public inbox for [email protected]help / color / mirror / Atom feed
[PATCH 01/12] Allow alternate compression methods for wal_compression 102+ messages / 2 participants [nested] [flat]
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 67 ++++++++++++++++--- src/backend/access/transam/xlogreader.c | 64 +++++++++++++++++- src/backend/utils/misc/guc.c | 11 +++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 16 +++++ src/include/access/xlogrecord.h | 11 ++- 10 files changed, 187 insertions(+), 13 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f4d1ce5dea..15da91a8dd 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..a93b33464f 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,8 +33,18 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +/* zlib compressBound is not a macro */ +#define ZLIB_MAX_BLCKSZ BLCKSZ + (BLCKSZ>>12) + (BLCKSZ>>14) + (BLCKSZ>>25) + 13 +#else +#define ZLIB_MAX_BLCKSZ 0 +#endif + /* Buffer size required to store a compressed version of backup block image */ -#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) +#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) + +#define COMPRESS_BUFSIZE Max(PGLZ_MAX_BLCKSZ, ZLIB_MAX_BLCKSZ) /* * For each block reference registered with XLogRegisterBuffer, we fill in @@ -58,7 +68,7 @@ typedef struct * backup block data in XLogRecordAssemble() */ /* buffer to store a compressed version of backup block image */ - char compressed_page[PGLZ_MAX_BLCKSZ]; + char compressed_page[COMPRESS_BUFSIZE]; } registered_buffer; static registered_buffer *registered_buffers; @@ -113,7 +123,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -625,16 +636,26 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, cbimg.hole_length = 0; } + bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; + /* * Try to compress a block image if wal_compression is enabled */ if (wal_compression) { + int compression; + /* The current compression is stored in the WAL record */ + wal_compression_name(wal_compression_method); /* Range check */ + compression = walmethods[wal_compression_method].walmethod; + Assert(compression < (1 << BKPIMAGE_COMPRESS_BITS)); + bimg.bimg_info |= + compression << BKPIMAGE_COMPRESS_OFFSET_BITS; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, + wal_compression_method); } /* @@ -652,8 +673,6 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, rdt_datas_last->next = ®buf->bkp_rdatas[0]; rdt_datas_last = rdt_datas_last->next; - bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; - /* * If WAL consistency checking is enabled for the resource manager * of this WAL record, a full-page image is included in the record @@ -827,7 +846,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +872,42 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = COMPRESS_BUFSIZE; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + len_l = -1; + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..0d8830fc50 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1535,6 +1539,30 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* This is a mapping indexed by wal_compression */ +// XXX: maybe this is better done as a GUC hook to assign the 1) method; and 2) level +struct walcompression walmethods[] = { + {"pglz", WAL_COMPRESSION_PGLZ}, + {"zlib", WAL_COMPRESSION_ZLIB}, +}; + +/* + * Return a statically allocated string associated with the given compression + * method. + * This is here to be visible to frontend tools like pg_rewind. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + if (compression < sizeof(walmethods)/sizeof(*walmethods)) + return walmethods[compression].name; + return "???"; +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1557,9 +1585,41 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { + int compression_method = BKPIMAGE_COMPRESSION(bkpb->bimg_info); /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + compression_method, + wal_compression_name(compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..b000a21557 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,20 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +struct walcompression +{ + char *name; + int walmethod; /* Compression method to be stored in WAL */ +}; + +extern struct walcompression walmethods[]; + +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..7107cf6186 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -147,6 +147,15 @@ typedef struct XLogRecordBlockImageHeader #define BKPIMAGE_IS_COMPRESSED 0x02 /* page image is compressed */ #define BKPIMAGE_APPLY 0x04 /* page image should be restored during * replay */ +#define BKPIMAGE_COMPRESS_METHOD1 0x08 /* bits to encode compression method */ +#define BKPIMAGE_COMPRESS_METHOD2 0x10 /* 0=pglz; 1=zlib; */ + +/* How many bits to shift to extract compression */ +#define BKPIMAGE_COMPRESS_OFFSET_BITS 3 +/* How many bits are for compression */ +#define BKPIMAGE_COMPRESS_BITS 2 +/* Extract the compression from the bimg_info */ +#define BKPIMAGE_COMPRESSION(info) ((info >> BKPIMAGE_COMPRESS_OFFSET_BITS) & ((1<<BKPIMAGE_COMPRESS_BITS) - 1)) /* * Extra header information used when page image has "hole" and -- 2.17.0 --jozmn01XJZjDjM3N Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 3/8] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 ++++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++++++++-- src/backend/access/transam/xlogreader.c | 62 ++++++++++++++++++++++++- src/backend/utils/misc/guc.c | 11 +++++ src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 ++++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 ++-- 9 files changed, 144 insertions(+), 12 deletions(-) diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index e3128564e1..0183589b4d 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..ee73bc3afd 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, + Z_DEFAULT_COMPRESSION); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..143df55fcb 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,38 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l = 0; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..b80759ed45 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ = 0, + WAL_COMPRESSION_ZLIB = 1, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --f0KYrhQ4vYSV2aJu Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0004-wal_compression_method-default-to-zlib.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 3/8] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 ++++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++++++++-- src/backend/access/transam/xlogreader.c | 62 ++++++++++++++++++++++++- src/backend/utils/misc/guc.c | 11 +++++ src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 ++++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 ++-- 9 files changed, 144 insertions(+), 12 deletions(-) diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index e3128564e1..0183589b4d 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..ee73bc3afd 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, + Z_DEFAULT_COMPRESSION); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..143df55fcb 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,38 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l = 0; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..b80759ed45 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ = 0, + WAL_COMPRESSION_ZLIB = 1, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --f0KYrhQ4vYSV2aJu Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0004-wal_compression_method-default-to-zlib.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++-- src/backend/access/transam/xlogreader.c | 63 ++++++++++++++++++- src/backend/utils/misc/guc.c | 11 ++++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 +++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 +-- 11 files changed, 163 insertions(+), 12 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f4d1ce5dea..15da91a8dd 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..34e1227381 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..afca22a26c 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,39 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..d653839b97 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --XsQoSWH+UP9D9v3l Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 3/8] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 ++++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++++++++-- src/backend/access/transam/xlogreader.c | 62 ++++++++++++++++++++++++- src/backend/utils/misc/guc.c | 11 +++++ src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 ++++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 ++-- 9 files changed, 144 insertions(+), 12 deletions(-) diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index e3128564e1..0183589b4d 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..ee73bc3afd 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, + Z_DEFAULT_COMPRESSION); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..143df55fcb 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,38 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l = 0; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..b80759ed45 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ = 0, + WAL_COMPRESSION_ZLIB = 1, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --f0KYrhQ4vYSV2aJu Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0004-wal_compression_method-default-to-zlib.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 67 ++++++++++++++++--- src/backend/access/transam/xlogreader.c | 64 +++++++++++++++++- src/backend/utils/misc/guc.c | 11 +++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 16 +++++ src/include/access/xlogrecord.h | 11 ++- 10 files changed, 187 insertions(+), 13 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f4d1ce5dea..15da91a8dd 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..a93b33464f 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,8 +33,18 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +/* zlib compressBound is not a macro */ +#define ZLIB_MAX_BLCKSZ BLCKSZ + (BLCKSZ>>12) + (BLCKSZ>>14) + (BLCKSZ>>25) + 13 +#else +#define ZLIB_MAX_BLCKSZ 0 +#endif + /* Buffer size required to store a compressed version of backup block image */ -#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) +#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) + +#define COMPRESS_BUFSIZE Max(PGLZ_MAX_BLCKSZ, ZLIB_MAX_BLCKSZ) /* * For each block reference registered with XLogRegisterBuffer, we fill in @@ -58,7 +68,7 @@ typedef struct * backup block data in XLogRecordAssemble() */ /* buffer to store a compressed version of backup block image */ - char compressed_page[PGLZ_MAX_BLCKSZ]; + char compressed_page[COMPRESS_BUFSIZE]; } registered_buffer; static registered_buffer *registered_buffers; @@ -113,7 +123,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -625,16 +636,26 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, cbimg.hole_length = 0; } + bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; + /* * Try to compress a block image if wal_compression is enabled */ if (wal_compression) { + int compression; + /* The current compression is stored in the WAL record */ + wal_compression_name(wal_compression_method); /* Range check */ + compression = walmethods[wal_compression_method].walmethod; + Assert(compression < (1 << BKPIMAGE_COMPRESS_BITS)); + bimg.bimg_info |= + compression << BKPIMAGE_COMPRESS_OFFSET_BITS; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, + wal_compression_method); } /* @@ -652,8 +673,6 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, rdt_datas_last->next = ®buf->bkp_rdatas[0]; rdt_datas_last = rdt_datas_last->next; - bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; - /* * If WAL consistency checking is enabled for the resource manager * of this WAL record, a full-page image is included in the record @@ -827,7 +846,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +872,42 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = COMPRESS_BUFSIZE; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + len_l = -1; + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..0d8830fc50 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1535,6 +1539,30 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* This is a mapping indexed by wal_compression */ +// XXX: maybe this is better done as a GUC hook to assign the 1) method; and 2) level +struct walcompression walmethods[] = { + {"pglz", WAL_COMPRESSION_PGLZ}, + {"zlib", WAL_COMPRESSION_ZLIB}, +}; + +/* + * Return a statically allocated string associated with the given compression + * method. + * This is here to be visible to frontend tools like pg_rewind. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + if (compression < sizeof(walmethods)/sizeof(*walmethods)) + return walmethods[compression].name; + return "???"; +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1557,9 +1585,41 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { + int compression_method = BKPIMAGE_COMPRESSION(bkpb->bimg_info); /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + compression_method, + wal_compression_name(compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..b000a21557 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,20 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +struct walcompression +{ + char *name; + int walmethod; /* Compression method to be stored in WAL */ +}; + +extern struct walcompression walmethods[]; + +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..7107cf6186 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -147,6 +147,15 @@ typedef struct XLogRecordBlockImageHeader #define BKPIMAGE_IS_COMPRESSED 0x02 /* page image is compressed */ #define BKPIMAGE_APPLY 0x04 /* page image should be restored during * replay */ +#define BKPIMAGE_COMPRESS_METHOD1 0x08 /* bits to encode compression method */ +#define BKPIMAGE_COMPRESS_METHOD2 0x10 /* 0=pglz; 1=zlib; */ + +/* How many bits to shift to extract compression */ +#define BKPIMAGE_COMPRESS_OFFSET_BITS 3 +/* How many bits are for compression */ +#define BKPIMAGE_COMPRESS_BITS 2 +/* Extract the compression from the bimg_info */ +#define BKPIMAGE_COMPRESSION(info) ((info >> BKPIMAGE_COMPRESS_OFFSET_BITS) & ((1<<BKPIMAGE_COMPRESS_BITS) - 1)) /* * Extra header information used when page image has "hole" and -- 2.17.0 --jozmn01XJZjDjM3N Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 3/8] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 ++++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++++++++-- src/backend/access/transam/xlogreader.c | 62 ++++++++++++++++++++++++- src/backend/utils/misc/guc.c | 11 +++++ src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 ++++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 ++-- 9 files changed, 144 insertions(+), 12 deletions(-) diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index e3128564e1..0183589b4d 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..ee73bc3afd 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, + Z_DEFAULT_COMPRESSION); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..143df55fcb 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,38 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l = 0; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..b80759ed45 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ = 0, + WAL_COMPRESSION_ZLIB = 1, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --f0KYrhQ4vYSV2aJu Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0004-wal_compression_method-default-to-zlib.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 67 ++++++++++++++++--- src/backend/access/transam/xlogreader.c | 64 +++++++++++++++++- src/backend/utils/misc/guc.c | 11 +++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 16 +++++ src/include/access/xlogrecord.h | 11 ++- 10 files changed, 187 insertions(+), 13 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f4d1ce5dea..15da91a8dd 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..a93b33464f 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,8 +33,18 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +/* zlib compressBound is not a macro */ +#define ZLIB_MAX_BLCKSZ BLCKSZ + (BLCKSZ>>12) + (BLCKSZ>>14) + (BLCKSZ>>25) + 13 +#else +#define ZLIB_MAX_BLCKSZ 0 +#endif + /* Buffer size required to store a compressed version of backup block image */ -#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) +#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) + +#define COMPRESS_BUFSIZE Max(PGLZ_MAX_BLCKSZ, ZLIB_MAX_BLCKSZ) /* * For each block reference registered with XLogRegisterBuffer, we fill in @@ -58,7 +68,7 @@ typedef struct * backup block data in XLogRecordAssemble() */ /* buffer to store a compressed version of backup block image */ - char compressed_page[PGLZ_MAX_BLCKSZ]; + char compressed_page[COMPRESS_BUFSIZE]; } registered_buffer; static registered_buffer *registered_buffers; @@ -113,7 +123,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -625,16 +636,26 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, cbimg.hole_length = 0; } + bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; + /* * Try to compress a block image if wal_compression is enabled */ if (wal_compression) { + int compression; + /* The current compression is stored in the WAL record */ + wal_compression_name(wal_compression_method); /* Range check */ + compression = walmethods[wal_compression_method].walmethod; + Assert(compression < (1 << BKPIMAGE_COMPRESS_BITS)); + bimg.bimg_info |= + compression << BKPIMAGE_COMPRESS_OFFSET_BITS; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, + wal_compression_method); } /* @@ -652,8 +673,6 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, rdt_datas_last->next = ®buf->bkp_rdatas[0]; rdt_datas_last = rdt_datas_last->next; - bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; - /* * If WAL consistency checking is enabled for the resource manager * of this WAL record, a full-page image is included in the record @@ -827,7 +846,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +872,42 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = COMPRESS_BUFSIZE; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + len_l = -1; + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..0d8830fc50 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1535,6 +1539,30 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* This is a mapping indexed by wal_compression */ +// XXX: maybe this is better done as a GUC hook to assign the 1) method; and 2) level +struct walcompression walmethods[] = { + {"pglz", WAL_COMPRESSION_PGLZ}, + {"zlib", WAL_COMPRESSION_ZLIB}, +}; + +/* + * Return a statically allocated string associated with the given compression + * method. + * This is here to be visible to frontend tools like pg_rewind. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + if (compression < sizeof(walmethods)/sizeof(*walmethods)) + return walmethods[compression].name; + return "???"; +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1557,9 +1585,41 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { + int compression_method = BKPIMAGE_COMPRESSION(bkpb->bimg_info); /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + compression_method, + wal_compression_name(compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..b000a21557 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,20 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +struct walcompression +{ + char *name; + int walmethod; /* Compression method to be stored in WAL */ +}; + +extern struct walcompression walmethods[]; + +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..7107cf6186 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -147,6 +147,15 @@ typedef struct XLogRecordBlockImageHeader #define BKPIMAGE_IS_COMPRESSED 0x02 /* page image is compressed */ #define BKPIMAGE_APPLY 0x04 /* page image should be restored during * replay */ +#define BKPIMAGE_COMPRESS_METHOD1 0x08 /* bits to encode compression method */ +#define BKPIMAGE_COMPRESS_METHOD2 0x10 /* 0=pglz; 1=zlib; */ + +/* How many bits to shift to extract compression */ +#define BKPIMAGE_COMPRESS_OFFSET_BITS 3 +/* How many bits are for compression */ +#define BKPIMAGE_COMPRESS_BITS 2 +/* Extract the compression from the bimg_info */ +#define BKPIMAGE_COMPRESSION(info) ((info >> BKPIMAGE_COMPRESS_OFFSET_BITS) & ((1<<BKPIMAGE_COMPRESS_BITS) - 1)) /* * Extra header information used when page image has "hole" and -- 2.17.0 --jozmn01XJZjDjM3N Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++-- src/backend/access/transam/xlogreader.c | 63 ++++++++++++++++++- src/backend/utils/misc/guc.c | 11 ++++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 +++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 +-- 11 files changed, 163 insertions(+), 12 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f4d1ce5dea..15da91a8dd 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..34e1227381 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..afca22a26c 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,39 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..d653839b97 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --XsQoSWH+UP9D9v3l Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++-- src/backend/access/transam/xlogreader.c | 63 ++++++++++++++++++- src/backend/utils/misc/guc.c | 11 ++++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 +++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 +-- 11 files changed, 163 insertions(+), 12 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f4d1ce5dea..15da91a8dd 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..34e1227381 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..afca22a26c 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,39 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..d653839b97 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --XsQoSWH+UP9D9v3l Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 3/8] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 ++++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++++++++-- src/backend/access/transam/xlogreader.c | 62 ++++++++++++++++++++++++- src/backend/utils/misc/guc.c | 11 +++++ src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 ++++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 ++-- 9 files changed, 144 insertions(+), 12 deletions(-) diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index e3128564e1..0183589b4d 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..ee73bc3afd 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, + Z_DEFAULT_COMPRESSION); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..143df55fcb 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,38 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l = 0; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..b80759ed45 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ = 0, + WAL_COMPRESSION_ZLIB = 1, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --f0KYrhQ4vYSV2aJu Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0004-wal_compression_method-default-to-zlib.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 3/8] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 ++++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++++++++-- src/backend/access/transam/xlogreader.c | 62 ++++++++++++++++++++++++- src/backend/utils/misc/guc.c | 11 +++++ src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 ++++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 ++-- 9 files changed, 144 insertions(+), 12 deletions(-) diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index e3128564e1..0183589b4d 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..ee73bc3afd 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, + Z_DEFAULT_COMPRESSION); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..143df55fcb 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,38 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l = 0; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..b80759ed45 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ = 0, + WAL_COMPRESSION_ZLIB = 1, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --f0KYrhQ4vYSV2aJu Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0004-wal_compression_method-default-to-zlib.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++-- src/backend/access/transam/xlogreader.c | 63 ++++++++++++++++++- src/backend/utils/misc/guc.c | 11 ++++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 +++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 +-- 11 files changed, 163 insertions(+), 12 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index e04250f4e9..04192b7add 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..34e1227381 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..afca22a26c 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,39 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..d653839b97 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --0qVF/w3MHQqLSynd Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 67 ++++++++++++++++--- src/backend/access/transam/xlogreader.c | 64 +++++++++++++++++- src/backend/utils/misc/guc.c | 11 +++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 16 +++++ src/include/access/xlogrecord.h | 11 ++- 10 files changed, 187 insertions(+), 13 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f4d1ce5dea..15da91a8dd 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..a93b33464f 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,8 +33,18 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +/* zlib compressBound is not a macro */ +#define ZLIB_MAX_BLCKSZ BLCKSZ + (BLCKSZ>>12) + (BLCKSZ>>14) + (BLCKSZ>>25) + 13 +#else +#define ZLIB_MAX_BLCKSZ 0 +#endif + /* Buffer size required to store a compressed version of backup block image */ -#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) +#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) + +#define COMPRESS_BUFSIZE Max(PGLZ_MAX_BLCKSZ, ZLIB_MAX_BLCKSZ) /* * For each block reference registered with XLogRegisterBuffer, we fill in @@ -58,7 +68,7 @@ typedef struct * backup block data in XLogRecordAssemble() */ /* buffer to store a compressed version of backup block image */ - char compressed_page[PGLZ_MAX_BLCKSZ]; + char compressed_page[COMPRESS_BUFSIZE]; } registered_buffer; static registered_buffer *registered_buffers; @@ -113,7 +123,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -625,16 +636,26 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, cbimg.hole_length = 0; } + bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; + /* * Try to compress a block image if wal_compression is enabled */ if (wal_compression) { + int compression; + /* The current compression is stored in the WAL record */ + wal_compression_name(wal_compression_method); /* Range check */ + compression = walmethods[wal_compression_method].walmethod; + Assert(compression < (1 << BKPIMAGE_COMPRESS_BITS)); + bimg.bimg_info |= + compression << BKPIMAGE_COMPRESS_OFFSET_BITS; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, + wal_compression_method); } /* @@ -652,8 +673,6 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, rdt_datas_last->next = ®buf->bkp_rdatas[0]; rdt_datas_last = rdt_datas_last->next; - bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; - /* * If WAL consistency checking is enabled for the resource manager * of this WAL record, a full-page image is included in the record @@ -827,7 +846,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +872,42 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = COMPRESS_BUFSIZE; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + len_l = -1; + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..0d8830fc50 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1535,6 +1539,30 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* This is a mapping indexed by wal_compression */ +// XXX: maybe this is better done as a GUC hook to assign the 1) method; and 2) level +struct walcompression walmethods[] = { + {"pglz", WAL_COMPRESSION_PGLZ}, + {"zlib", WAL_COMPRESSION_ZLIB}, +}; + +/* + * Return a statically allocated string associated with the given compression + * method. + * This is here to be visible to frontend tools like pg_rewind. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + if (compression < sizeof(walmethods)/sizeof(*walmethods)) + return walmethods[compression].name; + return "???"; +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1557,9 +1585,41 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { + int compression_method = BKPIMAGE_COMPRESSION(bkpb->bimg_info); /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + compression_method, + wal_compression_name(compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..b000a21557 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,20 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +struct walcompression +{ + char *name; + int walmethod; /* Compression method to be stored in WAL */ +}; + +extern struct walcompression walmethods[]; + +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..7107cf6186 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -147,6 +147,15 @@ typedef struct XLogRecordBlockImageHeader #define BKPIMAGE_IS_COMPRESSED 0x02 /* page image is compressed */ #define BKPIMAGE_APPLY 0x04 /* page image should be restored during * replay */ +#define BKPIMAGE_COMPRESS_METHOD1 0x08 /* bits to encode compression method */ +#define BKPIMAGE_COMPRESS_METHOD2 0x10 /* 0=pglz; 1=zlib; */ + +/* How many bits to shift to extract compression */ +#define BKPIMAGE_COMPRESS_OFFSET_BITS 3 +/* How many bits are for compression */ +#define BKPIMAGE_COMPRESS_BITS 2 +/* Extract the compression from the bimg_info */ +#define BKPIMAGE_COMPRESSION(info) ((info >> BKPIMAGE_COMPRESS_OFFSET_BITS) & ((1<<BKPIMAGE_COMPRESS_BITS) - 1)) /* * Extra header information used when page image has "hole" and -- 2.17.0 --jozmn01XJZjDjM3N Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH v7 1/9] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 67 ++++++++++++++++--- src/backend/access/transam/xlogreader.c | 64 +++++++++++++++++- src/backend/utils/misc/guc.c | 11 +++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 16 +++++ src/include/access/xlogrecord.h | 11 ++- 10 files changed, 187 insertions(+), 13 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index 7e32b0686c..70effa6345 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3137,6 +3137,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 8d163f190f..b7f8e12aea 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 32b4cc84e7..5ab07621d6 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,8 +33,18 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +/* zlib compressBound is not a macro */ +#define ZLIB_MAX_BLCKSZ BLCKSZ + (BLCKSZ>>12) + (BLCKSZ>>14) + (BLCKSZ>>25) + 13 +#else +#define ZLIB_MAX_BLCKSZ 0 +#endif + /* Buffer size required to store a compressed version of backup block image */ -#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) +#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) + +#define COMPRESS_BUFSIZE Max(PGLZ_MAX_BLCKSZ, ZLIB_MAX_BLCKSZ) /* * For each block reference registered with XLogRegisterBuffer, we fill in @@ -58,7 +68,7 @@ typedef struct * backup block data in XLogRecordAssemble() */ /* buffer to store a compressed version of backup block image */ - char compressed_page[PGLZ_MAX_BLCKSZ]; + char compressed_page[COMPRESS_BUFSIZE]; } registered_buffer; static registered_buffer *registered_buffers; @@ -113,7 +123,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -625,16 +636,26 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, cbimg.hole_length = 0; } + bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; + /* * Try to compress a block image if wal_compression is enabled */ if (wal_compression) { + int compression; + /* The current compression is stored in the WAL record */ + wal_compression_name(wal_compression_method); /* Range check */ + compression = walmethods[wal_compression_method].walmethod; + Assert(compression < (1 << BKPIMAGE_COMPRESS_BITS)); + bimg.bimg_info |= + compression << BKPIMAGE_COMPRESS_OFFSET_BITS; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, + wal_compression_method); } /* @@ -652,8 +673,6 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, rdt_datas_last->next = ®buf->bkp_rdatas[0]; rdt_datas_last = rdt_datas_last->next; - bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; - /* * If WAL consistency checking is enabled for the resource manager * of this WAL record, a full-page image is included in the record @@ -827,7 +846,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +872,42 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = COMPRESS_BUFSIZE; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + len_l = -1; + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..0d8830fc50 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1535,6 +1539,30 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* This is a mapping indexed by wal_compression */ +// XXX: maybe this is better done as a GUC hook to assign the 1) method; and 2) level +struct walcompression walmethods[] = { + {"pglz", WAL_COMPRESSION_PGLZ}, + {"zlib", WAL_COMPRESSION_ZLIB}, +}; + +/* + * Return a statically allocated string associated with the given compression + * method. + * This is here to be visible to frontend tools like pg_rewind. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + if (compression < sizeof(walmethods)/sizeof(*walmethods)) + return walmethods[compression].name; + return "???"; +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1557,9 +1585,41 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { + int compression_method = BKPIMAGE_COMPRESSION(bkpb->bimg_info); /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + compression_method, + wal_compression_name(compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index ee731044b6..99932582ba 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -548,6 +548,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4825,6 +4826,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SUSET, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index 6e36e4c2ef..baed4d9228 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -219,6 +219,7 @@ #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_init_zero = on # zero-fill new WAL files #wal_recycle = on # recycle WAL files #wal_buffers = -1 # min 32kB, -1 sets based on shared_buffers diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 77187c12be..2f3886431a 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 26a743b6b6..ceca0f7189 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,20 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +struct walcompression +{ + char *name; + int walmethod; /* Compression method to be stored in WAL */ +}; + +extern struct walcompression walmethods[]; + +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..7107cf6186 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -147,6 +147,15 @@ typedef struct XLogRecordBlockImageHeader #define BKPIMAGE_IS_COMPRESSED 0x02 /* page image is compressed */ #define BKPIMAGE_APPLY 0x04 /* page image should be restored during * replay */ +#define BKPIMAGE_COMPRESS_METHOD1 0x08 /* bits to encode compression method */ +#define BKPIMAGE_COMPRESS_METHOD2 0x10 /* 0=pglz; 1=zlib; */ + +/* How many bits to shift to extract compression */ +#define BKPIMAGE_COMPRESS_OFFSET_BITS 3 +/* How many bits are for compression */ +#define BKPIMAGE_COMPRESS_BITS 2 +/* Extract the compression from the bimg_info */ +#define BKPIMAGE_COMPRESSION(info) ((info >> BKPIMAGE_COMPRESS_OFFSET_BITS) & ((1<<BKPIMAGE_COMPRESS_BITS) - 1)) /* * Extra header information used when page image has "hole" and -- 2.17.0 --JYK4vJDZwFMowpUq Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="v7-0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 3/8] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 ++++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++++++++-- src/backend/access/transam/xlogreader.c | 62 ++++++++++++++++++++++++- src/backend/utils/misc/guc.c | 11 +++++ src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 ++++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 ++-- 9 files changed, 144 insertions(+), 12 deletions(-) diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index e3128564e1..0183589b4d 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..ee73bc3afd 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, + Z_DEFAULT_COMPRESSION); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..143df55fcb 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,38 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l = 0; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..b80759ed45 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ = 0, + WAL_COMPRESSION_ZLIB = 1, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --f0KYrhQ4vYSV2aJu Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0004-wal_compression_method-default-to-zlib.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 3/8] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 ++++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++++++++-- src/backend/access/transam/xlogreader.c | 62 ++++++++++++++++++++++++- src/backend/utils/misc/guc.c | 11 +++++ src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 ++++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 ++-- 9 files changed, 144 insertions(+), 12 deletions(-) diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index e3128564e1..0183589b4d 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..ee73bc3afd 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, + Z_DEFAULT_COMPRESSION); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..143df55fcb 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,38 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l = 0; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..b80759ed45 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ = 0, + WAL_COMPRESSION_ZLIB = 1, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --f0KYrhQ4vYSV2aJu Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0004-wal_compression_method-default-to-zlib.