append_conf('postgresql.conf', qq{ +recovery_target_lsn = '$lsn' +recovery_target_action = 'promote' +archive_mode = 'off' +}); +$pitr2->start(); + +# Wait until both servers exit recovery. +$pitr1->poll_query_until('postgres', + "SELECT NOT pg_is_in_recovery();") + or die "Timed out while waiting apply to reach LSN $lsn"; +$pitr2->poll_query_until('postgres', + "SELECT NOT pg_is_in_recovery();") + or die "Timed out while waiting apply to reach LSN $lsn"; + +# Perform a logical dump of each server, and check that they match. +# It would be much nicer if we could physically compare the data files, but +# that doesn't really work. The contents of the page hole aren't guaranteed to +# be identical, and there can be other discrepancies as well. To make this work +# we'd need the equivalent of each AM's rm_mask functon written or at least +# callable from Perl, and that doesn't seem practical. +# +# NB: We're just using the primary's backup directory for scratch space here. +# This could equally well be any other directory we wanted to pick. +my $backupdir = $primary->backup_dir; +my $dump1 = $backupdir . '/pitr1.dump'; +my $dump2 = $backupdir . '/pitr2.dump'; +$pitr1->command_ok([ + 'pg_dumpall', '-f', $dump1, '--no-sync', '--no-unlogged-table-data', + '-d', $pitr1->connstr('postgres'), + ], + 'dump from PITR 1'); +$pitr1->command_ok([ + 'pg_dumpall', '-f', $dump2, '--no-sync', '--no-unlogged-table-data', + '-d', $pitr1->connstr('postgres'), + ], + 'dump from PITR 2'); + +# Compare the two dumps, there should be no differences. +my $compare_res = compare($dump1, $dump2); +note($dump1); +note($dump2); +is($compare_res, 0, "dumps are identical"); + +# Provide more context if the dumps do not match. +if ($compare_res != 0) +{ + my ($stdout, $stderr) = + run_command([ 'diff', '-u', $dump1, $dump2 ]); + print "=== diff of $dump1 and $dump2\n"; + print "=== stdout ===\n"; + print $stdout; + print "=== stderr ===\n"; + print $stderr; + print "=== EOF ===\n"; +} + +done_testing(); diff --git a/src/bin/pg_combinebackup/write_manifest.c b/src/bin/pg_combinebackup/write_manifest.c new file mode 100644 index 0000000000..82160134d8 --- /dev/null +++ b/src/bin/pg_combinebackup/write_manifest.c @@ -0,0 +1,293 @@ +/*------------------------------------------------------------------------- + * + * Write a new backup manifest. + * + * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group + * Portions Copyright (c) 1994, Regents of the University of California + * + * src/bin/pg_combinebackup/write_manifest.c + * + *------------------------------------------------------------------------- + */ + +#include "postgres_fe.h" + +#include <fcntl.h> +#include <time.h> +#include <unistd.h> + +#include "common/checksum_helper.h" +#include "common/file_perm.h" +#include "common/logging.h" +#include "lib/stringinfo.h" +#include "load_manifest.h" +#include "mb/pg_wchar.h" +#include "write_manifest.h" + +struct manifest_writer +{ + char pathname[MAXPGPATH]; + int fd; + StringInfoData buf; + bool first_file; + bool still_checksumming; + pg_checksum_context manifest_ctx; +}; + +static void escape_json(StringInfo buf, const char *str); +static void flush_manifest(manifest_writer *mwriter); +static size_t hex_encode(const uint8 *src, size_t len, char *dst); + +/* + * Create a new backup manifest writer. + * + * The backup manifest will be written into a file named backup_manifest + * in the specified directory. + */ +manifest_writer * +create_manifest_writer(char *directory) +{ + manifest_writer *mwriter = pg_malloc(sizeof(manifest_writer)); + + snprintf(mwriter->pathname, MAXPGPATH, "%s/backup_manifest", directory); + mwriter->fd = -1; + initStringInfo(&mwriter->buf); + mwriter->first_file = true; + mwriter->still_checksumming = true; + pg_checksum_init(&mwriter->manifest_ctx, CHECKSUM_TYPE_SHA256); + + appendStringInfo(&mwriter->buf, + "{ \"PostgreSQL-Backup-Manifest-Version\": 1,\n" + "\"Files\": ["); + + return mwriter; +} + +/* + * Add an entry for a file to a backup manifest. + * + * This is very similar to the backend's AddFileToBackupManifest, but + * various adjustments are required due to frontend/backend differences + * and other details. + */ +void +add_file_to_manifest(manifest_writer *mwriter, const char *manifest_path, + size_t size, pg_time_t mtime, + pg_checksum_type checksum_type, + int checksum_length, + uint8 *checksum_payload) +{ + int pathlen = strlen(manifest_path); + + if (mwriter->first_file) + { + appendStringInfoChar(&mwriter->buf, '\n'); + mwriter->first_file = false; + } + else + appendStringInfoString(&mwriter->buf, ",\n"); + + if (pg_encoding_verifymbstr(PG_UTF8, manifest_path, pathlen) == pathlen) + { + appendStringInfoString(&mwriter->buf, "{ \"Path\": "); + escape_json(&mwriter->buf, manifest_path); + appendStringInfoString(&mwriter->buf, ", "); + } + else + { + appendStringInfoString(&mwriter->buf, "{ \"Encoded-Path\": \""); + enlargeStringInfo(&mwriter->buf, 2 * pathlen); + mwriter->buf.len += hex_encode((const uint8 *) manifest_path, pathlen, + &mwriter->buf.data[mwriter->buf.len]); + appendStringInfoString(&mwriter->buf, "\", "); + } + + appendStringInfo(&mwriter->buf, "\"Size\": %zu, ", size); + + appendStringInfoString(&mwriter->buf, "\"Last-Modified\": \""); + enlargeStringInfo(&mwriter->buf, 128); + mwriter->buf.len += strftime(&mwriter->buf.data[mwriter->buf.len], 128, + "%Y-%m-%d %H:%M:%S %Z", + gmtime(&mtime)); + appendStringInfoChar(&mwriter->buf, '"'); + + if (mwriter->buf.len > 128 * 1024) + flush_manifest(mwriter); + + if (checksum_length > 0) + { + appendStringInfo(&mwriter->buf, + ", \"Checksum-Algorithm\": \"%s\", \"Checksum\": \"", + pg_checksum_type_name(checksum_type)); + + enlargeStringInfo(&mwriter->buf, 2 * checksum_length); + mwriter->buf.len += hex_encode(checksum_payload, checksum_length, + &mwriter->buf.data[mwriter->buf.len]); + + appendStringInfoChar(&mwriter->buf, '"'); + } + + appendStringInfoString(&mwriter->buf, " }"); + + if (mwriter->buf.len > 128 * 1024) + flush_manifest(mwriter); +} + +/* + * Finalize the backup_manifest. + */ +void +finalize_manifest(manifest_writer *mwriter, + manifest_wal_range *first_wal_range) +{ + uint8 checksumbuf[PG_SHA256_DIGEST_LENGTH]; + int len; + manifest_wal_range *wal_range; + + /* Terminate the list of files. */ + appendStringInfoString(&mwriter->buf, "\n],\n"); + + /* Start a list of LSN ranges. */ + appendStringInfoString(&mwriter->buf, "\"WAL-Ranges\": [\n"); + + for (wal_range = first_wal_range; wal_range != NULL; + wal_range = wal_range->next) + appendStringInfo(&mwriter->buf, + "%s{ \"Timeline\": %u, \"Start-LSN\": \"%X/%X\", \"End-LSN\": \"%X/%X\" }", + wal_range == first_wal_range ? "" : ",\n", + wal_range->tli, + LSN_FORMAT_ARGS(wal_range->start_lsn), + LSN_FORMAT_ARGS(wal_range->end_lsn)); + + /* Terminate the list of WAL ranges. */ + appendStringInfoString(&mwriter->buf, "\n],\n"); + + /* Flush accumulated data and update checksum calculation. */ + flush_manifest(mwriter); + + /* Checksum only includes data up to this point. */ + mwriter->still_checksumming = false; + + /* Compute and insert manifest checksum. */ + appendStringInfoString(&mwriter->buf, "\"Manifest-Checksum\": \""); + enlargeStringInfo(&mwriter->buf, 2 * PG_SHA256_DIGEST_STRING_LENGTH); + len = pg_checksum_final(&mwriter->manifest_ctx, checksumbuf); + Assert(len == PG_SHA256_DIGEST_LENGTH); + mwriter->buf.len += + hex_encode(checksumbuf, len, &mwriter->buf.data[mwriter->buf.len]); + appendStringInfoString(&mwriter->buf, "\"}\n"); + + /* Flush the last manifest checksum itself. */ + flush_manifest(mwriter); + + /* Close the file. */ + if (close(mwriter->fd) != 0) + pg_fatal("could not close \"%s\": %m", mwriter->pathname); + mwriter->fd = -1; +} + +/* + * Produce a JSON string literal, properly escaping characters in the text. + */ +static void +escape_json(StringInfo buf, const char *str) +{ + const char *p; + + appendStringInfoCharMacro(buf, '"'); + for (p = str; *p; p++) + { + switch (*p) + { + case '\b': + appendStringInfoString(buf, "\\b"); + break; + case '\f': + appendStringInfoString(buf, "\\f"); + break; + case '\n': + appendStringInfoString(buf, "\\n"); + break; + case '\r': + appendStringInfoString(buf, "\\r"); + break; + case '\t': + appendStringInfoString(buf, "\\t"); + break; + case '"': + appendStringInfoString(buf, "\\\""); + break; + case '\\': + appendStringInfoString(buf, "\\\\"); + break; + default: + if ((unsigned char) *p < ' ') + appendStringInfo(buf, "\\u%04x", (int) *p); + else + appendStringInfoCharMacro(buf, *p); + break; + } + } + appendStringInfoCharMacro(buf, '"'); +} + +/* + * Flush whatever portion of the backup manifest we have generated and + * buffered in memory out to a file on disk. + * + * The first call to this function will create the file. After that, we + * keep it open and just append more data. + */ +static void +flush_manifest(manifest_writer *mwriter) +{ + char pathname[MAXPGPATH]; + + if (mwriter->fd == -1 && + (mwriter->fd = open(mwriter->pathname, + O_WRONLY | O_CREAT | O_EXCL | PG_BINARY, + pg_file_create_mode)) < 0) + pg_fatal("could not open file \"%s\": %m", mwriter->pathname); + + if (mwriter->buf.len > 0) + { + ssize_t wb; + + wb = write(mwriter->fd, mwriter->buf.data, mwriter->buf.len); + if (wb != mwriter->buf.len) + { + if (wb < 0) + pg_fatal("could not write \"%s\": %m", mwriter->pathname); + else + pg_fatal("could not write file \"%s\": wrote only %d of %d bytes", + pathname, (int) wb, mwriter->buf.len); + } + + if (mwriter->still_checksumming) + pg_checksum_update(&mwriter->manifest_ctx, + (uint8 *) mwriter->buf.data, + mwriter->buf.len); + resetStringInfo(&mwriter->buf); + } +} + +/* + * Encode bytes using two hexademical digits for each one. + */ +static size_t +hex_encode(const uint8 *src, size_t len, char *dst) +{ + const uint8 *end = src + len; + + while (src < end) + { + unsigned n1 = (*src >> 4) & 0xF; + unsigned n2 = *src & 0xF; + + *dst++ = n1 < 10 ? '0' + n1 : 'a' + n1 - 10; + *dst++ = n2 < 10 ? '0' + n2 : 'a' + n2 - 10; + ++src; + } + + return len * 2; +} diff --git a/src/bin/pg_combinebackup/write_manifest.h b/src/bin/pg_combinebackup/write_manifest.h new file mode 100644 index 0000000000..8fd7fe02c8 --- /dev/null +++ b/src/bin/pg_combinebackup/write_manifest.h @@ -0,0 +1,33 @@ +/*------------------------------------------------------------------------- + * + * Write a new backup manifest. + * + * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group + * Portions Copyright (c) 1994, Regents of the University of California + * + * src/bin/pg_combinebackup/write_manifest.h + * + *------------------------------------------------------------------------- + */ +#ifndef WRITE_MANIFEST_H +#define WRITE_MANIFEST_H + +#include "common/checksum_helper.h" +#include "pgtime.h" + +struct manifest_wal_range; + +struct manifest_writer; +typedef struct manifest_writer manifest_writer; + +extern manifest_writer *create_manifest_writer(char *directory); +extern void add_file_to_manifest(manifest_writer *mwriter, + const char *manifest_path, + size_t size, pg_time_t mtime, + pg_checksum_type checksum_type, + int checksum_length, + uint8 *checksum_payload); +extern void finalize_manifest(manifest_writer *mwriter, + struct manifest_wal_range *first_wal_range); + +#endif /* WRITE_MANIFEST_H */ diff --git a/src/bin/pg_resetwal/pg_resetwal.c b/src/bin/pg_resetwal/pg_resetwal.c index 3ae3fc06df..5407f51a4e 100644 --- a/src/bin/pg_resetwal/pg_resetwal.c +++ b/src/bin/pg_resetwal/pg_resetwal.c @@ -85,6 +85,7 @@ static void RewriteControlFile(void); static void FindEndOfXLOG(void); static void KillExistingXLOG(void); static void KillExistingArchiveStatus(void); +static void KillExistingWALSummaries(void); static void WriteEmptyXLOG(void); static void usage(void); @@ -493,6 +494,7 @@ main(int argc, char *argv[]) RewriteControlFile(); KillExistingXLOG(); KillExistingArchiveStatus(); + KillExistingWALSummaries(); WriteEmptyXLOG(); printf(_("Write-ahead log reset\n")); @@ -1034,6 +1036,40 @@ KillExistingArchiveStatus(void) pg_fatal("could not close directory \"%s\": %m", ARCHSTATDIR); } +/* + * Remove existing WAL summary files + */ +static void +KillExistingWALSummaries(void) +{ +#define WALSUMMARYDIR XLOGDIR "/summaries" +#define WALSUMMARY_NHEXCHARS 40 + + DIR *xldir; + struct dirent *xlde; + char path[MAXPGPATH + sizeof(WALSUMMARYDIR)]; + + xldir = opendir(WALSUMMARYDIR); + if (xldir == NULL) + pg_fatal("could not open directory \"%s\": %m", WALSUMMARYDIR); + + while (errno = 0, (xlde = readdir(xldir)) != NULL) + { + if (strspn(xlde->d_name, "0123456789ABCDEF") == WALSUMMARY_NHEXCHARS && + strcmp(xlde->d_name + WALSUMMARY_NHEXCHARS, ".summary") == 0) + { + snprintf(path, sizeof(path), "%s/%s", WALSUMMARYDIR, xlde->d_name); + if (unlink(path) < 0) + pg_fatal("could not delete file \"%s\": %m", path); + } + } + + if (errno) + pg_fatal("could not read directory \"%s\": %m", WALSUMMARYDIR); + + if (closedir(xldir)) + pg_fatal("could not close directory \"%s\": %m", ARCHSTATDIR); +} /* * Write an empty XLOG file, containing only the checkpoint record diff --git a/src/common/Makefile b/src/common/Makefile index 3c8effc533..2b41dd1839 100644 --- a/src/common/Makefile +++ b/src/common/Makefile @@ -49,6 +49,7 @@ OBJS_COMMON = \ archive.o \ base64.o \ binaryheap.o \ + blkreftable.o \ checksum_helper.o \ compression.o \ config_info.o \ diff --git a/src/common/blkreftable.c b/src/common/blkreftable.c new file mode 100644 index 0000000000..012a443584 --- /dev/null +++ b/src/common/blkreftable.c @@ -0,0 +1,1309 @@ +/*------------------------------------------------------------------------- + * + * blkreftable.c + * Block reference tables. + * + * A block reference table is used to keep track of which blocks have + * been modified by WAL records within a certain LSN range. + * + * For each relation fork, we keep track of all blocks that have appeared + * in block reference in the WAL. We also keep track of the "limit block", + * which is the smallest relation length in blocks known to have occurred + * during that range of WAL records. This should be set to 0 if the relation + * fork is created or destroyed, and to the post-truncation length if + * truncated. + * + * Whenever we set the limit block, we also forget about any modified blocks + * beyond that point. Those blocks don't exist any more. Such blocks can + * later be marked as modified again; if that happens, it means the relation + * was re-extended. + * + * Portions Copyright (c) 2010-2022, PostgreSQL Global Development Group + * + * src/common/blkreftable.c + * + *------------------------------------------------------------------------- + */ + + +#ifndef FRONTEND +#include "postgres.h" +#else +#include "postgres_fe.h" +#endif + +#ifdef FRONTEND +#include "common/logging.h" +#endif + +#include "common/blkreftable.h" +#include "common/hashfn.h" +#include "port/pg_crc32c.h" + +/* + * A block reference table keeps track of the status of each relation + * fork individually. + */ +typedef struct BlockRefTableKey +{ + RelFileLocator rlocator; + ForkNumber forknum; +} BlockRefTableKey; + +/* + * We could need to store data either for a relation in which only a + * tiny fraction of the blocks have been modified or for a relation in + * which nearly every block has been modified, and we want a + * space-efficient representation in both cases. To accomplish this, + * we divide the relation into chunks of 2^16 blocks and choose between + * an array representation and a bitmap representation for each chunk. + * + * When the number of modified blocks in a given chunk is small, we + * essentially store an array of block numbers, but we need not store the + * entire block number: instead, we store each block number as a 2-byte + * offset from the start of the chunk. + * + * When the number of modified blocks in a given chunk is large, we switch + * to a bitmap representation. + * + * These same basic representational choices are used both when a block + * reference table is stored in memory and when it is serialized to disk. + * + * In the in-memory representation, we initially allocate each chunk with + * space for a number of entries given by INITIAL_ENTRIES_PER_CHUNK and + * increase that as necessary until we reach MAX_ENTRIES_PER_CHUNK. + * Any chunk whose allocated size reaches MAX_ENTRIES_PER_CHUNK is converted + * to a bitmap, and thus never needs to grow further. + */ +#define BLOCKS_PER_CHUNK (1 << 16) +#define BLOCKS_PER_ENTRY (BITS_PER_BYTE * sizeof(uint16)) +#define MAX_ENTRIES_PER_CHUNK (BLOCKS_PER_CHUNK / BLOCKS_PER_ENTRY) +#define INITIAL_ENTRIES_PER_CHUNK 16 +typedef uint16 *BlockRefTableChunk; + +/* + * State for one relation fork. + * + * 'rlocator' and 'forknum' identify the relation fork to which this entry + * pertains. + * + * 'limit_block' is the shortest known length of the relation in blocks + * within the LSN range covered by a particular block reference table. + * It should be set to 0 if the relation fork is created or dropped. If the + * relation fork is truncated, it should be set to the number of blocks that + * remain after truncation. + * + * 'nchunks' is the allocated length of each of the three arrays that follow. + * We can only represent the status of block numbers less than nchunks * + * BLOCKS_PER_CHUNK. + * + * 'chunk_size' is an array storing the allocated size of each chunk. + * + * 'chunk_usage' is an array storing the number of elements used in each + * chunk. If that value is less than MAX_ENTRIES_PER_CHUNK, the corresonding + * chunk is used as an array; else the corresponding chunk is used as a bitmap. + * When used as a bitmap, the least significant bit of the first array element + * is the status of the lowest-numbered block covered by this chunk. + * + * 'chunk_data' is the array of chunks. + */ +struct BlockRefTableEntry +{ + BlockRefTableKey key; + BlockNumber limit_block; + char status; + uint32 nchunks; + uint16 *chunk_size; + uint16 *chunk_usage; + BlockRefTableChunk *chunk_data; +}; + +/* Declare and define a hash table over type BlockRefTableEntry. */ +#define SH_PREFIX blockreftable +#define SH_ELEMENT_TYPE BlockRefTableEntry +#define SH_KEY_TYPE BlockRefTableKey +#define SH_KEY key +#define SH_HASH_KEY(tb, key) \ + hash_bytes((const unsigned char *) &key, sizeof(BlockRefTableKey)) +#define SH_EQUAL(tb, a, b) memcmp(&a, &b, sizeof(BlockRefTableKey)) == 0 +#define SH_SCOPE static inline +#ifdef FRONTEND +#define SH_RAW_ALLOCATOR pg_malloc0 +#endif +#define SH_DEFINE +#define SH_DECLARE +#include "lib/simplehash.h" + +/* + * A block reference table is basically just the hash table, but we don't + * want to expose that to outside callers. + * + * We keep track of the memory context in use explicitly too, so that it's + * easy to place all of our allocations in the same context. + */ +struct BlockRefTable +{ + blockreftable_hash *hash; +#ifndef FRONTEND + MemoryContext mcxt; +#endif +}; + +/* + * On-disk serialization format for block reference table entries. + */ +typedef struct BlockRefTableSerializedEntry +{ + RelFileLocator rlocator; + ForkNumber forknum; + BlockNumber limit_block; + uint32 nchunks; +} BlockRefTableSerializedEntry; + +/* + * Buffer size, so that we avoid doing many small I/Os. + */ +#define BUFSIZE 65536 + +/* + * Ad-hoc buffer for file I/O. + */ +typedef struct BlockRefTableBuffer +{ + io_callback_fn io_callback; + void *io_callback_arg; + char data[BUFSIZE]; + int used; + int cursor; + pg_crc32c crc; +} BlockRefTableBuffer; + +/* + * State for keeping track of progress while incrementally reading a block + * table reference file from disk. + * + * total_chunks means the number of chunks for the RelFileLocator/ForkNumber + * combination that is curently being read, and consumed_chunks is the number + * of those that have been read. (We always read all the information for + * a single chunk at one time, so we don't need to be able to represent the + * state where a chunk has been partially read.) + * + * chunk_size is the array of chunk sizes. The length is given by total_chunks. + * + * chunk_data holds the current chunk. + * + * chunk_position helps us figure out how much progress we've made in returning + * the block numbers for the current chunk to the caller. If the chunk is a + * bitmap, it's the number of bits we've scanned; otherwise, it's the number + * of chunk entries we've scanned. + */ +struct BlockRefTableReader +{ + BlockRefTableBuffer buffer; + char *error_filename; + report_error_fn error_callback; + void *error_callback_arg; + uint32 total_chunks; + uint32 consumed_chunks; + uint16 *chunk_size; + uint16 chunk_data[MAX_ENTRIES_PER_CHUNK]; + uint32 chunk_position; +}; + +/* + * State for keeping track of progress while incrementally writing a block + * reference table file to disk. + */ +struct BlockRefTableWriter +{ + BlockRefTableBuffer buffer; +}; + +/* Function prototypes. */ +static int BlockRefTableComparator(const void *a, const void *b); +static void BlockRefTableFlush(BlockRefTableBuffer *buffer); +static void BlockRefTableRead(BlockRefTableReader *reader, void *data, + int length); +static void BlockRefTableWrite(BlockRefTableBuffer *buffer, void *data, + int length); +static void BlockRefTableFileTerminate(BlockRefTableBuffer *buffer); + +/* + * Create an empty block reference table. + */ +BlockRefTable * +CreateEmptyBlockRefTable(void) +{ + BlockRefTable *brtab = palloc(sizeof(BlockRefTable)); + + /* + * Even completely empty database has a few hundred relation forks, so it + * seems best to size the hash on the assumption that we're going to have + * at least a few thousand entries. + */ +#ifdef FRONTEND + brtab->hash = blockreftable_create(4096, NULL); +#else + brtab->mcxt = CurrentMemoryContext; + brtab->hash = blockreftable_create(brtab->mcxt, 4096, NULL); +#endif + + return brtab; +} + +/* + * Set the "limit block" for a relation fork and forget any modified blocks + * with equal or higher block numbers. + * + * The "limit block" is the shortest known length of the relation within the + * range of WAL records covered by this block reference table. + */ +void +BlockRefTableSetLimitBlock(BlockRefTable *brtab, + const RelFileLocator *rlocator, + ForkNumber forknum, + BlockNumber limit_block) +{ + BlockRefTableEntry *brtentry; + BlockRefTableKey key; + bool found; + + memcpy(&key.rlocator, rlocator, sizeof(RelFileLocator)); + key.forknum = forknum; + brtentry = blockreftable_insert(brtab->hash, key, &found); + + if (!found) + { + /* + * We have no existing data about this relation fork, so just record + * the limit_block value supplied by the caller, and make sure other + * parts of the entry are properly initialized. + */ + brtentry->limit_block = limit_block; + brtentry->nchunks = 0; + brtentry->chunk_size = NULL; + brtentry->chunk_usage = NULL; + brtentry->chunk_data = NULL; + return; + } + + BlockRefTableEntrySetLimitBlock(brtentry, limit_block); +} + +/* + * Mark a block in a given relation fork as known to have been modified. + */ +void +BlockRefTableMarkBlockModified(BlockRefTable *brtab, + const RelFileLocator *rlocator, + ForkNumber forknum, + BlockNumber blknum) +{ + BlockRefTableEntry *brtentry; + BlockRefTableKey key; + bool found; +#ifndef FRONTEND + MemoryContext oldcontext = MemoryContextSwitchTo(brtab->mcxt); +#endif + + memcpy(&key.rlocator, rlocator, sizeof(RelFileLocator)); + key.forknum = forknum; + brtentry = blockreftable_insert(brtab->hash, key, &found); + + if (!found) + { + /* + * We want to set the initial limit block value to something higher + * than any legal block number. InvalidBlockNumber fits the bill. + */ + brtentry->limit_block = InvalidBlockNumber; + brtentry->nchunks = 0; + brtentry->chunk_size = NULL; + brtentry->chunk_usage = NULL; + brtentry->chunk_data = NULL; + } + + BlockRefTableEntryMarkBlockModified(brtentry, forknum, blknum); + +#ifndef FRONTEND + MemoryContextSwitchTo(oldcontext); +#endif +} + +/* + * Get an entry from a block reference table. + * + * If the entry does not exist, this function returns NULL. Otherwise, it + * returns the entry and sets *limit_block to the value from the entry. + */ +BlockRefTableEntry * +BlockRefTableGetEntry(BlockRefTable *brtab, const RelFileLocator *rlocator, + ForkNumber forknum, BlockNumber *limit_block) +{ + BlockRefTableKey key; + BlockRefTableEntry *entry; + + Assert(limit_block != NULL); + + memcpy(&key.rlocator, rlocator, sizeof(RelFileLocator)); + key.forknum = forknum; + entry = blockreftable_lookup(brtab->hash, key); + + if (entry != NULL) + *limit_block = entry->limit_block; + + return entry; +} + +/* + * Get block numbers from a table entry. + * + * 'blocks' must point to enough space to hold at least 'nblocks' block + * numbers, and any block numbers we manage to get will be written there. + * The return value is the number of block numbers actually written. + * + * We do not return block numbers unless they are greater than or equal to + * start_blkno and strictly less than stop_blkno. + */ +int +BlockRefTableEntryGetBlocks(BlockRefTableEntry *entry, + BlockNumber start_blkno, + BlockNumber stop_blkno, + BlockNumber *blocks, + int nblocks) +{ + uint32 start_chunkno; + uint32 stop_chunkno; + uint32 chunkno; + int nresults = 0; + + Assert(entry != NULL); + + /* + * Figure out which chunks could potentially contain blocks of interest. + * + * We need to be careful about overflow here, because stop_blkno could be + * InvalidBlockNumber or something very close to it. + */ + start_chunkno = start_blkno / BLOCKS_PER_CHUNK; + stop_chunkno = stop_blkno / BLOCKS_PER_CHUNK; + if ((stop_blkno % BLOCKS_PER_CHUNK) != 0) + ++stop_chunkno; + if (stop_chunkno > entry->nchunks) + stop_chunkno = entry->nchunks; + + /* + * Loop over chunks. + */ + for (chunkno = start_chunkno; chunkno < stop_chunkno; ++chunkno) + { + uint16 chunk_usage = entry->chunk_usage[chunkno]; + BlockRefTableChunk chunk_data = entry->chunk_data[chunkno]; + unsigned start_offset = 0; + unsigned stop_offset = BLOCKS_PER_CHUNK; + + /* + * If the start and/or stop block number falls within this chunk, the + * whole chunk may not be of interest. Figure out which portion we + * care about, if it's not the whole thing. + */ + if (chunkno == start_chunkno) + start_offset = start_blkno % BLOCKS_PER_CHUNK; + if (chunkno == stop_chunkno) + stop_offset = stop_blkno % BLOCKS_PER_CHUNK; + + /* + * Handling differs depending on whether this is an array of offsets + * or a bitmap. + */ + if (chunk_usage == MAX_ENTRIES_PER_CHUNK) + { + unsigned i; + + /* It's a bitmap, so test every relevant bit. */ + for (i = start_offset; i < BLOCKS_PER_CHUNK; ++i) + { + uint16 w = chunk_data[i / BLOCKS_PER_ENTRY]; + + if ((w & (1 << (i % BLOCKS_PER_ENTRY))) != 0) + { + BlockNumber blkno = chunkno * BLOCKS_PER_CHUNK + i; + + blocks[nresults++] = blkno; + + /* Early exit if we run out of output space. */ + if (nresults == nblocks) + return nresults; + } + } + } + else + { + unsigned i; + + /* It's an array of offsets, so check each one. */ + for (i = 0; i < chunk_usage; ++i) + { + uint16 offset = chunk_data[i]; + + if (offset >= start_offset && offset < stop_offset) + { + BlockNumber blkno = chunkno * BLOCKS_PER_CHUNK + offset; + + blocks[nresults++] = blkno; + + /* Early exit if we run out of output space. */ + if (nresults == nblocks) + return nresults; + } + } + } + } + + return nresults; +} + +/* + * Serialize a block reference table to a file. + */ +void +WriteBlockRefTable(BlockRefTable *brtab, + io_callback_fn write_callback, + void *write_callback_arg) +{ + BlockRefTableSerializedEntry *sdata = NULL; + BlockRefTableBuffer buffer; + uint32 magic = BLOCKREFTABLE_MAGIC; + + /* Prepare buffer. */ + memset(&buffer, 0, sizeof(BlockRefTableBuffer)); + buffer.io_callback = write_callback; + buffer.io_callback_arg = write_callback_arg; + INIT_CRC32C(buffer.crc); + + /* Write magic number. */ + BlockRefTableWrite(&buffer, &magic, sizeof(uint32)); + + /* Write the entries, assuming there are some. */ + if (brtab->hash->members > 0) + { + unsigned i = 0; + blockreftable_iterator it; + BlockRefTableEntry *brtentry; + + /* Extract entries into serializable format and sort them. */ + sdata = + palloc(brtab->hash->members * sizeof(BlockRefTableSerializedEntry)); + blockreftable_start_iterate(brtab->hash, &it); + while ((brtentry = blockreftable_iterate(brtab->hash, &it)) != NULL) + { + BlockRefTableSerializedEntry *sentry = &sdata[i++]; + + sentry->rlocator = brtentry->key.rlocator; + sentry->forknum = brtentry->key.forknum; + sentry->limit_block = brtentry->limit_block; + sentry->nchunks = brtentry->nchunks; + + /* trim trailing zero entries */ + while (sentry->nchunks > 0 && + brtentry->chunk_usage[sentry->nchunks - 1] == 0) + sentry->nchunks--; + } + Assert(i == brtab->hash->members); + qsort(sdata, i, sizeof(BlockRefTableSerializedEntry), + BlockRefTableComparator); + + /* Loop over entries in sorted order and serialize each one. */ + for (i = 0; i < brtab->hash->members; ++i) + { + BlockRefTableSerializedEntry *sentry = &sdata[i]; + BlockRefTableEntry *brtentry; + BlockRefTableKey key; + unsigned j; + + /* Write the serialized entry itself. */ + BlockRefTableWrite(&buffer, sentry, + sizeof(BlockRefTableSerializedEntry)); + + /* Look up the original entry so we can access the chunks. */ + memcpy(&key.rlocator, &sentry->rlocator, sizeof(RelFileLocator)); + key.forknum = sentry->forknum; + brtentry = blockreftable_lookup(brtab->hash, key); + Assert(brtentry != NULL); + + /* Write the untruncated portion of the chunk length array. */ + if (sentry->nchunks != 0) + BlockRefTableWrite(&buffer, brtentry->chunk_usage, + sentry->nchunks * sizeof(uint16)); + + /* Write the contents of each chunk. */ + for (j = 0; j < brtentry->nchunks; ++j) + { + if (brtentry->chunk_usage[j] == 0) + continue; + BlockRefTableWrite(&buffer, brtentry->chunk_data[j], + brtentry->chunk_usage[j] * sizeof(uint16)); + } + } + } + + /* Write out appropriate terminator and CRC and flush buffer. */ + BlockRefTableFileTerminate(&buffer); +} + +/* + * Prepare to incrementally read a block reference table file. + * + * 'read_callback' is a function that can be called to read data from the + * underlying file (or other data source) into our internal buffer. + * + * 'read_callback_arg' is an opaque argument to be passed to read_callback. + * + * 'error_filename' is the filename that should be included in error messages + * if the file is found to be malformed. The value is not copied, so the + * caller should ensure that it remains valid until done with this + * BlockRefTableReader. + * + * 'error_callback' is a function to be called if the file is found to be + * malformed. This is not used for I/O errors, which must be handled internally + * by read_callback. + * + * 'error_callback_arg' is an opaque arguent to be passed to error_callback. + */ +BlockRefTableReader * +CreateBlockRefTableReader(io_callback_fn read_callback, + void *read_callback_arg, + char *error_filename, + report_error_fn error_callback, + void *error_callback_arg) +{ + BlockRefTableReader *reader; + uint32 magic; + + /* Initialize data structure. */ + reader = palloc0(sizeof(BlockRefTableReader)); + reader->buffer.io_callback = read_callback; + reader->buffer.io_callback_arg = read_callback_arg; + reader->error_filename = error_filename; + reader->error_callback = error_callback; + reader->error_callback_arg = error_callback_arg; + INIT_CRC32C(reader->buffer.crc); + + /* Verify magic number. */ + BlockRefTableRead(reader, &magic, sizeof(uint32)); + if (magic != BLOCKREFTABLE_MAGIC) + error_callback(error_callback_arg, + "file \"%s\" has wrong magic number: expected %u, found %u", + error_filename, + BLOCKREFTABLE_MAGIC, magic); + + return reader; +} + +/* + * Read next relation fork covered by this block reference table file. + * + * After calling this function, you must call BlockRefTableReaderGetBlocks + * until it returns 0 before calling it again. + */ +bool +BlockRefTableReaderNextRelation(BlockRefTableReader *reader, + RelFileLocator *rlocator, + ForkNumber *forknum, + BlockNumber *limit_block) +{ + BlockRefTableSerializedEntry sentry; + BlockRefTableSerializedEntry zentry = {0}; + + /* + * Sanity check: caller must read all blocks from all chunks before moving + * on to the next relation. + */ + Assert(reader->total_chunks == reader->consumed_chunks); + + /* Read serialized entry. */ + BlockRefTableRead(reader, &sentry, + sizeof(BlockRefTableSerializedEntry)); + + /* + * If we just read the sentinel entry indicating that we've reached the + * end, read and check the CRC. + */ + if (memcmp(&sentry, &zentry, sizeof(BlockRefTableSerializedEntry)) == 0) + { + pg_crc32c expected_crc; + pg_crc32c actual_crc; + + /* + * We want to know the CRC of the file excluding the 4-byte CRC + * itself, so copy the current value of the CRC accumulator before + * reading those bytes, and use the copy to finalize the calculation. + */ + expected_crc = reader->buffer.crc; + FIN_CRC32C(expected_crc); + + /* Now we can read the actual value. */ + BlockRefTableRead(reader, &actual_crc, sizeof(pg_crc32c)); + + /* Throw an error if there is a mismatch. */ + if (!EQ_CRC32C(expected_crc, actual_crc)) + reader->error_callback(reader->error_callback_arg, + "file \"%s\" has wrong checksum: expected %08X, found %08X", + reader->error_filename, expected_crc, actual_crc); + + return false; + } + + /* Read chunk size array. */ + if (reader->chunk_size != NULL) + pfree(reader->chunk_size); + reader->chunk_size = palloc(sentry.nchunks * sizeof(uint16)); + BlockRefTableRead(reader, reader->chunk_size, + sentry.nchunks * sizeof(uint16)); + + /* Set up for chunk scan. */ + reader->total_chunks = sentry.nchunks; + reader->consumed_chunks = 0; + + /* Return data to caller. */ + memcpy(rlocator, &sentry.rlocator, sizeof(RelFileLocator)); + *forknum = sentry.forknum; + *limit_block = sentry.limit_block; + return true; +} + +/* + * Get modified blocks associated with the relation fork returned by + * the most recent call to BlockRefTableReaderNextRelation. + * + * On return, block numbers will be written into the 'blocks' array, whose + * length should be passed via 'nblocks'. The return value is the number of + * entries actually written into the 'blocks' array, which may be less than + * 'nblocks' if we run out of modified blocks in the relation fork before + * we run out of room in the array. + */ +unsigned +BlockRefTableReaderGetBlocks(BlockRefTableReader *reader, + BlockNumber *blocks, + int nblocks) +{ + unsigned blocks_found = 0; + + /* Must provide space for at least one block number to be returned. */ + Assert(nblocks > 0); + + /* Loop collecting blocks to return to caller. */ + for (;;) + { + uint16 next_chunk_size; + + /* + * If we've read at least one chunk, maybe it contains some block + * numbers that could satisfy caller's request. + */ + if (reader->consumed_chunks > 0) + { + uint32 chunkno = reader->consumed_chunks - 1; + uint16 chunk_size = reader->chunk_size[chunkno]; + + if (chunk_size == MAX_ENTRIES_PER_CHUNK) + { + /* Bitmap format, so search for bits that are set. */ + while (reader->chunk_position < BLOCKS_PER_CHUNK && + blocks_found < nblocks) + { + uint16 chunkoffset = reader->chunk_position; + uint16 w; + + w = reader->chunk_data[chunkoffset / BLOCKS_PER_ENTRY]; + if ((w & (1u << (chunkoffset % BLOCKS_PER_ENTRY))) != 0) + blocks[blocks_found++] = + chunkno * BLOCKS_PER_CHUNK + chunkoffset; + ++reader->chunk_position; + } + } + else + { + /* Not in bitmap format, so each entry is a 2-byte offset. */ + while (reader->chunk_position < chunk_size && + blocks_found < nblocks) + { + blocks[blocks_found++] = chunkno * BLOCKS_PER_CHUNK + + reader->chunk_data[reader->chunk_position]; + ++reader->chunk_position; + } + } + } + + /* We found enough blocks, so we're done. */ + if (blocks_found >= nblocks) + break; + + /* + * We didn't find enough blocks, so we must need the next chunk. If + * there are none left, though, then we're done anyway. + */ + if (reader->consumed_chunks == reader->total_chunks) + break; + + /* + * Read data for next chunk and reset scan position to beginning of + * chunk. Note that the next chunk might be empty, in which case we + * consume the chunk without actually consuming any bytes from the + * underlying file. + */ + next_chunk_size = reader->chunk_size[reader->consumed_chunks]; + if (next_chunk_size > 0) + BlockRefTableRead(reader, reader->chunk_data, + next_chunk_size * sizeof(uint16)); + ++reader->consumed_chunks; + reader->chunk_position = 0; + } + + return blocks_found; +} + +/* + * Release memory used while reading a block reference table from a file. + */ +void +DestroyBlockRefTableReader(BlockRefTableReader *reader) +{ + if (reader->chunk_size != NULL) + { + pfree(reader->chunk_size); + reader->chunk_size = NULL; + } + pfree(reader); +} + +/* + * Prepare to write a block reference table file incrementally. + * + * Caller must be able to supply BlockRefTableEntry objects sorted in the + * appropriate order. + */ +BlockRefTableWriter * +CreateBlockRefTableWriter(io_callback_fn write_callback, + void *write_callback_arg) +{ + BlockRefTableWriter *writer; + uint32 magic = BLOCKREFTABLE_MAGIC; + + /* Prepare buffer and CRC check and save callbacks. */ + writer = palloc0(sizeof(BlockRefTableWriter)); + writer->buffer.io_callback = write_callback; + writer->buffer.io_callback_arg = write_callback_arg; + INIT_CRC32C(writer->buffer.crc); + + /* Write magic number. */ + BlockRefTableWrite(&writer->buffer, &magic, sizeof(uint32)); + + return writer; +} + +/* + * Append one entry to a block reference table file. + * + * Note that entries must be written in the proper order, that is, sorted by + * tablespace, then database, then relfilenumber, then fork number. Caller + * is responsible for supplying data in the correct order. If that seems hard, + * use an in-memory BlockRefTable instead. + */ +void +BlockRefTableWriteEntry(BlockRefTableWriter *writer, BlockRefTableEntry *entry) +{ + BlockRefTableSerializedEntry sentry; + unsigned j; + + /* Convert to serialized entry format. */ + sentry.rlocator = entry->key.rlocator; + sentry.forknum = entry->key.forknum; + sentry.limit_block = entry->limit_block; + sentry.nchunks = entry->nchunks; + + /* Trim trailing zero entries. */ + while (sentry.nchunks > 0 && entry->chunk_usage[sentry.nchunks - 1] == 0) + sentry.nchunks--; + + /* Write the serialized entry itself. */ + BlockRefTableWrite(&writer->buffer, &sentry, + sizeof(BlockRefTableSerializedEntry)); + + /* Write the untruncated portion of the chunk length array. */ + if (sentry.nchunks != 0) + BlockRefTableWrite(&writer->buffer, entry->chunk_usage, + sentry.nchunks * sizeof(uint16)); + + /* Write the contents of each chunk. */ + for (j = 0; j < entry->nchunks; ++j) + { + if (entry->chunk_usage[j] == 0) + continue; + BlockRefTableWrite(&writer->buffer, entry->chunk_data[j], + entry->chunk_usage[j] * sizeof(uint16)); + } +} + +/* + * Finalize an incremental write of a block reference table file. + */ +void +DestroyBlockRefTableWriter(BlockRefTableWriter *writer) +{ + BlockRefTableFileTerminate(&writer->buffer); + pfree(writer); +} + +/* + * Allocate a standalone BlockRefTableEntry. + * + * When we're manipulating a full in-memory BlockRefTable, the entries are + * part of the hash table and are allocated by simplehash. This routine is + * used by callers that want to write out a BlockRefTable to a file without + * needing to store the whole thing in memory at once. + * + * Entries allocated by this function can be manipulated using the functions + * BlockRefTableEntrySetLimitBlock and BlockRefTableEntryMarkBlockModified + * and then written using BlockRefTableWriteEntry and freed using + * BlockRefTableFreeEntry. + */ +BlockRefTableEntry * +CreateBlockRefTableEntry(RelFileLocator rlocator, ForkNumber forknum) +{ + BlockRefTableEntry *entry = palloc0(sizeof(BlockRefTableEntry)); + + memcpy(&entry->key.rlocator, &rlocator, sizeof(RelFileLocator)); + entry->key.forknum = forknum; + entry->limit_block = InvalidBlockNumber; + + return entry; +} + +/* + * Update a BlockRefTableEntry with a new value for the "limit block" and + * forget any equal-or-higher-numbered modified blocks. + * + * The "limit block" is the shortest known length of the relation within the + * range of WAL records covered by this block reference table. + */ +void +BlockRefTableEntrySetLimitBlock(BlockRefTableEntry *entry, + BlockNumber limit_block) +{ + unsigned chunkno; + unsigned limit_chunkno; + unsigned limit_chunkoffset; + BlockRefTableChunk limit_chunk; + + /* If we already have an equal or lower limit block, do nothing. */ + if (limit_block >= entry->limit_block) + return; + + /* Record the new limit block value. */ + entry->limit_block = limit_block; + + /* + * Figure out which chunk would store the state of the new limit block, + * and which offset within that chunk. + */ + limit_chunkno = limit_block / BLOCKS_PER_CHUNK; + limit_chunkoffset = limit_block % BLOCKS_PER_CHUNK; + + /* + * If the number of chunks is not large enough for any blocks with equal + * or higher block numbers to exist, then there is nothing further to do. + */ + if (limit_chunkno >= entry->nchunks) + return; + + /* Discard entire contents of any higher-numbered chunks. */ + for (chunkno = limit_chunkno + 1; chunkno < entry->nchunks; ++chunkno) + entry->chunk_usage[chunkno] = 0; + + /* + * Next, we need to discard any offsets within the chunk that would + * contain the limit_block. We must handle this differenly depending on + * whether the chunk that would contain limit_block is a bitmap or an + * array of offsets. + */ + limit_chunk = entry->chunk_data[limit_chunkno]; + if (entry->chunk_usage[limit_chunkno] == MAX_ENTRIES_PER_CHUNK) + { + unsigned chunkoffset; + + /* It's a bitmap. Unset bits. */ + for (chunkoffset = limit_chunkoffset; chunkoffset < BLOCKS_PER_CHUNK; + ++chunkoffset) + limit_chunk[chunkoffset / BLOCKS_PER_ENTRY] &= + ~(1 << (chunkoffset % BLOCKS_PER_ENTRY)); + } + else + { + unsigned i, + j = 0; + + /* It's an offset array. Filter out large offsets. */ + for (i = 0; i < entry->chunk_usage[limit_chunkno]; ++i) + { + Assert(j <= i); + if (limit_chunk[i] < limit_chunkoffset) + limit_chunk[j++] = limit_chunk[i]; + } + Assert(j <= entry->chunk_usage[limit_chunkno]); + entry->chunk_usage[limit_chunkno] = j; + } +} + +/* + * Mark a block in a given BlkRefTableEntry as known to have been modified. + */ +void +BlockRefTableEntryMarkBlockModified(BlockRefTableEntry *entry, + ForkNumber forknum, + BlockNumber blknum) +{ + unsigned chunkno; + unsigned chunkoffset; + unsigned i; + + /* + * Which chunk should store the state of this block? And what is the + * offset of this block relative to the start of that chunk? + */ + chunkno = blknum / BLOCKS_PER_CHUNK; + chunkoffset = blknum % BLOCKS_PER_CHUNK; + + /* + * If 'nchunks' isn't big enough for us to be able to represent the state + * of this block, we need to enlarge our arrays. + */ + if (chunkno >= entry->nchunks) + { + unsigned max_chunks; + unsigned extra_chunks; + + /* + * New array size is a power of 2, at least 16, big enough so that + * chunkno will be a valid array index. + */ + max_chunks = Max(16, entry->nchunks); + while (max_chunks < chunkno + 1) + chunkno *= 2; + Assert(max_chunks > chunkno); + extra_chunks = max_chunks - entry->nchunks; + + if (entry->nchunks == 0) + { + entry->chunk_size = palloc0(sizeof(uint16) * max_chunks); + entry->chunk_usage = palloc0(sizeof(uint16) * max_chunks); + entry->chunk_data = + palloc0(sizeof(BlockRefTableChunk) * max_chunks); + } + else + { + entry->chunk_size = repalloc(entry->chunk_size, + sizeof(uint16) * max_chunks); + memset(&entry->chunk_size[entry->nchunks], 0, + extra_chunks * sizeof(uint16)); + entry->chunk_usage = repalloc(entry->chunk_usage, + sizeof(uint16) * max_chunks); + memset(&entry->chunk_usage[entry->nchunks], 0, + extra_chunks * sizeof(uint16)); + entry->chunk_data = repalloc(entry->chunk_data, + sizeof(BlockRefTableChunk) * max_chunks); + memset(&entry->chunk_data[entry->nchunks], 0, + extra_chunks * sizeof(BlockRefTableChunk)); + } + entry->nchunks = max_chunks; + } + + /* + * If the chunk that covers this block number doesn't exist yet, create it + * as an array and add the appropriate offset to it. We make it pretty + * small initially, because there might only be 1 or a few block + * references in this chunk and we don't want to use up too much memory. + */ + if (entry->chunk_size[chunkno] == 0) + { + entry->chunk_data[chunkno] = + palloc(sizeof(uint16) * INITIAL_ENTRIES_PER_CHUNK); + entry->chunk_size[chunkno] = INITIAL_ENTRIES_PER_CHUNK; + entry->chunk_data[chunkno][0] = chunkoffset; + entry->chunk_usage[chunkno] = 1; + return; + } + + /* + * If the number of entries in this chunk is already maximum, it must be a + * bitmap. Just set the appropriate bit. + */ + if (entry->chunk_usage[chunkno] == MAX_ENTRIES_PER_CHUNK) + { + BlockRefTableChunk chunk = entry->chunk_data[chunkno]; + + chunk[chunkoffset / BLOCKS_PER_ENTRY] |= + 1 << (chunkoffset % BLOCKS_PER_ENTRY); + return; + } + + /* + * There is an existing chunk and it's in array format. Let's find out + * whether it already has an entry for this block. If so, we do not need + * to do anything. + */ + for (i = 0; i < entry->chunk_usage[chunkno]; ++i) + { + if (entry->chunk_data[chunkno][i] == chunkoffset) + return; + } + + /* + * If the number of entries currently used is one less than the maximum, + * it's time to convert to bitmap format. + */ + if (entry->chunk_usage[chunkno] == MAX_ENTRIES_PER_CHUNK - 1) + { + BlockRefTableChunk newchunk; + unsigned j; + + /* Allocate a new chunk. */ + newchunk = palloc0(MAX_ENTRIES_PER_CHUNK * sizeof(uint16)); + + /* Set the bit for each existing entry. */ + for (j = 0; j < entry->chunk_usage[chunkno]; ++j) + { + unsigned coff = entry->chunk_data[chunkno][j]; + + newchunk[coff / BLOCKS_PER_ENTRY] |= + 1 << (coff % BLOCKS_PER_ENTRY); + } + + /* Set the bit for the new entry. */ + newchunk[chunkoffset / BLOCKS_PER_ENTRY] |= + 1 << (chunkoffset % BLOCKS_PER_ENTRY); + + /* Swap the new chunk into place and update metadata. */ + pfree(entry->chunk_data[chunkno]); + entry->chunk_data[chunkno] = newchunk; + entry->chunk_size[chunkno] = MAX_ENTRIES_PER_CHUNK; + entry->chunk_usage[chunkno] = MAX_ENTRIES_PER_CHUNK; + return; + } + + /* + * OK, we currently have an array, and we don't need to convert to a + * bitmap, but we do need to add a new element. If there's not enough + * room, we'll have to expand the array. + */ + if (entry->chunk_usage[chunkno] == entry->chunk_size[chunkno]) + { + unsigned newsize = entry->chunk_size[chunkno] * 2; + + Assert(newsize <= MAX_ENTRIES_PER_CHUNK); + entry->chunk_data[chunkno] = repalloc(entry->chunk_data[chunkno], + newsize * sizeof(uint16)); + entry->chunk_size[chunkno] = newsize; + } + + /* Now we can add the new entry. */ + entry->chunk_data[chunkno][entry->chunk_usage[chunkno]] = + chunkoffset; + entry->chunk_usage[chunkno]++; +} + +/* + * Release memory for a BlockRefTablEntry that was created by + * CreateBlockRefTableEntry. + */ +void +BlockRefTableFreeEntry(BlockRefTableEntry *entry) +{ + if (entry->chunk_size != NULL) + { + pfree(entry->chunk_size); + entry->chunk_size = NULL; + } + + if (entry->chunk_usage != NULL) + { + pfree(entry->chunk_usage); + entry->chunk_usage = NULL; + } + + if (entry->chunk_data != NULL) + { + pfree(entry->chunk_data); + entry->chunk_data = NULL; + } + + pfree(entry); +} + +/* + * Comparator for BlockRefTableSerializedEntry objects. + * + * We make the tablespace OID the first column of the sort key to match + * the on-disk tree structure. + */ +static int +BlockRefTableComparator(const void *a, const void *b) +{ + const BlockRefTableSerializedEntry *sa = a; + const BlockRefTableSerializedEntry *sb = b; + + if (sa->rlocator.spcOid > sb->rlocator.spcOid) + return 1; + if (sa->rlocator.spcOid < sb->rlocator.spcOid) + return -1; + + if (sa->rlocator.dbOid > sb->rlocator.dbOid) + return 1; + if (sa->rlocator.dbOid < sb->rlocator.dbOid) + return -1; + + if (sa->rlocator.relNumber > sb->rlocator.relNumber) + return 1; + if (sa->rlocator.relNumber < sb->rlocator.relNumber) + return -1; + + if (sa->forknum > sb->forknum) + return 1; + if (sa->forknum < sb->forknum) + return -1; + + return 0; +} + +/* + * Flush any buffered data out of a BlockRefTableBuffer. + */ +static void +BlockRefTableFlush(BlockRefTableBuffer *buffer) +{ + buffer->io_callback(buffer->io_callback_arg, buffer->data, buffer->used); + buffer->used = 0; +} + +/* + * Read data from a BlockRefTableBuffer, and update the running CRC + * calculation for the returned data (but not any data that we may have + * buffered but not yet actually returned). + */ +static void +BlockRefTableRead(BlockRefTableReader *reader, void *data, int length) +{ + BlockRefTableBuffer *buffer = &reader->buffer; + + /* Loop until read is fully satisfied. */ + while (length > 0) + { + if (buffer->cursor < buffer->used) + { + /* + * If any buffered data is available, use that to satisfy as much + * of the request as possible. + */ + int bytes_to_copy = Min(length, buffer->used - buffer->cursor); + + memcpy(data, &buffer->data[buffer->cursor], bytes_to_copy); + COMP_CRC32C(buffer->crc, &buffer->data[buffer->cursor], + bytes_to_copy); + buffer->cursor += bytes_to_copy; + data = ((char *) data) + bytes_to_copy; + length -= bytes_to_copy; + } + else if (length >= BUFSIZE) + { + /* + * If the request length is long, read directly into caller's + * buffer. + */ + int bytes_read; + + bytes_read = buffer->io_callback(buffer->io_callback_arg, + data, length); + COMP_CRC32C(buffer->crc, data, bytes_read); + data = ((char *) data) + bytes_read; + length -= bytes_read; + + /* If we didn't get anything, that's bad. */ + if (bytes_read == 0) + reader->error_callback(reader->error_callback_arg, + "file \"%s\" ends unexpectedly", + reader->error_filename); + } + else + { + /* + * Refill our buffer. + */ + buffer->used = buffer->io_callback(buffer->io_callback_arg, + buffer->data, BUFSIZE); + buffer->cursor = 0; + + /* If we didn't get anything, that's bad. */ + if (buffer->used == 0) + reader->error_callback(reader->error_callback_arg, + "file \"%s\" ends unexpectedly", + reader->error_filename); + } + } +} + +/* + * Supply data to a BlockRefTableBuffer for write to the underlying File, + * and update the running CRC calculation for that data. + */ +static void +BlockRefTableWrite(BlockRefTableBuffer *buffer, void *data, int length) +{ + /* Update running CRC calculation. */ + COMP_CRC32C(buffer->crc, data, length); + + /* If the new data can't fit into the buffer, flush the buffer. */ + if (buffer->used + length > BUFSIZE) + { + buffer->io_callback(buffer->io_callback_arg, buffer->data, + buffer->used); + buffer->used = 0; + } + + /* If the new data would fill the buffer, or more, write it directly. */ + if (length >= BUFSIZE) + { + buffer->io_callback(buffer->io_callback_arg, data, length); + return; + } + + /* Otherwise, copy the new data into the buffer. */ + memcpy(&buffer->data[buffer->used], data, length); + buffer->used += length; + Assert(buffer->used <= BUFSIZE); +} + +/* + * Generate the sentinel and CRC required at the end of a block reference + * table file and flush them out of our internal buffer. + */ +static void +BlockRefTableFileTerminate(BlockRefTableBuffer *buffer) +{ + BlockRefTableSerializedEntry zentry = {0}; + pg_crc32c crc; + + /* Write a sentinel indicating that there are no more entries. */ + BlockRefTableWrite(buffer, &zentry, + sizeof(BlockRefTableSerializedEntry)); + + /* + * Writing the checksum will perturb the ongoing checksum calculation, so + * copy the state first and finalize the computation using the copy. + */ + crc = buffer->crc; + FIN_CRC32C(crc); + BlockRefTableWrite(buffer, &crc, sizeof(pg_crc32c)); + + /* Flush any leftover data out of our buffer. */ + BlockRefTableFlush(buffer); +} diff --git a/src/common/meson.build b/src/common/meson.build index aa646f96a3..6348d60ec4 100644 --- a/src/common/meson.build +++ b/src/common/meson.build @@ -4,6 +4,7 @@ common_sources = files( 'archive.c', 'base64.c', 'binaryheap.c', + 'blkreftable.c', 'checksum_helper.c', 'compression.c', 'controldata_utils.c', diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 4ad572cb87..9d1e4ab57b 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -209,6 +209,7 @@ extern int XLogFileOpen(XLogSegNo segno, TimeLineID tli); extern void CheckXLogRemoved(XLogSegNo segno, TimeLineID tli); extern XLogSegNo XLogGetLastRemovedSegno(void); +extern XLogSegNo XLogGetOldestSegno(TimeLineID tli); extern void XLogSetAsyncXactLSN(XLogRecPtr asyncXactLSN); extern void XLogSetReplicationSlotMinimumLSN(XLogRecPtr lsn); diff --git a/src/include/access/xlogbackup.h b/src/include/access/xlogbackup.h index 1611358137..90e04cad56 100644 --- a/src/include/access/xlogbackup.h +++ b/src/include/access/xlogbackup.h @@ -28,6 +28,8 @@ typedef struct BackupState XLogRecPtr checkpointloc; /* last checkpoint location */ pg_time_t starttime; /* backup start time */ bool started_in_recovery; /* backup started in recovery? */ + XLogRecPtr istartpoint; /* incremental based on backup at this LSN */ + TimeLineID istarttli; /* incremental based on backup on this TLI */ /* Fields saved at the end of backup */ XLogRecPtr stoppoint; /* backup stop WAL location */ diff --git a/src/include/backup/basebackup.h b/src/include/backup/basebackup.h index 1432d9c206..345bd22534 100644 --- a/src/include/backup/basebackup.h +++ b/src/include/backup/basebackup.h @@ -34,6 +34,9 @@ typedef struct int64 size; /* total size as sent; -1 if not known */ } tablespaceinfo; -extern void SendBaseBackup(BaseBackupCmd *cmd); +struct IncrementalBackupInfo; + +extern void SendBaseBackup(BaseBackupCmd *cmd, + struct IncrementalBackupInfo *ib); #endif /* _BASEBACKUP_H */ diff --git a/src/include/backup/basebackup_incremental.h b/src/include/backup/basebackup_incremental.h new file mode 100644 index 0000000000..c300235a2f --- /dev/null +++ b/src/include/backup/basebackup_incremental.h @@ -0,0 +1,56 @@ +/*------------------------------------------------------------------------- + * + * basebackup_incremental.h + * API for incremental backup support + * + * Portions Copyright (c) 2010-2022, PostgreSQL Global Development Group + * + * src/include/backup/basebackup_incremental.h + * + *------------------------------------------------------------------------- + */ +#ifndef BASEBACKUP_INCREMENTAL_H +#define BASEBACKUP_INCREMENTAL_H + +#include "access/xlogbackup.h" +#include "common/relpath.h" +#include "storage/block.h" +#include "utils/palloc.h" + +#define INCREMENTAL_MAGIC 0xd3ae1f0d + +typedef enum +{ + BACK_UP_FILE_FULLY, + BACK_UP_FILE_INCREMENTALLY, + DO_NOT_BACK_UP_FILE +} FileBackupMethod; + +struct IncrementalBackupInfo; +typedef struct IncrementalBackupInfo IncrementalBackupInfo; + +extern IncrementalBackupInfo *CreateIncrementalBackupInfo(MemoryContext); + +extern void AppendIncrementalManifestData(IncrementalBackupInfo *ib, + const char *data, + int len); +extern void FinalizeIncrementalManifest(IncrementalBackupInfo *ib); + +extern void PrepareForIncrementalBackup(IncrementalBackupInfo *ib, + BackupState *backup_state); + +extern char *GetIncrementalFilePath(Oid dboid, Oid spcoid, + RelFileNumber relfilenumber, + ForkNumber forknum, unsigned segno); +extern FileBackupMethod GetFileBackupMethod(IncrementalBackupInfo *ib, + char *path, + Oid dboid, Oid spcoid, + RelFileNumber relfilenumber, + ForkNumber forknum, + unsigned segno, size_t size, + unsigned *num_blocks_required, + BlockNumber *relative_block_numbers, + unsigned *truncation_block_length); +extern size_t GetIncrementalFileSize(unsigned num_blocks_required); + +#endif diff --git a/src/include/backup/walsummary.h b/src/include/backup/walsummary.h new file mode 100644 index 0000000000..d086e64019 --- /dev/null +++ b/src/include/backup/walsummary.h @@ -0,0 +1,49 @@ +/*------------------------------------------------------------------------- + * + * walsummary.h + * WAL summary management + * + * Portions Copyright (c) 2010-2022, PostgreSQL Global Development Group + * + * src/include/backup/walsummary.h + * + *------------------------------------------------------------------------- + */ +#ifndef WALSUMMARY_H +#define WALSUMMARY_H + +#include <time.h> + +#include "access/xlogdefs.h" +#include "nodes/pg_list.h" +#include "storage/fd.h" + +typedef struct WalSummaryIO +{ + File file; + off_t filepos; +} WalSummaryIO; + +typedef struct WalSummaryFile +{ + XLogRecPtr start_lsn; + XLogRecPtr end_lsn; + TimeLineID tli; +} WalSummaryFile; + +extern List *GetWalSummaries(TimeLineID tli, XLogRecPtr start_lsn, + XLogRecPtr end_lsn); +extern List *FilterWalSummaries(List *wslist, TimeLineID tli, + XLogRecPtr start_lsn, XLogRecPtr end_lsn); +extern bool WalSummariesAreComplete(List *wslist, + XLogRecPtr start_lsn, XLogRecPtr end_lsn, + XLogRecPtr *missing_lsn); +extern File OpenWalSummaryFile(WalSummaryFile *ws, bool missing_ok); +extern void RemoveWalSummaryIfOlderThan(WalSummaryFile *ws, + time_t cutoff_time); + +extern int ReadWalSummary(void *wal_summary_io, void *data, int length); +extern int WriteWalSummary(void *wal_summary_io, void *data, int length); +extern void ReportWalSummaryError(void *callback_arg, char *fmt,...); + +#endif /* WALSUMMARY_H */ diff --git a/src/include/catalog/pg_proc.dat b/src/include/catalog/pg_proc.dat index c92d0631a0..9717c4630e 100644 --- a/src/include/catalog/pg_proc.dat +++ b/src/include/catalog/pg_proc.dat @@ -12071,4 +12071,23 @@ proname => 'any_value_transfn', prorettype => 'anyelement', proargtypes => 'anyelement anyelement', prosrc => 'any_value_transfn' }, +{ oid => '8436', + descr => 'list of available WAL summary files', + proname => 'pg_available_wal_summaries', prorows => '100', + proretset => 't', provolatile => 'v', proparallel => 's', + prorettype => 'record', proargtypes => '', + proallargtypes => '{int8,pg_lsn,pg_lsn}', + proargmodes => '{o,o,o}', + proargnames => '{tli,start_lsn,end_lsn}', + prosrc => 'pg_available_wal_summaries' }, +{ oid => '8437', + descr => 'contents of a WAL sumamry file', + proname => 'pg_wal_summary_contents', prorows => '100', + proretset => 't', provolatile => 'v', proparallel => 's', + prorettype => 'record', proargtypes => 'int8 pg_lsn pg_lsn', + proallargtypes => '{int8,pg_lsn,pg_lsn,oid,oid,oid,int2,int8,bool}', + proargmodes => '{i,i,i,o,o,o,o,o,o}', + proargnames => '{tli,start_lsn,end_lsn,relfilenode,reltablespace,reldatabase,relforknumber,relblocknumber,is_limit_block}', + prosrc => 'pg_wal_summary_contents' }, + ] diff --git a/src/include/common/blkreftable.h b/src/include/common/blkreftable.h new file mode 100644 index 0000000000..22d9883dc5 --- /dev/null +++ b/src/include/common/blkreftable.h @@ -0,0 +1,120 @@ +/*------------------------------------------------------------------------- + * + * blkreftable.h + * Block reference tables. + * + * A block reference table is used to keep track of which blocks have + * been modified by WAL records within a certain LSN range. + * + * For each relation fork, there is a "limit block number". All existing + * blocks greater than or equal to the limit block number must be + * considered modified; for those less than the limit block number, + * we maintain a bitmap. When a relation fork is created or dropped, + * the limit block number should be set to 0. When it's truncated, + * the limit block number should be set to the length in blocks to + * which it was truncated. + * + * Portions Copyright (c) 2010-2022, PostgreSQL Global Development Group + * + * src/include/common/blkreftable.h + * + *------------------------------------------------------------------------- + */ +#ifndef BLKREFTABLE_H +#define BLKREFTABLE_H + +#include "storage/block.h" +#include "storage/relfilelocator.h" + +/* Magic number for serialization file format. */ +#define BLOCKREFTABLE_MAGIC 0x652b137b + +struct BlockRefTable; +struct BlockRefTableEntry; +struct BlockRefTableReader; +struct BlockRefTableWriter; +typedef struct BlockRefTable BlockRefTable; +typedef struct BlockRefTableEntry BlockRefTableEntry; +typedef struct BlockRefTableReader BlockRefTableReader; +typedef struct BlockRefTableWriter BlockRefTableWriter; + +/* + * The return value of io_callback_fn should be the number of bytes read + * or written. If an error occurs, the functions should report it and + * not return. When used as a write callback, short writes should be retried + * or treated as errors, so that if the callback returns, the return value + * is always the request length. + * + * report_error_fn should not return. + */ +typedef int (*io_callback_fn) (void *callback_arg, void *data, int length); +typedef void (*report_error_fn) (void *calblack_arg, char *msg,...); + + +/* + * Functions for manipulating an entire in-memory block reference table. + */ +extern BlockRefTable *CreateEmptyBlockRefTable(void); +extern void BlockRefTableSetLimitBlock(BlockRefTable *brtab, + const RelFileLocator *rlocator, + ForkNumber forknum, + BlockNumber limit_block); +extern void BlockRefTableMarkBlockModified(BlockRefTable *brtab, + const RelFileLocator *rlocator, + ForkNumber forknum, + BlockNumber blknum); +extern void WriteBlockRefTable(BlockRefTable *brtab, + io_callback_fn write_callback, + void *write_callback_arg); + +extern BlockRefTableEntry *BlockRefTableGetEntry(BlockRefTable *brtab, + const RelFileLocator *rlocator, + ForkNumber forknum, + BlockNumber *limit_block); +extern int BlockRefTableEntryGetBlocks(BlockRefTableEntry *entry, + BlockNumber start_blkno, + BlockNumber stop_blkno, + BlockNumber *blocks, + int nblocks); + +/* + * Functions for reading a block reference table incrementally from disk. + */ +extern BlockRefTableReader *CreateBlockRefTableReader(io_callback_fn read_callback, + void *read_callback_arg, + char *error_filename, + report_error_fn error_callback, + void *error_callback_arg); +extern bool BlockRefTableReaderNextRelation(BlockRefTableReader *reader, + RelFileLocator *rlocator, + ForkNumber *forknum, + BlockNumber *limit_block); +extern unsigned BlockRefTableReaderGetBlocks(BlockRefTableReader *reader, + BlockNumber *blocks, + int nblocks); +extern void DestroyBlockRefTableReader(BlockRefTableReader *reader); + +/* + * Functions for writing a block reference table incrementally to disk. + * + * Note that entries must be written in the proper order, that is, sorted by + * database, then tablespace, then relfilenumber, then fork number. Caller + * is responsible for supplying data in the correct order. If that seems hard, + * use an in-memory BlockRefTable instead. + */ +extern BlockRefTableWriter *CreateBlockRefTableWriter(io_callback_fn write_callback, + void *write_callback_arg); +extern void BlockRefTableWriteEntry(BlockRefTableWriter *writer, + BlockRefTableEntry *entry); +extern void DestroyBlockRefTableWriter(BlockRefTableWriter *writer); + +extern BlockRefTableEntry *CreateBlockRefTableEntry(RelFileLocator rlocator, + ForkNumber forknum); +extern void BlockRefTableEntrySetLimitBlock(BlockRefTableEntry *entry, + BlockNumber limit_block); +extern void BlockRefTableEntryMarkBlockModified(BlockRefTableEntry *entry, + ForkNumber forknum, + BlockNumber blknum); +extern void BlockRefTableFreeEntry(BlockRefTableEntry *entry); + +#endif /* BLKREFTABLE_H */ diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 7232b03e37..042fdc6ca1 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -340,6 +340,7 @@ typedef enum BackendType B_STARTUP, B_WAL_RECEIVER, B_WAL_SENDER, + B_WAL_SUMMARIZER, B_WAL_WRITER, } BackendType; @@ -446,6 +447,7 @@ typedef enum CheckpointerProcess, WalWriterProcess, WalReceiverProcess, + WalSummarizerProcess, NUM_AUXPROCTYPES /* Must be last! */ } AuxProcType; @@ -458,6 +460,7 @@ extern PGDLLIMPORT AuxProcType MyAuxProcType; #define AmCheckpointerProcess() (MyAuxProcType == CheckpointerProcess) #define AmWalWriterProcess() (MyAuxProcType == WalWriterProcess) #define AmWalReceiverProcess() (MyAuxProcType == WalReceiverProcess) +#define AmWalSummarizerProcess() (MyAuxProcType == WalSummarizerProcess) /***************************************************************************** diff --git a/src/include/nodes/replnodes.h b/src/include/nodes/replnodes.h index 4321ba8f86..856491eecd 100644 --- a/src/include/nodes/replnodes.h +++ b/src/include/nodes/replnodes.h @@ -108,4 +108,13 @@ typedef struct TimeLineHistoryCmd TimeLineID timeline; } TimeLineHistoryCmd; +/* ---------------------- + * UPLOAD_MANIFEST command + * ---------------------- + */ +typedef struct UploadManifestCmd +{ + NodeTag type; +} UploadManifestCmd; + #endif /* REPLNODES_H */ diff --git a/src/include/postmaster/walsummarizer.h b/src/include/postmaster/walsummarizer.h new file mode 100644 index 0000000000..7584cb69a7 --- /dev/null +++ b/src/include/postmaster/walsummarizer.h @@ -0,0 +1,31 @@ +/*------------------------------------------------------------------------- + * + * walsummarizer.h + * + * Header file for background WAL summarization process. + * + * Portions Copyright (c) 1996-2022, PostgreSQL Global Development Group + * + * IDENTIFICATION + * src/include/postmaster/walsummarizer.h + * + *------------------------------------------------------------------------- + */ +#ifndef WALSUMMARIZER_H +#define WALSUMMARIZER_H + +#include "access/xlogdefs.h" + +extern int wal_summarize_mb; +extern int wal_summarize_keep_time; + +extern Size WalSummarizerShmemSize(void); +extern void WalSummarizerShmemInit(void); +extern void WalSummarizerMain(void) pg_attribute_noreturn(); + +extern XLogRecPtr GetOldestUnsummarizedLSN(TimeLineID *tli, + bool *lsn_is_exact); +extern void SetWalSummarizerLatch(void); +extern XLogRecPtr WaitForWalSummarization(XLogRecPtr lsn, long timeout); + +#endif diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index ef74f32693..ee55008082 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -417,11 +417,12 @@ extern PGDLLIMPORT PGPROC *PreparedXactProcs; * We set aside some extra PGPROC structures for auxiliary processes, * ie things that aren't full-fledged backends but need shmem access. * - * Background writer, checkpointer, WAL writer and archiver run during normal - * operation. Startup process and WAL receiver also consume 2 slots, but WAL - * writer is launched only after startup has exited, so we only need 5 slots. + * Background writer, checkpointer, WAL writer, WAL summarizer, and archiver + * run during normal operation. Startup process and WAL receiver also consume + * 2 slots, but WAL writer is launched only after startup has exited, so we + * only need 6 slots. */ -#define NUM_AUXILIARY_PROCS 5 +#define NUM_AUXILIARY_PROCS 6 /* configurable options */ extern PGDLLIMPORT int DeadlockTimeout; diff --git a/src/include/utils/guc_tables.h b/src/include/utils/guc_tables.h index d5a0880678..7d3bc0f671 100644 --- a/src/include/utils/guc_tables.h +++ b/src/include/utils/guc_tables.h @@ -72,6 +72,7 @@ enum config_group WAL_RECOVERY, WAL_ARCHIVE_RECOVERY, WAL_RECOVERY_TARGET, + WAL_SUMMARIZATION, REPLICATION_SENDING, REPLICATION_PRIMARY, REPLICATION_STANDBY, diff --git a/src/test/perl/PostgreSQL/Test/Cluster.pm b/src/test/perl/PostgreSQL/Test/Cluster.pm index c3d46c7c70..b711d60fc4 100644 --- a/src/test/perl/PostgreSQL/Test/Cluster.pm +++ b/src/test/perl/PostgreSQL/Test/Cluster.pm @@ -779,6 +779,10 @@ a tar-format backup, pass the name of the tar program to use in the keyword parameter tar_program. Note that tablespace tar files aren't handled here. +To restore from an incremental backup, pass the parameter combine_with_prior +as a reference to an array of prior backup names with which this backup +is to be combined using pg_combinebackup. + Streaming replication can be enabled on this node by passing the keyword parameter has_streaming => 1. This is disabled by default. @@ -816,7 +820,22 @@ sub init_from_backup mkdir $self->archive_dir; my $data_path = $self->data_dir; - if (defined $params{tar_program}) + if (defined $params{combine_with_prior}) + { + my @prior_backups = @{$params{combine_with_prior}}; + my @prior_backup_path; + + for my $prior_backup_name (@prior_backups) + { + push @prior_backup_path, + $root_node->backup_dir . '/' . $prior_backup_name; + } + + local %ENV = $self->_get_env(); + PostgreSQL::Test::Utils::system_or_bail('pg_combinebackup', + @prior_backup_path, $backup_path, '-o', $data_path); + } + elsif (defined $params{tar_program}) { mkdir($data_path); PostgreSQL::Test::Utils::system_or_bail($params{tar_program}, 'xf', diff --git a/src/test/recovery/t/001_stream_rep.pl b/src/test/recovery/t/001_stream_rep.pl index 95f9b0d772..ad11be4664 100644 --- a/src/test/recovery/t/001_stream_rep.pl +++ b/src/test/recovery/t/001_stream_rep.pl @@ -15,6 +15,8 @@ my $node_primary = PostgreSQL::Test::Cluster->new('primary'); $node_primary->init( allows_streaming => 1, auth_extra => [ '--create-role', 'repl_role' ]); +# WAL summarization can postpone WAL recycling, leading to test failures +$node_primary->append_conf('postgresql.conf', "wal_summarize_mb = 0"); $node_primary->start; my $backup_name = 'my_backup'; diff --git a/src/test/recovery/t/019_replslot_limit.pl b/src/test/recovery/t/019_replslot_limit.pl index 7d94f15778..4f52ddbe79 100644 --- a/src/test/recovery/t/019_replslot_limit.pl +++ b/src/test/recovery/t/019_replslot_limit.pl @@ -22,6 +22,7 @@ $node_primary->append_conf( min_wal_size = 2MB max_wal_size = 4MB log_checkpoints = yes +wal_summarize_mb = 0 )); $node_primary->start; $node_primary->safe_psql('postgres', @@ -256,6 +257,7 @@ $node_primary2->append_conf( min_wal_size = 32MB max_wal_size = 32MB log_checkpoints = yes +wal_summarize_mb = 0 )); $node_primary2->start; $node_primary2->safe_psql('postgres', @@ -310,6 +312,7 @@ $node_primary3->append_conf( max_wal_size = 2MB log_checkpoints = yes max_slot_wal_keep_size = 1MB + wal_summarize_mb = 0 )); $node_primary3->start; $node_primary3->safe_psql('postgres', diff --git a/src/test/recovery/t/035_standby_logical_decoding.pl b/src/test/recovery/t/035_standby_logical_decoding.pl index 9c34c0d36c..5fe4faf1be 100644 --- a/src/test/recovery/t/035_standby_logical_decoding.pl +++ b/src/test/recovery/t/035_standby_logical_decoding.pl @@ -250,6 +250,7 @@ $node_primary->append_conf( wal_level = 'logical' max_replication_slots = 4 max_wal_senders = 4 +wal_summarize_mb = 0 }); $node_primary->dump_info; $node_primary->start; diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 06b25617bc..8bccec66c3 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -3993,3 +3993,26 @@ yyscan_t z_stream z_streamp zic_t +BlockRefTable +BlockRefTableBuffer +BlockRefTableEntry +BlockRefTableKey +BlockRefTableReader +BlockRefTableSerializedEntry +BlockRefTableWriter +FileBackupMethod +IncrementalBackupInfo +SummarizerReadLocalXLogPrivate +UploadManifestCmd +WalSummarizerData +WalSummaryFile +WalSummaryIO +backup_file_entry +backup_wal_range +cb_cleanup_dir +cb_options +cb_tablespace +cb_tablespace_mapping +manifest_data +manifest_writer +rfile -- 2.37.1 (Apple Git-137.1) ^ permalink raw reply [nested|flat] 22+ messages in thread
* Re: trying again to get incremental backup @ 2023-10-24 14:53 Peter Eisentraut <[email protected]> parent: Robert Haas <[email protected]> 2 siblings, 0 replies; 22+ messages in thread From: Peter Eisentraut @ 2023-10-24 14:53 UTC (permalink / raw) To: Robert Haas <[email protected]>; Dilip Kumar <[email protected]>; +Cc: Andres Freund <[email protected]>; pgsql-hackers On 04.10.23 22:08, Robert Haas wrote: > - I would like some feedback on the generation of WAL summary files. > Right now, I have it enabled by default, and summaries are kept for a > week. That means that, with no additional setup, you can take an > incremental backup as long as the reference backup was taken in the > last week. File removal is governed by mtimes, so if you change the > mtimes of your summary files or whack your system clock around, weird > things might happen. But obviously this might be inconvenient. Some > people might not want WAL summary files to be generated at all because > they don't care about incremental backup, and other people might want > them retained for longer, and still other people might want them to be > not removed automatically or removed automatically based on some > criteria other than mtime. I don't really know what's best here. I > don't think the default policy that the patches implement is > especially terrible, but it's just something that I made up and I > don't have any real confidence that it's wonderful. The easiest answer is to have it off by default. Let people figure out what works for them. There are various factors like storage, network, server performance, RTO that will determine what combination of full backup, incremental backup, and WAL replay will satisfy someone's requirements. I suppose tests could be set up to determine this to some degree. But we could also start slow and let people figure it out themselves. When pg_basebackup was added, it was also disabled by default. If we think that 7d is a good setting, then I would suggest to consider, like 10d. Otherwise, if you do a weekly incremental backup and you have a time change or a hiccup of some kind one day, you lose your backup sequence. Another possible answer is, like, 400 days? Because why not? What is a reasonable upper limit for this? > - It's regrettable that we don't have incremental JSON parsing; I > think that means anyone who has a backup manifest that is bigger than > 1GB can't use this feature. However, that's also a problem for the > existing backup manifest feature, and as far as I can see, we have no > complaints about it. So maybe people just don't have databases with > enough relations for that to be much of a live issue yet. I'm inclined > to treat this as a non-blocker, It looks like each file entry in the manifest takes about 150 bytes, so 1 GB would allow for 1024**3/150 = 7158278 files. That seems fine for now? > - Right now, I have a hard-coded 60 second timeout for WAL > summarization. If you try to take an incremental backup and the WAL > summaries you need don't show up within 60 seconds, the backup times > out. I think that's a reasonable default, but should it be > configurable? If yes, should that be a GUC or, perhaps better, a > pg_basebackup option? The current user experience of pg_basebackup is that it waits possibly a long time for a checkpoint, and there is an option to make it go faster, but there is no timeout AFAICT. Is this substantially different? Could we just let it wait forever? Also, does waiting for checkpoint and WAL summarization happen in parallel? If so, what if it starts a checkpoint that might take 15 min to complete, and then after 60 seconds it kicks you off because the WAL summarization isn't ready. That might be wasteful. > - I'm curious what people think about the pg_walsummary tool that is > included in 0006. I think it's going to be fairly important for > debugging, but it does feel a little bit bad to add a new binary for > something pretty niche. This seems fine. Is the WAL summary file format documented anywhere in your patch set yet? My only thought was, maybe the file format could be human-readable (more like backup_label) to avoid this. But maybe not. ^ permalink raw reply [nested|flat] 22+ messages in thread