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++-- src/backend/access/transam/xlogreader.c | 63 ++++++++++++++++++- src/backend/utils/misc/guc.c | 11 ++++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 +++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 +-- 11 files changed, 163 insertions(+), 12 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index e04250f4e9..04192b7add 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..34e1227381 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..afca22a26c 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,39 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..d653839b97 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --0qVF/w3MHQqLSynd Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH v9 1/9] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 9 +- doc/src/sgml/installation.sgml | 4 +- src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 14 ++- src/backend/access/transam/xloginsert.c | 65 ++++++++++-- src/backend/access/transam/xlogreader.c | 99 ++++++++++++++++--- src/backend/utils/misc/guc.c | 21 ++-- src/backend/utils/misc/postgresql.conf.sample | 2 +- src/bin/pg_waldump/pg_waldump.c | 13 ++- src/include/access/xlog.h | 2 +- src/include/access/xlog_internal.h | 10 ++ src/include/access/xlogrecord.h | 15 ++- 12 files changed, 207 insertions(+), 49 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index aa3e178240..1df56d8034 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3127,23 +3127,26 @@ include_dir 'conf.d' </varlistentry> <varlistentry id="guc-wal-compression" xreflabel="wal_compression"> - <term><varname>wal_compression</varname> (<type>boolean</type>) + <term><varname>wal_compression</varname> (<type>enum</type>) <indexterm> <primary><varname>wal_compression</varname> configuration parameter</primary> </indexterm> </term> <listitem> <para> - When this parameter is <literal>on</literal>, the <productname>PostgreSQL</productname> + This parameter enables compression of WAL using the specified + compression method. + When enabled, the <productname>PostgreSQL</productname> server compresses full page images written to WAL when <xref linkend="guc-full-page-writes"/> is on or during a base backup. A compressed page image will be decompressed during WAL replay. + The supported methods are pglz and zlib. The default value is <literal>off</literal>. Only superusers can change this setting. </para> <para> - Turning this parameter on can reduce the WAL volume without + Enabling compression can reduce the WAL volume without increasing the risk of unrecoverable data corruption, but at the cost of some extra CPU spent on the compression during WAL logging and on the decompression during WAL replay. diff --git a/doc/src/sgml/installation.sgml b/doc/src/sgml/installation.sgml index 3c0aa118c7..073d5089f7 100644 --- a/doc/src/sgml/installation.sgml +++ b/doc/src/sgml/installation.sgml @@ -147,7 +147,7 @@ su - postgres specify the <option>--without-zlib</option> option to <filename>configure</filename>. Using this option disables support for compressed archives in <application>pg_dump</application> and - <application>pg_restore</application>. + <application>pg_restore</application>, and compressed WAL. </para> </listitem> </itemizedlist> @@ -1236,7 +1236,7 @@ build-postgresql: Prevents use of the <application>Zlib</application> library. This disables support for compressed archives in <application>pg_dump</application> - and <application>pg_restore</application>. + and <application>pg_restore</application> and compressed WAL. </para> </listitem> </varlistentry> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 17eeff0720..1ccc51575a 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -98,7 +98,7 @@ char *XLogArchiveCommand = NULL; bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; -bool wal_compression = false; +int wal_compression = WAL_COMPRESSION_NONE; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -10470,7 +10470,17 @@ xlog_block_info(StringInfo buf, XLogReaderState *record) rnode.spcNode, rnode.dbNode, rnode.relNode, blk); if (XLogRecHasBlockImage(record, block_id)) - appendStringInfoString(buf, " FPW"); + { + int compression = + BKPIMAGE_IS_COMPRESSED(record->blocks[block_id].bimg_info) ? + BKPIMAGE_COMPRESSION(record->blocks[block_id].bimg_info) : -1; + if (compression == -1) + appendStringInfoString(buf, " FPW"); + else + appendStringInfo(buf, " FPW, compression method %d/%s", + compression, wal_compression_name(compression)); + } + } } diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 32b4cc84e7..4f81f19c49 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,8 +33,18 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +/* zlib compressBound is not a macro */ +#define ZLIB_MAX_BLCKSZ BLCKSZ + (BLCKSZ>>12) + (BLCKSZ>>14) + (BLCKSZ>>25) + 13 +#else +#define ZLIB_MAX_BLCKSZ 0 +#endif + /* Buffer size required to store a compressed version of backup block image */ -#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) +#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) + +#define COMPRESS_BUFSIZE Max(PGLZ_MAX_BLCKSZ, ZLIB_MAX_BLCKSZ) /* * For each block reference registered with XLogRegisterBuffer, we fill in @@ -58,7 +68,7 @@ typedef struct * backup block data in XLogRecordAssemble() */ /* buffer to store a compressed version of backup block image */ - char compressed_page[PGLZ_MAX_BLCKSZ]; + char compressed_page[COMPRESS_BUFSIZE]; } registered_buffer; static registered_buffer *registered_buffers; @@ -113,7 +123,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -628,13 +639,14 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, /* * Try to compress a block image if wal_compression is enabled */ - if (wal_compression) + if (wal_compression != WAL_COMPRESSION_NONE) { is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, + wal_compression); } /* @@ -665,8 +677,13 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, if (is_compressed) { + /* The current compression is stored in the WAL record */ + wal_compression_name(wal_compression); /* Range check */ + Assert(wal_compression < (1 << BKPIMAGE_COMPRESS_BITS)); + bimg.length = compressed_len; - bimg.bimg_info |= BKPIMAGE_IS_COMPRESSED; + bimg.bimg_info |= + wal_compression << BKPIMAGE_COMPRESS_OFFSET_BITS; rdt_datas_last->data = regbuf->compressed_page; rdt_datas_last->len = compressed_len; @@ -827,7 +844,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +870,42 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = COMPRESS_BUFSIZE; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + len_l = -1; + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d/%s", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..3a922caaeb 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -26,6 +26,7 @@ #include "catalog/pg_control.h" #include "common/pg_lzcompress.h" #include "replication/origin.h" +#include "utils/guc.h" #ifndef FRONTEND #include "miscadmin.h" @@ -33,6 +34,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -50,6 +55,29 @@ static void WALOpenSegmentInit(WALOpenSegment *seg, WALSegmentContext *segcxt, /* size of the buffer allocated for error message. */ #define MAX_ERRORMSG_LEN 1000 +/* + * Accept the likely variants for none and pglz, for compatibility with old + * server versions where wal_compression was a boolean. + */ +const struct config_enum_entry wal_compression_options[] = { + {"off", WAL_COMPRESSION_NONE, false}, + {"none", WAL_COMPRESSION_NONE, false}, + {"false", WAL_COMPRESSION_NONE, true}, + {"no", WAL_COMPRESSION_NONE, true}, + {"0", WAL_COMPRESSION_NONE, true}, + {"pglz", WAL_COMPRESSION_PGLZ, false}, + {"true", WAL_COMPRESSION_PGLZ, true}, + {"yes", WAL_COMPRESSION_PGLZ, true}, + {"on", WAL_COMPRESSION_PGLZ, true}, + {"1", WAL_COMPRESSION_PGLZ, true}, + +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + + {NULL, 0, false} +}; + /* * Construct a string in state->errormsg_buf explaining what's wrong with * the current record being read. @@ -1290,7 +1318,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); - if (blk->bimg_info & BKPIMAGE_IS_COMPRESSED) + if (BKPIMAGE_IS_COMPRESSED(blk->bimg_info)) { if (blk->bimg_info & BKPIMAGE_HAS_HOLE) COPY_HEADER_FIELD(&blk->hole_length, sizeof(uint16)); @@ -1335,29 +1363,28 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) } /* - * cross-check that bimg_len < BLCKSZ if the IS_COMPRESSED - * flag is set. + * cross-check that bimg_len < BLCKSZ if it's compressed */ - if ((blk->bimg_info & BKPIMAGE_IS_COMPRESSED) && + if (BKPIMAGE_IS_COMPRESSED(blk->bimg_info) && blk->bimg_len == BLCKSZ) { report_invalid_record(state, - "BKPIMAGE_IS_COMPRESSED set, but block image length %u at %X/%X", + "BKPIMAGE_IS_COMPRESSED, but block image length %u at %X/%X", (unsigned int) blk->bimg_len, LSN_FORMAT_ARGS(state->ReadRecPtr)); goto err; } /* - * cross-check that bimg_len = BLCKSZ if neither HAS_HOLE nor - * IS_COMPRESSED flag is set. + * cross-check that bimg_len = BLCKSZ if neither HAS_HOLE is + * set nor IS_COMPRESSED(). */ if (!(blk->bimg_info & BKPIMAGE_HAS_HOLE) && - !(blk->bimg_info & BKPIMAGE_IS_COMPRESSED) && + !BKPIMAGE_IS_COMPRESSED(blk->bimg_info) && blk->bimg_len != BLCKSZ) { report_invalid_record(state, - "neither BKPIMAGE_HAS_HOLE nor BKPIMAGE_IS_COMPRESSED set, but block image length is %u at %X/%X", + "neither BKPIMAGE_HAS_HOLE nor BKPIMAGE_IS_COMPRESSED, but block image length is %u at %X/%X", (unsigned int) blk->data_len, LSN_FORMAT_ARGS(state->ReadRecPtr)); goto err; @@ -1535,6 +1562,22 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. + */ +const char * +wal_compression_name(WalCompression compression) +{ + for (int i=0; wal_compression_options[i].name != NULL; ++i) + { + if (wal_compression_options[i].val == compression) + return wal_compression_options[i].name; + } + + return "???"; +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1555,11 +1598,43 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) bkpb = &record->blocks[block_id]; ptr = bkpb->bkp_image; - if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) + if (BKPIMAGE_IS_COMPRESSED(bkpb->bimg_info)) { + int compression_method = BKPIMAGE_COMPRESSION(bkpb->bimg_info); /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + compression_method, + wal_compression_name(compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 68b62d523d..ce1149bed5 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -548,6 +548,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -1304,16 +1305,6 @@ static struct config_bool ConfigureNamesBool[] = NULL, NULL, NULL }, - { - {"wal_compression", PGC_SUSET, WAL_SETTINGS, - gettext_noop("Compresses full-page writes written in WAL file."), - NULL - }, - &wal_compression, - false, - NULL, NULL, NULL - }, - { {"wal_init_zero", PGC_SUSET, WAL_SETTINGS, gettext_noop("Writes zeroes to new WAL files before first use."), @@ -4825,6 +4816,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression", PGC_SUSET, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression, + WAL_COMPRESSION_NONE, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index ddbb6dc2be..61786c6e07 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -218,7 +218,7 @@ #full_page_writes = on # recover from partial page writes #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) -#wal_compression = off # enable compression of full-page writes +#wal_compression = off # enable compression of full-page writes: off, pglz, zlib #wal_init_zero = on # zero-fill new WAL files #wal_recycle = on # recycle WAL files #wal_buffers = -1 # min 32kB, -1 sets based on shared_buffers diff --git a/src/bin/pg_waldump/pg_waldump.c b/src/bin/pg_waldump/pg_waldump.c index f8b8afe4a7..1cd71ac2f7 100644 --- a/src/bin/pg_waldump/pg_waldump.c +++ b/src/bin/pg_waldump/pg_waldump.c @@ -537,18 +537,21 @@ XLogDumpDisplayRecord(XLogDumpConfig *config, XLogReaderState *record) blk); if (XLogRecHasBlockImage(record, block_id)) { - if (record->blocks[block_id].bimg_info & - BKPIMAGE_IS_COMPRESSED) + if (BKPIMAGE_IS_COMPRESSED(record->blocks[block_id].bimg_info)) { + int compression = BKPIMAGE_COMPRESSION( + record->blocks[block_id].bimg_info); + printf(" (FPW%s); hole: offset: %u, length: %u, " - "compression saved: %u", + "compression method %d/%s, saved: %u", XLogRecBlockImageApply(record, block_id) ? "" : " for WAL verification", record->blocks[block_id].hole_offset, record->blocks[block_id].hole_length, + compression, wal_compression_name(compression), BLCKSZ - - record->blocks[block_id].hole_length - - record->blocks[block_id].bimg_len); + record->blocks[block_id].hole_length - + record->blocks[block_id].bimg_len); } else { diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 77187c12be..e8b2c53784 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -116,7 +116,7 @@ extern char *XLogArchiveCommand; extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; -extern bool wal_compression; +extern int wal_compression; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 26a743b6b6..8b740af66d 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,14 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +/* These are the compression IDs written into bimg_info */ +typedef enum WalCompression +{ + WAL_COMPRESSION_NONE, + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..2a60c0fb92 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -144,9 +144,18 @@ typedef struct XLogRecordBlockImageHeader /* Information stored in bimg_info */ #define BKPIMAGE_HAS_HOLE 0x01 /* page image has "hole" */ -#define BKPIMAGE_IS_COMPRESSED 0x02 /* page image is compressed */ -#define BKPIMAGE_APPLY 0x04 /* page image should be restored during +#define BKPIMAGE_APPLY 0x02 /* page image should be restored during * replay */ +#define BKPIMAGE_COMPRESS_METHOD1 0x04 /* bits to encode compression method */ +#define BKPIMAGE_COMPRESS_METHOD2 0x08 /* 0=none, 1=pglz, 2=zlib */ + +/* How many bits to shift to extract compression */ +#define BKPIMAGE_COMPRESS_OFFSET_BITS 2 +/* How many bits are for compression */ +#define BKPIMAGE_COMPRESS_BITS 2 +/* Extract the compression from the bimg_info */ +#define BKPIMAGE_COMPRESSION(info) ((info >> BKPIMAGE_COMPRESS_OFFSET_BITS) & ((1<<BKPIMAGE_COMPRESS_BITS) - 1)) +#define BKPIMAGE_IS_COMPRESSED(info) (BKPIMAGE_COMPRESSION(info) != 0) /* * Extra header information used when page image has "hole" and -- 2.17.0 --neYutvxvOLaeuPCA Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="v9-0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 67 ++++++++++++++++--- src/backend/access/transam/xlogreader.c | 64 +++++++++++++++++- src/backend/utils/misc/guc.c | 11 +++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 16 +++++ src/include/access/xlogrecord.h | 11 ++- 10 files changed, 187 insertions(+), 13 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f4d1ce5dea..15da91a8dd 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..a93b33464f 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,8 +33,18 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +/* zlib compressBound is not a macro */ +#define ZLIB_MAX_BLCKSZ BLCKSZ + (BLCKSZ>>12) + (BLCKSZ>>14) + (BLCKSZ>>25) + 13 +#else +#define ZLIB_MAX_BLCKSZ 0 +#endif + /* Buffer size required to store a compressed version of backup block image */ -#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) +#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) + +#define COMPRESS_BUFSIZE Max(PGLZ_MAX_BLCKSZ, ZLIB_MAX_BLCKSZ) /* * For each block reference registered with XLogRegisterBuffer, we fill in @@ -58,7 +68,7 @@ typedef struct * backup block data in XLogRecordAssemble() */ /* buffer to store a compressed version of backup block image */ - char compressed_page[PGLZ_MAX_BLCKSZ]; + char compressed_page[COMPRESS_BUFSIZE]; } registered_buffer; static registered_buffer *registered_buffers; @@ -113,7 +123,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -625,16 +636,26 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, cbimg.hole_length = 0; } + bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; + /* * Try to compress a block image if wal_compression is enabled */ if (wal_compression) { + int compression; + /* The current compression is stored in the WAL record */ + wal_compression_name(wal_compression_method); /* Range check */ + compression = walmethods[wal_compression_method].walmethod; + Assert(compression < (1 << BKPIMAGE_COMPRESS_BITS)); + bimg.bimg_info |= + compression << BKPIMAGE_COMPRESS_OFFSET_BITS; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, + wal_compression_method); } /* @@ -652,8 +673,6 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, rdt_datas_last->next = ®buf->bkp_rdatas[0]; rdt_datas_last = rdt_datas_last->next; - bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; - /* * If WAL consistency checking is enabled for the resource manager * of this WAL record, a full-page image is included in the record @@ -827,7 +846,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +872,42 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = COMPRESS_BUFSIZE; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + len_l = -1; + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..0d8830fc50 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1535,6 +1539,30 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* This is a mapping indexed by wal_compression */ +// XXX: maybe this is better done as a GUC hook to assign the 1) method; and 2) level +struct walcompression walmethods[] = { + {"pglz", WAL_COMPRESSION_PGLZ}, + {"zlib", WAL_COMPRESSION_ZLIB}, +}; + +/* + * Return a statically allocated string associated with the given compression + * method. + * This is here to be visible to frontend tools like pg_rewind. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + if (compression < sizeof(walmethods)/sizeof(*walmethods)) + return walmethods[compression].name; + return "???"; +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1557,9 +1585,41 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { + int compression_method = BKPIMAGE_COMPRESSION(bkpb->bimg_info); /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + compression_method, + wal_compression_name(compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..b000a21557 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,20 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +struct walcompression +{ + char *name; + int walmethod; /* Compression method to be stored in WAL */ +}; + +extern struct walcompression walmethods[]; + +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..7107cf6186 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -147,6 +147,15 @@ typedef struct XLogRecordBlockImageHeader #define BKPIMAGE_IS_COMPRESSED 0x02 /* page image is compressed */ #define BKPIMAGE_APPLY 0x04 /* page image should be restored during * replay */ +#define BKPIMAGE_COMPRESS_METHOD1 0x08 /* bits to encode compression method */ +#define BKPIMAGE_COMPRESS_METHOD2 0x10 /* 0=pglz; 1=zlib; */ + +/* How many bits to shift to extract compression */ +#define BKPIMAGE_COMPRESS_OFFSET_BITS 3 +/* How many bits are for compression */ +#define BKPIMAGE_COMPRESS_BITS 2 +/* Extract the compression from the bimg_info */ +#define BKPIMAGE_COMPRESSION(info) ((info >> BKPIMAGE_COMPRESS_OFFSET_BITS) & ((1<<BKPIMAGE_COMPRESS_BITS) - 1)) /* * Extra header information used when page image has "hole" and -- 2.17.0 --jozmn01XJZjDjM3N Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++-- src/backend/access/transam/xlogreader.c | 63 ++++++++++++++++++- src/backend/utils/misc/guc.c | 11 ++++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 +++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 +-- 11 files changed, 163 insertions(+), 12 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index e04250f4e9..04192b7add 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..34e1227381 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..afca22a26c 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,39 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..d653839b97 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --0qVF/w3MHQqLSynd Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++-- src/backend/access/transam/xlogreader.c | 63 ++++++++++++++++++- src/backend/utils/misc/guc.c | 11 ++++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 +++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 +-- 11 files changed, 163 insertions(+), 12 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index e04250f4e9..04192b7add 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..34e1227381 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..afca22a26c 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,39 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..d653839b97 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --0qVF/w3MHQqLSynd Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++-- src/backend/access/transam/xlogreader.c | 63 ++++++++++++++++++- src/backend/utils/misc/guc.c | 11 ++++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 +++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 +-- 11 files changed, 163 insertions(+), 12 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f4d1ce5dea..15da91a8dd 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..34e1227381 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..afca22a26c 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,39 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..d653839b97 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --XsQoSWH+UP9D9v3l Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 67 ++++++++++++++++--- src/backend/access/transam/xlogreader.c | 64 +++++++++++++++++- src/backend/utils/misc/guc.c | 11 +++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 16 +++++ src/include/access/xlogrecord.h | 11 ++- 10 files changed, 187 insertions(+), 13 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f4d1ce5dea..15da91a8dd 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..a93b33464f 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,8 +33,18 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +/* zlib compressBound is not a macro */ +#define ZLIB_MAX_BLCKSZ BLCKSZ + (BLCKSZ>>12) + (BLCKSZ>>14) + (BLCKSZ>>25) + 13 +#else +#define ZLIB_MAX_BLCKSZ 0 +#endif + /* Buffer size required to store a compressed version of backup block image */ -#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) +#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) + +#define COMPRESS_BUFSIZE Max(PGLZ_MAX_BLCKSZ, ZLIB_MAX_BLCKSZ) /* * For each block reference registered with XLogRegisterBuffer, we fill in @@ -58,7 +68,7 @@ typedef struct * backup block data in XLogRecordAssemble() */ /* buffer to store a compressed version of backup block image */ - char compressed_page[PGLZ_MAX_BLCKSZ]; + char compressed_page[COMPRESS_BUFSIZE]; } registered_buffer; static registered_buffer *registered_buffers; @@ -113,7 +123,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -625,16 +636,26 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, cbimg.hole_length = 0; } + bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; + /* * Try to compress a block image if wal_compression is enabled */ if (wal_compression) { + int compression; + /* The current compression is stored in the WAL record */ + wal_compression_name(wal_compression_method); /* Range check */ + compression = walmethods[wal_compression_method].walmethod; + Assert(compression < (1 << BKPIMAGE_COMPRESS_BITS)); + bimg.bimg_info |= + compression << BKPIMAGE_COMPRESS_OFFSET_BITS; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, + wal_compression_method); } /* @@ -652,8 +673,6 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, rdt_datas_last->next = ®buf->bkp_rdatas[0]; rdt_datas_last = rdt_datas_last->next; - bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; - /* * If WAL consistency checking is enabled for the resource manager * of this WAL record, a full-page image is included in the record @@ -827,7 +846,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +872,42 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = COMPRESS_BUFSIZE; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + len_l = -1; + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..0d8830fc50 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1535,6 +1539,30 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* This is a mapping indexed by wal_compression */ +// XXX: maybe this is better done as a GUC hook to assign the 1) method; and 2) level +struct walcompression walmethods[] = { + {"pglz", WAL_COMPRESSION_PGLZ}, + {"zlib", WAL_COMPRESSION_ZLIB}, +}; + +/* + * Return a statically allocated string associated with the given compression + * method. + * This is here to be visible to frontend tools like pg_rewind. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + if (compression < sizeof(walmethods)/sizeof(*walmethods)) + return walmethods[compression].name; + return "???"; +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1557,9 +1585,41 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { + int compression_method = BKPIMAGE_COMPRESSION(bkpb->bimg_info); /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + compression_method, + wal_compression_name(compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..b000a21557 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,20 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +struct walcompression +{ + char *name; + int walmethod; /* Compression method to be stored in WAL */ +}; + +extern struct walcompression walmethods[]; + +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..7107cf6186 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -147,6 +147,15 @@ typedef struct XLogRecordBlockImageHeader #define BKPIMAGE_IS_COMPRESSED 0x02 /* page image is compressed */ #define BKPIMAGE_APPLY 0x04 /* page image should be restored during * replay */ +#define BKPIMAGE_COMPRESS_METHOD1 0x08 /* bits to encode compression method */ +#define BKPIMAGE_COMPRESS_METHOD2 0x10 /* 0=pglz; 1=zlib; */ + +/* How many bits to shift to extract compression */ +#define BKPIMAGE_COMPRESS_OFFSET_BITS 3 +/* How many bits are for compression */ +#define BKPIMAGE_COMPRESS_BITS 2 +/* Extract the compression from the bimg_info */ +#define BKPIMAGE_COMPRESSION(info) ((info >> BKPIMAGE_COMPRESS_OFFSET_BITS) & ((1<<BKPIMAGE_COMPRESS_BITS) - 1)) /* * Extra header information used when page image has "hole" and -- 2.17.0 --jozmn01XJZjDjM3N Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 3/8] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 ++++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++++++++-- src/backend/access/transam/xlogreader.c | 62 ++++++++++++++++++++++++- src/backend/utils/misc/guc.c | 11 +++++ src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 ++++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 ++-- 9 files changed, 144 insertions(+), 12 deletions(-) diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index e3128564e1..0183589b4d 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..ee73bc3afd 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, + Z_DEFAULT_COMPRESSION); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..143df55fcb 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,38 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l = 0; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..b80759ed45 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ = 0, + WAL_COMPRESSION_ZLIB = 1, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --f0KYrhQ4vYSV2aJu Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0004-wal_compression_method-default-to-zlib.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++-- src/backend/access/transam/xlogreader.c | 63 ++++++++++++++++++- src/backend/utils/misc/guc.c | 11 ++++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 +++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 +-- 11 files changed, 163 insertions(+), 12 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index e04250f4e9..04192b7add 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..34e1227381 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..afca22a26c 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,39 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..d653839b97 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --0qVF/w3MHQqLSynd Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 3/8] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 ++++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++++++++-- src/backend/access/transam/xlogreader.c | 62 ++++++++++++++++++++++++- src/backend/utils/misc/guc.c | 11 +++++ src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 ++++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 ++-- 9 files changed, 144 insertions(+), 12 deletions(-) diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index e3128564e1..0183589b4d 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..ee73bc3afd 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, + Z_DEFAULT_COMPRESSION); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..143df55fcb 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,38 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l = 0; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..b80759ed45 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ = 0, + WAL_COMPRESSION_ZLIB = 1, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --f0KYrhQ4vYSV2aJu Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0004-wal_compression_method-default-to-zlib.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++-- src/backend/access/transam/xlogreader.c | 63 ++++++++++++++++++- src/backend/utils/misc/guc.c | 11 ++++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 +++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 +-- 11 files changed, 163 insertions(+), 12 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index e04250f4e9..04192b7add 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..34e1227381 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..afca22a26c 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,39 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..d653839b97 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --0qVF/w3MHQqLSynd Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 67 ++++++++++++++++--- src/backend/access/transam/xlogreader.c | 64 +++++++++++++++++- src/backend/utils/misc/guc.c | 11 +++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 16 +++++ src/include/access/xlogrecord.h | 11 ++- 10 files changed, 187 insertions(+), 13 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f4d1ce5dea..15da91a8dd 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..a93b33464f 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,8 +33,18 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +/* zlib compressBound is not a macro */ +#define ZLIB_MAX_BLCKSZ BLCKSZ + (BLCKSZ>>12) + (BLCKSZ>>14) + (BLCKSZ>>25) + 13 +#else +#define ZLIB_MAX_BLCKSZ 0 +#endif + /* Buffer size required to store a compressed version of backup block image */ -#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) +#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) + +#define COMPRESS_BUFSIZE Max(PGLZ_MAX_BLCKSZ, ZLIB_MAX_BLCKSZ) /* * For each block reference registered with XLogRegisterBuffer, we fill in @@ -58,7 +68,7 @@ typedef struct * backup block data in XLogRecordAssemble() */ /* buffer to store a compressed version of backup block image */ - char compressed_page[PGLZ_MAX_BLCKSZ]; + char compressed_page[COMPRESS_BUFSIZE]; } registered_buffer; static registered_buffer *registered_buffers; @@ -113,7 +123,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -625,16 +636,26 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, cbimg.hole_length = 0; } + bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; + /* * Try to compress a block image if wal_compression is enabled */ if (wal_compression) { + int compression; + /* The current compression is stored in the WAL record */ + wal_compression_name(wal_compression_method); /* Range check */ + compression = walmethods[wal_compression_method].walmethod; + Assert(compression < (1 << BKPIMAGE_COMPRESS_BITS)); + bimg.bimg_info |= + compression << BKPIMAGE_COMPRESS_OFFSET_BITS; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, + wal_compression_method); } /* @@ -652,8 +673,6 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, rdt_datas_last->next = ®buf->bkp_rdatas[0]; rdt_datas_last = rdt_datas_last->next; - bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; - /* * If WAL consistency checking is enabled for the resource manager * of this WAL record, a full-page image is included in the record @@ -827,7 +846,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +872,42 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = COMPRESS_BUFSIZE; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + len_l = -1; + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..0d8830fc50 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1535,6 +1539,30 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* This is a mapping indexed by wal_compression */ +// XXX: maybe this is better done as a GUC hook to assign the 1) method; and 2) level +struct walcompression walmethods[] = { + {"pglz", WAL_COMPRESSION_PGLZ}, + {"zlib", WAL_COMPRESSION_ZLIB}, +}; + +/* + * Return a statically allocated string associated with the given compression + * method. + * This is here to be visible to frontend tools like pg_rewind. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + if (compression < sizeof(walmethods)/sizeof(*walmethods)) + return walmethods[compression].name; + return "???"; +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1557,9 +1585,41 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { + int compression_method = BKPIMAGE_COMPRESSION(bkpb->bimg_info); /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + compression_method, + wal_compression_name(compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..b000a21557 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,20 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +struct walcompression +{ + char *name; + int walmethod; /* Compression method to be stored in WAL */ +}; + +extern struct walcompression walmethods[]; + +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..7107cf6186 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -147,6 +147,15 @@ typedef struct XLogRecordBlockImageHeader #define BKPIMAGE_IS_COMPRESSED 0x02 /* page image is compressed */ #define BKPIMAGE_APPLY 0x04 /* page image should be restored during * replay */ +#define BKPIMAGE_COMPRESS_METHOD1 0x08 /* bits to encode compression method */ +#define BKPIMAGE_COMPRESS_METHOD2 0x10 /* 0=pglz; 1=zlib; */ + +/* How many bits to shift to extract compression */ +#define BKPIMAGE_COMPRESS_OFFSET_BITS 3 +/* How many bits are for compression */ +#define BKPIMAGE_COMPRESS_BITS 2 +/* Extract the compression from the bimg_info */ +#define BKPIMAGE_COMPRESSION(info) ((info >> BKPIMAGE_COMPRESS_OFFSET_BITS) & ((1<<BKPIMAGE_COMPRESS_BITS) - 1)) /* * Extra header information used when page image has "hole" and -- 2.17.0 --jozmn01XJZjDjM3N Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 3/8] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 ++++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++++++++-- src/backend/access/transam/xlogreader.c | 62 ++++++++++++++++++++++++- src/backend/utils/misc/guc.c | 11 +++++ src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 ++++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 ++-- 9 files changed, 144 insertions(+), 12 deletions(-) diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index e3128564e1..0183589b4d 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..ee73bc3afd 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, + Z_DEFAULT_COMPRESSION); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..143df55fcb 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,38 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l = 0; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..b80759ed45 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ = 0, + WAL_COMPRESSION_ZLIB = 1, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --f0KYrhQ4vYSV2aJu Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0004-wal_compression_method-default-to-zlib.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++-- src/backend/access/transam/xlogreader.c | 63 ++++++++++++++++++- src/backend/utils/misc/guc.c | 11 ++++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 +++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 +-- 11 files changed, 163 insertions(+), 12 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f4d1ce5dea..15da91a8dd 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..34e1227381 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..afca22a26c 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,39 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..d653839b97 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --XsQoSWH+UP9D9v3l Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++-- src/backend/access/transam/xlogreader.c | 63 ++++++++++++++++++- src/backend/utils/misc/guc.c | 11 ++++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 +++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 +-- 11 files changed, 163 insertions(+), 12 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f4d1ce5dea..15da91a8dd 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..34e1227381 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..afca22a26c 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,39 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..d653839b97 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --XsQoSWH+UP9D9v3l Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++-- src/backend/access/transam/xlogreader.c | 63 ++++++++++++++++++- src/backend/utils/misc/guc.c | 11 ++++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 +++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 +-- 11 files changed, 163 insertions(+), 12 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f4d1ce5dea..15da91a8dd 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..34e1227381 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..afca22a26c 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,39 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..d653839b97 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --XsQoSWH+UP9D9v3l Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++-- src/backend/access/transam/xlogreader.c | 63 ++++++++++++++++++- src/backend/utils/misc/guc.c | 11 ++++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 +++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 +-- 11 files changed, 163 insertions(+), 12 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f4d1ce5dea..15da91a8dd 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..34e1227381 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..afca22a26c 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,39 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..d653839b97 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --XsQoSWH+UP9D9v3l Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++-- src/backend/access/transam/xlogreader.c | 63 ++++++++++++++++++- src/backend/utils/misc/guc.c | 11 ++++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 +++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 +-- 11 files changed, 163 insertions(+), 12 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index e04250f4e9..04192b7add 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..34e1227381 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..afca22a26c 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,39 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..d653839b97 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --0qVF/w3MHQqLSynd Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++-- src/backend/access/transam/xlogreader.c | 63 ++++++++++++++++++- src/backend/utils/misc/guc.c | 11 ++++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 +++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 +-- 11 files changed, 163 insertions(+), 12 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index e04250f4e9..04192b7add 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..34e1227381 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..afca22a26c 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,39 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..d653839b97 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --0qVF/w3MHQqLSynd Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++-- src/backend/access/transam/xlogreader.c | 63 ++++++++++++++++++- src/backend/utils/misc/guc.c | 11 ++++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 +++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 +-- 11 files changed, 163 insertions(+), 12 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f4d1ce5dea..15da91a8dd 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..34e1227381 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..afca22a26c 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,39 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..d653839b97 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --XsQoSWH+UP9D9v3l Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++-- src/backend/access/transam/xlogreader.c | 63 ++++++++++++++++++- src/backend/utils/misc/guc.c | 11 ++++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 +++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 +-- 11 files changed, 163 insertions(+), 12 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index e04250f4e9..04192b7add 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..34e1227381 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..afca22a26c 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,39 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..d653839b97 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --0qVF/w3MHQqLSynd Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++-- src/backend/access/transam/xlogreader.c | 63 ++++++++++++++++++- src/backend/utils/misc/guc.c | 11 ++++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 +++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 +-- 11 files changed, 163 insertions(+), 12 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index e04250f4e9..04192b7add 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..34e1227381 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..afca22a26c 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,39 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..d653839b97 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --0qVF/w3MHQqLSynd Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 3/8] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 ++++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++++++++-- src/backend/access/transam/xlogreader.c | 62 ++++++++++++++++++++++++- src/backend/utils/misc/guc.c | 11 +++++ src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 ++++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 ++-- 9 files changed, 144 insertions(+), 12 deletions(-) diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index e3128564e1..0183589b4d 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..ee73bc3afd 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, + Z_DEFAULT_COMPRESSION); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..143df55fcb 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,38 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l = 0; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..b80759ed45 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ = 0, + WAL_COMPRESSION_ZLIB = 1, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --f0KYrhQ4vYSV2aJu Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0004-wal_compression_method-default-to-zlib.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++-- src/backend/access/transam/xlogreader.c | 63 ++++++++++++++++++- src/backend/utils/misc/guc.c | 11 ++++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 +++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 +-- 11 files changed, 163 insertions(+), 12 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index e04250f4e9..04192b7add 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..34e1227381 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..afca22a26c 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,39 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..d653839b97 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --0qVF/w3MHQqLSynd Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 67 ++++++++++++++++--- src/backend/access/transam/xlogreader.c | 64 +++++++++++++++++- src/backend/utils/misc/guc.c | 11 +++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 16 +++++ src/include/access/xlogrecord.h | 11 ++- 10 files changed, 187 insertions(+), 13 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f4d1ce5dea..15da91a8dd 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..a93b33464f 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,8 +33,18 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +/* zlib compressBound is not a macro */ +#define ZLIB_MAX_BLCKSZ BLCKSZ + (BLCKSZ>>12) + (BLCKSZ>>14) + (BLCKSZ>>25) + 13 +#else +#define ZLIB_MAX_BLCKSZ 0 +#endif + /* Buffer size required to store a compressed version of backup block image */ -#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) +#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) + +#define COMPRESS_BUFSIZE Max(PGLZ_MAX_BLCKSZ, ZLIB_MAX_BLCKSZ) /* * For each block reference registered with XLogRegisterBuffer, we fill in @@ -58,7 +68,7 @@ typedef struct * backup block data in XLogRecordAssemble() */ /* buffer to store a compressed version of backup block image */ - char compressed_page[PGLZ_MAX_BLCKSZ]; + char compressed_page[COMPRESS_BUFSIZE]; } registered_buffer; static registered_buffer *registered_buffers; @@ -113,7 +123,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -625,16 +636,26 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, cbimg.hole_length = 0; } + bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; + /* * Try to compress a block image if wal_compression is enabled */ if (wal_compression) { + int compression; + /* The current compression is stored in the WAL record */ + wal_compression_name(wal_compression_method); /* Range check */ + compression = walmethods[wal_compression_method].walmethod; + Assert(compression < (1 << BKPIMAGE_COMPRESS_BITS)); + bimg.bimg_info |= + compression << BKPIMAGE_COMPRESS_OFFSET_BITS; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, + wal_compression_method); } /* @@ -652,8 +673,6 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, rdt_datas_last->next = ®buf->bkp_rdatas[0]; rdt_datas_last = rdt_datas_last->next; - bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; - /* * If WAL consistency checking is enabled for the resource manager * of this WAL record, a full-page image is included in the record @@ -827,7 +846,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +872,42 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = COMPRESS_BUFSIZE; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + len_l = -1; + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..0d8830fc50 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1535,6 +1539,30 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* This is a mapping indexed by wal_compression */ +// XXX: maybe this is better done as a GUC hook to assign the 1) method; and 2) level +struct walcompression walmethods[] = { + {"pglz", WAL_COMPRESSION_PGLZ}, + {"zlib", WAL_COMPRESSION_ZLIB}, +}; + +/* + * Return a statically allocated string associated with the given compression + * method. + * This is here to be visible to frontend tools like pg_rewind. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + if (compression < sizeof(walmethods)/sizeof(*walmethods)) + return walmethods[compression].name; + return "???"; +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1557,9 +1585,41 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { + int compression_method = BKPIMAGE_COMPRESSION(bkpb->bimg_info); /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + compression_method, + wal_compression_name(compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..b000a21557 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,20 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +struct walcompression +{ + char *name; + int walmethod; /* Compression method to be stored in WAL */ +}; + +extern struct walcompression walmethods[]; + +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..7107cf6186 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -147,6 +147,15 @@ typedef struct XLogRecordBlockImageHeader #define BKPIMAGE_IS_COMPRESSED 0x02 /* page image is compressed */ #define BKPIMAGE_APPLY 0x04 /* page image should be restored during * replay */ +#define BKPIMAGE_COMPRESS_METHOD1 0x08 /* bits to encode compression method */ +#define BKPIMAGE_COMPRESS_METHOD2 0x10 /* 0=pglz; 1=zlib; */ + +/* How many bits to shift to extract compression */ +#define BKPIMAGE_COMPRESS_OFFSET_BITS 3 +/* How many bits are for compression */ +#define BKPIMAGE_COMPRESS_BITS 2 +/* Extract the compression from the bimg_info */ +#define BKPIMAGE_COMPRESSION(info) ((info >> BKPIMAGE_COMPRESS_OFFSET_BITS) & ((1<<BKPIMAGE_COMPRESS_BITS) - 1)) /* * Extra header information used when page image has "hole" and -- 2.17.0 --jozmn01XJZjDjM3N Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++-- src/backend/access/transam/xlogreader.c | 63 ++++++++++++++++++- src/backend/utils/misc/guc.c | 11 ++++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 +++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 +-- 11 files changed, 163 insertions(+), 12 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index e04250f4e9..04192b7add 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..34e1227381 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..afca22a26c 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,39 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..d653839b97 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --0qVF/w3MHQqLSynd Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++-- src/backend/access/transam/xlogreader.c | 63 ++++++++++++++++++- src/backend/utils/misc/guc.c | 11 ++++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 +++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 +-- 11 files changed, 163 insertions(+), 12 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index e04250f4e9..04192b7add 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..34e1227381 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..afca22a26c 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,39 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..d653839b97 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --0qVF/w3MHQqLSynd Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 67 ++++++++++++++++--- src/backend/access/transam/xlogreader.c | 64 +++++++++++++++++- src/backend/utils/misc/guc.c | 11 +++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 16 +++++ src/include/access/xlogrecord.h | 11 ++- 10 files changed, 187 insertions(+), 13 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f4d1ce5dea..15da91a8dd 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..a93b33464f 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,8 +33,18 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +/* zlib compressBound is not a macro */ +#define ZLIB_MAX_BLCKSZ BLCKSZ + (BLCKSZ>>12) + (BLCKSZ>>14) + (BLCKSZ>>25) + 13 +#else +#define ZLIB_MAX_BLCKSZ 0 +#endif + /* Buffer size required to store a compressed version of backup block image */ -#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) +#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) + +#define COMPRESS_BUFSIZE Max(PGLZ_MAX_BLCKSZ, ZLIB_MAX_BLCKSZ) /* * For each block reference registered with XLogRegisterBuffer, we fill in @@ -58,7 +68,7 @@ typedef struct * backup block data in XLogRecordAssemble() */ /* buffer to store a compressed version of backup block image */ - char compressed_page[PGLZ_MAX_BLCKSZ]; + char compressed_page[COMPRESS_BUFSIZE]; } registered_buffer; static registered_buffer *registered_buffers; @@ -113,7 +123,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -625,16 +636,26 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, cbimg.hole_length = 0; } + bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; + /* * Try to compress a block image if wal_compression is enabled */ if (wal_compression) { + int compression; + /* The current compression is stored in the WAL record */ + wal_compression_name(wal_compression_method); /* Range check */ + compression = walmethods[wal_compression_method].walmethod; + Assert(compression < (1 << BKPIMAGE_COMPRESS_BITS)); + bimg.bimg_info |= + compression << BKPIMAGE_COMPRESS_OFFSET_BITS; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, + wal_compression_method); } /* @@ -652,8 +673,6 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, rdt_datas_last->next = ®buf->bkp_rdatas[0]; rdt_datas_last = rdt_datas_last->next; - bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; - /* * If WAL consistency checking is enabled for the resource manager * of this WAL record, a full-page image is included in the record @@ -827,7 +846,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +872,42 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = COMPRESS_BUFSIZE; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + len_l = -1; + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..0d8830fc50 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1535,6 +1539,30 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* This is a mapping indexed by wal_compression */ +// XXX: maybe this is better done as a GUC hook to assign the 1) method; and 2) level +struct walcompression walmethods[] = { + {"pglz", WAL_COMPRESSION_PGLZ}, + {"zlib", WAL_COMPRESSION_ZLIB}, +}; + +/* + * Return a statically allocated string associated with the given compression + * method. + * This is here to be visible to frontend tools like pg_rewind. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + if (compression < sizeof(walmethods)/sizeof(*walmethods)) + return walmethods[compression].name; + return "???"; +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1557,9 +1585,41 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { + int compression_method = BKPIMAGE_COMPRESSION(bkpb->bimg_info); /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + compression_method, + wal_compression_name(compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..b000a21557 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,20 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +struct walcompression +{ + char *name; + int walmethod; /* Compression method to be stored in WAL */ +}; + +extern struct walcompression walmethods[]; + +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..7107cf6186 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -147,6 +147,15 @@ typedef struct XLogRecordBlockImageHeader #define BKPIMAGE_IS_COMPRESSED 0x02 /* page image is compressed */ #define BKPIMAGE_APPLY 0x04 /* page image should be restored during * replay */ +#define BKPIMAGE_COMPRESS_METHOD1 0x08 /* bits to encode compression method */ +#define BKPIMAGE_COMPRESS_METHOD2 0x10 /* 0=pglz; 1=zlib; */ + +/* How many bits to shift to extract compression */ +#define BKPIMAGE_COMPRESS_OFFSET_BITS 3 +/* How many bits are for compression */ +#define BKPIMAGE_COMPRESS_BITS 2 +/* Extract the compression from the bimg_info */ +#define BKPIMAGE_COMPRESSION(info) ((info >> BKPIMAGE_COMPRESS_OFFSET_BITS) & ((1<<BKPIMAGE_COMPRESS_BITS) - 1)) /* * Extra header information used when page image has "hole" and -- 2.17.0 --jozmn01XJZjDjM3N Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++-- src/backend/access/transam/xlogreader.c | 63 ++++++++++++++++++- src/backend/utils/misc/guc.c | 11 ++++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 +++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 +-- 11 files changed, 163 insertions(+), 12 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index e04250f4e9..04192b7add 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..34e1227381 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..afca22a26c 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,39 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..d653839b97 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --0qVF/w3MHQqLSynd Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 67 ++++++++++++++++--- src/backend/access/transam/xlogreader.c | 64 +++++++++++++++++- src/backend/utils/misc/guc.c | 11 +++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 16 +++++ src/include/access/xlogrecord.h | 11 ++- 10 files changed, 187 insertions(+), 13 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f4d1ce5dea..15da91a8dd 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..a93b33464f 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,8 +33,18 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +/* zlib compressBound is not a macro */ +#define ZLIB_MAX_BLCKSZ BLCKSZ + (BLCKSZ>>12) + (BLCKSZ>>14) + (BLCKSZ>>25) + 13 +#else +#define ZLIB_MAX_BLCKSZ 0 +#endif + /* Buffer size required to store a compressed version of backup block image */ -#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) +#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) + +#define COMPRESS_BUFSIZE Max(PGLZ_MAX_BLCKSZ, ZLIB_MAX_BLCKSZ) /* * For each block reference registered with XLogRegisterBuffer, we fill in @@ -58,7 +68,7 @@ typedef struct * backup block data in XLogRecordAssemble() */ /* buffer to store a compressed version of backup block image */ - char compressed_page[PGLZ_MAX_BLCKSZ]; + char compressed_page[COMPRESS_BUFSIZE]; } registered_buffer; static registered_buffer *registered_buffers; @@ -113,7 +123,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -625,16 +636,26 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, cbimg.hole_length = 0; } + bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; + /* * Try to compress a block image if wal_compression is enabled */ if (wal_compression) { + int compression; + /* The current compression is stored in the WAL record */ + wal_compression_name(wal_compression_method); /* Range check */ + compression = walmethods[wal_compression_method].walmethod; + Assert(compression < (1 << BKPIMAGE_COMPRESS_BITS)); + bimg.bimg_info |= + compression << BKPIMAGE_COMPRESS_OFFSET_BITS; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, + wal_compression_method); } /* @@ -652,8 +673,6 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, rdt_datas_last->next = ®buf->bkp_rdatas[0]; rdt_datas_last = rdt_datas_last->next; - bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; - /* * If WAL consistency checking is enabled for the resource manager * of this WAL record, a full-page image is included in the record @@ -827,7 +846,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +872,42 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = COMPRESS_BUFSIZE; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + len_l = -1; + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..0d8830fc50 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1535,6 +1539,30 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* This is a mapping indexed by wal_compression */ +// XXX: maybe this is better done as a GUC hook to assign the 1) method; and 2) level +struct walcompression walmethods[] = { + {"pglz", WAL_COMPRESSION_PGLZ}, + {"zlib", WAL_COMPRESSION_ZLIB}, +}; + +/* + * Return a statically allocated string associated with the given compression + * method. + * This is here to be visible to frontend tools like pg_rewind. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + if (compression < sizeof(walmethods)/sizeof(*walmethods)) + return walmethods[compression].name; + return "???"; +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1557,9 +1585,41 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { + int compression_method = BKPIMAGE_COMPRESSION(bkpb->bimg_info); /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + compression_method, + wal_compression_name(compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..b000a21557 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,20 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +struct walcompression +{ + char *name; + int walmethod; /* Compression method to be stored in WAL */ +}; + +extern struct walcompression walmethods[]; + +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..7107cf6186 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -147,6 +147,15 @@ typedef struct XLogRecordBlockImageHeader #define BKPIMAGE_IS_COMPRESSED 0x02 /* page image is compressed */ #define BKPIMAGE_APPLY 0x04 /* page image should be restored during * replay */ +#define BKPIMAGE_COMPRESS_METHOD1 0x08 /* bits to encode compression method */ +#define BKPIMAGE_COMPRESS_METHOD2 0x10 /* 0=pglz; 1=zlib; */ + +/* How many bits to shift to extract compression */ +#define BKPIMAGE_COMPRESS_OFFSET_BITS 3 +/* How many bits are for compression */ +#define BKPIMAGE_COMPRESS_BITS 2 +/* Extract the compression from the bimg_info */ +#define BKPIMAGE_COMPRESSION(info) ((info >> BKPIMAGE_COMPRESS_OFFSET_BITS) & ((1<<BKPIMAGE_COMPRESS_BITS) - 1)) /* * Extra header information used when page image has "hole" and -- 2.17.0 --jozmn01XJZjDjM3N Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 67 ++++++++++++++++--- src/backend/access/transam/xlogreader.c | 64 +++++++++++++++++- src/backend/utils/misc/guc.c | 11 +++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 16 +++++ src/include/access/xlogrecord.h | 11 ++- 10 files changed, 187 insertions(+), 13 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f4d1ce5dea..15da91a8dd 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..a93b33464f 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,8 +33,18 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +/* zlib compressBound is not a macro */ +#define ZLIB_MAX_BLCKSZ BLCKSZ + (BLCKSZ>>12) + (BLCKSZ>>14) + (BLCKSZ>>25) + 13 +#else +#define ZLIB_MAX_BLCKSZ 0 +#endif + /* Buffer size required to store a compressed version of backup block image */ -#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) +#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) + +#define COMPRESS_BUFSIZE Max(PGLZ_MAX_BLCKSZ, ZLIB_MAX_BLCKSZ) /* * For each block reference registered with XLogRegisterBuffer, we fill in @@ -58,7 +68,7 @@ typedef struct * backup block data in XLogRecordAssemble() */ /* buffer to store a compressed version of backup block image */ - char compressed_page[PGLZ_MAX_BLCKSZ]; + char compressed_page[COMPRESS_BUFSIZE]; } registered_buffer; static registered_buffer *registered_buffers; @@ -113,7 +123,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -625,16 +636,26 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, cbimg.hole_length = 0; } + bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; + /* * Try to compress a block image if wal_compression is enabled */ if (wal_compression) { + int compression; + /* The current compression is stored in the WAL record */ + wal_compression_name(wal_compression_method); /* Range check */ + compression = walmethods[wal_compression_method].walmethod; + Assert(compression < (1 << BKPIMAGE_COMPRESS_BITS)); + bimg.bimg_info |= + compression << BKPIMAGE_COMPRESS_OFFSET_BITS; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, + wal_compression_method); } /* @@ -652,8 +673,6 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, rdt_datas_last->next = ®buf->bkp_rdatas[0]; rdt_datas_last = rdt_datas_last->next; - bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; - /* * If WAL consistency checking is enabled for the resource manager * of this WAL record, a full-page image is included in the record @@ -827,7 +846,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +872,42 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = COMPRESS_BUFSIZE; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + len_l = -1; + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..0d8830fc50 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1535,6 +1539,30 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* This is a mapping indexed by wal_compression */ +// XXX: maybe this is better done as a GUC hook to assign the 1) method; and 2) level +struct walcompression walmethods[] = { + {"pglz", WAL_COMPRESSION_PGLZ}, + {"zlib", WAL_COMPRESSION_ZLIB}, +}; + +/* + * Return a statically allocated string associated with the given compression + * method. + * This is here to be visible to frontend tools like pg_rewind. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + if (compression < sizeof(walmethods)/sizeof(*walmethods)) + return walmethods[compression].name; + return "???"; +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1557,9 +1585,41 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { + int compression_method = BKPIMAGE_COMPRESSION(bkpb->bimg_info); /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + compression_method, + wal_compression_name(compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..b000a21557 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,20 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +struct walcompression +{ + char *name; + int walmethod; /* Compression method to be stored in WAL */ +}; + +extern struct walcompression walmethods[]; + +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..7107cf6186 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -147,6 +147,15 @@ typedef struct XLogRecordBlockImageHeader #define BKPIMAGE_IS_COMPRESSED 0x02 /* page image is compressed */ #define BKPIMAGE_APPLY 0x04 /* page image should be restored during * replay */ +#define BKPIMAGE_COMPRESS_METHOD1 0x08 /* bits to encode compression method */ +#define BKPIMAGE_COMPRESS_METHOD2 0x10 /* 0=pglz; 1=zlib; */ + +/* How many bits to shift to extract compression */ +#define BKPIMAGE_COMPRESS_OFFSET_BITS 3 +/* How many bits are for compression */ +#define BKPIMAGE_COMPRESS_BITS 2 +/* Extract the compression from the bimg_info */ +#define BKPIMAGE_COMPRESSION(info) ((info >> BKPIMAGE_COMPRESS_OFFSET_BITS) & ((1<<BKPIMAGE_COMPRESS_BITS) - 1)) /* * Extra header information used when page image has "hole" and -- 2.17.0 --jozmn01XJZjDjM3N Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++-- src/backend/access/transam/xlogreader.c | 63 ++++++++++++++++++- src/backend/utils/misc/guc.c | 11 ++++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 +++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 +-- 11 files changed, 163 insertions(+), 12 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f4d1ce5dea..15da91a8dd 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..34e1227381 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..afca22a26c 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,39 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..d653839b97 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --XsQoSWH+UP9D9v3l Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 67 ++++++++++++++++--- src/backend/access/transam/xlogreader.c | 64 +++++++++++++++++- src/backend/utils/misc/guc.c | 11 +++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 16 +++++ src/include/access/xlogrecord.h | 11 ++- 10 files changed, 187 insertions(+), 13 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f4d1ce5dea..15da91a8dd 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..a93b33464f 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,8 +33,18 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +/* zlib compressBound is not a macro */ +#define ZLIB_MAX_BLCKSZ BLCKSZ + (BLCKSZ>>12) + (BLCKSZ>>14) + (BLCKSZ>>25) + 13 +#else +#define ZLIB_MAX_BLCKSZ 0 +#endif + /* Buffer size required to store a compressed version of backup block image */ -#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) +#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) + +#define COMPRESS_BUFSIZE Max(PGLZ_MAX_BLCKSZ, ZLIB_MAX_BLCKSZ) /* * For each block reference registered with XLogRegisterBuffer, we fill in @@ -58,7 +68,7 @@ typedef struct * backup block data in XLogRecordAssemble() */ /* buffer to store a compressed version of backup block image */ - char compressed_page[PGLZ_MAX_BLCKSZ]; + char compressed_page[COMPRESS_BUFSIZE]; } registered_buffer; static registered_buffer *registered_buffers; @@ -113,7 +123,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -625,16 +636,26 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, cbimg.hole_length = 0; } + bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; + /* * Try to compress a block image if wal_compression is enabled */ if (wal_compression) { + int compression; + /* The current compression is stored in the WAL record */ + wal_compression_name(wal_compression_method); /* Range check */ + compression = walmethods[wal_compression_method].walmethod; + Assert(compression < (1 << BKPIMAGE_COMPRESS_BITS)); + bimg.bimg_info |= + compression << BKPIMAGE_COMPRESS_OFFSET_BITS; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, + wal_compression_method); } /* @@ -652,8 +673,6 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, rdt_datas_last->next = ®buf->bkp_rdatas[0]; rdt_datas_last = rdt_datas_last->next; - bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; - /* * If WAL consistency checking is enabled for the resource manager * of this WAL record, a full-page image is included in the record @@ -827,7 +846,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +872,42 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = COMPRESS_BUFSIZE; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + len_l = -1; + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..0d8830fc50 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1535,6 +1539,30 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* This is a mapping indexed by wal_compression */ +// XXX: maybe this is better done as a GUC hook to assign the 1) method; and 2) level +struct walcompression walmethods[] = { + {"pglz", WAL_COMPRESSION_PGLZ}, + {"zlib", WAL_COMPRESSION_ZLIB}, +}; + +/* + * Return a statically allocated string associated with the given compression + * method. + * This is here to be visible to frontend tools like pg_rewind. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + if (compression < sizeof(walmethods)/sizeof(*walmethods)) + return walmethods[compression].name; + return "???"; +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1557,9 +1585,41 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { + int compression_method = BKPIMAGE_COMPRESSION(bkpb->bimg_info); /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + compression_method, + wal_compression_name(compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..b000a21557 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,20 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +struct walcompression +{ + char *name; + int walmethod; /* Compression method to be stored in WAL */ +}; + +extern struct walcompression walmethods[]; + +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..7107cf6186 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -147,6 +147,15 @@ typedef struct XLogRecordBlockImageHeader #define BKPIMAGE_IS_COMPRESSED 0x02 /* page image is compressed */ #define BKPIMAGE_APPLY 0x04 /* page image should be restored during * replay */ +#define BKPIMAGE_COMPRESS_METHOD1 0x08 /* bits to encode compression method */ +#define BKPIMAGE_COMPRESS_METHOD2 0x10 /* 0=pglz; 1=zlib; */ + +/* How many bits to shift to extract compression */ +#define BKPIMAGE_COMPRESS_OFFSET_BITS 3 +/* How many bits are for compression */ +#define BKPIMAGE_COMPRESS_BITS 2 +/* Extract the compression from the bimg_info */ +#define BKPIMAGE_COMPRESSION(info) ((info >> BKPIMAGE_COMPRESS_OFFSET_BITS) & ((1<<BKPIMAGE_COMPRESS_BITS) - 1)) /* * Extra header information used when page image has "hole" and -- 2.17.0 --jozmn01XJZjDjM3N Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 3/8] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 ++++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++++++++-- src/backend/access/transam/xlogreader.c | 62 ++++++++++++++++++++++++- src/backend/utils/misc/guc.c | 11 +++++ src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 ++++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 ++-- 9 files changed, 144 insertions(+), 12 deletions(-) diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index e3128564e1..0183589b4d 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..ee73bc3afd 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, + Z_DEFAULT_COMPRESSION); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..143df55fcb 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,38 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l = 0; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..b80759ed45 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ = 0, + WAL_COMPRESSION_ZLIB = 1, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --f0KYrhQ4vYSV2aJu Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0004-wal_compression_method-default-to-zlib.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++-- src/backend/access/transam/xlogreader.c | 63 ++++++++++++++++++- src/backend/utils/misc/guc.c | 11 ++++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 +++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 +-- 11 files changed, 163 insertions(+), 12 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f4d1ce5dea..15da91a8dd 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..34e1227381 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..afca22a26c 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,39 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..d653839b97 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --XsQoSWH+UP9D9v3l Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++-- src/backend/access/transam/xlogreader.c | 63 ++++++++++++++++++- src/backend/utils/misc/guc.c | 11 ++++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 +++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 +-- 11 files changed, 163 insertions(+), 12 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f4d1ce5dea..15da91a8dd 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..34e1227381 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..afca22a26c 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,39 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..d653839b97 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --XsQoSWH+UP9D9v3l Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 67 ++++++++++++++++--- src/backend/access/transam/xlogreader.c | 64 +++++++++++++++++- src/backend/utils/misc/guc.c | 11 +++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 16 +++++ src/include/access/xlogrecord.h | 11 ++- 10 files changed, 187 insertions(+), 13 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f4d1ce5dea..15da91a8dd 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..a93b33464f 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,8 +33,18 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +/* zlib compressBound is not a macro */ +#define ZLIB_MAX_BLCKSZ BLCKSZ + (BLCKSZ>>12) + (BLCKSZ>>14) + (BLCKSZ>>25) + 13 +#else +#define ZLIB_MAX_BLCKSZ 0 +#endif + /* Buffer size required to store a compressed version of backup block image */ -#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) +#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) + +#define COMPRESS_BUFSIZE Max(PGLZ_MAX_BLCKSZ, ZLIB_MAX_BLCKSZ) /* * For each block reference registered with XLogRegisterBuffer, we fill in @@ -58,7 +68,7 @@ typedef struct * backup block data in XLogRecordAssemble() */ /* buffer to store a compressed version of backup block image */ - char compressed_page[PGLZ_MAX_BLCKSZ]; + char compressed_page[COMPRESS_BUFSIZE]; } registered_buffer; static registered_buffer *registered_buffers; @@ -113,7 +123,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -625,16 +636,26 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, cbimg.hole_length = 0; } + bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; + /* * Try to compress a block image if wal_compression is enabled */ if (wal_compression) { + int compression; + /* The current compression is stored in the WAL record */ + wal_compression_name(wal_compression_method); /* Range check */ + compression = walmethods[wal_compression_method].walmethod; + Assert(compression < (1 << BKPIMAGE_COMPRESS_BITS)); + bimg.bimg_info |= + compression << BKPIMAGE_COMPRESS_OFFSET_BITS; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, + wal_compression_method); } /* @@ -652,8 +673,6 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, rdt_datas_last->next = ®buf->bkp_rdatas[0]; rdt_datas_last = rdt_datas_last->next; - bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; - /* * If WAL consistency checking is enabled for the resource manager * of this WAL record, a full-page image is included in the record @@ -827,7 +846,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +872,42 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = COMPRESS_BUFSIZE; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + len_l = -1; + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..0d8830fc50 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1535,6 +1539,30 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* This is a mapping indexed by wal_compression */ +// XXX: maybe this is better done as a GUC hook to assign the 1) method; and 2) level +struct walcompression walmethods[] = { + {"pglz", WAL_COMPRESSION_PGLZ}, + {"zlib", WAL_COMPRESSION_ZLIB}, +}; + +/* + * Return a statically allocated string associated with the given compression + * method. + * This is here to be visible to frontend tools like pg_rewind. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + if (compression < sizeof(walmethods)/sizeof(*walmethods)) + return walmethods[compression].name; + return "???"; +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1557,9 +1585,41 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { + int compression_method = BKPIMAGE_COMPRESSION(bkpb->bimg_info); /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + compression_method, + wal_compression_name(compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..b000a21557 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,20 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +struct walcompression +{ + char *name; + int walmethod; /* Compression method to be stored in WAL */ +}; + +extern struct walcompression walmethods[]; + +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..7107cf6186 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -147,6 +147,15 @@ typedef struct XLogRecordBlockImageHeader #define BKPIMAGE_IS_COMPRESSED 0x02 /* page image is compressed */ #define BKPIMAGE_APPLY 0x04 /* page image should be restored during * replay */ +#define BKPIMAGE_COMPRESS_METHOD1 0x08 /* bits to encode compression method */ +#define BKPIMAGE_COMPRESS_METHOD2 0x10 /* 0=pglz; 1=zlib; */ + +/* How many bits to shift to extract compression */ +#define BKPIMAGE_COMPRESS_OFFSET_BITS 3 +/* How many bits are for compression */ +#define BKPIMAGE_COMPRESS_BITS 2 +/* Extract the compression from the bimg_info */ +#define BKPIMAGE_COMPRESSION(info) ((info >> BKPIMAGE_COMPRESS_OFFSET_BITS) & ((1<<BKPIMAGE_COMPRESS_BITS) - 1)) /* * Extra header information used when page image has "hole" and -- 2.17.0 --jozmn01XJZjDjM3N Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 3/8] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 ++++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++++++++-- src/backend/access/transam/xlogreader.c | 62 ++++++++++++++++++++++++- src/backend/utils/misc/guc.c | 11 +++++ src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 ++++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 ++-- 9 files changed, 144 insertions(+), 12 deletions(-) diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index e3128564e1..0183589b4d 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..ee73bc3afd 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, + Z_DEFAULT_COMPRESSION); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..143df55fcb 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,38 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l = 0; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..b80759ed45 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ = 0, + WAL_COMPRESSION_ZLIB = 1, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --f0KYrhQ4vYSV2aJu Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0004-wal_compression_method-default-to-zlib.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 3/8] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 ++++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++++++++-- src/backend/access/transam/xlogreader.c | 62 ++++++++++++++++++++++++- src/backend/utils/misc/guc.c | 11 +++++ src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 ++++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 ++-- 9 files changed, 144 insertions(+), 12 deletions(-) diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index e3128564e1..0183589b4d 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..ee73bc3afd 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, + Z_DEFAULT_COMPRESSION); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..143df55fcb 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,38 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l = 0; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..b80759ed45 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ = 0, + WAL_COMPRESSION_ZLIB = 1, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --f0KYrhQ4vYSV2aJu Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0004-wal_compression_method-default-to-zlib.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++-- src/backend/access/transam/xlogreader.c | 63 ++++++++++++++++++- src/backend/utils/misc/guc.c | 11 ++++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 +++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 +-- 11 files changed, 163 insertions(+), 12 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index e04250f4e9..04192b7add 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..34e1227381 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..afca22a26c 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,39 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..d653839b97 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --0qVF/w3MHQqLSynd Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++-- src/backend/access/transam/xlogreader.c | 63 ++++++++++++++++++- src/backend/utils/misc/guc.c | 11 ++++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 +++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 +-- 11 files changed, 163 insertions(+), 12 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index e04250f4e9..04192b7add 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..34e1227381 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..afca22a26c 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,39 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..d653839b97 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --0qVF/w3MHQqLSynd Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++-- src/backend/access/transam/xlogreader.c | 63 ++++++++++++++++++- src/backend/utils/misc/guc.c | 11 ++++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 +++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 +-- 11 files changed, 163 insertions(+), 12 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f4d1ce5dea..15da91a8dd 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..34e1227381 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..afca22a26c 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,39 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..d653839b97 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --XsQoSWH+UP9D9v3l Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 67 ++++++++++++++++--- src/backend/access/transam/xlogreader.c | 64 +++++++++++++++++- src/backend/utils/misc/guc.c | 11 +++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 16 +++++ src/include/access/xlogrecord.h | 11 ++- 10 files changed, 187 insertions(+), 13 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f4d1ce5dea..15da91a8dd 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..a93b33464f 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,8 +33,18 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +/* zlib compressBound is not a macro */ +#define ZLIB_MAX_BLCKSZ BLCKSZ + (BLCKSZ>>12) + (BLCKSZ>>14) + (BLCKSZ>>25) + 13 +#else +#define ZLIB_MAX_BLCKSZ 0 +#endif + /* Buffer size required to store a compressed version of backup block image */ -#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) +#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) + +#define COMPRESS_BUFSIZE Max(PGLZ_MAX_BLCKSZ, ZLIB_MAX_BLCKSZ) /* * For each block reference registered with XLogRegisterBuffer, we fill in @@ -58,7 +68,7 @@ typedef struct * backup block data in XLogRecordAssemble() */ /* buffer to store a compressed version of backup block image */ - char compressed_page[PGLZ_MAX_BLCKSZ]; + char compressed_page[COMPRESS_BUFSIZE]; } registered_buffer; static registered_buffer *registered_buffers; @@ -113,7 +123,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -625,16 +636,26 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, cbimg.hole_length = 0; } + bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; + /* * Try to compress a block image if wal_compression is enabled */ if (wal_compression) { + int compression; + /* The current compression is stored in the WAL record */ + wal_compression_name(wal_compression_method); /* Range check */ + compression = walmethods[wal_compression_method].walmethod; + Assert(compression < (1 << BKPIMAGE_COMPRESS_BITS)); + bimg.bimg_info |= + compression << BKPIMAGE_COMPRESS_OFFSET_BITS; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, + wal_compression_method); } /* @@ -652,8 +673,6 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, rdt_datas_last->next = ®buf->bkp_rdatas[0]; rdt_datas_last = rdt_datas_last->next; - bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; - /* * If WAL consistency checking is enabled for the resource manager * of this WAL record, a full-page image is included in the record @@ -827,7 +846,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +872,42 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = COMPRESS_BUFSIZE; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + len_l = -1; + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..0d8830fc50 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1535,6 +1539,30 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* This is a mapping indexed by wal_compression */ +// XXX: maybe this is better done as a GUC hook to assign the 1) method; and 2) level +struct walcompression walmethods[] = { + {"pglz", WAL_COMPRESSION_PGLZ}, + {"zlib", WAL_COMPRESSION_ZLIB}, +}; + +/* + * Return a statically allocated string associated with the given compression + * method. + * This is here to be visible to frontend tools like pg_rewind. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + if (compression < sizeof(walmethods)/sizeof(*walmethods)) + return walmethods[compression].name; + return "???"; +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1557,9 +1585,41 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { + int compression_method = BKPIMAGE_COMPRESSION(bkpb->bimg_info); /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + compression_method, + wal_compression_name(compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..b000a21557 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,20 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +struct walcompression +{ + char *name; + int walmethod; /* Compression method to be stored in WAL */ +}; + +extern struct walcompression walmethods[]; + +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..7107cf6186 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -147,6 +147,15 @@ typedef struct XLogRecordBlockImageHeader #define BKPIMAGE_IS_COMPRESSED 0x02 /* page image is compressed */ #define BKPIMAGE_APPLY 0x04 /* page image should be restored during * replay */ +#define BKPIMAGE_COMPRESS_METHOD1 0x08 /* bits to encode compression method */ +#define BKPIMAGE_COMPRESS_METHOD2 0x10 /* 0=pglz; 1=zlib; */ + +/* How many bits to shift to extract compression */ +#define BKPIMAGE_COMPRESS_OFFSET_BITS 3 +/* How many bits are for compression */ +#define BKPIMAGE_COMPRESS_BITS 2 +/* Extract the compression from the bimg_info */ +#define BKPIMAGE_COMPRESSION(info) ((info >> BKPIMAGE_COMPRESS_OFFSET_BITS) & ((1<<BKPIMAGE_COMPRESS_BITS) - 1)) /* * Extra header information used when page image has "hole" and -- 2.17.0 --jozmn01XJZjDjM3N Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++-- src/backend/access/transam/xlogreader.c | 63 ++++++++++++++++++- src/backend/utils/misc/guc.c | 11 ++++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 +++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 +-- 11 files changed, 163 insertions(+), 12 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f4d1ce5dea..15da91a8dd 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..34e1227381 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..afca22a26c 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,39 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..d653839b97 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --XsQoSWH+UP9D9v3l Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++-- src/backend/access/transam/xlogreader.c | 63 ++++++++++++++++++- src/backend/utils/misc/guc.c | 11 ++++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 +++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 +-- 11 files changed, 163 insertions(+), 12 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f4d1ce5dea..15da91a8dd 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..34e1227381 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..afca22a26c 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,39 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..d653839b97 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --XsQoSWH+UP9D9v3l Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 67 ++++++++++++++++--- src/backend/access/transam/xlogreader.c | 64 +++++++++++++++++- src/backend/utils/misc/guc.c | 11 +++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 16 +++++ src/include/access/xlogrecord.h | 11 ++- 10 files changed, 187 insertions(+), 13 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f4d1ce5dea..15da91a8dd 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..a93b33464f 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,8 +33,18 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +/* zlib compressBound is not a macro */ +#define ZLIB_MAX_BLCKSZ BLCKSZ + (BLCKSZ>>12) + (BLCKSZ>>14) + (BLCKSZ>>25) + 13 +#else +#define ZLIB_MAX_BLCKSZ 0 +#endif + /* Buffer size required to store a compressed version of backup block image */ -#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) +#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) + +#define COMPRESS_BUFSIZE Max(PGLZ_MAX_BLCKSZ, ZLIB_MAX_BLCKSZ) /* * For each block reference registered with XLogRegisterBuffer, we fill in @@ -58,7 +68,7 @@ typedef struct * backup block data in XLogRecordAssemble() */ /* buffer to store a compressed version of backup block image */ - char compressed_page[PGLZ_MAX_BLCKSZ]; + char compressed_page[COMPRESS_BUFSIZE]; } registered_buffer; static registered_buffer *registered_buffers; @@ -113,7 +123,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -625,16 +636,26 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, cbimg.hole_length = 0; } + bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; + /* * Try to compress a block image if wal_compression is enabled */ if (wal_compression) { + int compression; + /* The current compression is stored in the WAL record */ + wal_compression_name(wal_compression_method); /* Range check */ + compression = walmethods[wal_compression_method].walmethod; + Assert(compression < (1 << BKPIMAGE_COMPRESS_BITS)); + bimg.bimg_info |= + compression << BKPIMAGE_COMPRESS_OFFSET_BITS; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, + wal_compression_method); } /* @@ -652,8 +673,6 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, rdt_datas_last->next = ®buf->bkp_rdatas[0]; rdt_datas_last = rdt_datas_last->next; - bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; - /* * If WAL consistency checking is enabled for the resource manager * of this WAL record, a full-page image is included in the record @@ -827,7 +846,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +872,42 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = COMPRESS_BUFSIZE; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + len_l = -1; + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..0d8830fc50 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1535,6 +1539,30 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* This is a mapping indexed by wal_compression */ +// XXX: maybe this is better done as a GUC hook to assign the 1) method; and 2) level +struct walcompression walmethods[] = { + {"pglz", WAL_COMPRESSION_PGLZ}, + {"zlib", WAL_COMPRESSION_ZLIB}, +}; + +/* + * Return a statically allocated string associated with the given compression + * method. + * This is here to be visible to frontend tools like pg_rewind. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + if (compression < sizeof(walmethods)/sizeof(*walmethods)) + return walmethods[compression].name; + return "???"; +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1557,9 +1585,41 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { + int compression_method = BKPIMAGE_COMPRESSION(bkpb->bimg_info); /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + compression_method, + wal_compression_name(compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..b000a21557 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,20 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +struct walcompression +{ + char *name; + int walmethod; /* Compression method to be stored in WAL */ +}; + +extern struct walcompression walmethods[]; + +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..7107cf6186 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -147,6 +147,15 @@ typedef struct XLogRecordBlockImageHeader #define BKPIMAGE_IS_COMPRESSED 0x02 /* page image is compressed */ #define BKPIMAGE_APPLY 0x04 /* page image should be restored during * replay */ +#define BKPIMAGE_COMPRESS_METHOD1 0x08 /* bits to encode compression method */ +#define BKPIMAGE_COMPRESS_METHOD2 0x10 /* 0=pglz; 1=zlib; */ + +/* How many bits to shift to extract compression */ +#define BKPIMAGE_COMPRESS_OFFSET_BITS 3 +/* How many bits are for compression */ +#define BKPIMAGE_COMPRESS_BITS 2 +/* Extract the compression from the bimg_info */ +#define BKPIMAGE_COMPRESSION(info) ((info >> BKPIMAGE_COMPRESS_OFFSET_BITS) & ((1<<BKPIMAGE_COMPRESS_BITS) - 1)) /* * Extra header information used when page image has "hole" and -- 2.17.0 --jozmn01XJZjDjM3N Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++-- src/backend/access/transam/xlogreader.c | 63 ++++++++++++++++++- src/backend/utils/misc/guc.c | 11 ++++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 +++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 +-- 11 files changed, 163 insertions(+), 12 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index e04250f4e9..04192b7add 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..34e1227381 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..afca22a26c 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,39 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..d653839b97 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --0qVF/w3MHQqLSynd Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++-- src/backend/access/transam/xlogreader.c | 63 ++++++++++++++++++- src/backend/utils/misc/guc.c | 11 ++++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 +++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 +-- 11 files changed, 163 insertions(+), 12 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f4d1ce5dea..15da91a8dd 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..34e1227381 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..afca22a26c 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,39 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..d653839b97 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --XsQoSWH+UP9D9v3l Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 67 ++++++++++++++++--- src/backend/access/transam/xlogreader.c | 64 +++++++++++++++++- src/backend/utils/misc/guc.c | 11 +++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 16 +++++ src/include/access/xlogrecord.h | 11 ++- 10 files changed, 187 insertions(+), 13 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f4d1ce5dea..15da91a8dd 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..a93b33464f 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,8 +33,18 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +/* zlib compressBound is not a macro */ +#define ZLIB_MAX_BLCKSZ BLCKSZ + (BLCKSZ>>12) + (BLCKSZ>>14) + (BLCKSZ>>25) + 13 +#else +#define ZLIB_MAX_BLCKSZ 0 +#endif + /* Buffer size required to store a compressed version of backup block image */ -#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) +#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) + +#define COMPRESS_BUFSIZE Max(PGLZ_MAX_BLCKSZ, ZLIB_MAX_BLCKSZ) /* * For each block reference registered with XLogRegisterBuffer, we fill in @@ -58,7 +68,7 @@ typedef struct * backup block data in XLogRecordAssemble() */ /* buffer to store a compressed version of backup block image */ - char compressed_page[PGLZ_MAX_BLCKSZ]; + char compressed_page[COMPRESS_BUFSIZE]; } registered_buffer; static registered_buffer *registered_buffers; @@ -113,7 +123,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -625,16 +636,26 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, cbimg.hole_length = 0; } + bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; + /* * Try to compress a block image if wal_compression is enabled */ if (wal_compression) { + int compression; + /* The current compression is stored in the WAL record */ + wal_compression_name(wal_compression_method); /* Range check */ + compression = walmethods[wal_compression_method].walmethod; + Assert(compression < (1 << BKPIMAGE_COMPRESS_BITS)); + bimg.bimg_info |= + compression << BKPIMAGE_COMPRESS_OFFSET_BITS; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, + wal_compression_method); } /* @@ -652,8 +673,6 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, rdt_datas_last->next = ®buf->bkp_rdatas[0]; rdt_datas_last = rdt_datas_last->next; - bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; - /* * If WAL consistency checking is enabled for the resource manager * of this WAL record, a full-page image is included in the record @@ -827,7 +846,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +872,42 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = COMPRESS_BUFSIZE; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + len_l = -1; + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..0d8830fc50 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1535,6 +1539,30 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* This is a mapping indexed by wal_compression */ +// XXX: maybe this is better done as a GUC hook to assign the 1) method; and 2) level +struct walcompression walmethods[] = { + {"pglz", WAL_COMPRESSION_PGLZ}, + {"zlib", WAL_COMPRESSION_ZLIB}, +}; + +/* + * Return a statically allocated string associated with the given compression + * method. + * This is here to be visible to frontend tools like pg_rewind. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + if (compression < sizeof(walmethods)/sizeof(*walmethods)) + return walmethods[compression].name; + return "???"; +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1557,9 +1585,41 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { + int compression_method = BKPIMAGE_COMPRESSION(bkpb->bimg_info); /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + compression_method, + wal_compression_name(compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..b000a21557 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,20 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +struct walcompression +{ + char *name; + int walmethod; /* Compression method to be stored in WAL */ +}; + +extern struct walcompression walmethods[]; + +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..7107cf6186 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -147,6 +147,15 @@ typedef struct XLogRecordBlockImageHeader #define BKPIMAGE_IS_COMPRESSED 0x02 /* page image is compressed */ #define BKPIMAGE_APPLY 0x04 /* page image should be restored during * replay */ +#define BKPIMAGE_COMPRESS_METHOD1 0x08 /* bits to encode compression method */ +#define BKPIMAGE_COMPRESS_METHOD2 0x10 /* 0=pglz; 1=zlib; */ + +/* How many bits to shift to extract compression */ +#define BKPIMAGE_COMPRESS_OFFSET_BITS 3 +/* How many bits are for compression */ +#define BKPIMAGE_COMPRESS_BITS 2 +/* Extract the compression from the bimg_info */ +#define BKPIMAGE_COMPRESSION(info) ((info >> BKPIMAGE_COMPRESS_OFFSET_BITS) & ((1<<BKPIMAGE_COMPRESS_BITS) - 1)) /* * Extra header information used when page image has "hole" and -- 2.17.0 --jozmn01XJZjDjM3N Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 67 ++++++++++++++++--- src/backend/access/transam/xlogreader.c | 64 +++++++++++++++++- src/backend/utils/misc/guc.c | 11 +++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 16 +++++ src/include/access/xlogrecord.h | 11 ++- 10 files changed, 187 insertions(+), 13 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f4d1ce5dea..15da91a8dd 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..a93b33464f 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,8 +33,18 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +/* zlib compressBound is not a macro */ +#define ZLIB_MAX_BLCKSZ BLCKSZ + (BLCKSZ>>12) + (BLCKSZ>>14) + (BLCKSZ>>25) + 13 +#else +#define ZLIB_MAX_BLCKSZ 0 +#endif + /* Buffer size required to store a compressed version of backup block image */ -#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) +#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) + +#define COMPRESS_BUFSIZE Max(PGLZ_MAX_BLCKSZ, ZLIB_MAX_BLCKSZ) /* * For each block reference registered with XLogRegisterBuffer, we fill in @@ -58,7 +68,7 @@ typedef struct * backup block data in XLogRecordAssemble() */ /* buffer to store a compressed version of backup block image */ - char compressed_page[PGLZ_MAX_BLCKSZ]; + char compressed_page[COMPRESS_BUFSIZE]; } registered_buffer; static registered_buffer *registered_buffers; @@ -113,7 +123,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -625,16 +636,26 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, cbimg.hole_length = 0; } + bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; + /* * Try to compress a block image if wal_compression is enabled */ if (wal_compression) { + int compression; + /* The current compression is stored in the WAL record */ + wal_compression_name(wal_compression_method); /* Range check */ + compression = walmethods[wal_compression_method].walmethod; + Assert(compression < (1 << BKPIMAGE_COMPRESS_BITS)); + bimg.bimg_info |= + compression << BKPIMAGE_COMPRESS_OFFSET_BITS; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, + wal_compression_method); } /* @@ -652,8 +673,6 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, rdt_datas_last->next = ®buf->bkp_rdatas[0]; rdt_datas_last = rdt_datas_last->next; - bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; - /* * If WAL consistency checking is enabled for the resource manager * of this WAL record, a full-page image is included in the record @@ -827,7 +846,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +872,42 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = COMPRESS_BUFSIZE; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + len_l = -1; + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..0d8830fc50 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1535,6 +1539,30 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* This is a mapping indexed by wal_compression */ +// XXX: maybe this is better done as a GUC hook to assign the 1) method; and 2) level +struct walcompression walmethods[] = { + {"pglz", WAL_COMPRESSION_PGLZ}, + {"zlib", WAL_COMPRESSION_ZLIB}, +}; + +/* + * Return a statically allocated string associated with the given compression + * method. + * This is here to be visible to frontend tools like pg_rewind. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + if (compression < sizeof(walmethods)/sizeof(*walmethods)) + return walmethods[compression].name; + return "???"; +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1557,9 +1585,41 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { + int compression_method = BKPIMAGE_COMPRESSION(bkpb->bimg_info); /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + compression_method, + wal_compression_name(compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..b000a21557 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,20 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +struct walcompression +{ + char *name; + int walmethod; /* Compression method to be stored in WAL */ +}; + +extern struct walcompression walmethods[]; + +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..7107cf6186 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -147,6 +147,15 @@ typedef struct XLogRecordBlockImageHeader #define BKPIMAGE_IS_COMPRESSED 0x02 /* page image is compressed */ #define BKPIMAGE_APPLY 0x04 /* page image should be restored during * replay */ +#define BKPIMAGE_COMPRESS_METHOD1 0x08 /* bits to encode compression method */ +#define BKPIMAGE_COMPRESS_METHOD2 0x10 /* 0=pglz; 1=zlib; */ + +/* How many bits to shift to extract compression */ +#define BKPIMAGE_COMPRESS_OFFSET_BITS 3 +/* How many bits are for compression */ +#define BKPIMAGE_COMPRESS_BITS 2 +/* Extract the compression from the bimg_info */ +#define BKPIMAGE_COMPRESSION(info) ((info >> BKPIMAGE_COMPRESS_OFFSET_BITS) & ((1<<BKPIMAGE_COMPRESS_BITS) - 1)) /* * Extra header information used when page image has "hole" and -- 2.17.0 --jozmn01XJZjDjM3N Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 3/8] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 ++++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++++++++-- src/backend/access/transam/xlogreader.c | 62 ++++++++++++++++++++++++- src/backend/utils/misc/guc.c | 11 +++++ src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 ++++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 ++-- 9 files changed, 144 insertions(+), 12 deletions(-) diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index e3128564e1..0183589b4d 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..ee73bc3afd 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, + Z_DEFAULT_COMPRESSION); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..143df55fcb 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,38 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l = 0; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..b80759ed45 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ = 0, + WAL_COMPRESSION_ZLIB = 1, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --f0KYrhQ4vYSV2aJu Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0004-wal_compression_method-default-to-zlib.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 3/8] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 ++++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++++++++-- src/backend/access/transam/xlogreader.c | 62 ++++++++++++++++++++++++- src/backend/utils/misc/guc.c | 11 +++++ src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 ++++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 ++-- 9 files changed, 144 insertions(+), 12 deletions(-) diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index e3128564e1..0183589b4d 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..ee73bc3afd 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, + Z_DEFAULT_COMPRESSION); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..143df55fcb 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,38 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l = 0; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..b80759ed45 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ = 0, + WAL_COMPRESSION_ZLIB = 1, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --f0KYrhQ4vYSV2aJu Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0004-wal_compression_method-default-to-zlib.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 3/8] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 ++++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++++++++-- src/backend/access/transam/xlogreader.c | 62 ++++++++++++++++++++++++- src/backend/utils/misc/guc.c | 11 +++++ src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 ++++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 ++-- 9 files changed, 144 insertions(+), 12 deletions(-) diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index e3128564e1..0183589b4d 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..ee73bc3afd 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, + Z_DEFAULT_COMPRESSION); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..143df55fcb 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,38 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l = 0; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..b80759ed45 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ = 0, + WAL_COMPRESSION_ZLIB = 1, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --f0KYrhQ4vYSV2aJu Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0004-wal_compression_method-default-to-zlib.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++-- src/backend/access/transam/xlogreader.c | 63 ++++++++++++++++++- src/backend/utils/misc/guc.c | 11 ++++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 +++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 +-- 11 files changed, 163 insertions(+), 12 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f4d1ce5dea..15da91a8dd 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..34e1227381 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..afca22a26c 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,39 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..d653839b97 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --XsQoSWH+UP9D9v3l Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 3/8] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 ++++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++++++++-- src/backend/access/transam/xlogreader.c | 62 ++++++++++++++++++++++++- src/backend/utils/misc/guc.c | 11 +++++ src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 ++++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 ++-- 9 files changed, 144 insertions(+), 12 deletions(-) diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index e3128564e1..0183589b4d 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..ee73bc3afd 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, + Z_DEFAULT_COMPRESSION); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..143df55fcb 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,38 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l = 0; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..b80759ed45 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ = 0, + WAL_COMPRESSION_ZLIB = 1, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --f0KYrhQ4vYSV2aJu Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0004-wal_compression_method-default-to-zlib.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++-- src/backend/access/transam/xlogreader.c | 63 ++++++++++++++++++- src/backend/utils/misc/guc.c | 11 ++++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 +++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 +-- 11 files changed, 163 insertions(+), 12 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index e04250f4e9..04192b7add 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..34e1227381 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..afca22a26c 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,39 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..d653839b97 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --0qVF/w3MHQqLSynd Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 3/8] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 ++++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++++++++-- src/backend/access/transam/xlogreader.c | 62 ++++++++++++++++++++++++- src/backend/utils/misc/guc.c | 11 +++++ src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 ++++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 ++-- 9 files changed, 144 insertions(+), 12 deletions(-) diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index e3128564e1..0183589b4d 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..ee73bc3afd 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, + Z_DEFAULT_COMPRESSION); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..143df55fcb 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,38 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l = 0; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..b80759ed45 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ = 0, + WAL_COMPRESSION_ZLIB = 1, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --f0KYrhQ4vYSV2aJu Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0004-wal_compression_method-default-to-zlib.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++-- src/backend/access/transam/xlogreader.c | 63 ++++++++++++++++++- src/backend/utils/misc/guc.c | 11 ++++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 +++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 +-- 11 files changed, 163 insertions(+), 12 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index e04250f4e9..04192b7add 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..34e1227381 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..afca22a26c 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,39 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..d653839b97 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --0qVF/w3MHQqLSynd Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++-- src/backend/access/transam/xlogreader.c | 63 ++++++++++++++++++- src/backend/utils/misc/guc.c | 11 ++++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 +++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 +-- 11 files changed, 163 insertions(+), 12 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f4d1ce5dea..15da91a8dd 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..34e1227381 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..afca22a26c 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,39 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..d653839b97 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --XsQoSWH+UP9D9v3l Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++-- src/backend/access/transam/xlogreader.c | 63 ++++++++++++++++++- src/backend/utils/misc/guc.c | 11 ++++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 +++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 +-- 11 files changed, 163 insertions(+), 12 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index e04250f4e9..04192b7add 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..34e1227381 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..afca22a26c 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,39 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..d653839b97 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --0qVF/w3MHQqLSynd Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++-- src/backend/access/transam/xlogreader.c | 63 ++++++++++++++++++- src/backend/utils/misc/guc.c | 11 ++++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 +++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 +-- 11 files changed, 163 insertions(+), 12 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index e04250f4e9..04192b7add 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..34e1227381 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..afca22a26c 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,39 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..d653839b97 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --0qVF/w3MHQqLSynd Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 67 ++++++++++++++++--- src/backend/access/transam/xlogreader.c | 64 +++++++++++++++++- src/backend/utils/misc/guc.c | 11 +++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 16 +++++ src/include/access/xlogrecord.h | 11 ++- 10 files changed, 187 insertions(+), 13 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f4d1ce5dea..15da91a8dd 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..a93b33464f 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,8 +33,18 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +/* zlib compressBound is not a macro */ +#define ZLIB_MAX_BLCKSZ BLCKSZ + (BLCKSZ>>12) + (BLCKSZ>>14) + (BLCKSZ>>25) + 13 +#else +#define ZLIB_MAX_BLCKSZ 0 +#endif + /* Buffer size required to store a compressed version of backup block image */ -#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) +#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) + +#define COMPRESS_BUFSIZE Max(PGLZ_MAX_BLCKSZ, ZLIB_MAX_BLCKSZ) /* * For each block reference registered with XLogRegisterBuffer, we fill in @@ -58,7 +68,7 @@ typedef struct * backup block data in XLogRecordAssemble() */ /* buffer to store a compressed version of backup block image */ - char compressed_page[PGLZ_MAX_BLCKSZ]; + char compressed_page[COMPRESS_BUFSIZE]; } registered_buffer; static registered_buffer *registered_buffers; @@ -113,7 +123,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -625,16 +636,26 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, cbimg.hole_length = 0; } + bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; + /* * Try to compress a block image if wal_compression is enabled */ if (wal_compression) { + int compression; + /* The current compression is stored in the WAL record */ + wal_compression_name(wal_compression_method); /* Range check */ + compression = walmethods[wal_compression_method].walmethod; + Assert(compression < (1 << BKPIMAGE_COMPRESS_BITS)); + bimg.bimg_info |= + compression << BKPIMAGE_COMPRESS_OFFSET_BITS; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, + wal_compression_method); } /* @@ -652,8 +673,6 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, rdt_datas_last->next = ®buf->bkp_rdatas[0]; rdt_datas_last = rdt_datas_last->next; - bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; - /* * If WAL consistency checking is enabled for the resource manager * of this WAL record, a full-page image is included in the record @@ -827,7 +846,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +872,42 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = COMPRESS_BUFSIZE; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + len_l = -1; + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..0d8830fc50 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1535,6 +1539,30 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* This is a mapping indexed by wal_compression */ +// XXX: maybe this is better done as a GUC hook to assign the 1) method; and 2) level +struct walcompression walmethods[] = { + {"pglz", WAL_COMPRESSION_PGLZ}, + {"zlib", WAL_COMPRESSION_ZLIB}, +}; + +/* + * Return a statically allocated string associated with the given compression + * method. + * This is here to be visible to frontend tools like pg_rewind. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + if (compression < sizeof(walmethods)/sizeof(*walmethods)) + return walmethods[compression].name; + return "???"; +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1557,9 +1585,41 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { + int compression_method = BKPIMAGE_COMPRESSION(bkpb->bimg_info); /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + compression_method, + wal_compression_name(compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..b000a21557 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,20 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +struct walcompression +{ + char *name; + int walmethod; /* Compression method to be stored in WAL */ +}; + +extern struct walcompression walmethods[]; + +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..7107cf6186 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -147,6 +147,15 @@ typedef struct XLogRecordBlockImageHeader #define BKPIMAGE_IS_COMPRESSED 0x02 /* page image is compressed */ #define BKPIMAGE_APPLY 0x04 /* page image should be restored during * replay */ +#define BKPIMAGE_COMPRESS_METHOD1 0x08 /* bits to encode compression method */ +#define BKPIMAGE_COMPRESS_METHOD2 0x10 /* 0=pglz; 1=zlib; */ + +/* How many bits to shift to extract compression */ +#define BKPIMAGE_COMPRESS_OFFSET_BITS 3 +/* How many bits are for compression */ +#define BKPIMAGE_COMPRESS_BITS 2 +/* Extract the compression from the bimg_info */ +#define BKPIMAGE_COMPRESSION(info) ((info >> BKPIMAGE_COMPRESS_OFFSET_BITS) & ((1<<BKPIMAGE_COMPRESS_BITS) - 1)) /* * Extra header information used when page image has "hole" and -- 2.17.0 --jozmn01XJZjDjM3N Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 3/8] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 ++++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++++++++-- src/backend/access/transam/xlogreader.c | 62 ++++++++++++++++++++++++- src/backend/utils/misc/guc.c | 11 +++++ src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 ++++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 ++-- 9 files changed, 144 insertions(+), 12 deletions(-) diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index e3128564e1..0183589b4d 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..ee73bc3afd 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, + Z_DEFAULT_COMPRESSION); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..143df55fcb 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,38 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l = 0; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..b80759ed45 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ = 0, + WAL_COMPRESSION_ZLIB = 1, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --f0KYrhQ4vYSV2aJu Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0004-wal_compression_method-default-to-zlib.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++-- src/backend/access/transam/xlogreader.c | 63 ++++++++++++++++++- src/backend/utils/misc/guc.c | 11 ++++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 +++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 +-- 11 files changed, 163 insertions(+), 12 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index e04250f4e9..04192b7add 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..34e1227381 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..afca22a26c 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,39 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..d653839b97 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --0qVF/w3MHQqLSynd Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++-- src/backend/access/transam/xlogreader.c | 63 ++++++++++++++++++- src/backend/utils/misc/guc.c | 11 ++++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 +++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 +-- 11 files changed, 163 insertions(+), 12 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index e04250f4e9..04192b7add 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..34e1227381 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..afca22a26c 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,39 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..d653839b97 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --0qVF/w3MHQqLSynd Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 67 ++++++++++++++++--- src/backend/access/transam/xlogreader.c | 64 +++++++++++++++++- src/backend/utils/misc/guc.c | 11 +++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 16 +++++ src/include/access/xlogrecord.h | 11 ++- 10 files changed, 187 insertions(+), 13 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f4d1ce5dea..15da91a8dd 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..a93b33464f 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,8 +33,18 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +/* zlib compressBound is not a macro */ +#define ZLIB_MAX_BLCKSZ BLCKSZ + (BLCKSZ>>12) + (BLCKSZ>>14) + (BLCKSZ>>25) + 13 +#else +#define ZLIB_MAX_BLCKSZ 0 +#endif + /* Buffer size required to store a compressed version of backup block image */ -#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) +#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) + +#define COMPRESS_BUFSIZE Max(PGLZ_MAX_BLCKSZ, ZLIB_MAX_BLCKSZ) /* * For each block reference registered with XLogRegisterBuffer, we fill in @@ -58,7 +68,7 @@ typedef struct * backup block data in XLogRecordAssemble() */ /* buffer to store a compressed version of backup block image */ - char compressed_page[PGLZ_MAX_BLCKSZ]; + char compressed_page[COMPRESS_BUFSIZE]; } registered_buffer; static registered_buffer *registered_buffers; @@ -113,7 +123,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -625,16 +636,26 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, cbimg.hole_length = 0; } + bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; + /* * Try to compress a block image if wal_compression is enabled */ if (wal_compression) { + int compression; + /* The current compression is stored in the WAL record */ + wal_compression_name(wal_compression_method); /* Range check */ + compression = walmethods[wal_compression_method].walmethod; + Assert(compression < (1 << BKPIMAGE_COMPRESS_BITS)); + bimg.bimg_info |= + compression << BKPIMAGE_COMPRESS_OFFSET_BITS; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, + wal_compression_method); } /* @@ -652,8 +673,6 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, rdt_datas_last->next = ®buf->bkp_rdatas[0]; rdt_datas_last = rdt_datas_last->next; - bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; - /* * If WAL consistency checking is enabled for the resource manager * of this WAL record, a full-page image is included in the record @@ -827,7 +846,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +872,42 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = COMPRESS_BUFSIZE; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + len_l = -1; + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..0d8830fc50 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1535,6 +1539,30 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* This is a mapping indexed by wal_compression */ +// XXX: maybe this is better done as a GUC hook to assign the 1) method; and 2) level +struct walcompression walmethods[] = { + {"pglz", WAL_COMPRESSION_PGLZ}, + {"zlib", WAL_COMPRESSION_ZLIB}, +}; + +/* + * Return a statically allocated string associated with the given compression + * method. + * This is here to be visible to frontend tools like pg_rewind. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + if (compression < sizeof(walmethods)/sizeof(*walmethods)) + return walmethods[compression].name; + return "???"; +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1557,9 +1585,41 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { + int compression_method = BKPIMAGE_COMPRESSION(bkpb->bimg_info); /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + compression_method, + wal_compression_name(compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..b000a21557 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,20 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +struct walcompression +{ + char *name; + int walmethod; /* Compression method to be stored in WAL */ +}; + +extern struct walcompression walmethods[]; + +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..7107cf6186 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -147,6 +147,15 @@ typedef struct XLogRecordBlockImageHeader #define BKPIMAGE_IS_COMPRESSED 0x02 /* page image is compressed */ #define BKPIMAGE_APPLY 0x04 /* page image should be restored during * replay */ +#define BKPIMAGE_COMPRESS_METHOD1 0x08 /* bits to encode compression method */ +#define BKPIMAGE_COMPRESS_METHOD2 0x10 /* 0=pglz; 1=zlib; */ + +/* How many bits to shift to extract compression */ +#define BKPIMAGE_COMPRESS_OFFSET_BITS 3 +/* How many bits are for compression */ +#define BKPIMAGE_COMPRESS_BITS 2 +/* Extract the compression from the bimg_info */ +#define BKPIMAGE_COMPRESSION(info) ((info >> BKPIMAGE_COMPRESS_OFFSET_BITS) & ((1<<BKPIMAGE_COMPRESS_BITS) - 1)) /* * Extra header information used when page image has "hole" and -- 2.17.0 --jozmn01XJZjDjM3N Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 3/8] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 ++++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++++++++-- src/backend/access/transam/xlogreader.c | 62 ++++++++++++++++++++++++- src/backend/utils/misc/guc.c | 11 +++++ src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 ++++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 ++-- 9 files changed, 144 insertions(+), 12 deletions(-) diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index e3128564e1..0183589b4d 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..ee73bc3afd 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, + Z_DEFAULT_COMPRESSION); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..143df55fcb 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,38 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l = 0; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..b80759ed45 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ = 0, + WAL_COMPRESSION_ZLIB = 1, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --f0KYrhQ4vYSV2aJu Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0004-wal_compression_method-default-to-zlib.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++-- src/backend/access/transam/xlogreader.c | 63 ++++++++++++++++++- src/backend/utils/misc/guc.c | 11 ++++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 +++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 +-- 11 files changed, 163 insertions(+), 12 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f4d1ce5dea..15da91a8dd 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..34e1227381 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..afca22a26c 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,39 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..d653839b97 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --XsQoSWH+UP9D9v3l Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++-- src/backend/access/transam/xlogreader.c | 63 ++++++++++++++++++- src/backend/utils/misc/guc.c | 11 ++++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 +++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 +-- 11 files changed, 163 insertions(+), 12 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f4d1ce5dea..15da91a8dd 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..34e1227381 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..afca22a26c 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,39 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..d653839b97 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --XsQoSWH+UP9D9v3l Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/12] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 67 ++++++++++++++++--- src/backend/access/transam/xlogreader.c | 64 +++++++++++++++++- src/backend/utils/misc/guc.c | 11 +++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 16 +++++ src/include/access/xlogrecord.h | 11 ++- 10 files changed, 187 insertions(+), 13 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index ee4925d6d9..bed89d055d 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 6f8810e149..5b657ec724 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..a93b33464f 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,8 +33,18 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +/* zlib compressBound is not a macro */ +#define ZLIB_MAX_BLCKSZ BLCKSZ + (BLCKSZ>>12) + (BLCKSZ>>14) + (BLCKSZ>>25) + 13 +#else +#define ZLIB_MAX_BLCKSZ 0 +#endif + /* Buffer size required to store a compressed version of backup block image */ -#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) +#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) + +#define COMPRESS_BUFSIZE Max(PGLZ_MAX_BLCKSZ, ZLIB_MAX_BLCKSZ) /* * For each block reference registered with XLogRegisterBuffer, we fill in @@ -58,7 +68,7 @@ typedef struct * backup block data in XLogRecordAssemble() */ /* buffer to store a compressed version of backup block image */ - char compressed_page[PGLZ_MAX_BLCKSZ]; + char compressed_page[COMPRESS_BUFSIZE]; } registered_buffer; static registered_buffer *registered_buffers; @@ -113,7 +123,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -625,16 +636,26 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, cbimg.hole_length = 0; } + bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; + /* * Try to compress a block image if wal_compression is enabled */ if (wal_compression) { + int compression; + /* The current compression is stored in the WAL record */ + wal_compression_name(wal_compression_method); /* Range check */ + compression = walmethods[wal_compression_method].walmethod; + Assert(compression < (1 << BKPIMAGE_COMPRESS_BITS)); + bimg.bimg_info |= + compression << BKPIMAGE_COMPRESS_OFFSET_BITS; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, + wal_compression_method); } /* @@ -652,8 +673,6 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, rdt_datas_last->next = ®buf->bkp_rdatas[0]; rdt_datas_last = rdt_datas_last->next; - bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; - /* * If WAL consistency checking is enabled for the resource manager * of this WAL record, a full-page image is included in the record @@ -827,7 +846,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +872,42 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = COMPRESS_BUFSIZE; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + len_l = -1; + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..0d8830fc50 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1535,6 +1539,30 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* This is a mapping indexed by wal_compression */ +// XXX: maybe this is better done as a GUC hook to assign the 1) method; and 2) level +struct walcompression walmethods[] = { + {"pglz", WAL_COMPRESSION_PGLZ}, + {"zlib", WAL_COMPRESSION_ZLIB}, +}; + +/* + * Return a statically allocated string associated with the given compression + * method. + * This is here to be visible to frontend tools like pg_rewind. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + if (compression < sizeof(walmethods)/sizeof(*walmethods)) + return walmethods[compression].name; + return "???"; +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1557,9 +1585,41 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { + int compression_method = BKPIMAGE_COMPRESSION(bkpb->bimg_info); /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + compression_method, + wal_compression_name(compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 3b36a31a47..47d30d6cda 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -517,6 +517,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4760,6 +4761,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index 86425965d0..530007f55a 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 77187c12be..2f3886431a 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..b000a21557 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,20 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +struct walcompression +{ + char *name; + int walmethod; /* Compression method to be stored in WAL */ +}; + +extern struct walcompression walmethods[]; + +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..7107cf6186 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -147,6 +147,15 @@ typedef struct XLogRecordBlockImageHeader #define BKPIMAGE_IS_COMPRESSED 0x02 /* page image is compressed */ #define BKPIMAGE_APPLY 0x04 /* page image should be restored during * replay */ +#define BKPIMAGE_COMPRESS_METHOD1 0x08 /* bits to encode compression method */ +#define BKPIMAGE_COMPRESS_METHOD2 0x10 /* 0=pglz; 1=zlib; */ + +/* How many bits to shift to extract compression */ +#define BKPIMAGE_COMPRESS_OFFSET_BITS 3 +/* How many bits are for compression */ +#define BKPIMAGE_COMPRESS_BITS 2 +/* Extract the compression from the bimg_info */ +#define BKPIMAGE_COMPRESSION(info) ((info >> BKPIMAGE_COMPRESS_OFFSET_BITS) & ((1<<BKPIMAGE_COMPRESS_BITS) - 1)) /* * Extra header information used when page image has "hole" and -- 2.17.0 --45Z9DzgjV8m4Oswq Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH v8 1/9] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 9 +- doc/src/sgml/installation.sgml | 4 +- src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xloginsert.c | 65 +++++++++++-- src/backend/access/transam/xlogreader.c | 97 ++++++++++++++++--- src/backend/utils/misc/guc.c | 21 ++-- src/backend/utils/misc/postgresql.conf.sample | 2 +- src/bin/pg_waldump/pg_waldump.c | 13 ++- src/include/access/xlog.h | 2 +- src/include/access/xlog_internal.h | 10 ++ src/include/access/xlogrecord.h | 15 ++- 12 files changed, 194 insertions(+), 48 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index 7e32b0686c..218a98cfc6 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3113,23 +3113,26 @@ include_dir 'conf.d' </varlistentry> <varlistentry id="guc-wal-compression" xreflabel="wal_compression"> - <term><varname>wal_compression</varname> (<type>boolean</type>) + <term><varname>wal_compression</varname> (<type>enum</type>) <indexterm> <primary><varname>wal_compression</varname> configuration parameter</primary> </indexterm> </term> <listitem> <para> - When this parameter is <literal>on</literal>, the <productname>PostgreSQL</productname> + This parameter enables compression of WAL using the specified + compression method. + When enabled, the <productname>PostgreSQL</productname> server compresses full page images written to WAL when <xref linkend="guc-full-page-writes"/> is on or during a base backup. A compressed page image will be decompressed during WAL replay. + The supported methods are pglz and zlib. The default value is <literal>off</literal>. Only superusers can change this setting. </para> <para> - Turning this parameter on can reduce the WAL volume without + Enabling compression can reduce the WAL volume without increasing the risk of unrecoverable data corruption, but at the cost of some extra CPU spent on the compression during WAL logging and on the decompression during WAL replay. diff --git a/doc/src/sgml/installation.sgml b/doc/src/sgml/installation.sgml index 3c0aa118c7..073d5089f7 100644 --- a/doc/src/sgml/installation.sgml +++ b/doc/src/sgml/installation.sgml @@ -147,7 +147,7 @@ su - postgres specify the <option>--without-zlib</option> option to <filename>configure</filename>. Using this option disables support for compressed archives in <application>pg_dump</application> and - <application>pg_restore</application>. + <application>pg_restore</application>, and compressed WAL. </para> </listitem> </itemizedlist> @@ -1236,7 +1236,7 @@ build-postgresql: Prevents use of the <application>Zlib</application> library. This disables support for compressed archives in <application>pg_dump</application> - and <application>pg_restore</application>. + and <application>pg_restore</application> and compressed WAL. </para> </listitem> </varlistentry> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 441a9124cd..64094e7175 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -98,7 +98,7 @@ char *XLogArchiveCommand = NULL; bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; -bool wal_compression = false; +int wal_compression = WAL_COMPRESSION_NONE; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 32b4cc84e7..4f81f19c49 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,8 +33,18 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +/* zlib compressBound is not a macro */ +#define ZLIB_MAX_BLCKSZ BLCKSZ + (BLCKSZ>>12) + (BLCKSZ>>14) + (BLCKSZ>>25) + 13 +#else +#define ZLIB_MAX_BLCKSZ 0 +#endif + /* Buffer size required to store a compressed version of backup block image */ -#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) +#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) + +#define COMPRESS_BUFSIZE Max(PGLZ_MAX_BLCKSZ, ZLIB_MAX_BLCKSZ) /* * For each block reference registered with XLogRegisterBuffer, we fill in @@ -58,7 +68,7 @@ typedef struct * backup block data in XLogRecordAssemble() */ /* buffer to store a compressed version of backup block image */ - char compressed_page[PGLZ_MAX_BLCKSZ]; + char compressed_page[COMPRESS_BUFSIZE]; } registered_buffer; static registered_buffer *registered_buffers; @@ -113,7 +123,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -628,13 +639,14 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, /* * Try to compress a block image if wal_compression is enabled */ - if (wal_compression) + if (wal_compression != WAL_COMPRESSION_NONE) { is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, + wal_compression); } /* @@ -665,8 +677,13 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, if (is_compressed) { + /* The current compression is stored in the WAL record */ + wal_compression_name(wal_compression); /* Range check */ + Assert(wal_compression < (1 << BKPIMAGE_COMPRESS_BITS)); + bimg.length = compressed_len; - bimg.bimg_info |= BKPIMAGE_IS_COMPRESSED; + bimg.bimg_info |= + wal_compression << BKPIMAGE_COMPRESS_OFFSET_BITS; rdt_datas_last->data = regbuf->compressed_page; rdt_datas_last->len = compressed_len; @@ -827,7 +844,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +870,42 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = COMPRESS_BUFSIZE; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + len_l = -1; + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d/%s", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..822b2612cd 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -26,6 +26,7 @@ #include "catalog/pg_control.h" #include "common/pg_lzcompress.h" #include "replication/origin.h" +#include "utils/guc.h" #ifndef FRONTEND #include "miscadmin.h" @@ -33,6 +34,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -50,6 +55,27 @@ static void WALOpenSegmentInit(WALOpenSegment *seg, WALSegmentContext *segcxt, /* size of the buffer allocated for error message. */ #define MAX_ERRORMSG_LEN 1000 +/* + * Accept the likely variants for none and pglz, for compatibility with old + * server versions where wal_compression was a boolean. + */ +const struct config_enum_entry wal_compression_options[] = { + {"off", WAL_COMPRESSION_NONE, false}, + {"none", WAL_COMPRESSION_NONE, false}, + {"false", WAL_COMPRESSION_NONE, true}, + {"no", WAL_COMPRESSION_NONE, true}, + {"0", WAL_COMPRESSION_NONE, true}, + {"pglz", WAL_COMPRESSION_PGLZ, false}, + {"true", WAL_COMPRESSION_PGLZ, true}, + {"yes", WAL_COMPRESSION_PGLZ, true}, + {"on", WAL_COMPRESSION_PGLZ, true}, + {"1", WAL_COMPRESSION_PGLZ, true}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Construct a string in state->errormsg_buf explaining what's wrong with * the current record being read. @@ -1290,7 +1316,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); - if (blk->bimg_info & BKPIMAGE_IS_COMPRESSED) + if (BKPIMAGE_IS_COMPRESSED(blk->bimg_info)) { if (blk->bimg_info & BKPIMAGE_HAS_HOLE) COPY_HEADER_FIELD(&blk->hole_length, sizeof(uint16)); @@ -1335,29 +1361,28 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) } /* - * cross-check that bimg_len < BLCKSZ if the IS_COMPRESSED - * flag is set. + * cross-check that bimg_len < BLCKSZ if it's compressed */ - if ((blk->bimg_info & BKPIMAGE_IS_COMPRESSED) && + if (BKPIMAGE_IS_COMPRESSED(blk->bimg_info) && blk->bimg_len == BLCKSZ) { report_invalid_record(state, - "BKPIMAGE_IS_COMPRESSED set, but block image length %u at %X/%X", + "BKPIMAGE_IS_COMPRESSED, but block image length %u at %X/%X", (unsigned int) blk->bimg_len, LSN_FORMAT_ARGS(state->ReadRecPtr)); goto err; } /* - * cross-check that bimg_len = BLCKSZ if neither HAS_HOLE nor - * IS_COMPRESSED flag is set. + * cross-check that bimg_len = BLCKSZ if neither HAS_HOLE is + * set nor IS_COMPRESSED(). */ if (!(blk->bimg_info & BKPIMAGE_HAS_HOLE) && - !(blk->bimg_info & BKPIMAGE_IS_COMPRESSED) && + !BKPIMAGE_IS_COMPRESSED(blk->bimg_info) && blk->bimg_len != BLCKSZ) { report_invalid_record(state, - "neither BKPIMAGE_HAS_HOLE nor BKPIMAGE_IS_COMPRESSED set, but block image length is %u at %X/%X", + "neither BKPIMAGE_HAS_HOLE nor BKPIMAGE_IS_COMPRESSED, but block image length is %u at %X/%X", (unsigned int) blk->data_len, LSN_FORMAT_ARGS(state->ReadRecPtr)); goto err; @@ -1535,6 +1560,22 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. + */ +const char * +wal_compression_name(WalCompression compression) +{ + for (int i=0; wal_compression_options[i].name != NULL; ++i) + { + if (wal_compression_options[i].val == compression) + return wal_compression_options[i].name; + } + + return "???"; +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1555,11 +1596,43 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) bkpb = &record->blocks[block_id]; ptr = bkpb->bkp_image; - if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) + if (BKPIMAGE_IS_COMPRESSED(bkpb->bimg_info)) { + int compression_method = BKPIMAGE_COMPRESSION(bkpb->bimg_info); /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + compression_method, + wal_compression_name(compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index ee731044b6..8860deda2a 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -548,6 +548,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -1304,16 +1305,6 @@ static struct config_bool ConfigureNamesBool[] = NULL, NULL, NULL }, - { - {"wal_compression", PGC_SUSET, WAL_SETTINGS, - gettext_noop("Compresses full-page writes written in WAL file."), - NULL - }, - &wal_compression, - false, - NULL, NULL, NULL - }, - { {"wal_init_zero", PGC_SUSET, WAL_SETTINGS, gettext_noop("Writes zeroes to new WAL files before first use."), @@ -4825,6 +4816,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression", PGC_SUSET, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression, + WAL_COMPRESSION_NONE, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index 6e36e4c2ef..3991d35afd 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -218,7 +218,7 @@ #full_page_writes = on # recover from partial page writes #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) -#wal_compression = off # enable compression of full-page writes +#wal_compression = off # enable compression of full-page writes: off, pglz, zlib #wal_init_zero = on # zero-fill new WAL files #wal_recycle = on # recycle WAL files #wal_buffers = -1 # min 32kB, -1 sets based on shared_buffers diff --git a/src/bin/pg_waldump/pg_waldump.c b/src/bin/pg_waldump/pg_waldump.c index f8b8afe4a7..1cd71ac2f7 100644 --- a/src/bin/pg_waldump/pg_waldump.c +++ b/src/bin/pg_waldump/pg_waldump.c @@ -537,18 +537,21 @@ XLogDumpDisplayRecord(XLogDumpConfig *config, XLogReaderState *record) blk); if (XLogRecHasBlockImage(record, block_id)) { - if (record->blocks[block_id].bimg_info & - BKPIMAGE_IS_COMPRESSED) + if (BKPIMAGE_IS_COMPRESSED(record->blocks[block_id].bimg_info)) { + int compression = BKPIMAGE_COMPRESSION( + record->blocks[block_id].bimg_info); + printf(" (FPW%s); hole: offset: %u, length: %u, " - "compression saved: %u", + "compression method %d/%s, saved: %u", XLogRecBlockImageApply(record, block_id) ? "" : " for WAL verification", record->blocks[block_id].hole_offset, record->blocks[block_id].hole_length, + compression, wal_compression_name(compression), BLCKSZ - - record->blocks[block_id].hole_length - - record->blocks[block_id].bimg_len); + record->blocks[block_id].hole_length - + record->blocks[block_id].bimg_len); } else { diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 77187c12be..e8b2c53784 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -116,7 +116,7 @@ extern char *XLogArchiveCommand; extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; -extern bool wal_compression; +extern int wal_compression; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 26a743b6b6..8b740af66d 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,14 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +/* These are the compression IDs written into bimg_info */ +typedef enum WalCompression +{ + WAL_COMPRESSION_NONE, + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..2a60c0fb92 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -144,9 +144,18 @@ typedef struct XLogRecordBlockImageHeader /* Information stored in bimg_info */ #define BKPIMAGE_HAS_HOLE 0x01 /* page image has "hole" */ -#define BKPIMAGE_IS_COMPRESSED 0x02 /* page image is compressed */ -#define BKPIMAGE_APPLY 0x04 /* page image should be restored during +#define BKPIMAGE_APPLY 0x02 /* page image should be restored during * replay */ +#define BKPIMAGE_COMPRESS_METHOD1 0x04 /* bits to encode compression method */ +#define BKPIMAGE_COMPRESS_METHOD2 0x08 /* 0=none, 1=pglz, 2=zlib */ + +/* How many bits to shift to extract compression */ +#define BKPIMAGE_COMPRESS_OFFSET_BITS 2 +/* How many bits are for compression */ +#define BKPIMAGE_COMPRESS_BITS 2 +/* Extract the compression from the bimg_info */ +#define BKPIMAGE_COMPRESSION(info) ((info >> BKPIMAGE_COMPRESS_OFFSET_BITS) & ((1<<BKPIMAGE_COMPRESS_BITS) - 1)) +#define BKPIMAGE_IS_COMPRESSED(info) (BKPIMAGE_COMPRESSION(info) != 0) /* * Extra header information used when page image has "hole" and -- 2.17.0 --P+33d92oIH25kiaB Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="v8-0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++-- src/backend/access/transam/xlogreader.c | 63 ++++++++++++++++++- src/backend/utils/misc/guc.c | 11 ++++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 +++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 +-- 11 files changed, 163 insertions(+), 12 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f4d1ce5dea..15da91a8dd 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..34e1227381 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..afca22a26c 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,39 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..d653839b97 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --XsQoSWH+UP9D9v3l Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++-- src/backend/access/transam/xlogreader.c | 63 ++++++++++++++++++- src/backend/utils/misc/guc.c | 11 ++++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 +++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 +-- 11 files changed, 163 insertions(+), 12 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f4d1ce5dea..15da91a8dd 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..34e1227381 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..afca22a26c 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,39 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..d653839b97 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --XsQoSWH+UP9D9v3l Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 67 ++++++++++++++++--- src/backend/access/transam/xlogreader.c | 64 +++++++++++++++++- src/backend/utils/misc/guc.c | 11 +++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 16 +++++ src/include/access/xlogrecord.h | 11 ++- 10 files changed, 187 insertions(+), 13 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f4d1ce5dea..15da91a8dd 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..a93b33464f 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,8 +33,18 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +/* zlib compressBound is not a macro */ +#define ZLIB_MAX_BLCKSZ BLCKSZ + (BLCKSZ>>12) + (BLCKSZ>>14) + (BLCKSZ>>25) + 13 +#else +#define ZLIB_MAX_BLCKSZ 0 +#endif + /* Buffer size required to store a compressed version of backup block image */ -#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) +#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) + +#define COMPRESS_BUFSIZE Max(PGLZ_MAX_BLCKSZ, ZLIB_MAX_BLCKSZ) /* * For each block reference registered with XLogRegisterBuffer, we fill in @@ -58,7 +68,7 @@ typedef struct * backup block data in XLogRecordAssemble() */ /* buffer to store a compressed version of backup block image */ - char compressed_page[PGLZ_MAX_BLCKSZ]; + char compressed_page[COMPRESS_BUFSIZE]; } registered_buffer; static registered_buffer *registered_buffers; @@ -113,7 +123,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -625,16 +636,26 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, cbimg.hole_length = 0; } + bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; + /* * Try to compress a block image if wal_compression is enabled */ if (wal_compression) { + int compression; + /* The current compression is stored in the WAL record */ + wal_compression_name(wal_compression_method); /* Range check */ + compression = walmethods[wal_compression_method].walmethod; + Assert(compression < (1 << BKPIMAGE_COMPRESS_BITS)); + bimg.bimg_info |= + compression << BKPIMAGE_COMPRESS_OFFSET_BITS; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, + wal_compression_method); } /* @@ -652,8 +673,6 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, rdt_datas_last->next = ®buf->bkp_rdatas[0]; rdt_datas_last = rdt_datas_last->next; - bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; - /* * If WAL consistency checking is enabled for the resource manager * of this WAL record, a full-page image is included in the record @@ -827,7 +846,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +872,42 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = COMPRESS_BUFSIZE; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + len_l = -1; + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..0d8830fc50 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1535,6 +1539,30 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* This is a mapping indexed by wal_compression */ +// XXX: maybe this is better done as a GUC hook to assign the 1) method; and 2) level +struct walcompression walmethods[] = { + {"pglz", WAL_COMPRESSION_PGLZ}, + {"zlib", WAL_COMPRESSION_ZLIB}, +}; + +/* + * Return a statically allocated string associated with the given compression + * method. + * This is here to be visible to frontend tools like pg_rewind. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + if (compression < sizeof(walmethods)/sizeof(*walmethods)) + return walmethods[compression].name; + return "???"; +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1557,9 +1585,41 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { + int compression_method = BKPIMAGE_COMPRESSION(bkpb->bimg_info); /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + compression_method, + wal_compression_name(compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..b000a21557 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,20 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +struct walcompression +{ + char *name; + int walmethod; /* Compression method to be stored in WAL */ +}; + +extern struct walcompression walmethods[]; + +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..7107cf6186 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -147,6 +147,15 @@ typedef struct XLogRecordBlockImageHeader #define BKPIMAGE_IS_COMPRESSED 0x02 /* page image is compressed */ #define BKPIMAGE_APPLY 0x04 /* page image should be restored during * replay */ +#define BKPIMAGE_COMPRESS_METHOD1 0x08 /* bits to encode compression method */ +#define BKPIMAGE_COMPRESS_METHOD2 0x10 /* 0=pglz; 1=zlib; */ + +/* How many bits to shift to extract compression */ +#define BKPIMAGE_COMPRESS_OFFSET_BITS 3 +/* How many bits are for compression */ +#define BKPIMAGE_COMPRESS_BITS 2 +/* Extract the compression from the bimg_info */ +#define BKPIMAGE_COMPRESSION(info) ((info >> BKPIMAGE_COMPRESS_OFFSET_BITS) & ((1<<BKPIMAGE_COMPRESS_BITS) - 1)) /* * Extra header information used when page image has "hole" and -- 2.17.0 --jozmn01XJZjDjM3N Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/12] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 67 ++++++++++++++++--- src/backend/access/transam/xlogreader.c | 64 +++++++++++++++++- src/backend/utils/misc/guc.c | 11 +++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 16 +++++ src/include/access/xlogrecord.h | 11 ++- 10 files changed, 187 insertions(+), 13 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index ee4925d6d9..bed89d055d 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 6f8810e149..5b657ec724 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..a93b33464f 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,8 +33,18 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +/* zlib compressBound is not a macro */ +#define ZLIB_MAX_BLCKSZ BLCKSZ + (BLCKSZ>>12) + (BLCKSZ>>14) + (BLCKSZ>>25) + 13 +#else +#define ZLIB_MAX_BLCKSZ 0 +#endif + /* Buffer size required to store a compressed version of backup block image */ -#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) +#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) + +#define COMPRESS_BUFSIZE Max(PGLZ_MAX_BLCKSZ, ZLIB_MAX_BLCKSZ) /* * For each block reference registered with XLogRegisterBuffer, we fill in @@ -58,7 +68,7 @@ typedef struct * backup block data in XLogRecordAssemble() */ /* buffer to store a compressed version of backup block image */ - char compressed_page[PGLZ_MAX_BLCKSZ]; + char compressed_page[COMPRESS_BUFSIZE]; } registered_buffer; static registered_buffer *registered_buffers; @@ -113,7 +123,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -625,16 +636,26 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, cbimg.hole_length = 0; } + bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; + /* * Try to compress a block image if wal_compression is enabled */ if (wal_compression) { + int compression; + /* The current compression is stored in the WAL record */ + wal_compression_name(wal_compression_method); /* Range check */ + compression = walmethods[wal_compression_method].walmethod; + Assert(compression < (1 << BKPIMAGE_COMPRESS_BITS)); + bimg.bimg_info |= + compression << BKPIMAGE_COMPRESS_OFFSET_BITS; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, + wal_compression_method); } /* @@ -652,8 +673,6 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, rdt_datas_last->next = ®buf->bkp_rdatas[0]; rdt_datas_last = rdt_datas_last->next; - bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; - /* * If WAL consistency checking is enabled for the resource manager * of this WAL record, a full-page image is included in the record @@ -827,7 +846,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +872,42 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = COMPRESS_BUFSIZE; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + len_l = -1; + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..0d8830fc50 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1535,6 +1539,30 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* This is a mapping indexed by wal_compression */ +// XXX: maybe this is better done as a GUC hook to assign the 1) method; and 2) level +struct walcompression walmethods[] = { + {"pglz", WAL_COMPRESSION_PGLZ}, + {"zlib", WAL_COMPRESSION_ZLIB}, +}; + +/* + * Return a statically allocated string associated with the given compression + * method. + * This is here to be visible to frontend tools like pg_rewind. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + if (compression < sizeof(walmethods)/sizeof(*walmethods)) + return walmethods[compression].name; + return "???"; +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1557,9 +1585,41 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { + int compression_method = BKPIMAGE_COMPRESSION(bkpb->bimg_info); /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + compression_method, + wal_compression_name(compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 3b36a31a47..47d30d6cda 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -517,6 +517,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4760,6 +4761,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index 86425965d0..530007f55a 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 77187c12be..2f3886431a 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..b000a21557 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,20 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +struct walcompression +{ + char *name; + int walmethod; /* Compression method to be stored in WAL */ +}; + +extern struct walcompression walmethods[]; + +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..7107cf6186 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -147,6 +147,15 @@ typedef struct XLogRecordBlockImageHeader #define BKPIMAGE_IS_COMPRESSED 0x02 /* page image is compressed */ #define BKPIMAGE_APPLY 0x04 /* page image should be restored during * replay */ +#define BKPIMAGE_COMPRESS_METHOD1 0x08 /* bits to encode compression method */ +#define BKPIMAGE_COMPRESS_METHOD2 0x10 /* 0=pglz; 1=zlib; */ + +/* How many bits to shift to extract compression */ +#define BKPIMAGE_COMPRESS_OFFSET_BITS 3 +/* How many bits are for compression */ +#define BKPIMAGE_COMPRESS_BITS 2 +/* Extract the compression from the bimg_info */ +#define BKPIMAGE_COMPRESSION(info) ((info >> BKPIMAGE_COMPRESS_OFFSET_BITS) & ((1<<BKPIMAGE_COMPRESS_BITS) - 1)) /* * Extra header information used when page image has "hole" and -- 2.17.0 --45Z9DzgjV8m4Oswq Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 67 ++++++++++++++++--- src/backend/access/transam/xlogreader.c | 64 +++++++++++++++++- src/backend/utils/misc/guc.c | 11 +++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 16 +++++ src/include/access/xlogrecord.h | 11 ++- 10 files changed, 187 insertions(+), 13 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f4d1ce5dea..15da91a8dd 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..a93b33464f 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,8 +33,18 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +/* zlib compressBound is not a macro */ +#define ZLIB_MAX_BLCKSZ BLCKSZ + (BLCKSZ>>12) + (BLCKSZ>>14) + (BLCKSZ>>25) + 13 +#else +#define ZLIB_MAX_BLCKSZ 0 +#endif + /* Buffer size required to store a compressed version of backup block image */ -#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) +#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) + +#define COMPRESS_BUFSIZE Max(PGLZ_MAX_BLCKSZ, ZLIB_MAX_BLCKSZ) /* * For each block reference registered with XLogRegisterBuffer, we fill in @@ -58,7 +68,7 @@ typedef struct * backup block data in XLogRecordAssemble() */ /* buffer to store a compressed version of backup block image */ - char compressed_page[PGLZ_MAX_BLCKSZ]; + char compressed_page[COMPRESS_BUFSIZE]; } registered_buffer; static registered_buffer *registered_buffers; @@ -113,7 +123,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -625,16 +636,26 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, cbimg.hole_length = 0; } + bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; + /* * Try to compress a block image if wal_compression is enabled */ if (wal_compression) { + int compression; + /* The current compression is stored in the WAL record */ + wal_compression_name(wal_compression_method); /* Range check */ + compression = walmethods[wal_compression_method].walmethod; + Assert(compression < (1 << BKPIMAGE_COMPRESS_BITS)); + bimg.bimg_info |= + compression << BKPIMAGE_COMPRESS_OFFSET_BITS; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, + wal_compression_method); } /* @@ -652,8 +673,6 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, rdt_datas_last->next = ®buf->bkp_rdatas[0]; rdt_datas_last = rdt_datas_last->next; - bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; - /* * If WAL consistency checking is enabled for the resource manager * of this WAL record, a full-page image is included in the record @@ -827,7 +846,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +872,42 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = COMPRESS_BUFSIZE; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + len_l = -1; + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..0d8830fc50 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1535,6 +1539,30 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* This is a mapping indexed by wal_compression */ +// XXX: maybe this is better done as a GUC hook to assign the 1) method; and 2) level +struct walcompression walmethods[] = { + {"pglz", WAL_COMPRESSION_PGLZ}, + {"zlib", WAL_COMPRESSION_ZLIB}, +}; + +/* + * Return a statically allocated string associated with the given compression + * method. + * This is here to be visible to frontend tools like pg_rewind. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + if (compression < sizeof(walmethods)/sizeof(*walmethods)) + return walmethods[compression].name; + return "???"; +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1557,9 +1585,41 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { + int compression_method = BKPIMAGE_COMPRESSION(bkpb->bimg_info); /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + compression_method, + wal_compression_name(compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..b000a21557 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,20 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +struct walcompression +{ + char *name; + int walmethod; /* Compression method to be stored in WAL */ +}; + +extern struct walcompression walmethods[]; + +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..7107cf6186 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -147,6 +147,15 @@ typedef struct XLogRecordBlockImageHeader #define BKPIMAGE_IS_COMPRESSED 0x02 /* page image is compressed */ #define BKPIMAGE_APPLY 0x04 /* page image should be restored during * replay */ +#define BKPIMAGE_COMPRESS_METHOD1 0x08 /* bits to encode compression method */ +#define BKPIMAGE_COMPRESS_METHOD2 0x10 /* 0=pglz; 1=zlib; */ + +/* How many bits to shift to extract compression */ +#define BKPIMAGE_COMPRESS_OFFSET_BITS 3 +/* How many bits are for compression */ +#define BKPIMAGE_COMPRESS_BITS 2 +/* Extract the compression from the bimg_info */ +#define BKPIMAGE_COMPRESSION(info) ((info >> BKPIMAGE_COMPRESS_OFFSET_BITS) & ((1<<BKPIMAGE_COMPRESS_BITS) - 1)) /* * Extra header information used when page image has "hole" and -- 2.17.0 --jozmn01XJZjDjM3N Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++-- src/backend/access/transam/xlogreader.c | 63 ++++++++++++++++++- src/backend/utils/misc/guc.c | 11 ++++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 +++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 +-- 11 files changed, 163 insertions(+), 12 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f4d1ce5dea..15da91a8dd 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..34e1227381 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..afca22a26c 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,39 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..d653839b97 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --XsQoSWH+UP9D9v3l Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++-- src/backend/access/transam/xlogreader.c | 63 ++++++++++++++++++- src/backend/utils/misc/guc.c | 11 ++++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 +++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 +-- 11 files changed, 163 insertions(+), 12 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f4d1ce5dea..15da91a8dd 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..34e1227381 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..afca22a26c 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,39 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..d653839b97 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --XsQoSWH+UP9D9v3l Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 67 ++++++++++++++++--- src/backend/access/transam/xlogreader.c | 64 +++++++++++++++++- src/backend/utils/misc/guc.c | 11 +++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 16 +++++ src/include/access/xlogrecord.h | 11 ++- 10 files changed, 187 insertions(+), 13 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f4d1ce5dea..15da91a8dd 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..a93b33464f 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,8 +33,18 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +/* zlib compressBound is not a macro */ +#define ZLIB_MAX_BLCKSZ BLCKSZ + (BLCKSZ>>12) + (BLCKSZ>>14) + (BLCKSZ>>25) + 13 +#else +#define ZLIB_MAX_BLCKSZ 0 +#endif + /* Buffer size required to store a compressed version of backup block image */ -#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) +#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) + +#define COMPRESS_BUFSIZE Max(PGLZ_MAX_BLCKSZ, ZLIB_MAX_BLCKSZ) /* * For each block reference registered with XLogRegisterBuffer, we fill in @@ -58,7 +68,7 @@ typedef struct * backup block data in XLogRecordAssemble() */ /* buffer to store a compressed version of backup block image */ - char compressed_page[PGLZ_MAX_BLCKSZ]; + char compressed_page[COMPRESS_BUFSIZE]; } registered_buffer; static registered_buffer *registered_buffers; @@ -113,7 +123,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -625,16 +636,26 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, cbimg.hole_length = 0; } + bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; + /* * Try to compress a block image if wal_compression is enabled */ if (wal_compression) { + int compression; + /* The current compression is stored in the WAL record */ + wal_compression_name(wal_compression_method); /* Range check */ + compression = walmethods[wal_compression_method].walmethod; + Assert(compression < (1 << BKPIMAGE_COMPRESS_BITS)); + bimg.bimg_info |= + compression << BKPIMAGE_COMPRESS_OFFSET_BITS; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, + wal_compression_method); } /* @@ -652,8 +673,6 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, rdt_datas_last->next = ®buf->bkp_rdatas[0]; rdt_datas_last = rdt_datas_last->next; - bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; - /* * If WAL consistency checking is enabled for the resource manager * of this WAL record, a full-page image is included in the record @@ -827,7 +846,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +872,42 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = COMPRESS_BUFSIZE; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + len_l = -1; + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..0d8830fc50 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1535,6 +1539,30 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* This is a mapping indexed by wal_compression */ +// XXX: maybe this is better done as a GUC hook to assign the 1) method; and 2) level +struct walcompression walmethods[] = { + {"pglz", WAL_COMPRESSION_PGLZ}, + {"zlib", WAL_COMPRESSION_ZLIB}, +}; + +/* + * Return a statically allocated string associated with the given compression + * method. + * This is here to be visible to frontend tools like pg_rewind. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + if (compression < sizeof(walmethods)/sizeof(*walmethods)) + return walmethods[compression].name; + return "???"; +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1557,9 +1585,41 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { + int compression_method = BKPIMAGE_COMPRESSION(bkpb->bimg_info); /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + compression_method, + wal_compression_name(compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..b000a21557 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,20 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +struct walcompression +{ + char *name; + int walmethod; /* Compression method to be stored in WAL */ +}; + +extern struct walcompression walmethods[]; + +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..7107cf6186 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -147,6 +147,15 @@ typedef struct XLogRecordBlockImageHeader #define BKPIMAGE_IS_COMPRESSED 0x02 /* page image is compressed */ #define BKPIMAGE_APPLY 0x04 /* page image should be restored during * replay */ +#define BKPIMAGE_COMPRESS_METHOD1 0x08 /* bits to encode compression method */ +#define BKPIMAGE_COMPRESS_METHOD2 0x10 /* 0=pglz; 1=zlib; */ + +/* How many bits to shift to extract compression */ +#define BKPIMAGE_COMPRESS_OFFSET_BITS 3 +/* How many bits are for compression */ +#define BKPIMAGE_COMPRESS_BITS 2 +/* Extract the compression from the bimg_info */ +#define BKPIMAGE_COMPRESSION(info) ((info >> BKPIMAGE_COMPRESS_OFFSET_BITS) & ((1<<BKPIMAGE_COMPRESS_BITS) - 1)) /* * Extra header information used when page image has "hole" and -- 2.17.0 --jozmn01XJZjDjM3N Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 67 ++++++++++++++++--- src/backend/access/transam/xlogreader.c | 64 +++++++++++++++++- src/backend/utils/misc/guc.c | 11 +++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 16 +++++ src/include/access/xlogrecord.h | 11 ++- 10 files changed, 187 insertions(+), 13 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f4d1ce5dea..15da91a8dd 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..a93b33464f 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,8 +33,18 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +/* zlib compressBound is not a macro */ +#define ZLIB_MAX_BLCKSZ BLCKSZ + (BLCKSZ>>12) + (BLCKSZ>>14) + (BLCKSZ>>25) + 13 +#else +#define ZLIB_MAX_BLCKSZ 0 +#endif + /* Buffer size required to store a compressed version of backup block image */ -#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) +#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) + +#define COMPRESS_BUFSIZE Max(PGLZ_MAX_BLCKSZ, ZLIB_MAX_BLCKSZ) /* * For each block reference registered with XLogRegisterBuffer, we fill in @@ -58,7 +68,7 @@ typedef struct * backup block data in XLogRecordAssemble() */ /* buffer to store a compressed version of backup block image */ - char compressed_page[PGLZ_MAX_BLCKSZ]; + char compressed_page[COMPRESS_BUFSIZE]; } registered_buffer; static registered_buffer *registered_buffers; @@ -113,7 +123,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -625,16 +636,26 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, cbimg.hole_length = 0; } + bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; + /* * Try to compress a block image if wal_compression is enabled */ if (wal_compression) { + int compression; + /* The current compression is stored in the WAL record */ + wal_compression_name(wal_compression_method); /* Range check */ + compression = walmethods[wal_compression_method].walmethod; + Assert(compression < (1 << BKPIMAGE_COMPRESS_BITS)); + bimg.bimg_info |= + compression << BKPIMAGE_COMPRESS_OFFSET_BITS; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, + wal_compression_method); } /* @@ -652,8 +673,6 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, rdt_datas_last->next = ®buf->bkp_rdatas[0]; rdt_datas_last = rdt_datas_last->next; - bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; - /* * If WAL consistency checking is enabled for the resource manager * of this WAL record, a full-page image is included in the record @@ -827,7 +846,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +872,42 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = COMPRESS_BUFSIZE; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + len_l = -1; + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..0d8830fc50 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1535,6 +1539,30 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* This is a mapping indexed by wal_compression */ +// XXX: maybe this is better done as a GUC hook to assign the 1) method; and 2) level +struct walcompression walmethods[] = { + {"pglz", WAL_COMPRESSION_PGLZ}, + {"zlib", WAL_COMPRESSION_ZLIB}, +}; + +/* + * Return a statically allocated string associated with the given compression + * method. + * This is here to be visible to frontend tools like pg_rewind. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + if (compression < sizeof(walmethods)/sizeof(*walmethods)) + return walmethods[compression].name; + return "???"; +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1557,9 +1585,41 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { + int compression_method = BKPIMAGE_COMPRESSION(bkpb->bimg_info); /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + compression_method, + wal_compression_name(compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..b000a21557 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,20 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +struct walcompression +{ + char *name; + int walmethod; /* Compression method to be stored in WAL */ +}; + +extern struct walcompression walmethods[]; + +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..7107cf6186 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -147,6 +147,15 @@ typedef struct XLogRecordBlockImageHeader #define BKPIMAGE_IS_COMPRESSED 0x02 /* page image is compressed */ #define BKPIMAGE_APPLY 0x04 /* page image should be restored during * replay */ +#define BKPIMAGE_COMPRESS_METHOD1 0x08 /* bits to encode compression method */ +#define BKPIMAGE_COMPRESS_METHOD2 0x10 /* 0=pglz; 1=zlib; */ + +/* How many bits to shift to extract compression */ +#define BKPIMAGE_COMPRESS_OFFSET_BITS 3 +/* How many bits are for compression */ +#define BKPIMAGE_COMPRESS_BITS 2 +/* Extract the compression from the bimg_info */ +#define BKPIMAGE_COMPRESSION(info) ((info >> BKPIMAGE_COMPRESS_OFFSET_BITS) & ((1<<BKPIMAGE_COMPRESS_BITS) - 1)) /* * Extra header information used when page image has "hole" and -- 2.17.0 --jozmn01XJZjDjM3N Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 3/8] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 ++++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++++++++-- src/backend/access/transam/xlogreader.c | 62 ++++++++++++++++++++++++- src/backend/utils/misc/guc.c | 11 +++++ src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 ++++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 ++-- 9 files changed, 144 insertions(+), 12 deletions(-) diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index e3128564e1..0183589b4d 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..ee73bc3afd 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, + Z_DEFAULT_COMPRESSION); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..143df55fcb 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,38 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l = 0; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..b80759ed45 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ = 0, + WAL_COMPRESSION_ZLIB = 1, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --f0KYrhQ4vYSV2aJu Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0004-wal_compression_method-default-to-zlib.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++-- src/backend/access/transam/xlogreader.c | 63 ++++++++++++++++++- src/backend/utils/misc/guc.c | 11 ++++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 +++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 +-- 11 files changed, 163 insertions(+), 12 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index e04250f4e9..04192b7add 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..34e1227381 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..afca22a26c 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,39 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..d653839b97 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --0qVF/w3MHQqLSynd Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 3/8] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 ++++ src/backend/access/transam/xloginsert.c | 52 +++++++++++++++++++-- src/backend/access/transam/xlogreader.c | 62 ++++++++++++++++++++++++- src/backend/utils/misc/guc.c | 11 +++++ src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 8 ++++ src/include/access/xlogreader.h | 1 + src/include/access/xlogrecord.h | 9 ++-- 9 files changed, 144 insertions(+), 12 deletions(-) diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index e3128564e1..0183589b4d 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..ee73bc3afd 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,6 +33,10 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + /* Buffer size required to store a compressed version of backup block image */ #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) @@ -113,7 +117,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -630,11 +635,12 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ if (wal_compression) { + bimg.compression_method = wal_compression_method; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, bimg.compression_method); } /* @@ -827,7 +833,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +859,48 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = PGLZ_MAX_BLCKSZ; + int ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, + Z_DEFAULT_COMPRESSION); + if (ret != Z_OK) + { + // XXX: using an interface other than compress() would allow giving a better error message + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("failed compressing zlib (%d)", ret))); + len_l = -1; + } + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..143df55fcb 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1286,6 +1290,7 @@ DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg) { COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16)); COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16)); + COPY_HEADER_FIELD(&blk->compression_method, sizeof(uint8)); COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8)); blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0); @@ -1535,6 +1540,29 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* + * Return a statically allocated string associated with the given compression + * method. This is similar to the guc, but isn't subject to conditional + * compilation. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + return "pglz"; + case WAL_COMPRESSION_ZLIB: + return "zlib"; + default: + return "???"; + } +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1558,8 +1586,38 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (bkpb->compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l = 0; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + bkpb->compression_method, + wal_compression_name(bkpb->compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..b80759ed45 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,12 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ = 0, + WAL_COMPRESSION_ZLIB = 1, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogreader.h b/src/include/access/xlogreader.h index 21d200d3df..3d19c315d7 100644 --- a/src/include/access/xlogreader.h +++ b/src/include/access/xlogreader.h @@ -133,6 +133,7 @@ typedef struct bool apply_image; /* has image that should be restored */ char *bkp_image; uint16 hole_offset; + uint8 compression_method; uint16 hole_length; uint16 bimg_len; uint8 bimg_info; diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..0d4c212f15 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -129,9 +129,10 @@ typedef struct XLogRecordBlockHeader */ typedef struct XLogRecordBlockImageHeader { - uint16 length; /* number of page image bytes */ - uint16 hole_offset; /* number of bytes before "hole" */ - uint8 bimg_info; /* flag bits, see below */ + uint16 length; /* number of page image bytes */ + uint16 hole_offset; /* number of bytes before "hole" */ + uint8 compression_method; /* compression method used for image */ + uint8 bimg_info; /* flag bits, see below */ /* * If BKPIMAGE_HAS_HOLE and BKPIMAGE_IS_COMPRESSED, an -- 2.17.0 --f0KYrhQ4vYSV2aJu Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0004-wal_compression_method-default-to-zlib.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 01/10] Allow alternate compression methods for wal_compression @ 2021-02-27 04:03 Andrey Borodin <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Andrey Borodin @ 2021-02-27 04:03 UTC (permalink / raw) TODO: bump XLOG_PAGE_MAGIC --- doc/src/sgml/config.sgml | 17 +++++ src/backend/Makefile | 2 +- src/backend/access/transam/xlog.c | 10 +++ src/backend/access/transam/xloginsert.c | 67 ++++++++++++++++--- src/backend/access/transam/xlogreader.c | 64 +++++++++++++++++- src/backend/utils/misc/guc.c | 11 +++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/include/access/xlog.h | 1 + src/include/access/xlog_internal.h | 16 +++++ src/include/access/xlogrecord.h | 11 ++- 10 files changed, 187 insertions(+), 13 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a218d78bef..7fb2a84626 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -3072,6 +3072,23 @@ include_dir 'conf.d' </listitem> </varlistentry> + <varlistentry id="guc-wal-compression-method" xreflabel="wal_compression_method"> + <term><varname>wal_compressionion_method</varname> (<type>enum</type>) + <indexterm> + <primary><varname>wal_compression_method</varname> configuration parameter</primary> + </indexterm> + </term> + <listitem> + <para> + This parameter selects the compression method used to compress WAL when + <varname>wal_compression</varname> is enabled. + The supported methods are pglz and zlib. + The default value is <literal>pglz</literal>. + Only superusers can change this setting. + </para> + </listitem> + </varlistentry> + <varlistentry id="guc-wal-init-zero" xreflabel="wal_init_zero"> <term><varname>wal_init_zero</varname> (<type>boolean</type>) <indexterm> diff --git a/src/backend/Makefile b/src/backend/Makefile index 0da848b1fd..3af216ddfc 100644 --- a/src/backend/Makefile +++ b/src/backend/Makefile @@ -48,7 +48,7 @@ OBJS = \ LIBS := $(filter-out -lpgport -lpgcommon, $(LIBS)) $(LDAP_LIBS_BE) $(ICU_LIBS) # The backend doesn't need everything that's in LIBS, however -LIBS := $(filter-out -lz -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) +LIBS := $(filter-out -lreadline -ledit -ltermcap -lncurses -lcurses, $(LIBS)) ifeq ($(with_systemd),yes) LIBS += -lsystemd diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f4d1ce5dea..15da91a8dd 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -99,6 +99,7 @@ bool EnableHotStandby = false; bool fullPageWrites = true; bool wal_log_hints = false; bool wal_compression = false; +int wal_compression_method = WAL_COMPRESSION_PGLZ; char *wal_consistency_checking_string = NULL; bool *wal_consistency_checking = NULL; bool wal_init_zero = true; @@ -180,6 +181,15 @@ const struct config_enum_entry recovery_target_action_options[] = { {NULL, 0, false} }; +/* Note that due to conditional compilation, offsets within the array are not static */ +const struct config_enum_entry wal_compression_options[] = { + {"pglz", WAL_COMPRESSION_PGLZ, false}, +#ifdef HAVE_LIBZ + {"zlib", WAL_COMPRESSION_ZLIB, false}, +#endif + {NULL, 0, false} +}; + /* * Statistics for current checkpoint are collected in this global struct. * Because only the checkpointer or a stand-alone backend can perform diff --git a/src/backend/access/transam/xloginsert.c b/src/backend/access/transam/xloginsert.c index 7052dc245e..a93b33464f 100644 --- a/src/backend/access/transam/xloginsert.c +++ b/src/backend/access/transam/xloginsert.c @@ -33,8 +33,18 @@ #include "storage/proc.h" #include "utils/memutils.h" +#ifdef HAVE_LIBZ +#include <zlib.h> +/* zlib compressBound is not a macro */ +#define ZLIB_MAX_BLCKSZ BLCKSZ + (BLCKSZ>>12) + (BLCKSZ>>14) + (BLCKSZ>>25) + 13 +#else +#define ZLIB_MAX_BLCKSZ 0 +#endif + /* Buffer size required to store a compressed version of backup block image */ -#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) +#define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ) + +#define COMPRESS_BUFSIZE Max(PGLZ_MAX_BLCKSZ, ZLIB_MAX_BLCKSZ) /* * For each block reference registered with XLogRegisterBuffer, we fill in @@ -58,7 +68,7 @@ typedef struct * backup block data in XLogRecordAssemble() */ /* buffer to store a compressed version of backup block image */ - char compressed_page[PGLZ_MAX_BLCKSZ]; + char compressed_page[COMPRESS_BUFSIZE]; } registered_buffer; static registered_buffer *registered_buffers; @@ -113,7 +123,8 @@ static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info, XLogRecPtr RedoRecPtr, bool doPageWrites, XLogRecPtr *fpw_lsn, int *num_fpi); static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, - uint16 hole_length, char *dest, uint16 *dlen); + uint16 hole_length, char *dest, + uint16 *dlen, WalCompression compression); /* * Begin constructing a WAL record. This must be called before the @@ -625,16 +636,26 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, cbimg.hole_length = 0; } + bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; + /* * Try to compress a block image if wal_compression is enabled */ if (wal_compression) { + int compression; + /* The current compression is stored in the WAL record */ + wal_compression_name(wal_compression_method); /* Range check */ + compression = walmethods[wal_compression_method].walmethod; + Assert(compression < (1 << BKPIMAGE_COMPRESS_BITS)); + bimg.bimg_info |= + compression << BKPIMAGE_COMPRESS_OFFSET_BITS; is_compressed = XLogCompressBackupBlock(page, bimg.hole_offset, cbimg.hole_length, regbuf->compressed_page, - &compressed_len); + &compressed_len, + wal_compression_method); } /* @@ -652,8 +673,6 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, rdt_datas_last->next = ®buf->bkp_rdatas[0]; rdt_datas_last = rdt_datas_last->next; - bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE; - /* * If WAL consistency checking is enabled for the resource manager * of this WAL record, a full-page image is included in the record @@ -827,7 +846,7 @@ XLogRecordAssemble(RmgrId rmid, uint8 info, */ static bool XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, - char *dest, uint16 *dlen) + char *dest, uint16 *dlen, WalCompression compression) { int32 orig_len = BLCKSZ - hole_length; int32 len; @@ -853,12 +872,42 @@ XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length, else source = page; + switch (compression) + { + case WAL_COMPRESSION_PGLZ: + len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long len_l = COMPRESS_BUFSIZE; + int ret; + ret = compress2((Bytef*)dest, &len_l, (Bytef*)source, orig_len, 1); + if (ret != Z_OK) + len_l = -1; + len = len_l; + break; + } +#endif + + default: + /* + * It should be impossible to get here for unsupported algorithms, + * which cannot be assigned if they're not enabled at compile time. + */ + ereport(ERROR, + (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("unknown compression method requested: %d(%s)", + compression, wal_compression_name(compression)))); + + } + /* - * We recheck the actual size even if pglz_compress() reports success and + * We recheck the actual size even if compression reports success and * see if the number of bytes saved by compression is larger than the * length of extra data needed for the compressed version of block image. */ - len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default); if (len >= 0 && len + extra_bytes < orig_len) { diff --git a/src/backend/access/transam/xlogreader.c b/src/backend/access/transam/xlogreader.c index 42738eb940..0d8830fc50 100644 --- a/src/backend/access/transam/xlogreader.c +++ b/src/backend/access/transam/xlogreader.c @@ -33,6 +33,10 @@ #include "utils/memutils.h" #endif +#ifdef HAVE_LIBZ +#include <zlib.h> +#endif + static void report_invalid_record(XLogReaderState *state, const char *fmt,...) pg_attribute_printf(2, 3); static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength); @@ -1535,6 +1539,30 @@ XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len) } } +/* This is a mapping indexed by wal_compression */ +// XXX: maybe this is better done as a GUC hook to assign the 1) method; and 2) level +struct walcompression walmethods[] = { + {"pglz", WAL_COMPRESSION_PGLZ}, + {"zlib", WAL_COMPRESSION_ZLIB}, +}; + +/* + * Return a statically allocated string associated with the given compression + * method. + * This is here to be visible to frontend tools like pg_rewind. + */ +const char * +wal_compression_name(WalCompression compression) +{ + /* + * This could index into the guc array, except that it's compiled + * conditionally and unsupported methods are elided. + */ + if (compression < sizeof(walmethods)/sizeof(*walmethods)) + return walmethods[compression].name; + return "???"; +} + /* * Restore a full-page image from a backup block attached to an XLOG record. * @@ -1557,9 +1585,41 @@ RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page) if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED) { + int compression_method = BKPIMAGE_COMPRESSION(bkpb->bimg_info); /* If a backup block image is compressed, decompress it */ - if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data, - BLCKSZ - bkpb->hole_length, true) < 0) + int32 decomp_result = -1; + switch (compression_method) + { + case WAL_COMPRESSION_PGLZ: + decomp_result = pglz_decompress(ptr, bkpb->bimg_len, tmp.data, + BLCKSZ - bkpb->hole_length, true); + break; + +#ifdef HAVE_LIBZ + case WAL_COMPRESSION_ZLIB: + { + unsigned long decomp_result_l; + decomp_result_l = BLCKSZ - bkpb->hole_length; + if (uncompress((Bytef*)tmp.data, &decomp_result_l, + (Bytef*)ptr, bkpb->bimg_len) == Z_OK) + decomp_result = decomp_result_l; + else + decomp_result = -1; + break; + } +#endif + + default: + report_invalid_record(record, "image at %X/%X is compressed with unsupported codec, block %d (%d/%s)", + (uint32) (record->ReadRecPtr >> 32), + (uint32) record->ReadRecPtr, + block_id, + compression_method, + wal_compression_name(compression_method)); + return false; + } + + if (decomp_result < 0) { report_invalid_record(record, "invalid compressed image at %X/%X, block %d", LSN_FORMAT_ARGS(record->ReadRecPtr), diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c index 855076b1fd..8084027465 100644 --- a/src/backend/utils/misc/guc.c +++ b/src/backend/utils/misc/guc.c @@ -508,6 +508,7 @@ extern const struct config_enum_entry archive_mode_options[]; extern const struct config_enum_entry recovery_target_action_options[]; extern const struct config_enum_entry sync_method_options[]; extern const struct config_enum_entry dynamic_shared_memory_options[]; +extern const struct config_enum_entry wal_compression_options[]; /* * GUC option variables that are exported from this module @@ -4721,6 +4722,16 @@ static struct config_enum ConfigureNamesEnum[] = NULL, NULL, NULL }, + { + {"wal_compression_method", PGC_SIGHUP, WAL_SETTINGS, + gettext_noop("Set the method used to compress full page images in the WAL."), + NULL + }, + &wal_compression_method, + WAL_COMPRESSION_PGLZ, wal_compression_options, + NULL, NULL, NULL + }, + { {"dynamic_shared_memory_type", PGC_POSTMASTER, RESOURCES_MEM, gettext_noop("Selects the dynamic shared memory implementation used."), diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index f46c2dd7a8..ef69a94492 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -213,6 +213,7 @@ # open_sync #full_page_writes = on # recover from partial page writes #wal_compression = off # enable compression of full-page writes +#wal_compression_method = pglz # pglz, zlib #wal_log_hints = off # also do full page writes of non-critical updates # (change requires restart) #wal_init_zero = on # zero-fill new WAL files diff --git a/src/include/access/xlog.h b/src/include/access/xlog.h index 6d384d3ce6..fa2e5c611f 100644 --- a/src/include/access/xlog.h +++ b/src/include/access/xlog.h @@ -117,6 +117,7 @@ extern bool EnableHotStandby; extern bool fullPageWrites; extern bool wal_log_hints; extern bool wal_compression; +extern int wal_compression_method; extern bool wal_init_zero; extern bool wal_recycle; extern bool *wal_consistency_checking; diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index b23e286406..b000a21557 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -324,4 +324,20 @@ extern bool InArchiveRecovery; extern bool StandbyMode; extern char *recoveryRestoreCommand; +struct walcompression +{ + char *name; + int walmethod; /* Compression method to be stored in WAL */ +}; + +extern struct walcompression walmethods[]; + +typedef enum WalCompression +{ + WAL_COMPRESSION_PGLZ, + WAL_COMPRESSION_ZLIB, +} WalCompression; + +extern const char *wal_compression_name(WalCompression compression); + #endif /* XLOG_INTERNAL_H */ diff --git a/src/include/access/xlogrecord.h b/src/include/access/xlogrecord.h index 80c92a2498..7107cf6186 100644 --- a/src/include/access/xlogrecord.h +++ b/src/include/access/xlogrecord.h @@ -114,7 +114,7 @@ typedef struct XLogRecordBlockHeader * present is (BLCKSZ - <length of "hole" bytes>). * * Additionally, when wal_compression is enabled, we will try to compress full - * page images using the PGLZ compression algorithm, after removing the "hole". + * page images, after removing the "hole". * This can reduce the WAL volume, but at some extra cost of CPU spent * on the compression during WAL logging. In this case, since the "hole" * length cannot be calculated by subtracting the number of page image bytes @@ -147,6 +147,15 @@ typedef struct XLogRecordBlockImageHeader #define BKPIMAGE_IS_COMPRESSED 0x02 /* page image is compressed */ #define BKPIMAGE_APPLY 0x04 /* page image should be restored during * replay */ +#define BKPIMAGE_COMPRESS_METHOD1 0x08 /* bits to encode compression method */ +#define BKPIMAGE_COMPRESS_METHOD2 0x10 /* 0=pglz; 1=zlib; */ + +/* How many bits to shift to extract compression */ +#define BKPIMAGE_COMPRESS_OFFSET_BITS 3 +/* How many bits are for compression */ +#define BKPIMAGE_COMPRESS_BITS 2 +/* Extract the compression from the bimg_info */ +#define BKPIMAGE_COMPRESSION(info) ((info >> BKPIMAGE_COMPRESS_OFFSET_BITS) & ((1<<BKPIMAGE_COMPRESS_BITS) - 1)) /* * Extra header information used when page image has "hole" and -- 2.17.0 --jozmn01XJZjDjM3N Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="0002-Run-011_crash_recovery.pl-with-wal_level-minimal.patch" ^ permalink raw reply [nested|flat] 102+ messages in thread
* Re: [PoC] XMLCast (SQL/XML X025) @ 2026-01-13 09:31 Jim Jones <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Jim Jones @ 2026-01-13 09:31 UTC (permalink / raw) To: Marcos Pegoraro <[email protected]>; +Cc: Robert Haas <[email protected]>; [email protected]; Vik Fearing <[email protected]> rebase Jim Attachments: [text/x-patch] v18-0001-Add-XMLCast-function-SQL-XML-X025.patch (135.3K, ../../[email protected]/2-v18-0001-Add-XMLCast-function-SQL-XML-X025.patch) download | inline diff: From 3de0b6cb56a604d4a9be346235569936afdb0d27 Mon Sep 17 00:00:00 2001 From: Jim Jones <[email protected]> Date: Tue, 13 Jan 2026 10:25:12 +0100 Subject: [PATCH v18] Add XMLCast function (SQL/XML X025) This patch introduces support for the XMLCAST function, as specified in SQL/XML:2023 (ISO/IEC 9075-14:2023), Subclause 6.7 "<XML cast specification>". It enables standards-compliant conversion between SQL data types and XML, following the lexical rules defined by the W3C XML Schema Part 2. XMLCast provides an alternative to CAST when converting SQL values into XML content, ensuring the output uses canonical XML Schema lexical representations. For example, timestamp and interval values are rendered as xs:dateTime and xs:duration (e.g., "2024-01-01T12:00:00Z" or "P1Y2M"), conforming to ISO 8601 formats. Conversely, XMLCast also allows converting XML content back into SQL types (e.g., boolean, numeric, date/time), validating the input string according to XML Schema lexical forms. Supported casts include: * SQL -> XML: boolean, numeric, character, date/time, interval, binary * XML -> SQL: the inverse of the above, with lexical validation The BY REF and BY VALUE clauses are accepted for SQL/XML compatibility, but ignored. To support proper handling of infinity values in numeric types during XML casting (as required by XML Schema specifications), two helper functions (numeric_is_inf and numeric_is_positive_inf) have been added to numeric.c. --- doc/src/sgml/datatype.sgml | 89 +++- src/backend/catalog/sql_features.txt | 2 +- src/backend/executor/execExprInterp.c | 135 +++++- src/backend/nodes/nodeFuncs.c | 13 + src/backend/parser/gram.y | 22 +- src/backend/parser/parse_expr.c | 90 ++++ src/backend/parser/parse_target.c | 7 + src/backend/utils/adt/numeric.c | 22 + src/backend/utils/adt/ruleutils.c | 10 +- src/backend/utils/adt/xml.c | 121 ++++- src/include/nodes/parsenodes.h | 8 + src/include/nodes/primnodes.h | 3 + src/include/parser/kwlist.h | 1 + src/include/utils/numeric.h | 2 + src/include/utils/xml.h | 1 + src/test/regress/expected/xml.out | 607 +++++++++++++++++++++++++ src/test/regress/expected/xml_1.out | 453 +++++++++++++++++++ src/test/regress/expected/xml_2.out | 609 ++++++++++++++++++++++++++ src/test/regress/sql/xml.sql | 290 ++++++++++++ src/tools/pgindent/typedefs.list | 1 + 20 files changed, 2473 insertions(+), 13 deletions(-) diff --git a/doc/src/sgml/datatype.sgml b/doc/src/sgml/datatype.sgml index 3017c67404..1aa90f6bda 100644 --- a/doc/src/sgml/datatype.sgml +++ b/doc/src/sgml/datatype.sgml @@ -4495,14 +4495,95 @@ XMLPARSE ( { DOCUMENT | CONTENT } <replaceable>value</replaceable>) XMLPARSE (DOCUMENT '<?xml version="1.0"?><book><title>Manual</title><chapter>...</chapter></book>') XMLPARSE (CONTENT 'abc<foo>bar</foo><bar>foo</bar>') ]]></programlisting> - While this is the only way to convert character strings into XML - values according to the SQL standard, the PostgreSQL-specific - syntaxes: + + Another option for converting values to or from <type>xml</type> is the <function>xmlcast</function> function, + which is designed to cast SQL data types into <type>xml</type>, and vice versa, in a standards-compliant way. +<synopsis> +XMLCAST ( <replaceable>expression</replaceable> AS <replaceable>type</replaceable> [ BY REF | BY VALUE ] ) +</synopsis> + Similar to the SQL function <function>CAST</function>, this function converts an <replaceable>expression</replaceable> + into the specified <replaceable>type</replaceable>. It is primarily used for converting between SQL values + and <type>xml</type> values in a standards-compliant way. + + Unlike <function>CAST</function>, which may coerce SQL values into text or XML without enforcing a specific + lexical representation, <function>xmlcast</function> ensures that the conversion produces or expects a + canonical XML Schema lexical form appropriate for the target type. For example, an <type>interval</type> + value is rendered as <literal>P1Y2M</literal> (<type>xs:duration</type>), and a <type>timestamp</type> as + <literal>2023-05-19T14:30:00Z</literal> (xs:dateTime). Similarly, when converting from XML to SQL types, + <function>xmlcast</function> validates that the input string conforms to the lexical format required by the + corresponding SQL type. + + The function <function>xmlcast</function> follows these rules: + + <itemizedlist> + <listitem> + <para> + Either <replaceable>expression</replaceable> or <replaceable>type</replaceable> must be of type <type>xml</type>. + </para> + </listitem> + <listitem> + <para> + Casting is supported between <type>xml</type> and <link linkend="datatype-character-table">character</link>, + <link linkend="datatype-numeric">numeric</link>, <link linkend="datatype-datetime">date/time</link>, + <link linkend="datatype-boolean">boolean</link> and <link linkend="datatype-binary">binary</link> data types. + </para> + </listitem> + <listitem> + <para> + Similar to the function <function>xmltext</function>, <replaceable>expression</replaceable> + values containing XML predefined entities will be escaped (see examples below). + </para> + </listitem> + <listitem> + <para> + Values of type <type>date</type>, <type>time with time zone</type>, <type>timestamp with time zone</type>, + and <type>interval</type> are converted to their corresponding XML Schema types: <type>xs:date</type>, + <type>xs:time</type>, <type>xs:dateTime</type>, and <type>xs:duration</type>, respectively. + </para> + </listitem> + <listitem> + <para> + The <literal>BY REF</literal> and <literal>BY VALUE</literal> clauses + are accepted but ignored, as discussed in + <xref linkend="functions-xml-limits-postgresql"/>. + </para> + </listitem> + </itemizedlist> + + Examples: +<screen><![CDATA[ +SELECT xmlcast('<foo&bar>'::text AS xml); + xmlcast +--------------------- + <foo&bar> + +SELECT xmlcast('<foo&bar>'::xml AS text); + xmlcast +----------- + <foo&bar> + +SELECT xmlcast(CURRENT_TIMESTAMP AS xml); + xmlcast +--------------------------------- + 2024-06-02T00:29:40.92397+02:00 + +SELECT xmlcast('P1Y2M3W4DT5H6M7S'::xml AS interval); + xmlcast +-------------------------------- + 1 year 2 mons 25 days 05:06:07 + +SELECT xmlcast('1 year 2 months 3 weeks 4 days 5 hours 6 minutes 7 seconds'::interval AS xml); + xmlcast +----------------- + P1Y2M25DT5H6M7S +]]></screen> + + Alternatively, it is also possible to convert character strings into XML using PostgreSQL-specific cast syntaxes: <programlisting><![CDATA[ xml '<foo>bar</foo>' '<foo>bar</foo>'::xml ]]></programlisting> - can also be used. + </para> <para> diff --git a/src/backend/catalog/sql_features.txt b/src/backend/catalog/sql_features.txt index 3a8ad20160..412e1a4e37 100644 --- a/src/backend/catalog/sql_features.txt +++ b/src/backend/catalog/sql_features.txt @@ -624,7 +624,7 @@ X014 Attributes of XML type YES X015 Fields of XML type NO X016 Persistent XML values YES X020 XMLConcat YES -X025 XMLCast NO +X025 XMLCast YES X030 XMLDocument NO X031 XMLElement YES X032 XMLForest YES diff --git a/src/backend/executor/execExprInterp.c b/src/backend/executor/execExprInterp.c index a7a5ac1e83..eaf543c1a4 100644 --- a/src/backend/executor/execExprInterp.c +++ b/src/backend/executor/execExprInterp.c @@ -69,6 +69,7 @@ #include "utils/array.h" #include "utils/builtins.h" #include "utils/date.h" +#include "utils/datetime.h" #include "utils/datum.h" #include "utils/expandedrecord.h" #include "utils/json.h" @@ -4643,11 +4644,139 @@ ExecEvalXmlExpr(ExprState *state, ExprEvalStep *op) *op->resnull = false; } break; + case IS_XMLCAST: + { + Datum *argvalue = op->d.xmlexpr.argvalue; + bool *argnull = op->d.xmlexpr.argnull; + char *str; - default: - elog(ERROR, "unrecognized XML operation"); + Assert(list_length(xexpr->args) == 1); + + if (argnull[0]) + return; + + value = argvalue[0]; + + switch (xexpr->targetType) + { + case XMLOID: + /* + * Certain SQL data types must be mapped to XML Schema types when casting to XML: + * - DATE -> xs:date + * - TIME [WITH/WITHOUT TZ] -> xs:time + * - TIMESTAMP [WITH/WITHOUT TZ] -> xs:dateTime + * - BOOLEAN -> xs:boolean ("true"/"false") + * - BYTEA -> xs:base64Binary or xs:hexBinary + * - FLOAT/DOUBLE/NUMERIC -> xs:float/double/decimal (with INF/-INF handling) + * + * These mappings are defined in SQL/XML:2023 (ISO/IEC 9075-14:2023), + * Subclause 6.7 "<XML cast specification>", Syntax Rules, item 15.e.i–v. + * + * The corresponding XML Schema lexical formats follow ISO 8601 and are specified in + * W3C XML Schema Part 2: Primitive Datatypes: + * - §3.2.7 (dateTime): "2023-05-19T14:30:00+01:00" + * - §3.2.8 (time): "14:30:00Z" + * - §3.2.9 (date): "2023-05-19" + * - §3.2.2 (boolean): "true"/"false" + * - §3.2.16 (base64Binary/hexBinary): base64/hex encoded strings + * - §3.2.4 (float) and §3.2.5 (double): with "INF"/"-INF" for infinity + */ + if (xexpr->type == TIMESTAMPOID || xexpr->type == TIMESTAMPTZOID || + xexpr->type == DATEOID || xexpr->type == BYTEAOID || xexpr->type == BOOLOID || + xexpr->type == FLOAT4OID || xexpr->type == FLOAT8OID || xexpr->type == NUMERICOID) + { + char *strdt = map_sql_value_to_xml_value(value, xexpr->type, false); + text *mapped_value = cstring_to_text(strdt); + pfree(strdt); + *op->resvalue = PointerGetDatum(mapped_value); + } + /* + * SQL interval types must be mapped to XML Schema types when casting to XML: + * - Year-month intervals → xs:yearMonthDuration + * - Day-time intervals → xs:dayTimeDuration + * + * This behavior is required by SQL/XML:2023 (ISO/IEC 9075-14:2023), + * Subclause 6.7 "<XML cast specification>", General Rules, item 3.d.ii.1–2. + * + * These XML Schema types require ISO 8601-compatible lexical representations, + * such as: "P1Y2M", "P3DT4H5M", or "P1Y2M3DT4H5M6S", as defined in + * W3C XML Schema Part 2: Primitive Datatypes, §3.2.6 (duration) + */ + else if (xexpr->type == INTERVALOID) + { + Interval *in = DatumGetIntervalP(value); + + struct pg_itm tt, *itm = &tt; + char buf[MAXDATELEN + 1]; + + if (INTERVAL_NOT_FINITE(in)) + { + if (INTERVAL_IS_NOBEGIN(in)) + strcpy(buf, EARLY); + else if (INTERVAL_IS_NOEND(in)) + strcpy(buf, LATE); + else + elog(ERROR, "invalid interval argument"); + } + else + { + interval2itm(*in, itm); + EncodeInterval(itm, INTSTYLE_ISO_8601, buf); + } + + *op->resvalue = PointerGetDatum(cstring_to_text(buf)); + } + /* no need to escape the result, as the origin is also an XML */ + else if (xexpr->type == XMLOID) + *op->resvalue = PointerGetDatum(DatumGetXmlP(value)); + /* we make sure that potential predifined entitties are escaped */ + else + *op->resvalue = PointerGetDatum( + DatumGetXmlP((DirectFunctionCall1(xmltext, value)))); + break; + case TEXTOID: + case VARCHAROID: + case NAMEOID: + case BPCHAROID: + /* + * when casting from XML to a character string we make sure that + * all escaped xml characters are unescaped. + */ + str = text_to_cstring(DatumGetTextPP(value)); + { + char *unescaped = unescape_xml(str); + *op->resvalue = PointerGetDatum(cstring_to_text(unescaped)); + pfree(unescaped); + } + pfree(str); + break; + case INT2OID: + case INT4OID: + case INT8OID: + case NUMERICOID: + case FLOAT4OID: + case FLOAT8OID: + case BOOLOID: + case TIMESTAMPOID: + case TIMESTAMPTZOID: + case TIMEOID: + case TIMETZOID: + case DATEOID: + case BYTEAOID: + case INTERVALOID: + *op->resvalue = PointerGetDatum(DatumGetTextP(value)); + break; + default: + elog(ERROR, "unsupported target data type for XMLCast"); + } + + *op->resnull = false; + } break; - } + default: + elog(ERROR, "unrecognized XML operation"); + break; + } } /* diff --git a/src/backend/nodes/nodeFuncs.c b/src/backend/nodes/nodeFuncs.c index d29664ca5d..067658a778 100644 --- a/src/backend/nodes/nodeFuncs.c +++ b/src/backend/nodes/nodeFuncs.c @@ -1735,6 +1735,9 @@ exprLocation(const Node *expr) case T_FunctionParameter: loc = ((const FunctionParameter *) expr)->location; break; + case T_XmlCast: + loc = ((const XmlCast *) expr)->location; + break; case T_XmlSerialize: /* XMLSERIALIZE keyword should always be the first thing */ loc = ((const XmlSerialize *) expr)->location; @@ -4467,6 +4470,16 @@ raw_expression_tree_walker_impl(Node *node, return true; } break; + case T_XmlCast: + { + XmlCast *xc = (XmlCast *) node; + + if (WALK(xc->expr)) + return true; + if (WALK(xc->targetType)) + return true; + } + break; case T_CollateClause: return WALK(((CollateClause *) node)->arg); case T_SortBy: diff --git a/src/backend/parser/gram.y b/src/backend/parser/gram.y index 713ee5c10a..a4ac1382db 100644 --- a/src/backend/parser/gram.y +++ b/src/backend/parser/gram.y @@ -797,7 +797,7 @@ static Node *makeRecursiveViewSelect(char *relname, List *aliases, Node *query); WAIT WHEN WHERE WHITESPACE_P WINDOW WITH WITHIN WITHOUT WORK WRAPPER WRITE - XML_P XMLATTRIBUTES XMLCONCAT XMLELEMENT XMLEXISTS XMLFOREST XMLNAMESPACES + XML_P XMLATTRIBUTES XMLCAST XMLCONCAT XMLELEMENT XMLEXISTS XMLFOREST XMLNAMESPACES XMLPARSE XMLPI XMLROOT XMLSERIALIZE XMLTABLE YEAR_P YES_P @@ -16221,6 +16221,24 @@ func_expr_common_subexpr: v->location = @1; $$ = (Node *) v; } + | XMLCAST '(' a_expr AS Typename ')' + { + XmlCast *n = makeNode(XmlCast); + + n->expr = $3; + n->targetType = $5; + n->location = @1; + $$ = (Node *) n; + } + | XMLCAST '(' a_expr AS Typename xml_passing_mech')' + { + XmlCast *n = makeNode(XmlCast); + + n->expr = $3; + n->targetType = $5; + n->location = @1; + $$ = (Node *) n; + } | XMLCONCAT '(' expr_list ')' { $$ = makeXmlExpr(IS_XMLCONCAT, NULL, NIL, $3, @1); @@ -18296,6 +18314,7 @@ col_name_keyword: | VALUES | VARCHAR | XMLATTRIBUTES + | XMLCAST | XMLCONCAT | XMLELEMENT | XMLEXISTS @@ -18888,6 +18907,7 @@ bare_label_keyword: | WRITE | XML_P | XMLATTRIBUTES + | XMLCAST | XMLCONCAT | XMLELEMENT | XMLEXISTS diff --git a/src/backend/parser/parse_expr.c b/src/backend/parser/parse_expr.c index 56826db4c2..99c632ea6a 100644 --- a/src/backend/parser/parse_expr.c +++ b/src/backend/parser/parse_expr.c @@ -41,6 +41,18 @@ #include "utils/typcache.h" #include "utils/xml.h" +/* + * Determines whether a type category is allowed in XMLCAST conversions, + * either as a source (SQL -> XML) or target (XML -> SQL) type, per + * SQL/XML:2023 (ISO/IEC 9075-14:2023), Subclause 6.7 + * "<XML cast specification>", General Rules, item 3.d.ii: + */ +#define type_is_xmlcast_supported(x) ((x) == TYPCATEGORY_NUMERIC \ + || (x) == TYPCATEGORY_STRING || (x) == TYPCATEGORY_DATETIME \ + || (x) == TYPCATEGORY_BOOLEAN || (x) == TYPCATEGORY_UNKNOWN \ + || (x) == TYPCATEGORY_TIMESPAN) + + /* GUC parameters */ bool Transform_null_equals = false; @@ -68,6 +80,7 @@ static Node *transformMinMaxExpr(ParseState *pstate, MinMaxExpr *m); static Node *transformSQLValueFunction(ParseState *pstate, SQLValueFunction *svf); static Node *transformXmlExpr(ParseState *pstate, XmlExpr *x); +static Node *transformXmlCast(ParseState *pstate, XmlCast *xc); static Node *transformXmlSerialize(ParseState *pstate, XmlSerialize *xs); static Node *transformBooleanTest(ParseState *pstate, BooleanTest *b); static Node *transformCurrentOfExpr(ParseState *pstate, CurrentOfExpr *cexpr); @@ -276,6 +289,10 @@ transformExprRecurse(ParseState *pstate, Node *expr) (SQLValueFunction *) expr); break; + case T_XmlCast: + result = transformXmlCast(pstate, (XmlCast *) expr); + break; + case T_XmlExpr: result = transformXmlExpr(pstate, (XmlExpr *) expr); break; @@ -2468,6 +2485,10 @@ transformXmlExpr(ParseState *pstate, XmlExpr *x) newe = coerce_to_specific_type(pstate, newe, INT4OID, "XMLROOT"); break; + case IS_XMLCAST: + /* not handled here */ + Assert(false); + break; case IS_XMLSERIALIZE: /* not handled here */ Assert(false); @@ -2484,6 +2505,75 @@ transformXmlExpr(ParseState *pstate, XmlExpr *x) return (Node *) newx; } +static Node * +transformXmlCast(ParseState *pstate, XmlCast *xc) +{ + Node *result; + XmlExpr *xexpr; + int32 targetTypmod; + Oid targetType; + Oid inputType; + char inputTypcategory; + char targetTypcategory; + bool inputTypispreferred; + bool targetTypispreferred; + + inputType = exprType(transformExprRecurse(pstate, xc->expr)); + get_type_category_preferred(inputType, &inputTypcategory, &inputTypispreferred); + + typenameTypeIdAndMod(pstate, xc->targetType, &targetType, &targetTypmod); + get_type_category_preferred(targetType, &targetTypcategory, &targetTypispreferred); + + /* + * Ensure that either the cast operand or the data type is an XML, + * and check if the data types are supported. + */ + if ((inputType != XMLOID && targetType != XMLOID) || + (!type_is_xmlcast_supported(inputTypcategory) && inputType != XMLOID && inputType != BYTEAOID) || + (!type_is_xmlcast_supported(targetTypcategory) && targetType != XMLOID && targetType != BYTEAOID)) + ereport(ERROR, + (errcode(ERRCODE_CANNOT_COERCE), + errmsg("cannot cast from '%s' to '%s'", + format_type_be(inputType), format_type_be(targetType)), + parser_errposition(pstate, xc->location))); + + /* + * It is not possible to cast some supported data types directly to XML, + * so we first handle them as text and later on return them as XML. + */ + if ((inputTypcategory == TYPCATEGORY_NUMERIC && inputType != FLOAT4OID && + inputType != FLOAT8OID && inputType != NUMERICOID) || inputType == TIMEOID || + inputType == TIMETZOID || inputType == UNKNOWNOID || inputType == NAMEOID) + inputType = TEXTOID; + + xexpr = makeNode(XmlExpr); + xexpr->op = IS_XMLCAST; + xexpr->location = xc->location; + xexpr->type = inputType; + xexpr->targetType = targetType; + xexpr->name = "xmlcast"; + xexpr->args = list_make1(coerce_to_specific_type(pstate, + transformExprRecurse(pstate, xc->expr), + inputType, + "XMLCAST")); + + result = coerce_to_target_type(pstate, (Node *) xexpr, + targetType == TEXTOID ? XMLOID : TEXTOID, + targetType, targetTypmod, + COERCION_EXPLICIT, + COERCE_EXPLICIT_CAST, + -1); + + if (result == NULL) + ereport(ERROR, + (errcode(ERRCODE_CANNOT_COERCE), + errmsg("cannot cast XMLCAST result to %s", + format_type_be(targetType)), + parser_errposition(pstate, xexpr->location))); + + return result; +} + static Node * transformXmlSerialize(ParseState *pstate, XmlSerialize *xs) { diff --git a/src/backend/parser/parse_target.c b/src/backend/parser/parse_target.c index b5a2f915b6..5e66b0bccb 100644 --- a/src/backend/parser/parse_target.c +++ b/src/backend/parser/parse_target.c @@ -1955,6 +1955,9 @@ FigureColnameInternal(Node *node, char **name) /* make SQL/XML functions act like a regular function */ switch (((XmlExpr *) node)->op) { + case IS_XMLCAST: + *name = "xmlcast"; + return 2; case IS_XMLCONCAT: *name = "xmlconcat"; return 2; @@ -1981,6 +1984,10 @@ FigureColnameInternal(Node *node, char **name) break; } break; + case T_XmlCast: + /* make XMLCAST act like a regular function */ + *name = "xmlcast"; + return 2; case T_XmlSerialize: /* make XMLSERIALIZE act like a regular function */ *name = "xmlserialize"; diff --git a/src/backend/utils/adt/numeric.c b/src/backend/utils/adt/numeric.c index 891ae6ba7f..d2fb9db49e 100644 --- a/src/backend/utils/adt/numeric.c +++ b/src/backend/utils/adt/numeric.c @@ -847,6 +847,28 @@ numeric_is_inf(Numeric num) return NUMERIC_IS_INF(num); } +/* + * numeric_is_positive_inf() - + * + * Is Numeric value positive infinity? + */ +bool +numeric_is_positive_inf(Numeric num) +{ + return NUMERIC_IS_PINF(num); +} + +/* + * numeric_is_negative_inf() - + * + * Is Numeric value negative infinity? + */ +bool +numeric_is_negative_inf(Numeric num) +{ + return NUMERIC_IS_NINF(num); +} + /* * numeric_is_integral() - * diff --git a/src/backend/utils/adt/ruleutils.c b/src/backend/utils/adt/ruleutils.c index 033b625f3f..a3d4674136 100644 --- a/src/backend/utils/adt/ruleutils.c +++ b/src/backend/utils/adt/ruleutils.c @@ -10119,6 +10119,9 @@ get_rule_expr(Node *node, deparse_context *context, case IS_XMLSERIALIZE: appendStringInfoString(buf, "XMLSERIALIZE("); break; + case IS_XMLCAST: + appendStringInfoString(buf, "XMLCAST("); + break; case IS_DOCUMENT: break; } @@ -10129,7 +10132,7 @@ get_rule_expr(Node *node, deparse_context *context, else appendStringInfoString(buf, "CONTENT "); } - if (xexpr->name) + if (xexpr->name && xexpr->op != IS_XMLCAST) { appendStringInfo(buf, "NAME %s", quote_identifier(map_xml_name_to_sql_identifier(xexpr->name))); @@ -10165,6 +10168,7 @@ get_rule_expr(Node *node, deparse_context *context, appendStringInfoString(buf, ", "); switch (xexpr->op) { + case IS_XMLCAST: case IS_XMLCONCAT: case IS_XMLELEMENT: case IS_XMLFOREST: @@ -10242,6 +10246,10 @@ get_rule_expr(Node *node, deparse_context *context, appendStringInfoString(buf, " NO INDENT"); } + if (xexpr->op == IS_XMLCAST) + appendStringInfo(buf, " AS %s", + format_type_be(xexpr->targetType)); + if (xexpr->op == IS_DOCUMENT) appendStringInfoString(buf, " IS DOCUMENT"); else diff --git a/src/backend/utils/adt/xml.c b/src/backend/utils/adt/xml.c index f69dc68286..e496291da9 100644 --- a/src/backend/utils/adt/xml.c +++ b/src/backend/utils/adt/xml.c @@ -99,6 +99,7 @@ #include "utils/date.h" #include "utils/datetime.h" #include "utils/lsyscache.h" +#include "utils/numeric.h" #include "utils/rel.h" #include "utils/syscache.h" #include "utils/xml.h" @@ -2574,9 +2575,9 @@ map_sql_value_to_xml_value(Datum value, Oid type, bool xml_escape_strings) { case BOOLOID: if (DatumGetBool(value)) - return "true"; + return pstrdup("true"); else - return "false"; + return pstrdup("false"); case DATEOID: { @@ -2706,8 +2707,60 @@ map_sql_value_to_xml_value(Datum value, Oid type, bool xml_escape_strings) } #endif /* USE_LIBXML */ - } + case FLOAT4OID: + /* + * Special handling for infinity values in floating-point and + * numeric types. The XML Schema specification requires that + * positive infinity be represented as "INF" and negative + * infinity as "-INF" in XML documents for float and double + * types. While decimal types in XML Schema do not support + * infinity, PostgreSQL's NUMERIC type can represent it, so + * we use the same representation for consistency. + * + * See: XML Schema Part 2: Datatypes Second Edition, sections + * 3.2.4 (float) and 3.2.5 (double). + */ + { + float4 val = DatumGetFloat4(value); + + if (isinf(val)) + { + if (val < 0) + return pstrdup("-INF"); + else + return pstrdup("INF"); + } + } + break; + case FLOAT8OID: + { + float8 val = DatumGetFloat8(value); + + if (isinf(val)) + { + if (val < 0) + return pstrdup("-INF"); + else + return pstrdup("INF"); + } + } + break; + + case NUMERICOID: + { + Numeric num = DatumGetNumeric(value); + + if (numeric_is_inf(num)) + { + if (numeric_is_positive_inf(num)) + return pstrdup("INF"); + else + return pstrdup("-INF"); + } + } + break; + } /* * otherwise, just use the type's native text representation */ @@ -2762,6 +2815,67 @@ escape_xml(const char *str) return buf.data; } +/* + * Unescape XML escaped characters. + * + * In order to keep it consistent with "escape_xml(const char*)", + * this function intentionally does not depend on libxml2. + */ +char * +unescape_xml(const char *str) +{ + StringInfoData buf; + size_t p = 0; + size_t len; + + if (!str) + return NULL; + + len = strlen(str); + + initStringInfo(&buf); + + while (p < len) + { + if (p + 4 <= len && strncmp(str + p, "<", 4) == 0) + { + appendStringInfoChar(&buf, '<'); + p += 4; + } + else if (p + 4 <= len && strncmp(str + p, ">", 4) == 0) + { + appendStringInfoChar(&buf, '>'); + p += 4; + } + else if (p + 5 <= len && strncmp(str + p, "&", 5) == 0) + { + appendStringInfoChar(&buf, '&'); + p += 5; + } + else if (p + 5 <= len && strncmp(str + p, " ", 5) == 0) + { + appendStringInfoChar(&buf, '\r'); + p += 5; + } + else if (p + 6 <= len && strncmp(str + p, """, 6) == 0) + { + appendStringInfoChar(&buf, '"'); + p += 6; + } + else if (p + 6 <= len && strncmp(str + p, "
", 6) == 0) + { + appendStringInfoChar(&buf, '\r'); + p += 6; + } + else + { + appendStringInfoChar(&buf, *(str + p)); + p++; + } + } + + return buf.data; +} static char * _SPI_strdup(const char *s) @@ -4350,6 +4464,7 @@ xml_xpathobjtoxmlarray(xmlXPathObjectPtr xpathobj, datum = PointerGetDatum(cstring_to_xmltype(result_str)); (void) accumArrayResult(astate, datum, false, XMLOID, CurrentMemoryContext); + pfree(result_str); return 1; } diff --git a/src/include/nodes/parsenodes.h b/src/include/nodes/parsenodes.h index aac4bfc70d..81bbf79565 100644 --- a/src/include/nodes/parsenodes.h +++ b/src/include/nodes/parsenodes.h @@ -878,6 +878,14 @@ typedef struct XmlSerialize ParseLoc location; /* token location, or -1 if unknown */ } XmlSerialize; +typedef struct XmlCast +{ + NodeTag type; + Node *expr; + TypeName *targetType; + ParseLoc location; /* token location, or -1 if unknown */ +} XmlCast; + /* Partitioning related definitions */ /* diff --git a/src/include/nodes/primnodes.h b/src/include/nodes/primnodes.h index 5211cadc25..cc1d3541b5 100644 --- a/src/include/nodes/primnodes.h +++ b/src/include/nodes/primnodes.h @@ -1602,6 +1602,7 @@ typedef struct SQLValueFunction */ typedef enum XmlExprOp { + IS_XMLCAST, /* XMLCAST(op AS datatype) */ IS_XMLCONCAT, /* XMLCONCAT(args) */ IS_XMLELEMENT, /* XMLELEMENT(name, xml_attributes, args) */ IS_XMLFOREST, /* XMLFOREST(xml_attributes) */ @@ -1638,6 +1639,8 @@ typedef struct XmlExpr /* target type/typmod for XMLSERIALIZE */ Oid type pg_node_attr(query_jumble_ignore); int32 typmod pg_node_attr(query_jumble_ignore); + /* option for XMLCAST */ + Oid targetType; /* token location, or -1 if unknown */ ParseLoc location; } XmlExpr; diff --git a/src/include/parser/kwlist.h b/src/include/parser/kwlist.h index f7753c5c8a..12892e32aa 100644 --- a/src/include/parser/kwlist.h +++ b/src/include/parser/kwlist.h @@ -512,6 +512,7 @@ PG_KEYWORD("wrapper", WRAPPER, UNRESERVED_KEYWORD, BARE_LABEL) PG_KEYWORD("write", WRITE, UNRESERVED_KEYWORD, BARE_LABEL) PG_KEYWORD("xml", XML_P, UNRESERVED_KEYWORD, BARE_LABEL) PG_KEYWORD("xmlattributes", XMLATTRIBUTES, COL_NAME_KEYWORD, BARE_LABEL) +PG_KEYWORD("xmlcast", XMLCAST, COL_NAME_KEYWORD, BARE_LABEL) PG_KEYWORD("xmlconcat", XMLCONCAT, COL_NAME_KEYWORD, BARE_LABEL) PG_KEYWORD("xmlelement", XMLELEMENT, COL_NAME_KEYWORD, BARE_LABEL) PG_KEYWORD("xmlexists", XMLEXISTS, COL_NAME_KEYWORD, BARE_LABEL) diff --git a/src/include/utils/numeric.h b/src/include/utils/numeric.h index b1cf40ed9f..79f0548ea9 100644 --- a/src/include/utils/numeric.h +++ b/src/include/utils/numeric.h @@ -87,6 +87,8 @@ NumericGetDatum(Numeric X) */ extern bool numeric_is_nan(Numeric num); extern bool numeric_is_inf(Numeric num); +extern bool numeric_is_positive_inf(Numeric num); +extern bool numeric_is_negative_inf(Numeric num); extern int32 numeric_maximum_size(int32 typmod); extern char *numeric_out_sci(Numeric num, int scale); extern char *numeric_normalize(Numeric num); diff --git a/src/include/utils/xml.h b/src/include/utils/xml.h index 03acb25544..7fc0a8d6f8 100644 --- a/src/include/utils/xml.h +++ b/src/include/utils/xml.h @@ -80,6 +80,7 @@ extern bool xml_is_document(xmltype *arg); extern text *xmltotext_with_options(xmltype *data, XmlOptionType xmloption_arg, bool indent); extern char *escape_xml(const char *str); +extern char *unescape_xml(const char *str); extern char *map_sql_identifier_to_xml_name(const char *ident, bool fully_escaped, bool escape_period); extern char *map_xml_name_to_sql_identifier(const char *name); diff --git a/src/test/regress/expected/xml.out b/src/test/regress/expected/xml.out index 103a22a3b1..a5a56c5b74 100644 --- a/src/test/regress/expected/xml.out +++ b/src/test/regress/expected/xml.out @@ -1881,3 +1881,610 @@ SELECT xmltext('x'|| '<P>73</P>'::xml || .42 || true || 'j'::char); x<P>73</P>0.42truej (1 row) +-- for xmlcast() tests +INSERT INTO xmltest + VALUES (42, +'<?xml version="1.0" encoding="utf-8"?> + <xmlcast> + <period1>P1Y2M3DT4H5M6S</period1> + <period2>1 year 2 mons 3 days 4 hours 5 minutes 6 seconds</period2> + <date1>2002-09-24</date1> + <date2>2002-09-24+06:00</date2> + <time>09:30:10.5</time> + <time_tz1>09:30:10Z</time_tz1> + <time_tz2>09:30:10-06:00</time_tz2> + <time_tz3>09:30:10+06:00</time_tz3> + <timestamp1>2002-05-30T09:00:00</timestamp1> + <timestamp2>2002-05-30T09:30:10.5</timestamp2> + <timestamp_tz1>2002-05-30T09:30:10Z</timestamp_tz1> + <timestamp_tz2>2002-05-30T09:30:10-06:00</timestamp_tz2> + <timestamp_tz3>2002-05-30T09:30:10+06:00</timestamp_tz3> + <text1>foo bar</text1> + <text2> foo bar </text2> + <text3>foo & <"bar"></text3> + <decimal1>42.7312345678910</decimal1> + <decimal2>+42.7312345678910</decimal2> + <decimal3>-42.7312345678910</decimal3> + <decimal4>INF</decimal4> + <decimal5>-INF</decimal5> + <integer1>42</integer1> + <integer2>+42</integer2> + <integer3>-42</integer3> + <long1>4273535420162021</long1> + <long2>+4273535420162021</long2> + <long3>-4273535420162021</long3> + <bool1 att="true">42</bool1> + <bool2 att="false">73</bool2> + <empty></empty> + </xmlcast>'::xml +); +-- This prevents the xmlcast regression tests from failing if the system's timezone has been changed. +SET timezone TO 'America/Los_Angeles'; +-- xmlcast exceptions +\set VERBOSITY terse +SELECT xmlcast((xpath('//text1/text()', data))[1] AS text[]) FROM xmltest WHERE id = 42; +ERROR: cannot cast from 'xml' to 'text[]' at character 8 +SELECT xmlcast((xpath('//text1/integer1()', data))[1] AS int[]) FROM xmltest WHERE id = 42; +ERROR: cannot cast from 'xml' to 'integer[]' at character 8 +SELECT xmlcast(NULL AS text); +ERROR: cannot cast from 'unknown' to 'text' at character 8 +SELECT xmlcast('foo'::text AS varchar); +ERROR: cannot cast from 'text' to 'character varying' at character 8 +SELECT xmlcast(42 AS text); +ERROR: cannot cast from 'integer' to 'text' at character 8 +SELECT xmlcast(array['foo','bar'] AS xml); +ERROR: cannot cast from 'text[]' to 'xml' at character 8 +\set VERBOSITY default +-- xmlcast tests for "XML to non-XML" expressions +SELECT + xmlcast((xpath('//date1/text()', data))[1] AS date), pg_typeof(xmlcast((xpath('//date1/text()', data))[1] AS date)), + xmlcast((xpath('//date2/text()', data))[1] AS date), pg_typeof(xmlcast((xpath('//date2/text()', data))[1] AS date)) +FROM xmltest WHERE id = 42; + xmlcast | pg_typeof | xmlcast | pg_typeof +------------+-----------+------------+----------- + 09-24-2002 | date | 09-24-2002 | date +(1 row) + +SELECT + xmlcast((xpath('//period1/text()', data))[1] AS interval), pg_typeof(xmlcast((xpath('//period1/text()', data))[1] AS interval)), + xmlcast((xpath('//period2/text()', data))[1] AS interval), pg_typeof(xmlcast((xpath('//period2/text()', data))[1] AS interval)) +FROM xmltest WHERE id = 42; + xmlcast | pg_typeof | xmlcast | pg_typeof +----------------------------------------------+-----------+----------------------------------------------+----------- + @ 1 year 2 mons 3 days 4 hours 5 mins 6 secs | interval | @ 1 year 2 mons 3 days 4 hours 5 mins 6 secs | interval +(1 row) + +SELECT + xmlcast((xpath('//time/text()', data))[1] AS time), pg_typeof(xmlcast((xpath('//time/text()', data))[1] AS time)), + xmlcast((xpath('//time_tz1/text()', data))[1] AS time with time zone), pg_typeof(xmlcast((xpath('//time_tz1/text()', data))[1] AS time with time zone)), + xmlcast((xpath('//time_tz2/text()', data))[1] AS time with time zone), pg_typeof(xmlcast((xpath('//time_tz2/text()', data))[1] AS time with time zone)), + xmlcast((xpath('//time_tz3/text()', data))[1] AS time with time zone), pg_typeof(xmlcast((xpath('//time_tz3/text()', data))[1] AS time with time zone)) +FROM xmltest WHERE id = 42; + xmlcast | pg_typeof | xmlcast | pg_typeof | xmlcast | pg_typeof | xmlcast | pg_typeof +------------+------------------------+-------------+---------------------+-------------+---------------------+-------------+--------------------- + 09:30:10.5 | time without time zone | 09:30:10+00 | time with time zone | 09:30:10-06 | time with time zone | 09:30:10+06 | time with time zone +(1 row) + +SELECT + xmlcast((xpath('//text1/text()', data))[1] AS text), pg_typeof(xmlcast((xpath('//text1/text()', data))[1] AS text)), + xmlcast((xpath('//text2/text()', data))[1] AS text), pg_typeof(xmlcast((xpath('//text2/text()', data))[1] AS text)), + xmlcast((xpath('//text3/text()', data))[1] AS text), pg_typeof(xmlcast((xpath('//text3/text()', data))[1] AS text)) +FROM xmltest WHERE id = 42; + xmlcast | pg_typeof | xmlcast | pg_typeof | xmlcast | pg_typeof +---------+-----------+---------------------+-----------+---------------+----------- + foo bar | text | foo bar | text | foo & <"bar"> | text +(1 row) + +SELECT + xmlcast((xpath('//text1/text()', data))[1] AS varchar), pg_typeof(xmlcast((xpath('//text1/text()', data))[1] AS varchar)), + xmlcast((xpath('//text2/text()', data))[1] AS varchar), pg_typeof(xmlcast((xpath('//text2/text()', data))[1] AS varchar)), + xmlcast((xpath('//text3/text()', data))[1] AS varchar), pg_typeof(xmlcast((xpath('//text3/text()', data))[1] AS varchar)) +FROM xmltest WHERE id = 42; + xmlcast | pg_typeof | xmlcast | pg_typeof | xmlcast | pg_typeof +---------+-------------------+---------------------+-------------------+---------------+------------------- + foo bar | character varying | foo bar | character varying | foo & <"bar"> | character varying +(1 row) + +SELECT + xmlcast((xpath('//text1/text()', data))[1] AS name), pg_typeof(xmlcast((xpath('//text1/text()', data))[1] AS name)), + xmlcast((xpath('//text2/text()', data))[1] AS name), pg_typeof(xmlcast((xpath('//text2/text()', data))[1] AS name)), + xmlcast((xpath('//text3/text()', data))[1] AS name), pg_typeof(xmlcast((xpath('//text3/text()', data))[1] AS name)) +FROM xmltest WHERE id = 42; + xmlcast | pg_typeof | xmlcast | pg_typeof | xmlcast | pg_typeof +---------+-----------+---------------------+-----------+---------------+----------- + foo bar | name | foo bar | name | foo & <"bar"> | name +(1 row) + +SELECT + xmlcast((xpath('//text1/text()', data))[1] AS bpchar), pg_typeof(xmlcast((xpath('//text1/text()', data))[1] AS bpchar)), + xmlcast((xpath('//text2/text()', data))[1] AS bpchar), pg_typeof(xmlcast((xpath('//text2/text()', data))[1] AS bpchar)), + xmlcast((xpath('//text3/text()', data))[1] AS bpchar), pg_typeof(xmlcast((xpath('//text3/text()', data))[1] AS bpchar)) +FROM xmltest WHERE id = 42; + xmlcast | pg_typeof | xmlcast | pg_typeof | xmlcast | pg_typeof +---------+-----------+---------------------+-----------+---------------+----------- + foo bar | character | foo bar | character | foo & <"bar"> | character +(1 row) + +SELECT + xmlcast((xpath('//decimal1/text()', data))[1] AS numeric), pg_typeof(xmlcast((xpath('//decimal1/text()', data))[1] AS numeric)), + xmlcast((xpath('//decimal2/text()', data))[1] AS numeric), pg_typeof(xmlcast((xpath('//decimal2/text()', data))[1] AS numeric)), + xmlcast((xpath('//decimal3/text()', data))[1] AS numeric), pg_typeof(xmlcast((xpath('//decimal3/text()', data))[1] AS numeric)), + xmlcast((xpath('//decimal4/text()', data))[1] AS numeric), pg_typeof(xmlcast((xpath('//decimal4/text()', data))[1] AS numeric)), + xmlcast((xpath('//decimal5/text()', data))[1] AS numeric), pg_typeof(xmlcast((xpath('//decimal5/text()', data))[1] AS numeric)) +FROM xmltest WHERE id = 42; + xmlcast | pg_typeof | xmlcast | pg_typeof | xmlcast | pg_typeof | xmlcast | pg_typeof | xmlcast | pg_typeof +------------------+-----------+------------------+-----------+-------------------+-----------+----------+-----------+-----------+----------- + 42.7312345678910 | numeric | 42.7312345678910 | numeric | -42.7312345678910 | numeric | Infinity | numeric | -Infinity | numeric +(1 row) + +SELECT + xmlcast((xpath('//decimal1/text()', data))[1] AS double precision), pg_typeof(xmlcast((xpath('//decimal1/text()', data))[1] AS double precision)), + xmlcast((xpath('//decimal2/text()', data))[1] AS double precision), pg_typeof(xmlcast((xpath('//decimal2/text()', data))[1] AS double precision)), + xmlcast((xpath('//decimal3/text()', data))[1] AS double precision), pg_typeof(xmlcast((xpath('//decimal3/text()', data))[1] AS double precision)), + xmlcast((xpath('//decimal4/text()', data))[1] AS double precision), pg_typeof(xmlcast((xpath('//decimal4/text()', data))[1] AS double precision)), + xmlcast((xpath('//decimal5/text()', data))[1] AS double precision), pg_typeof(xmlcast((xpath('//decimal5/text()', data))[1] AS double precision)) +FROM xmltest WHERE id = 42; + xmlcast | pg_typeof | xmlcast | pg_typeof | xmlcast | pg_typeof | xmlcast | pg_typeof | xmlcast | pg_typeof +-----------------+------------------+-----------------+------------------+------------------+------------------+----------+------------------+-----------+------------------ + 42.731234567891 | double precision | 42.731234567891 | double precision | -42.731234567891 | double precision | Infinity | double precision | -Infinity | double precision +(1 row) + +SELECT + xmlcast((xpath('//integer1/text()', data))[1] AS int), pg_typeof(xmlcast((xpath('//integer1/text()', data))[1] AS int)), + xmlcast((xpath('//integer2/text()', data))[1] AS int), pg_typeof(xmlcast((xpath('//integer2/text()', data))[1] AS int)), + xmlcast((xpath('//integer3/text()', data))[1] AS int), pg_typeof(xmlcast((xpath('//integer3/text()', data))[1] AS int)) +FROM xmltest WHERE id = 42; + xmlcast | pg_typeof | xmlcast | pg_typeof | xmlcast | pg_typeof +---------+-----------+---------+-----------+---------+----------- + 42 | integer | 42 | integer | -42 | integer +(1 row) + +SELECT + xmlcast((xpath('//long1/text()', data))[1] AS bigint), pg_typeof(xmlcast((xpath('//long1/text()', data))[1] AS bigint)), + xmlcast((xpath('//long2/text()', data))[1] AS bigint), pg_typeof(xmlcast((xpath('//long2/text()', data))[1] AS bigint)), + xmlcast((xpath('//long3/text()', data))[1] AS bigint), pg_typeof(xmlcast((xpath('//long3/text()', data))[1] AS bigint)) +FROM xmltest WHERE id = 42; + xmlcast | pg_typeof | xmlcast | pg_typeof | xmlcast | pg_typeof +------------------+-----------+------------------+-----------+-------------------+----------- + 4273535420162021 | bigint | 4273535420162021 | bigint | -4273535420162021 | bigint +(1 row) + +SELECT + xmlcast((xpath('//bool1/@att', data))[1] AS boolean), pg_typeof(xmlcast((xpath('//bool1/@att', data))[1] AS boolean)), + xmlcast((xpath('//bool2/@att', data))[1] AS boolean), pg_typeof(xmlcast((xpath('//bool1/@att', data))[1] AS boolean)) +FROM xmltest WHERE id = 42; + xmlcast | pg_typeof | xmlcast | pg_typeof +---------+-----------+---------+----------- + t | boolean | f | boolean +(1 row) + +SELECT xmlcast((xpath('//empty/text()', data))[1] AS text), pg_typeof(xmlcast((xpath('//empty/text()', data))[1] AS text)) +FROM xmltest WHERE id = 42; + xmlcast | pg_typeof +---------+----------- + | text +(1 row) + +-- xmlcast tests for "XML to XML" expressions +SELECT + xmlcast((xpath('//text1/text()', data))[1] AS xml), pg_typeof(xmlcast((xpath('//text1/text()', data))[1] AS xml)), + xmlcast((xpath('//text2/text()', data))[1] AS xml), pg_typeof(xmlcast((xpath('//text2/text()', data))[1] AS xml)), + xmlcast((xpath('//text3/text()', data))[1] AS xml), pg_typeof(xmlcast((xpath('//text3/text()', data))[1] AS xml)) +FROM xmltest WHERE id = 42; + xmlcast | pg_typeof | xmlcast | pg_typeof | xmlcast | pg_typeof +---------+-----------+---------------------+-----------+-------------------------+----------- + foo bar | xml | foo bar | xml | foo & <"bar"> | xml +(1 row) + +-- xmlcast tests for "non-XML to XML" expressions +SELECT j, pg_typeof(j) FROM xmlcast(NULL AS xml) t(j); + j | pg_typeof +---+----------- + | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast('foo' AS xml) t(j); + j | pg_typeof +-----+----------- + foo | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast(''::text AS xml) t(j); + j | pg_typeof +---+----------- + | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast(NULL::text AS xml) t(j); + j | pg_typeof +---+----------- + | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast(''::xml AS text) t(j); + j | pg_typeof +---+----------- + | text +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast(NULL::xml AS text) t(j); + j | pg_typeof +---+----------- + | text +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast('foo & <"bar">'::text AS xml) t(j); + j | pg_typeof +-----------------------------------+----------- + foo & <"bar"> | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast('foo & <"bar">'::varchar AS xml) t(j); + j | pg_typeof +-----------------------------------+----------- + foo & <"bar"> | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast('foo & <"bar">'::name AS xml) t(j); + j | pg_typeof +-----------------------------------+----------- + foo & <"bar"> | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast(xmltext(E'foo & <"bar">\r') AS text) t(j); + j | pg_typeof +-----------------+----------- + foo & <"bar">\r | text +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast(xmlcast(E'foo & <"bar">\r' AS xml) AS text) t(j); + j | pg_typeof +-----------------+----------- + foo & <"bar">\r | text +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast(to_date('29/05/2024','dd/mm/yyyy') AS xml) t(j); + j | pg_typeof +------------+----------- + 2024-05-29 | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast('2024-05-29 12:04:10.703585+02'::timestamp with time zone at time zone 'Europe/Berlin' AS xml) t(j); + j | pg_typeof +----------------------------+----------- + 2024-05-29T12:04:10.703585 | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast('2024-05-29 12:04:10.703585+02'::timestamp without time zone AS xml) t(j); + j | pg_typeof +----------------------------+----------- + 2024-05-29T12:04:10.703585 | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast('1 year 2 months 3 days 4 hours 5 minutes 6 seconds'::interval AS xml) t(j); + j | pg_typeof +----------------+----------- + P1Y2M3DT4H5M6S | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast(427353542 AS xml) t(j); + j | pg_typeof +-----------+----------- + 427353542 | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast(4273535420162021 AS xml) t(j); + j | pg_typeof +------------------+----------- + 4273535420162021 | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast(42.007312345678910 AS xml) t(j); + j | pg_typeof +--------------------+----------- + 42.007312345678910 | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast(42.007312345678910::double precision AS xml) t(j); + j | pg_typeof +-------------------+----------- + 42.00731234567891 | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast('infinity'::double precision AS xml) t(j); + j | pg_typeof +-----+----------- + INF | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast('-infinity'::double precision AS xml) t(j); + j | pg_typeof +------+----------- + -INF | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast('nan'::double precision AS xml) t(j); + j | pg_typeof +-----+----------- + NaN | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast('infinity'::numeric AS xml) t(j); + j | pg_typeof +-----+----------- + INF | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast('-infinity'::numeric AS xml) t(j); + j | pg_typeof +------+----------- + -INF | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast('nan'::numeric AS xml) t(j); + j | pg_typeof +-----+----------- + NaN | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast(true AS xml) t(j); + j | pg_typeof +------+----------- + true | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast(false AS xml) t(j); + j | pg_typeof +-------+----------- + false | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast(42 = 73 AS xml) t(j); + j | pg_typeof +-------+----------- + false | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast(42 <> 73 AS xml) t(j); + j | pg_typeof +------+----------- + true | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast('11:11:11.5'::time AS xml) t(j); + j | pg_typeof +------------+----------- + 11:11:11.5 | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast('11:11:11.5'::time with time zone at time zone 'Europe/Berlin' AS xml) t(j); + j | pg_typeof +---------------+----------- + 20:11:11.5+01 | xml +(1 row) + +-- Convert an XML string to bytea and back to xml +SELECT xmlcast(convert_from(xmlcast('<"foo&bar">'::xml AS bytea),'UTF8')::xml AS text); + xmlcast +------------- + <"foo&bar"> +(1 row) + +SET xmlbinary TO hex; +SELECT xmlcast(E'\\xDEADBEEF'::bytea AS xml); + xmlcast +---------- + DEADBEEF +(1 row) + +SET xmlbinary TO base64; +SELECT xmlcast(E'\\xDEADBEEF'::bytea AS xml); + xmlcast +---------- + 3q2+7w== +(1 row) + +-- The BY REF and BY VALUE clauses are accepted but ignored. +-- This checks if the results are indeed the same as without the clauses. +SELECT + xmlcast('foo' AS xml)::text = xmlcast('foo' AS xml BY REF)::text, + xmlcast('foo' AS xml)::text = xmlcast('foo' AS xml BY VALUE)::text, + xmlcast('foo'::xml AS text) = xmlcast('foo'::xml AS text BY REF), + xmlcast('foo'::xml AS text) = xmlcast('foo'::xml AS text BY VALUE); + ?column? | ?column? | ?column? | ?column? +----------+----------+----------+---------- + t | t | t | t +(1 row) + +SELECT + xmlcast(42 AS xml)::text = xmlcast(42 AS xml BY REF)::text, + xmlcast(42 AS xml)::text = xmlcast(42 AS xml BY VALUE)::text, + xmlcast('42'::xml AS int) = xmlcast('42'::xml AS int BY REF), + xmlcast('42'::xml AS int) = xmlcast('42'::xml AS int BY VALUE); + ?column? | ?column? | ?column? | ?column? +----------+----------+----------+---------- + t | t | t | t +(1 row) + +SELECT + xmlcast(42.73 AS xml)::text = xmlcast(42.73 AS xml BY REF)::text, + xmlcast(42.73 AS xml)::text = xmlcast(42.73 AS xml BY VALUE)::text, + xmlcast('42.73'::xml AS numeric) = xmlcast('42.73'::xml AS numeric BY REF), + xmlcast('42.73'::xml AS numeric) = xmlcast('42.73'::xml AS numeric BY VALUE); + ?column? | ?column? | ?column? | ?column? +----------+----------+----------+---------- + t | t | t | t +(1 row) + +SELECT + xmlcast('2024-08-14'::date AS xml)::text = xmlcast('2024-08-14'::date AS xml BY REF)::text, + xmlcast('2024-08-14'::date AS xml)::text = xmlcast('2024-08-14'::date AS xml BY VALUE)::text, + xmlcast('2024-08-14'::xml AS date) = xmlcast('2024-08-14'::xml AS date BY REF), + xmlcast('2024-08-14'::xml AS date) = xmlcast('2024-08-14'::xml AS date BY VALUE); + ?column? | ?column? | ?column? | ?column? +----------+----------+----------+---------- + t | t | t | t +(1 row) + +SELECT + xmlcast('12:30:45'::time without time zone AS xml)::text = xmlcast('12:30:45'::time without time zone AS xml BY REF)::text, + xmlcast('12:30:45'::time without time zone AS xml)::text = xmlcast('12:30:45'::time without time zone AS xml BY VALUE)::text, + xmlcast('12:30:45'::xml AS time without time zone) = xmlcast('12:30:45'::xml AS time without time zone BY REF), + xmlcast('12:30:45'::xml AS time without time zone) = xmlcast('12:30:45'::xml AS time without time zone BY VALUE); + ?column? | ?column? | ?column? | ?column? +----------+----------+----------+---------- + t | t | t | t +(1 row) + +SELECT + xmlcast('09:30:10+06:00'::time with time zone AS xml)::text = xmlcast('09:30:10+06:00'::time with time zone AS xml BY REF)::text, + xmlcast('09:30:10+06:00'::time with time zone AS xml)::text = xmlcast('09:30:10+06:00'::time with time zone AS xml BY VALUE)::text, + xmlcast('09:30:10+06:00'::xml AS time with time zone) = xmlcast('09:30:10+06:00'::xml AS time with time zone BY REF), + xmlcast('09:30:10+06:00'::xml AS time with time zone) = xmlcast('09:30:10+06:00'::xml AS time with time zone BY VALUE); + ?column? | ?column? | ?column? | ?column? +----------+----------+----------+---------- + t | t | t | t +(1 row) + +SELECT + xmlcast('2002-05-30 09:30:10+06:00'::timestamp with time zone AS xml)::text = xmlcast('2002-05-30 09:30:10+06:00'::timestamp with time zone AS xml BY REF)::text, + xmlcast('2002-05-30 09:30:10+06:00'::timestamp with time zone AS xml)::text = xmlcast('2002-05-30 09:30:10+06:00'::timestamp with time zone AS xml BY VALUE)::text, + xmlcast('2002-05-30T09:30:10+06:00'::xml AS timestamp with time zone) = xmlcast('2002-05-30T09:30:10+06:00'::xml AS timestamp with time zone BY REF), + xmlcast('2002-05-30T09:30:10+06:00'::xml AS timestamp with time zone) = xmlcast('2002-05-30T09:30:10+06:00'::xml AS timestamp with time zone BY VALUE); + ?column? | ?column? | ?column? | ?column? +----------+----------+----------+---------- + t | t | t | t +(1 row) + +SELECT + xmlcast('1 year 2 mons'::interval AS xml)::text = xmlcast('1 year 2 mons'::interval AS xml BY REF)::text, + xmlcast('1 year 2 mons'::interval AS xml)::text = xmlcast('1 year 2 mons'::interval AS xml BY VALUE)::text, + xmlcast('1 year 2 mons'::xml AS interval) = xmlcast('1 year 2 mons'::xml AS interval BY REF), + xmlcast('1 year 2 mons'::xml AS interval) = xmlcast('1 year 2 mons'::xml AS interval BY VALUE); + ?column? | ?column? | ?column? | ?column? +----------+----------+----------+---------- + t | t | t | t +(1 row) + +CREATE VIEW view_xmlcast_to_xml AS +SELECT + xmlcast(NULL AS xml) AS c1, + xmlcast('foo' AS xml) AS c2, + xmlcast(''::text AS xml) AS c3, + xmlcast(NULL::text AS xml) AS c4, + xmlcast(''::xml AS text) AS c5, + xmlcast(NULL::xml AS text) c6, + xmlcast('foo & <"bar">'::text AS xml) AS c7, + xmlcast('foo & <"bar">'::varchar AS xml) AS c8, + xmlcast('foo & <"bar">'::name AS xml) AS c9, + xmlcast(xmltext(E'foo & <"bar">\r') AS text) AS c10, + xmlcast(xmlcast(E'foo & <"bar">\r' AS xml) AS text) AS c11, + xmlcast(to_date('29/05/2024','dd/mm/yyyy') AS xml) AS c12, + xmlcast('2024-05-29 12:04:10.703585+02'::timestamp with time zone at time zone 'Europe/Berlin' AS xml) AS c13, + xmlcast('2024-05-29 12:04:10.703585+02'::timestamp without time zone AS xml) AS c14, + xmlcast('1 year 2 months 3 days 4 hours 5 minutes 6 seconds'::interval AS xml) AS c15, + xmlcast(427353542 AS xml) AS c16, + xmlcast(4273535420162021 AS xml) AS c17, + xmlcast(42.007312345678910 AS xml) AS c18, + xmlcast(42.007312345678910::double precision AS xml) AS c19, + xmlcast(true AS xml) AS c20, + xmlcast(false AS xml) AS c21, + xmlcast(42 = 73 AS xml) AS c22, + xmlcast(42 <> 73 AS xml) AS c23, + xmlcast('11:11:11.5'::time AS xml) AS c24, + xmlcast('11:11:11.5'::time with time zone at time zone 'Europe/Berlin' AS xml) AS c25; +\sv view_xmlcast_to_xml +CREATE OR REPLACE VIEW public.view_xmlcast_to_xml AS + SELECT XMLCAST(NULL::text AS xml)::xml AS c1, + XMLCAST('foo'::text AS xml)::xml AS c2, + XMLCAST(''::text AS xml)::xml AS c3, + XMLCAST(NULL::text AS xml)::xml AS c4, + XMLCAST(''::xml AS text)::text AS c5, + XMLCAST(NULL::xml AS text)::text AS c6, + XMLCAST('foo & <"bar">'::text AS xml)::xml AS c7, + XMLCAST('foo & <"bar">'::character varying AS xml)::xml AS c8, + XMLCAST('foo & <"bar">'::name::text AS xml)::xml AS c9, + XMLCAST(xmltext('foo & <"bar"> '::text) AS text)::text AS c10, + XMLCAST(XMLCAST('foo & <"bar"> '::text AS xml)::xml AS text)::text AS c11, + XMLCAST(to_date('29/05/2024'::text, 'dd/mm/yyyy'::text) AS xml)::xml AS c12, + XMLCAST(('Wed May 29 03:04:10.703585 2024 PDT'::timestamp with time zone AT TIME ZONE 'Europe/Berlin'::text) AS xml)::xml AS c13, + XMLCAST('Wed May 29 12:04:10.703585 2024'::timestamp without time zone AS xml)::xml AS c14, + XMLCAST('@ 1 year 2 mons 3 days 4 hours 5 mins 6 secs'::interval AS xml)::xml AS c15, + XMLCAST(427353542::text AS xml)::xml AS c16, + XMLCAST('4273535420162021'::bigint::text AS xml)::xml AS c17, + XMLCAST(42.007312345678910 AS xml)::xml AS c18, + XMLCAST(42.007312345678910::double precision AS xml)::xml AS c19, + XMLCAST(true AS xml)::xml AS c20, + XMLCAST(false AS xml)::xml AS c21, + XMLCAST(42 = 73 AS xml)::xml AS c22, + XMLCAST(42 <> 73 AS xml)::xml AS c23, + XMLCAST('11:11:11.5'::time without time zone::text AS xml)::xml AS c24, + XMLCAST(('11:11:11.5-08'::time with time zone AT TIME ZONE 'Europe/Berlin'::text)::text AS xml)::xml AS c25 +SELECT * FROM view_xmlcast_to_xml; + c1 | c2 | c3 | c4 | c5 | c6 | c7 | c8 | c9 | c10 | c11 | c12 | c13 | c14 | c15 | c16 | c17 | c18 | c19 | c20 | c21 | c22 | c23 | c24 | c25 +----+-----+----+----+----+----+-----------------------------------+-----------------------------------+-----------------------------------+-----------------+-----------------+------------+----------------------------+----------------------------+----------------+-----------+------------------+--------------------+-------------------+------+-------+-------+------+------------+--------------- + | foo | | | | | foo & <"bar"> | foo & <"bar"> | foo & <"bar"> | foo & <"bar">\r | foo & <"bar">\r | 2024-05-29 | 2024-05-29T12:04:10.703585 | 2024-05-29T12:04:10.703585 | P1Y2M3DT4H5M6S | 427353542 | 4273535420162021 | 42.007312345678910 | 42.00731234567891 | true | false | false | true | 11:11:11.5 | 20:11:11.5+01 +(1 row) + +CREATE VIEW view_xmlcast_from_xml AS +SELECT + xmlcast('P1Y2M3DT4H5M6S'::xml AS interval) AS c1, + xmlcast('1 year 2 mons 3 days 4 hours 5 minutes 6 seconds'::xml AS interval) AS c2, + xmlcast('2002-09-24'::xml AS date) AS c3, + xmlcast('2002-09-24+06:00'::xml AS date) AS c4, + xmlcast('09:30:10Z'::xml AS time with time zone) at time zone 'Europe/Berlin' AS c5, + xmlcast('09:30:10-06:00'::xml AS time with time zone) at time zone 'Europe/Berlin' AS c6, + xmlcast('09:30:10+06:00'::xml AS time with time zone) at time zone 'Europe/Berlin' AS c7, + xmlcast('2002-05-30T09:30:10Z'::xml AS timestamp with time zone) at time zone 'Europe/Berlin' AS c8, + xmlcast('2002-05-30T09:30:10-06:00'::xml AS timestamp with time zone) at time zone 'Europe/Berlin' AS c9, + xmlcast('2002-05-30T09:30:10+06:00'::xml AS timestamp with time zone) at time zone 'Europe/Berlin' AS c10, + xmlcast('foo bar'::xml AS text) AS c11, + xmlcast(' foo bar '::xml AS varchar) AS c12, + xmlcast('foo & <"bar">'::xml AS text) AS c13, + xmlcast('42.7312345678910'::xml AS numeric) AS c14, + xmlcast('+42.7312345678910'::xml AS numeric) AS c15, + xmlcast('-42.7312345678910'::xml AS numeric) AS c16, + xmlcast('42'::xml AS integer) AS c17, + xmlcast('+42'::xml AS integer) AS c18, + xmlcast('-42'::xml AS integer) AS c19, + xmlcast('4273535420162021'::xml AS bigint) AS c20, + xmlcast('+4273535420162021'::xml AS bigint) AS c21, + xmlcast('-4273535420162021'::xml AS bigint) AS c22, + xmlcast('true'::xml AS boolean) AS c23, + xmlcast('false'::xml AS boolean) AS c24, + xmlcast(''::xml AS character varying) AS c25, + xmlcast(NULL::xml AS character varying) AS c26; +\sv view_xmlcast_from_xml +CREATE OR REPLACE VIEW public.view_xmlcast_from_xml AS + SELECT XMLCAST('P1Y2M3DT4H5M6S'::xml AS interval)::interval AS c1, + XMLCAST('1 year 2 mons 3 days 4 hours 5 minutes 6 seconds'::xml AS interval)::interval AS c2, + XMLCAST('2002-09-24'::xml AS date)::date AS c3, + XMLCAST('2002-09-24+06:00'::xml AS date)::date AS c4, + (XMLCAST('09:30:10Z'::xml AS time with time zone)::time with time zone AT TIME ZONE 'Europe/Berlin'::text) AS c5, + (XMLCAST('09:30:10-06:00'::xml AS time with time zone)::time with time zone AT TIME ZONE 'Europe/Berlin'::text) AS c6, + (XMLCAST('09:30:10+06:00'::xml AS time with time zone)::time with time zone AT TIME ZONE 'Europe/Berlin'::text) AS c7, + (XMLCAST('2002-05-30T09:30:10Z'::xml AS timestamp with time zone)::timestamp with time zone AT TIME ZONE 'Europe/Berlin'::text) AS c8, + (XMLCAST('2002-05-30T09:30:10-06:00'::xml AS timestamp with time zone)::timestamp with time zone AT TIME ZONE 'Europe/Berlin'::text) AS c9, + (XMLCAST('2002-05-30T09:30:10+06:00'::xml AS timestamp with time zone)::timestamp with time zone AT TIME ZONE 'Europe/Berlin'::text) AS c10, + XMLCAST('foo bar'::xml AS text)::text AS c11, + XMLCAST(' foo bar '::xml AS character varying)::character varying AS c12, + XMLCAST('foo & <"bar">'::xml AS text)::text AS c13, + XMLCAST('42.7312345678910'::xml AS numeric)::numeric AS c14, + XMLCAST('+42.7312345678910'::xml AS numeric)::numeric AS c15, + XMLCAST('-42.7312345678910'::xml AS numeric)::numeric AS c16, + XMLCAST('42'::xml AS integer)::integer AS c17, + XMLCAST('+42'::xml AS integer)::integer AS c18, + XMLCAST('-42'::xml AS integer)::integer AS c19, + XMLCAST('4273535420162021'::xml AS bigint)::bigint AS c20, + XMLCAST('+4273535420162021'::xml AS bigint)::bigint AS c21, + XMLCAST('-4273535420162021'::xml AS bigint)::bigint AS c22, + XMLCAST('true'::xml AS boolean)::boolean AS c23, + XMLCAST('false'::xml AS boolean)::boolean AS c24, + XMLCAST(''::xml AS character varying)::character varying AS c25, + XMLCAST(NULL::xml AS character varying)::character varying AS c26 +SELECT * FROM view_xmlcast_from_xml; + c1 | c2 | c3 | c4 | c5 | c6 | c7 | c8 | c9 | c10 | c11 | c12 | c13 | c14 | c15 | c16 | c17 | c18 | c19 | c20 | c21 | c22 | c23 | c24 | c25 | c26 +----------------------------------------------+----------------------------------------------+------------+------------+-------------+-------------+-------------+--------------------------+--------------------------+--------------------------+---------+---------------------+---------------+------------------+------------------+-------------------+-----+-----+-----+------------------+------------------+-------------------+-----+-----+-----+----- + @ 1 year 2 mons 3 days 4 hours 5 mins 6 secs | @ 1 year 2 mons 3 days 4 hours 5 mins 6 secs | 09-24-2002 | 09-24-2002 | 10:30:10+01 | 16:30:10+01 | 04:30:10+01 | Thu May 30 11:30:10 2002 | Thu May 30 17:30:10 2002 | Thu May 30 05:30:10 2002 | foo bar | foo bar | foo & <"bar"> | 42.7312345678910 | 42.7312345678910 | -42.7312345678910 | 42 | 42 | -42 | 4273535420162021 | 4273535420162021 | -4273535420162021 | t | f | | +(1 row) + diff --git a/src/test/regress/expected/xml_1.out b/src/test/regress/expected/xml_1.out index 73c411118a..98b2a2878e 100644 --- a/src/test/regress/expected/xml_1.out +++ b/src/test/regress/expected/xml_1.out @@ -1496,3 +1496,456 @@ ERROR: unsupported XML feature LINE 1: SELECT xmltext('x'|| '<P>73</P>'::xml || .42 || true || 'j':... ^ DETAIL: This functionality requires the server to be built with libxml support. +-- for xmlcast() tests +INSERT INTO xmltest + VALUES (42, +'<?xml version="1.0" encoding="utf-8"?> + <xmlcast> + <period1>P1Y2M3DT4H5M6S</period1> + <period2>1 year 2 mons 3 days 4 hours 5 minutes 6 seconds</period2> + <date1>2002-09-24</date1> + <date2>2002-09-24+06:00</date2> + <time>09:30:10.5</time> + <time_tz1>09:30:10Z</time_tz1> + <time_tz2>09:30:10-06:00</time_tz2> + <time_tz3>09:30:10+06:00</time_tz3> + <timestamp1>2002-05-30T09:00:00</timestamp1> + <timestamp2>2002-05-30T09:30:10.5</timestamp2> + <timestamp_tz1>2002-05-30T09:30:10Z</timestamp_tz1> + <timestamp_tz2>2002-05-30T09:30:10-06:00</timestamp_tz2> + <timestamp_tz3>2002-05-30T09:30:10+06:00</timestamp_tz3> + <text1>foo bar</text1> + <text2> foo bar </text2> + <text3>foo & <"bar"></text3> + <decimal1>42.7312345678910</decimal1> + <decimal2>+42.7312345678910</decimal2> + <decimal3>-42.7312345678910</decimal3> + <decimal4>INF</decimal4> + <decimal5>-INF</decimal5> + <integer1>42</integer1> + <integer2>+42</integer2> + <integer3>-42</integer3> + <long1>4273535420162021</long1> + <long2>+4273535420162021</long2> + <long3>-4273535420162021</long3> + <bool1 att="true">42</bool1> + <bool2 att="false">73</bool2> + <empty></empty> + </xmlcast>'::xml +); +ERROR: unsupported XML feature +LINE 3: '<?xml version="1.0" encoding="utf-8"?