* Re: trying again to get incremental backup @ 2023-10-30 17:45 Robert Haas <[email protected]> parent: Jakub Wartak <[email protected]> 1 sibling, 0 replies; 22+ messages in thread From: Robert Haas @ 2023-10-30 17:45 UTC (permalink / raw) To: Jakub Wartak <[email protected]>; +Cc: Andres Freund <[email protected]>; pgsql-hackers On Thu, Sep 28, 2023 at 6:22 AM Jakub Wartak <[email protected]> wrote: > If that is still an area open for discussion: wouldn't it be better to > just specify LSN as it would allow resyncing standby across major lag > where the WAL to replay would be enormous? Given that we had > primary->standby where standby would be stuck on some LSN, right now > it would be: > 1) calculate backup manifest of desynced 10TB standby (how? using > which tool?) - even if possible, that means reading 10TB of data > instead of just putting a number, isn't it? > 2) backup primary with such incremental backup >= LSN > 3) copy the incremental backup to standby > 4) apply it to the impaired standby > 5) restart the WAL replay As you may be able to tell from the flurry of posts and new patch sets, I'm trying hard to sort out the remaining open items that pertain to this patch set, and I'm now back to thinking about this one. TL;DR: I think the idea has some potential, but there are some pitfalls that I'm not sure how to address. I spent some time looking at how we currently use the data from the backup manifest. Currently, we do two things with it. First, when we're backing up each file, we check whether it's present in the backup manifest and, if not, we back it up in full. This actually feels fairly poor. If it makes any difference at all, then presumably the underlying algorithm is buggy and needs to be fixed. Maybe that should be ripped out altogether or turned into some kind of sanity check that causes a big explosion if it fails. Second, we check whether the WAL ranges reported by the client match up with the timeline history of the server (see PrepareForIncrementalBackup). This set of sanity checks seems fairly important to me, and I'd regret discarding them. I think there's some possibility that they might catch user error, like where somebody promotes multiple standbys and maybe they even get the same timeline on more than one of them, and then confusion might ensue. I also think that there's a real possibility that they might make it easier to track down bugs in my code, even if those bugs aren't necessarily timeline-related. If (or more realistically when) somebody ends up with a corrupted cluster after running pg_combinebackup, we're going to need to figure out whether that corruption is the result of bugs (and if so where they are) or user error (and if so what it was). The most obvious ways of ending up with a corrupted cluster are (1) taking an incremental backup against a prior backup that is not in the history of the server from which the backup is taken or (2) combining an incremental backup the wrong prior backup, so whatever sanity checks we can have that will tend to prevent those kinds of mistakes seem like a really good idea. And those kinds of checks seem relevant here, too. Consider that it wouldn't be valid to use pg_combinebackup to fast-forward a standby server if the incremental backup's backup-end-LSN preceded the standby server's minimum recovery point. Imagine that you have a standby whose last checkpoint's redo location was at LSN 2/48. Being the enterprising DBA that you are, you make a note of that LSN and go take an incremental backup based on it. You then stop the standby server and try to apply the incremental backup to fast-forward the standby. Well, it's possible that in the meanwhile the standby actually caught up, and now has a minimum recovery point that follows the backup-end-LSN of your incremental backup. In that case, you can't legally use that incremental backup to fast-forward that standby, but no code I've yet written would be smart enough to figure that out. Or, maybe you (or some other DBA on your team) got really excited and actually promoted that standby meanwhile, and now it's not even on the same timeline any more. In the "normal" case where you take an incremental backup based on an earlier base backup, these kinds of problems are detectable, and it seems to me that if we want to enable this kind of use case, it would be pretty smart to have a plan to detect similar mistakes here. I don't, currently, but maybe there is one. Another practical problem here is that, right now, pg_combinebackup doesn't have an in-place mode. It knows how to take a bunch of input backups and write out an output backup, but that output backup needs to go into a new, fresh directory (or directories plural, if there are user-defined tablespaces). I had previously considered adding such a mode, but the idea I had at the time wouldn't have worked for this case. I imagined that someone might want to run "pg_combinebackup --in-place full incr" and clobber the contents of the incr directory with the output, basically discarding the incremental backup you took in favor of a full backup that could have been taken at the same point in time. But here, you'd want to clobber the *first* input to pg_combinebackup, not the last one, so if we want to add something like this, the UI needs some thought. One thing that I find quite scary about such a mode is that if you crash mid-way through, you're in a lot of trouble. In the case that I had previous contemplated -- overwrite the last incremental with the reconstructed full backup -- you *might* be able to make it crash safe by writing out the full files for each incremental file, fsyncing everything, then removing all of the incremental files and fsyncing again. The idea would be that if you crash midway through it's OK to just repeat whatever you were trying to do before the crash and if it succeeds the second time then all is well. If, for a given file, there are both incremental and non-incremental versions, then the second attempt should remove and recreate the non-incremental version from the incremental version. If there's only a non-incremental version, it could be that the previous attempt got far enough to remove the incremental file, but in that case the full file that we now have should be the same thing that we would produce if we did the operation now. It all sounds a little scary, but maybe it's OK. And as long as you don't remove the this-is-an-incremental-backup markers from the backup_label file until you've done everything else, you can tell whether you've ever successfully completed the reassembly or not. But if you're using a hypothetical overwrite mode to overwrite the first input rather than the last one, well, it looks like a valid data directory already, and if you replace a bunch of files and then crash, it still does, but it's not any more, really. I'm not sure I've really wrapped my head around all of the cases here, but it does feel like there are some new ways to go wrong. One thing I also realized when thinking about this is that you could probably hose yourself with the patch set as it stands today by taking a full backup, downgrading to wal_level=minimal for a while, doing some WAL-skipping operations, upgrading to a higher WAL-level again, and then taking an incremental backup. I think the solution to that is probably for the WAL summarizer to refuse to run if wal_level=minimal. Then there would be a gap in the summary files which an incremental backup attempt would detect. -- Robert Haas EDB: http://www.enterprisedb.com ^ permalink raw reply [nested|flat] 22+ messages in thread
end of thread, other threads:[~2023-10-30 17:45 UTC | newest] Thread overview: 22+ messages (download: mbox mbox.gz follow: Atom feed) -- links below jump to the message on this page -- 2023-01-31 22:58 [PATCH v7 3/4] restructure archive modules API Nathan Bossart <[email protected]> 2023-06-14 20:40 Re: trying again to get incremental backup Andres Freund <[email protected]> 2023-06-15 07:44 ` Re: trying again to get incremental backup Dilip Kumar <[email protected]> 2023-06-19 13:46 ` Re: trying again to get incremental backup Robert Haas <[email protected]> 2023-06-19 15:51 ` Re: trying again to get incremental backup Andres Freund <[email protected]> 2023-08-30 14:49 ` Re: trying again to get incremental backup Robert Haas <[email protected]> 2023-09-04 12:41 ` Re: trying again to get incremental backup Dilip Kumar <[email protected]> 2023-09-05 15:05 ` Re: trying again to get incremental backup Robert Haas <[email protected]> 2023-09-12 09:56 ` Re: trying again to get incremental backup Dilip Kumar <[email protected]> 2023-10-03 18:21 ` Re: trying again to get incremental backup Robert Haas <[email protected]> 2023-10-04 20:08 ` Re: trying again to get incremental backup Robert Haas <[email protected]> 2023-10-19 16:05 ` Re: trying again to get incremental backup Robert Haas <[email protected]> 2023-10-19 19:18 ` Re: trying again to get incremental backup David Steele <[email protected]> 2023-10-19 20:00 ` Re: trying again to get incremental backup Robert Haas <[email protected]> 2023-10-20 15:30 ` Re: trying again to get incremental backup David Steele <[email protected]> 2023-10-23 15:44 ` Re: trying again to get incremental backup Robert Haas <[email protected]> 2023-10-20 13:20 ` Re: trying again to get incremental backup Jakub Wartak <[email protected]> 2023-10-23 19:34 ` Re: trying again to get incremental backup Robert Haas <[email protected]> 2023-10-24 14:53 ` Re: trying again to get incremental backup Peter Eisentraut <[email protected]> 2023-09-28 10:22 ` Re: trying again to get incremental backup Jakub Wartak <[email protected]> 2023-10-04 19:33 ` Re: trying again to get incremental backup Robert Haas <[email protected]> 2023-10-30 17:45 ` Re: trying again to get incremental backup Robert Haas <[email protected]>
This inbox is served by agora; see mirroring instructions for how to clone and mirror all data and code used for this inbox