> + ^ +DETAIL: This functionality requires the server to be built with libxml support. +-- This prevents the xmlcast regression tests from failing if the system's timezone has been changed. +SET timezone TO 'America/Los_Angeles'; +-- xmlcast exceptions +\set VERBOSITY terse +SELECT xmlcast((xpath('//text1/text()', data))[1] AS text[]) FROM xmltest WHERE id = 42; +ERROR: cannot cast from 'xml' to 'text[]' at character 8 +SELECT xmlcast((xpath('//text1/integer1()', data))[1] AS int[]) FROM xmltest WHERE id = 42; +ERROR: cannot cast from 'xml' to 'integer[]' at character 8 +SELECT xmlcast(NULL AS text); +ERROR: cannot cast from 'unknown' to 'text' at character 8 +SELECT xmlcast('foo'::text AS varchar); +ERROR: cannot cast from 'text' to 'character varying' at character 8 +SELECT xmlcast(42 AS text); +ERROR: cannot cast from 'integer' to 'text' at character 8 +SELECT xmlcast(array['foo','bar'] AS xml); +ERROR: cannot cast from 'text[]' to 'xml' at character 8 +\set VERBOSITY default +-- xmlcast tests for "XML to non-XML" expressions +SELECT + xmlcast((xpath('//date1/text()', data))[1] AS date), pg_typeof(xmlcast((xpath('//date1/text()', data))[1] AS date)), + xmlcast((xpath('//date2/text()', data))[1] AS date), pg_typeof(xmlcast((xpath('//date2/text()', data))[1] AS date)) +FROM xmltest WHERE id = 42; + xmlcast | pg_typeof | xmlcast | pg_typeof +---------+-----------+---------+----------- +(0 rows) + +SELECT + xmlcast((xpath('//period1/text()', data))[1] AS interval), pg_typeof(xmlcast((xpath('//period1/text()', data))[1] AS interval)), + xmlcast((xpath('//period2/text()', data))[1] AS interval), pg_typeof(xmlcast((xpath('//period2/text()', data))[1] AS interval)) +FROM xmltest WHERE id = 42; + xmlcast | pg_typeof | xmlcast | pg_typeof +---------+-----------+---------+----------- +(0 rows) + +SELECT + xmlcast((xpath('//time/text()', data))[1] AS time), pg_typeof(xmlcast((xpath('//time/text()', data))[1] AS time)), + xmlcast((xpath('//time_tz1/text()', data))[1] AS time with time zone), pg_typeof(xmlcast((xpath('//time_tz1/text()', data))[1] AS time with time zone)), + xmlcast((xpath('//time_tz2/text()', data))[1] AS time with time zone), pg_typeof(xmlcast((xpath('//time_tz2/text()', data))[1] AS time with time zone)), + xmlcast((xpath('//time_tz3/text()', data))[1] AS time with time zone), pg_typeof(xmlcast((xpath('//time_tz3/text()', data))[1] AS time with time zone)) +FROM xmltest WHERE id = 42; + xmlcast | pg_typeof | xmlcast | pg_typeof | xmlcast | pg_typeof | xmlcast | pg_typeof +---------+-----------+---------+-----------+---------+-----------+---------+----------- +(0 rows) + +SELECT + xmlcast((xpath('//text1/text()', data))[1] AS text), pg_typeof(xmlcast((xpath('//text1/text()', data))[1] AS text)), + xmlcast((xpath('//text2/text()', data))[1] AS text), pg_typeof(xmlcast((xpath('//text2/text()', data))[1] AS text)), + xmlcast((xpath('//text3/text()', data))[1] AS text), pg_typeof(xmlcast((xpath('//text3/text()', data))[1] AS text)) +FROM xmltest WHERE id = 42; + xmlcast | pg_typeof | xmlcast | pg_typeof | xmlcast | pg_typeof +---------+-----------+---------+-----------+---------+----------- +(0 rows) + +SELECT + xmlcast((xpath('//text1/text()', data))[1] AS varchar), pg_typeof(xmlcast((xpath('//text1/text()', data))[1] AS varchar)), + xmlcast((xpath('//text2/text()', data))[1] AS varchar), pg_typeof(xmlcast((xpath('//text2/text()', data))[1] AS varchar)), + xmlcast((xpath('//text3/text()', data))[1] AS varchar), pg_typeof(xmlcast((xpath('//text3/text()', data))[1] AS varchar)) +FROM xmltest WHERE id = 42; + xmlcast | pg_typeof | xmlcast | pg_typeof | xmlcast | pg_typeof +---------+-----------+---------+-----------+---------+----------- +(0 rows) + +SELECT + xmlcast((xpath('//text1/text()', data))[1] AS name), pg_typeof(xmlcast((xpath('//text1/text()', data))[1] AS name)), + xmlcast((xpath('//text2/text()', data))[1] AS name), pg_typeof(xmlcast((xpath('//text2/text()', data))[1] AS name)), + xmlcast((xpath('//text3/text()', data))[1] AS name), pg_typeof(xmlcast((xpath('//text3/text()', data))[1] AS name)) +FROM xmltest WHERE id = 42; + xmlcast | pg_typeof | xmlcast | pg_typeof | xmlcast | pg_typeof +---------+-----------+---------+-----------+---------+----------- +(0 rows) + +SELECT + xmlcast((xpath('//text1/text()', data))[1] AS bpchar), pg_typeof(xmlcast((xpath('//text1/text()', data))[1] AS bpchar)), + xmlcast((xpath('//text2/text()', data))[1] AS bpchar), pg_typeof(xmlcast((xpath('//text2/text()', data))[1] AS bpchar)), + xmlcast((xpath('//text3/text()', data))[1] AS bpchar), pg_typeof(xmlcast((xpath('//text3/text()', data))[1] AS bpchar)) +FROM xmltest WHERE id = 42; + xmlcast | pg_typeof | xmlcast | pg_typeof | xmlcast | pg_typeof +---------+-----------+---------+-----------+---------+----------- +(0 rows) + +SELECT + xmlcast((xpath('//decimal1/text()', data))[1] AS numeric), pg_typeof(xmlcast((xpath('//decimal1/text()', data))[1] AS numeric)), + xmlcast((xpath('//decimal2/text()', data))[1] AS numeric), pg_typeof(xmlcast((xpath('//decimal2/text()', data))[1] AS numeric)), + xmlcast((xpath('//decimal3/text()', data))[1] AS numeric), pg_typeof(xmlcast((xpath('//decimal3/text()', data))[1] AS numeric)), + xmlcast((xpath('//decimal4/text()', data))[1] AS numeric), pg_typeof(xmlcast((xpath('//decimal4/text()', data))[1] AS numeric)), + xmlcast((xpath('//decimal5/text()', data))[1] AS numeric), pg_typeof(xmlcast((xpath('//decimal5/text()', data))[1] AS numeric)) +FROM xmltest WHERE id = 42; + xmlcast | pg_typeof | xmlcast | pg_typeof | xmlcast | pg_typeof | xmlcast | pg_typeof | xmlcast | pg_typeof +---------+-----------+---------+-----------+---------+-----------+---------+-----------+---------+----------- +(0 rows) + +SELECT + xmlcast((xpath('//decimal1/text()', data))[1] AS double precision), pg_typeof(xmlcast((xpath('//decimal1/text()', data))[1] AS double precision)), + xmlcast((xpath('//decimal2/text()', data))[1] AS double precision), pg_typeof(xmlcast((xpath('//decimal2/text()', data))[1] AS double precision)), + xmlcast((xpath('//decimal3/text()', data))[1] AS double precision), pg_typeof(xmlcast((xpath('//decimal3/text()', data))[1] AS double precision)), + xmlcast((xpath('//decimal4/text()', data))[1] AS double precision), pg_typeof(xmlcast((xpath('//decimal4/text()', data))[1] AS double precision)), + xmlcast((xpath('//decimal5/text()', data))[1] AS double precision), pg_typeof(xmlcast((xpath('//decimal5/text()', data))[1] AS double precision)) +FROM xmltest WHERE id = 42; + xmlcast | pg_typeof | xmlcast | pg_typeof | xmlcast | pg_typeof | xmlcast | pg_typeof | xmlcast | pg_typeof +---------+-----------+---------+-----------+---------+-----------+---------+-----------+---------+----------- +(0 rows) + +SELECT + xmlcast((xpath('//integer1/text()', data))[1] AS int), pg_typeof(xmlcast((xpath('//integer1/text()', data))[1] AS int)), + xmlcast((xpath('//integer2/text()', data))[1] AS int), pg_typeof(xmlcast((xpath('//integer2/text()', data))[1] AS int)), + xmlcast((xpath('//integer3/text()', data))[1] AS int), pg_typeof(xmlcast((xpath('//integer3/text()', data))[1] AS int)) +FROM xmltest WHERE id = 42; + xmlcast | pg_typeof | xmlcast | pg_typeof | xmlcast | pg_typeof +---------+-----------+---------+-----------+---------+----------- +(0 rows) + +SELECT + xmlcast((xpath('//long1/text()', data))[1] AS bigint), pg_typeof(xmlcast((xpath('//long1/text()', data))[1] AS bigint)), + xmlcast((xpath('//long2/text()', data))[1] AS bigint), pg_typeof(xmlcast((xpath('//long2/text()', data))[1] AS bigint)), + xmlcast((xpath('//long3/text()', data))[1] AS bigint), pg_typeof(xmlcast((xpath('//long3/text()', data))[1] AS bigint)) +FROM xmltest WHERE id = 42; + xmlcast | pg_typeof | xmlcast | pg_typeof | xmlcast | pg_typeof +---------+-----------+---------+-----------+---------+----------- +(0 rows) + +SELECT + xmlcast((xpath('//bool1/@att', data))[1] AS boolean), pg_typeof(xmlcast((xpath('//bool1/@att', data))[1] AS boolean)), + xmlcast((xpath('//bool2/@att', data))[1] AS boolean), pg_typeof(xmlcast((xpath('//bool1/@att', data))[1] AS boolean)) +FROM xmltest WHERE id = 42; + xmlcast | pg_typeof | xmlcast | pg_typeof +---------+-----------+---------+----------- +(0 rows) + +SELECT xmlcast((xpath('//empty/text()', data))[1] AS text), pg_typeof(xmlcast((xpath('//empty/text()', data))[1] AS text)) +FROM xmltest WHERE id = 42; + xmlcast | pg_typeof +---------+----------- +(0 rows) + +-- xmlcast tests for "XML to XML" expressions +SELECT + xmlcast((xpath('//text1/text()', data))[1] AS xml), pg_typeof(xmlcast((xpath('//text1/text()', data))[1] AS xml)), + xmlcast((xpath('//text2/text()', data))[1] AS xml), pg_typeof(xmlcast((xpath('//text2/text()', data))[1] AS xml)), + xmlcast((xpath('//text3/text()', data))[1] AS xml), pg_typeof(xmlcast((xpath('//text3/text()', data))[1] AS xml)) +FROM xmltest WHERE id = 42; + xmlcast | pg_typeof | xmlcast | pg_typeof | xmlcast | pg_typeof +---------+-----------+---------+-----------+---------+----------- +(0 rows) + +-- xmlcast tests for "non-XML to XML" expressions +SELECT j, pg_typeof(j) FROM xmlcast(NULL AS xml) t(j); + j | pg_typeof +---+----------- + | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast('foo' AS xml) t(j); +ERROR: unsupported XML feature +DETAIL: This functionality requires the server to be built with libxml support. +SELECT j, pg_typeof(j) FROM xmlcast(''::text AS xml) t(j); +ERROR: unsupported XML feature +DETAIL: This functionality requires the server to be built with libxml support. +SELECT j, pg_typeof(j) FROM xmlcast(NULL::text AS xml) t(j); + j | pg_typeof +---+----------- + | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast(''::xml AS text) t(j); +ERROR: unsupported XML feature +LINE 1: SELECT j, pg_typeof(j) FROM xmlcast(''::xml AS text) t(j); + ^ +DETAIL: This functionality requires the server to be built with libxml support. +SELECT j, pg_typeof(j) FROM xmlcast(NULL::xml AS text) t(j); + j | pg_typeof +---+----------- + | text +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast('foo & <"bar">'::text AS xml) t(j); +ERROR: unsupported XML feature +DETAIL: This functionality requires the server to be built with libxml support. +SELECT j, pg_typeof(j) FROM xmlcast('foo & <"bar">'::varchar AS xml) t(j); +ERROR: unsupported XML feature +DETAIL: This functionality requires the server to be built with libxml support. +SELECT j, pg_typeof(j) FROM xmlcast('foo & <"bar">'::name AS xml) t(j); +ERROR: unsupported XML feature +DETAIL: This functionality requires the server to be built with libxml support. +SELECT j, pg_typeof(j) FROM xmlcast(xmltext(E'foo & <"bar">\r') AS text) t(j); +ERROR: unsupported XML feature +DETAIL: This functionality requires the server to be built with libxml support. +SELECT j, pg_typeof(j) FROM xmlcast(xmlcast(E'foo & <"bar">\r' AS xml) AS text) t(j); +ERROR: unsupported XML feature +DETAIL: This functionality requires the server to be built with libxml support. +SELECT j, pg_typeof(j) FROM xmlcast(to_date('29/05/2024','dd/mm/yyyy') AS xml) t(j); +ERROR: unsupported XML feature +DETAIL: This functionality requires the server to be built with libxml support. +SELECT j, pg_typeof(j) FROM xmlcast('2024-05-29 12:04:10.703585+02'::timestamp with time zone at time zone 'Europe/Berlin' AS xml) t(j); +ERROR: unsupported XML feature +DETAIL: This functionality requires the server to be built with libxml support. +SELECT j, pg_typeof(j) FROM xmlcast('2024-05-29 12:04:10.703585+02'::timestamp without time zone AS xml) t(j); +ERROR: unsupported XML feature +DETAIL: This functionality requires the server to be built with libxml support. +SELECT j, pg_typeof(j) FROM xmlcast('1 year 2 months 3 days 4 hours 5 minutes 6 seconds'::interval AS xml) t(j); +ERROR: unsupported XML feature +DETAIL: This functionality requires the server to be built with libxml support. +SELECT j, pg_typeof(j) FROM xmlcast(427353542 AS xml) t(j); +ERROR: unsupported XML feature +DETAIL: This functionality requires the server to be built with libxml support. +SELECT j, pg_typeof(j) FROM xmlcast(4273535420162021 AS xml) t(j); +ERROR: unsupported XML feature +DETAIL: This functionality requires the server to be built with libxml support. +SELECT j, pg_typeof(j) FROM xmlcast(42.007312345678910 AS xml) t(j); +ERROR: unsupported XML feature +DETAIL: This functionality requires the server to be built with libxml support. +SELECT j, pg_typeof(j) FROM xmlcast(42.007312345678910::double precision AS xml) t(j); +ERROR: unsupported XML feature +DETAIL: This functionality requires the server to be built with libxml support. +SELECT j, pg_typeof(j) FROM xmlcast('infinity'::double precision AS xml) t(j); +ERROR: unsupported XML feature +DETAIL: This functionality requires the server to be built with libxml support. +SELECT j, pg_typeof(j) FROM xmlcast('-infinity'::double precision AS xml) t(j); +ERROR: unsupported XML feature +DETAIL: This functionality requires the server to be built with libxml support. +SELECT j, pg_typeof(j) FROM xmlcast('nan'::double precision AS xml) t(j); +ERROR: unsupported XML feature +DETAIL: This functionality requires the server to be built with libxml support. +SELECT j, pg_typeof(j) FROM xmlcast('infinity'::numeric AS xml) t(j); +ERROR: unsupported XML feature +DETAIL: This functionality requires the server to be built with libxml support. +SELECT j, pg_typeof(j) FROM xmlcast('-infinity'::numeric AS xml) t(j); +ERROR: unsupported XML feature +DETAIL: This functionality requires the server to be built with libxml support. +SELECT j, pg_typeof(j) FROM xmlcast('nan'::numeric AS xml) t(j); +ERROR: unsupported XML feature +DETAIL: This functionality requires the server to be built with libxml support. +SELECT j, pg_typeof(j) FROM xmlcast(true AS xml) t(j); +ERROR: unsupported XML feature +DETAIL: This functionality requires the server to be built with libxml support. +SELECT j, pg_typeof(j) FROM xmlcast(false AS xml) t(j); +ERROR: unsupported XML feature +DETAIL: This functionality requires the server to be built with libxml support. +SELECT j, pg_typeof(j) FROM xmlcast(42 = 73 AS xml) t(j); +ERROR: unsupported XML feature +DETAIL: This functionality requires the server to be built with libxml support. +SELECT j, pg_typeof(j) FROM xmlcast(42 <> 73 AS xml) t(j); +ERROR: unsupported XML feature +DETAIL: This functionality requires the server to be built with libxml support. +SELECT j, pg_typeof(j) FROM xmlcast('11:11:11.5'::time AS xml) t(j); +ERROR: unsupported XML feature +DETAIL: This functionality requires the server to be built with libxml support. +SELECT j, pg_typeof(j) FROM xmlcast('11:11:11.5'::time with time zone at time zone 'Europe/Berlin' AS xml) t(j); +ERROR: unsupported XML feature +DETAIL: This functionality requires the server to be built with libxml support. +-- Convert an XML string to bytea and back to xml +SELECT xmlcast(convert_from(xmlcast('<"foo&bar">'::xml AS bytea),'UTF8')::xml AS text); +ERROR: unsupported XML feature +LINE 1: SELECT xmlcast(convert_from(xmlcast('<"foo&bar&q... + ^ +DETAIL: This functionality requires the server to be built with libxml support. +SET xmlbinary TO hex; +SELECT xmlcast(E'\\xDEADBEEF'::bytea AS xml); +ERROR: unsupported XML feature +DETAIL: This functionality requires the server to be built with libxml support. +SET xmlbinary TO base64; +SELECT xmlcast(E'\\xDEADBEEF'::bytea AS xml); +ERROR: unsupported XML feature +DETAIL: This functionality requires the server to be built with libxml support. +-- The BY REF and BY VALUE clauses are accepted but ignored. +-- This checks if the results are indeed the same as without the clauses. +SELECT + xmlcast('foo' AS xml)::text = xmlcast('foo' AS xml BY REF)::text, + xmlcast('foo' AS xml)::text = xmlcast('foo' AS xml BY VALUE)::text, + xmlcast('foo'::xml AS text) = xmlcast('foo'::xml AS text BY REF), + xmlcast('foo'::xml AS text) = xmlcast('foo'::xml AS text BY VALUE); +ERROR: unsupported XML feature +LINE 4: xmlcast('foo'::xml AS text) = xmlcast('foo'::xml AS text B... + ^ +DETAIL: This functionality requires the server to be built with libxml support. +SELECT + xmlcast(42 AS xml)::text = xmlcast(42 AS xml BY REF)::text, + xmlcast(42 AS xml)::text = xmlcast(42 AS xml BY VALUE)::text, + xmlcast('42'::xml AS int) = xmlcast('42'::xml AS int BY REF), + xmlcast('42'::xml AS int) = xmlcast('42'::xml AS int BY VALUE); +ERROR: unsupported XML feature +LINE 4: xmlcast('42'::xml AS int) = xmlcast('42'::xml AS int BY RE... + ^ +DETAIL: This functionality requires the server to be built with libxml support. +SELECT + xmlcast(42.73 AS xml)::text = xmlcast(42.73 AS xml BY REF)::text, + xmlcast(42.73 AS xml)::text = xmlcast(42.73 AS xml BY VALUE)::text, + xmlcast('42.73'::xml AS numeric) = xmlcast('42.73'::xml AS numeric BY REF), + xmlcast('42.73'::xml AS numeric) = xmlcast('42.73'::xml AS numeric BY VALUE); +ERROR: unsupported XML feature +LINE 4: xmlcast('42.73'::xml AS numeric) = xmlcast('42.73'::xml AS... + ^ +DETAIL: This functionality requires the server to be built with libxml support. +SELECT + xmlcast('2024-08-14'::date AS xml)::text = xmlcast('2024-08-14'::date AS xml BY REF)::text, + xmlcast('2024-08-14'::date AS xml)::text = xmlcast('2024-08-14'::date AS xml BY VALUE)::text, + xmlcast('2024-08-14'::xml AS date) = xmlcast('2024-08-14'::xml AS date BY REF), + xmlcast('2024-08-14'::xml AS date) = xmlcast('2024-08-14'::xml AS date BY VALUE); +ERROR: unsupported XML feature +LINE 4: xmlcast('2024-08-14'::xml AS date) = xmlcast('2024-08-14':... + ^ +DETAIL: This functionality requires the server to be built with libxml support. +SELECT + xmlcast('12:30:45'::time without time zone AS xml)::text = xmlcast('12:30:45'::time without time zone AS xml BY REF)::text, + xmlcast('12:30:45'::time without time zone AS xml)::text = xmlcast('12:30:45'::time without time zone AS xml BY VALUE)::text, + xmlcast('12:30:45'::xml AS time without time zone) = xmlcast('12:30:45'::xml AS time without time zone BY REF), + xmlcast('12:30:45'::xml AS time without time zone) = xmlcast('12:30:45'::xml AS time without time zone BY VALUE); +ERROR: unsupported XML feature +LINE 4: xmlcast('12:30:45'::xml AS time without time zone) = xmlca... + ^ +DETAIL: This functionality requires the server to be built with libxml support. +SELECT + xmlcast('09:30:10+06:00'::time with time zone AS xml)::text = xmlcast('09:30:10+06:00'::time with time zone AS xml BY REF)::text, + xmlcast('09:30:10+06:00'::time with time zone AS xml)::text = xmlcast('09:30:10+06:00'::time with time zone AS xml BY VALUE)::text, + xmlcast('09:30:10+06:00'::xml AS time with time zone) = xmlcast('09:30:10+06:00'::xml AS time with time zone BY REF), + xmlcast('09:30:10+06:00'::xml AS time with time zone) = xmlcast('09:30:10+06:00'::xml AS time with time zone BY VALUE); +ERROR: unsupported XML feature +LINE 4: xmlcast('09:30:10+06:00'::xml AS time with time zone) = xm... + ^ +DETAIL: This functionality requires the server to be built with libxml support. +SELECT + xmlcast('2002-05-30 09:30:10+06:00'::timestamp with time zone AS xml)::text = xmlcast('2002-05-30 09:30:10+06:00'::timestamp with time zone AS xml BY REF)::text, + xmlcast('2002-05-30 09:30:10+06:00'::timestamp with time zone AS xml)::text = xmlcast('2002-05-30 09:30:10+06:00'::timestamp with time zone AS xml BY VALUE)::text, + xmlcast('2002-05-30T09:30:10+06:00'::xml AS timestamp with time zone) = xmlcast('2002-05-30T09:30:10+06:00'::xml AS timestamp with time zone BY REF), + xmlcast('2002-05-30T09:30:10+06:00'::xml AS timestamp with time zone) = xmlcast('2002-05-30T09:30:10+06:00'::xml AS timestamp with time zone BY VALUE); +ERROR: unsupported XML feature +LINE 4: xmlcast('2002-05-30T09:30:10+06:00'::xml AS timestamp with... + ^ +DETAIL: This functionality requires the server to be built with libxml support. +SELECT + xmlcast('1 year 2 mons'::interval AS xml)::text = xmlcast('1 year 2 mons'::interval AS xml BY REF)::text, + xmlcast('1 year 2 mons'::interval AS xml)::text = xmlcast('1 year 2 mons'::interval AS xml BY VALUE)::text, + xmlcast('1 year 2 mons'::xml AS interval) = xmlcast('1 year 2 mons'::xml AS interval BY REF), + xmlcast('1 year 2 mons'::xml AS interval) = xmlcast('1 year 2 mons'::xml AS interval BY VALUE); +ERROR: unsupported XML feature +LINE 4: xmlcast('1 year 2 mons'::xml AS interval) = xmlcast('1 yea... + ^ +DETAIL: This functionality requires the server to be built with libxml support. +CREATE VIEW view_xmlcast_to_xml AS +SELECT + xmlcast(NULL AS xml) AS c1, + xmlcast('foo' AS xml) AS c2, + xmlcast(''::text AS xml) AS c3, + xmlcast(NULL::text AS xml) AS c4, + xmlcast(''::xml AS text) AS c5, + xmlcast(NULL::xml AS text) c6, + xmlcast('foo & <"bar">'::text AS xml) AS c7, + xmlcast('foo & <"bar">'::varchar AS xml) AS c8, + xmlcast('foo & <"bar">'::name AS xml) AS c9, + xmlcast(xmltext(E'foo & <"bar">\r') AS text) AS c10, + xmlcast(xmlcast(E'foo & <"bar">\r' AS xml) AS text) AS c11, + xmlcast(to_date('29/05/2024','dd/mm/yyyy') AS xml) AS c12, + xmlcast('2024-05-29 12:04:10.703585+02'::timestamp with time zone at time zone 'Europe/Berlin' AS xml) AS c13, + xmlcast('2024-05-29 12:04:10.703585+02'::timestamp without time zone AS xml) AS c14, + xmlcast('1 year 2 months 3 days 4 hours 5 minutes 6 seconds'::interval AS xml) AS c15, + xmlcast(427353542 AS xml) AS c16, + xmlcast(4273535420162021 AS xml) AS c17, + xmlcast(42.007312345678910 AS xml) AS c18, + xmlcast(42.007312345678910::double precision AS xml) AS c19, + xmlcast(true AS xml) AS c20, + xmlcast(false AS xml) AS c21, + xmlcast(42 = 73 AS xml) AS c22, + xmlcast(42 <> 73 AS xml) AS c23, + xmlcast('11:11:11.5'::time AS xml) AS c24, + xmlcast('11:11:11.5'::time with time zone at time zone 'Europe/Berlin' AS xml) AS c25; +ERROR: unsupported XML feature +LINE 7: xmlcast(''::xml AS text) AS c5, + ^ +DETAIL: This functionality requires the server to be built with libxml support. +\sv view_xmlcast_to_xml +ERROR: relation "view_xmlcast_to_xml" does not exist +SELECT * FROM view_xmlcast_to_xml; +ERROR: relation "view_xmlcast_to_xml" does not exist +LINE 1: SELECT * FROM view_xmlcast_to_xml; + ^ +CREATE VIEW view_xmlcast_from_xml AS +SELECT + xmlcast('P1Y2M3DT4H5M6S'::xml AS interval) AS c1, + xmlcast('1 year 2 mons 3 days 4 hours 5 minutes 6 seconds'::xml AS interval) AS c2, + xmlcast('2002-09-24'::xml AS date) AS c3, + xmlcast('2002-09-24+06:00'::xml AS date) AS c4, + xmlcast('09:30:10Z'::xml AS time with time zone) at time zone 'Europe/Berlin' AS c5, + xmlcast('09:30:10-06:00'::xml AS time with time zone) at time zone 'Europe/Berlin' AS c6, + xmlcast('09:30:10+06:00'::xml AS time with time zone) at time zone 'Europe/Berlin' AS c7, + xmlcast('2002-05-30T09:30:10Z'::xml AS timestamp with time zone) at time zone 'Europe/Berlin' AS c8, + xmlcast('2002-05-30T09:30:10-06:00'::xml AS timestamp with time zone) at time zone 'Europe/Berlin' AS c9, + xmlcast('2002-05-30T09:30:10+06:00'::xml AS timestamp with time zone) at time zone 'Europe/Berlin' AS c10, + xmlcast('foo bar'::xml AS text) AS c11, + xmlcast(' foo bar '::xml AS varchar) AS c12, + xmlcast('foo & <"bar">'::xml AS text) AS c13, + xmlcast('42.7312345678910'::xml AS numeric) AS c14, + xmlcast('+42.7312345678910'::xml AS numeric) AS c15, + xmlcast('-42.7312345678910'::xml AS numeric) AS c16, + xmlcast('42'::xml AS integer) AS c17, + xmlcast('+42'::xml AS integer) AS c18, + xmlcast('-42'::xml AS integer) AS c19, + xmlcast('4273535420162021'::xml AS bigint) AS c20, + xmlcast('+4273535420162021'::xml AS bigint) AS c21, + xmlcast('-4273535420162021'::xml AS bigint) AS c22, + xmlcast('true'::xml AS boolean) AS c23, + xmlcast('false'::xml AS boolean) AS c24, + xmlcast(''::xml AS character varying) AS c25, + xmlcast(NULL::xml AS character varying) AS c26; +ERROR: unsupported XML feature +LINE 3: xmlcast('P1Y2M3DT4H5M6S'::xml AS interval) AS c1, + ^ +DETAIL: This functionality requires the server to be built with libxml support. +\sv view_xmlcast_from_xml +ERROR: relation "view_xmlcast_from_xml" does not exist +SELECT * FROM view_xmlcast_from_xml; +ERROR: relation "view_xmlcast_from_xml" does not exist +LINE 1: SELECT * FROM view_xmlcast_from_xml; + ^ diff --git a/src/test/regress/expected/xml_2.out b/src/test/regress/expected/xml_2.out index a85d95358d..5ec57ae5ef 100644 --- a/src/test/regress/expected/xml_2.out +++ b/src/test/regress/expected/xml_2.out @@ -1867,3 +1867,612 @@ SELECT xmltext('x'|| '<P>73</P>'::xml || .42 || true || 'j'::char); x<P>73</P>0.42truej (1 row) +-- for xmlcast() tests +INSERT INTO xmltest + VALUES (42, +'<?xml version="1.0" encoding="utf-8"?> + <xmlcast> + <period1>P1Y2M3DT4H5M6S</period1> + <period2>1 year 2 mons 3 days 4 hours 5 minutes 6 seconds</period2> + <date1>2002-09-24</date1> + <date2>2002-09-24+06:00</date2> + <time>09:30:10.5</time> + <time_tz1>09:30:10Z</time_tz1> + <time_tz2>09:30:10-06:00</time_tz2> + <time_tz3>09:30:10+06:00</time_tz3> + <timestamp1>2002-05-30T09:00:00</timestamp1> + <timestamp2>2002-05-30T09:30:10.5</timestamp2> + <timestamp_tz1>2002-05-30T09:30:10Z</timestamp_tz1> + <timestamp_tz2>2002-05-30T09:30:10-06:00</timestamp_tz2> + <timestamp_tz3>2002-05-30T09:30:10+06:00</timestamp_tz3> + <text1>foo bar</text1> + <text2> foo bar </text2> + <text3>foo & <"bar"></text3> + <decimal1>42.7312345678910</decimal1> + <decimal2>+42.7312345678910</decimal2> + <decimal3>-42.7312345678910</decimal3> + <decimal4>INF</decimal4> + <decimal5>-INF</decimal5> + <integer1>42</integer1> + <integer2>+42</integer2> + <integer3>-42</integer3> + <long1>4273535420162021</long1> + <long2>+4273535420162021</long2> + <long3>-4273535420162021</long3> + <bool1 att="true">42</bool1> + <bool2 att="false">73</bool2> + <empty></empty> + </xmlcast>'::xml +); +-- This prevents the xmlcast regression tests from failing if the system's timezone has been changed. +SET timezone TO 'America/Los_Angeles'; +-- xmlcast exceptions +\set VERBOSITY terse +SELECT xmlcast((xpath('//text1/text()', data))[1] AS text[]) FROM xmltest WHERE id = 42; +ERROR: cannot cast from 'xml' to 'text[]' at character 8 +SELECT xmlcast((xpath('//text1/integer1()', data))[1] AS int[]) FROM xmltest WHERE id = 42; +ERROR: cannot cast from 'xml' to 'integer[]' at character 8 +SELECT xmlcast(NULL AS text); +ERROR: cannot cast from 'unknown' to 'text' at character 8 +SELECT xmlcast('foo'::text AS varchar); +ERROR: cannot cast from 'text' to 'character varying' at character 8 +SELECT xmlcast(42 AS text); +ERROR: cannot cast from 'integer' to 'text' at character 8 +SELECT xmlcast(array['foo','bar'] AS xml); +ERROR: cannot cast from 'text[]' to 'xml' at character 8 +\set VERBOSITY default +-- xmlcast tests for "XML to non-XML" expressions +SELECT + xmlcast((xpath('//date1/text()', data))[1] AS date), pg_typeof(xmlcast((xpath('//date1/text()', data))[1] AS date)), + xmlcast((xpath('//date2/text()', data))[1] AS date), pg_typeof(xmlcast((xpath('//date2/text()', data))[1] AS date)) +FROM xmltest WHERE id = 42; + xmlcast | pg_typeof | xmlcast | pg_typeof +------------+-----------+------------+----------- + 09-24-2002 | date | 09-24-2002 | date +(1 row) + +SELECT + xmlcast((xpath('//period1/text()', data))[1] AS interval), pg_typeof(xmlcast((xpath('//period1/text()', data))[1] AS interval)), + xmlcast((xpath('//period2/text()', data))[1] AS interval), pg_typeof(xmlcast((xpath('//period2/text()', data))[1] AS interval)) +FROM xmltest WHERE id = 42; + xmlcast | pg_typeof | xmlcast | pg_typeof +----------------------------------------------+-----------+----------------------------------------------+----------- + @ 1 year 2 mons 3 days 4 hours 5 mins 6 secs | interval | @ 1 year 2 mons 3 days 4 hours 5 mins 6 secs | interval +(1 row) + +SELECT + xmlcast((xpath('//time/text()', data))[1] AS time), pg_typeof(xmlcast((xpath('//time/text()', data))[1] AS time)), + xmlcast((xpath('//time_tz1/text()', data))[1] AS time with time zone), pg_typeof(xmlcast((xpath('//time_tz1/text()', data))[1] AS time with time zone)), + xmlcast((xpath('//time_tz2/text()', data))[1] AS time with time zone), pg_typeof(xmlcast((xpath('//time_tz2/text()', data))[1] AS time with time zone)), + xmlcast((xpath('//time_tz3/text()', data))[1] AS time with time zone), pg_typeof(xmlcast((xpath('//time_tz3/text()', data))[1] AS time with time zone)) +FROM xmltest WHERE id = 42; + xmlcast | pg_typeof | xmlcast | pg_typeof | xmlcast | pg_typeof | xmlcast | pg_typeof +------------+------------------------+-------------+---------------------+-------------+---------------------+-------------+--------------------- + 09:30:10.5 | time without time zone | 09:30:10+00 | time with time zone | 09:30:10-06 | time with time zone | 09:30:10+06 | time with time zone +(1 row) + +SELECT + xmlcast((xpath('//text1/text()', data))[1] AS text), pg_typeof(xmlcast((xpath('//text1/text()', data))[1] AS text)), + xmlcast((xpath('//text2/text()', data))[1] AS text), pg_typeof(xmlcast((xpath('//text2/text()', data))[1] AS text)), + xmlcast((xpath('//text3/text()', data))[1] AS text), pg_typeof(xmlcast((xpath('//text3/text()', data))[1] AS text)) +FROM xmltest WHERE id = 42; + xmlcast | pg_typeof | xmlcast | pg_typeof | xmlcast | pg_typeof +---------+-----------+---------------------+-----------+---------------+----------- + foo bar | text | foo bar | text | foo & <"bar"> | text +(1 row) + +SELECT + xmlcast((xpath('//text1/text()', data))[1] AS varchar), pg_typeof(xmlcast((xpath('//text1/text()', data))[1] AS varchar)), + xmlcast((xpath('//text2/text()', data))[1] AS varchar), pg_typeof(xmlcast((xpath('//text2/text()', data))[1] AS varchar)), + xmlcast((xpath('//text3/text()', data))[1] AS varchar), pg_typeof(xmlcast((xpath('//text3/text()', data))[1] AS varchar)) +FROM xmltest WHERE id = 42; + xmlcast | pg_typeof | xmlcast | pg_typeof | xmlcast | pg_typeof +---------+-------------------+---------------------+-------------------+---------------+------------------- + foo bar | character varying | foo bar | character varying | foo & <"bar"> | character varying +(1 row) + +SELECT + xmlcast((xpath('//text1/text()', data))[1] AS name), pg_typeof(xmlcast((xpath('//text1/text()', data))[1] AS name)), + xmlcast((xpath('//text2/text()', data))[1] AS name), pg_typeof(xmlcast((xpath('//text2/text()', data))[1] AS name)), + xmlcast((xpath('//text3/text()', data))[1] AS name), pg_typeof(xmlcast((xpath('//text3/text()', data))[1] AS name)) +FROM xmltest WHERE id = 42; + xmlcast | pg_typeof | xmlcast | pg_typeof | xmlcast | pg_typeof +---------+-----------+---------------------+-----------+---------------+----------- + foo bar | name | foo bar | name | foo & <"bar"> | name +(1 row) + +SELECT + xmlcast((xpath('//text1/text()', data))[1] AS bpchar), pg_typeof(xmlcast((xpath('//text1/text()', data))[1] AS bpchar)), + xmlcast((xpath('//text2/text()', data))[1] AS bpchar), pg_typeof(xmlcast((xpath('//text2/text()', data))[1] AS bpchar)), + xmlcast((xpath('//text3/text()', data))[1] AS bpchar), pg_typeof(xmlcast((xpath('//text3/text()', data))[1] AS bpchar)) +FROM xmltest WHERE id = 42; + xmlcast | pg_typeof | xmlcast | pg_typeof | xmlcast | pg_typeof +---------+-----------+---------------------+-----------+---------------+----------- + foo bar | character | foo bar | character | foo & <"bar"> | character +(1 row) + +SELECT + xmlcast((xpath('//decimal1/text()', data))[1] AS numeric), pg_typeof(xmlcast((xpath('//decimal1/text()', data))[1] AS numeric)), + xmlcast((xpath('//decimal2/text()', data))[1] AS numeric), pg_typeof(xmlcast((xpath('//decimal2/text()', data))[1] AS numeric)), + xmlcast((xpath('//decimal3/text()', data))[1] AS numeric), pg_typeof(xmlcast((xpath('//decimal3/text()', data))[1] AS numeric)), + xmlcast((xpath('//decimal4/text()', data))[1] AS numeric), pg_typeof(xmlcast((xpath('//decimal4/text()', data))[1] AS numeric)), + xmlcast((xpath('//decimal5/text()', data))[1] AS numeric), pg_typeof(xmlcast((xpath('//decimal5/text()', data))[1] AS numeric)) +FROM xmltest WHERE id = 42; + xmlcast | pg_typeof | xmlcast | pg_typeof | xmlcast | pg_typeof | xmlcast | pg_typeof | xmlcast | pg_typeof +------------------+-----------+------------------+-----------+-------------------+-----------+----------+-----------+-----------+----------- + 42.7312345678910 | numeric | 42.7312345678910 | numeric | -42.7312345678910 | numeric | Infinity | numeric | -Infinity | numeric +(1 row) + +SELECT + xmlcast((xpath('//decimal1/text()', data))[1] AS double precision), pg_typeof(xmlcast((xpath('//decimal1/text()', data))[1] AS double precision)), + xmlcast((xpath('//decimal2/text()', data))[1] AS double precision), pg_typeof(xmlcast((xpath('//decimal2/text()', data))[1] AS double precision)), + xmlcast((xpath('//decimal3/text()', data))[1] AS double precision), pg_typeof(xmlcast((xpath('//decimal3/text()', data))[1] AS double precision)), + xmlcast((xpath('//decimal4/text()', data))[1] AS double precision), pg_typeof(xmlcast((xpath('//decimal4/text()', data))[1] AS double precision)), + xmlcast((xpath('//decimal5/text()', data))[1] AS double precision), pg_typeof(xmlcast((xpath('//decimal5/text()', data))[1] AS double precision)) +FROM xmltest WHERE id = 42; + xmlcast | pg_typeof | xmlcast | pg_typeof | xmlcast | pg_typeof | xmlcast | pg_typeof | xmlcast | pg_typeof +-----------------+------------------+-----------------+------------------+------------------+------------------+----------+------------------+-----------+------------------ + 42.731234567891 | double precision | 42.731234567891 | double precision | -42.731234567891 | double precision | Infinity | double precision | -Infinity | double precision +(1 row) + +SELECT + xmlcast((xpath('//integer1/text()', data))[1] AS int), pg_typeof(xmlcast((xpath('//integer1/text()', data))[1] AS int)), + xmlcast((xpath('//integer2/text()', data))[1] AS int), pg_typeof(xmlcast((xpath('//integer2/text()', data))[1] AS int)), + xmlcast((xpath('//integer3/text()', data))[1] AS int), pg_typeof(xmlcast((xpath('//integer3/text()', data))[1] AS int)) +FROM xmltest WHERE id = 42; + xmlcast | pg_typeof | xmlcast | pg_typeof | xmlcast | pg_typeof +---------+-----------+---------+-----------+---------+----------- + 42 | integer | 42 | integer | -42 | integer +(1 row) + +SELECT + xmlcast((xpath('//long1/text()', data))[1] AS bigint), pg_typeof(xmlcast((xpath('//long1/text()', data))[1] AS bigint)), + xmlcast((xpath('//long2/text()', data))[1] AS bigint), pg_typeof(xmlcast((xpath('//long2/text()', data))[1] AS bigint)), + xmlcast((xpath('//long3/text()', data))[1] AS bigint), pg_typeof(xmlcast((xpath('//long3/text()', data))[1] AS bigint)) +FROM xmltest WHERE id = 42; + xmlcast | pg_typeof | xmlcast | pg_typeof | xmlcast | pg_typeof +------------------+-----------+------------------+-----------+-------------------+----------- + 4273535420162021 | bigint | 4273535420162021 | bigint | -4273535420162021 | bigint +(1 row) + +SELECT + xmlcast((xpath('//bool1/@att', data))[1] AS boolean), pg_typeof(xmlcast((xpath('//bool1/@att', data))[1] AS boolean)), + xmlcast((xpath('//bool2/@att', data))[1] AS boolean), pg_typeof(xmlcast((xpath('//bool1/@att', data))[1] AS boolean)) +FROM xmltest WHERE id = 42; + xmlcast | pg_typeof | xmlcast | pg_typeof +---------+-----------+---------+----------- + t | boolean | f | boolean +(1 row) + +SELECT xmlcast((xpath('//empty/text()', data))[1] AS text), pg_typeof(xmlcast((xpath('//empty/text()', data))[1] AS text)) +FROM xmltest WHERE id = 42; + xmlcast | pg_typeof +---------+----------- + | text +(1 row) + +-- xmlcast tests for "XML to XML" expressions +SELECT + xmlcast((xpath('//text1/text()', data))[1] AS xml), pg_typeof(xmlcast((xpath('//text1/text()', data))[1] AS xml)), + xmlcast((xpath('//text2/text()', data))[1] AS xml), pg_typeof(xmlcast((xpath('//text2/text()', data))[1] AS xml)), + xmlcast((xpath('//text3/text()', data))[1] AS xml), pg_typeof(xmlcast((xpath('//text3/text()', data))[1] AS xml)) +FROM xmltest WHERE id = 42; + xmlcast | pg_typeof | xmlcast | pg_typeof | xmlcast | pg_typeof +---------+-----------+---------------------+-----------+-------------------------+----------- + foo bar | xml | foo bar | xml | foo & <"bar"> | xml +(1 row) + +-- xmlcast tests for "non-XML to XML" expressions +SELECT j, pg_typeof(j) FROM xmlcast(NULL AS xml) t(j); + j | pg_typeof +---+----------- + | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast('foo' AS xml) t(j); + j | pg_typeof +-----+----------- + foo | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast(''::text AS xml) t(j); + j | pg_typeof +---+----------- + | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast(NULL::text AS xml) t(j); + j | pg_typeof +---+----------- + | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast(''::xml AS text) t(j); + j | pg_typeof +---+----------- + | text +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast(NULL::xml AS text) t(j); + j | pg_typeof +---+----------- + | text +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast('foo & <"bar">'::text AS xml) t(j); + j | pg_typeof +-----------------------------------+----------- + foo & <"bar"> | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast('foo & <"bar">'::varchar AS xml) t(j); + j | pg_typeof +-----------------------------------+----------- + foo & <"bar"> | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast('foo & <"bar">'::name AS xml) t(j); + j | pg_typeof +-----------------------------------+----------- + foo & <"bar"> | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast(xmltext(E'foo & <"bar">\r') AS text) t(j); + j | pg_typeof +-----------------+----------- + foo & <"bar">\r | text +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast(xmlcast(E'foo & <"bar">\r' AS xml) AS text) t(j); + j | pg_typeof +-----------------+----------- + foo & <"bar">\r | text +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast(to_date('29/05/2024','dd/mm/yyyy') AS xml) t(j); + j | pg_typeof +------------+----------- + 2024-05-29 | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast('2024-05-29 12:04:10.703585+02'::timestamp with time zone at time zone 'Europe/Berlin' AS xml) t(j); + j | pg_typeof +----------------------------+----------- + 2024-05-29T12:04:10.703585 | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast('2024-05-29 12:04:10.703585+02'::timestamp without time zone AS xml) t(j); + j | pg_typeof +----------------------------+----------- + 2024-05-29T12:04:10.703585 | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast('1 year 2 months 3 days 4 hours 5 minutes 6 seconds'::interval AS xml) t(j); + j | pg_typeof +----------------+----------- + P1Y2M3DT4H5M6S | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast(427353542 AS xml) t(j); + j | pg_typeof +-----------+----------- + 427353542 | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast(4273535420162021 AS xml) t(j); + j | pg_typeof +------------------+----------- + 4273535420162021 | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast(42.007312345678910 AS xml) t(j); + j | pg_typeof +--------------------+----------- + 42.007312345678910 | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast(42.007312345678910::double precision AS xml) t(j); + j | pg_typeof +-------------------+----------- + 42.00731234567891 | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast('infinity'::double precision AS xml) t(j); + j | pg_typeof +-----+----------- + INF | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast('-infinity'::double precision AS xml) t(j); + j | pg_typeof +------+----------- + -INF | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast('nan'::double precision AS xml) t(j); + j | pg_typeof +-----+----------- + NaN | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast('infinity'::numeric AS xml) t(j); + j | pg_typeof +-----+----------- + INF | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast('-infinity'::numeric AS xml) t(j); + j | pg_typeof +------+----------- + -INF | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast('nan'::numeric AS xml) t(j); + j | pg_typeof +-----+----------- + NaN | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast(true AS xml) t(j); + j | pg_typeof +------+----------- + true | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast(false AS xml) t(j); + j | pg_typeof +-------+----------- + false | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast(42 = 73 AS xml) t(j); + j | pg_typeof +-------+----------- + false | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast(42 <> 73 AS xml) t(j); + j | pg_typeof +------+----------- + true | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast('11:11:11.5'::time AS xml) t(j); + j | pg_typeof +------------+----------- + 11:11:11.5 | xml +(1 row) + +SELECT j, pg_typeof(j) FROM xmlcast('11:11:11.5'::time with time zone at time zone 'Europe/Berlin' AS xml) t(j); + j | pg_typeof +---------------+----------- + 20:11:11.5+01 | xml +(1 row) + +-- Convert an XML string to bytea and back to xml +SELECT xmlcast(convert_from(xmlcast('<"foo&bar">'::xml AS bytea),'UTF8')::xml AS text); + xmlcast +------------- + <"foo&bar"> +(1 row) + +SET xmlbinary TO hex; +SELECT xmlcast(E'\\xDEADBEEF'::bytea AS xml); + xmlcast +---------- + DEADBEEF +(1 row) + +SET xmlbinary TO base64; +SELECT xmlcast(E'\\xDEADBEEF'::bytea AS xml); + xmlcast +---------- + 3q2+7w== +(1 row) + +-- The BY REF and BY VALUE clauses are accepted but ignored. +-- This checks if the results are indeed the same as without the clauses. +SELECT + xmlcast('foo' AS xml)::text = xmlcast('foo' AS xml BY REF)::text, + xmlcast('foo' AS xml)::text = xmlcast('foo' AS xml BY VALUE)::text, + xmlcast('foo'::xml AS text) = xmlcast('foo'::xml AS text BY REF), + xmlcast('foo'::xml AS text) = xmlcast('foo'::xml AS text BY VALUE); + ?column? | ?column? | ?column? | ?column? +----------+----------+----------+---------- + t | t | t | t +(1 row) + +SELECT + xmlcast(42 AS xml)::text = xmlcast(42 AS xml BY REF)::text, + xmlcast(42 AS xml)::text = xmlcast(42 AS xml BY VALUE)::text, + xmlcast('42'::xml AS int) = xmlcast('42'::xml AS int BY REF), + xmlcast('42'::xml AS int) = xmlcast('42'::xml AS int BY VALUE); + ?column? | ?column? | ?column? | ?column? +----------+----------+----------+---------- + t | t | t | t +(1 row) + +SELECT + xmlcast(42.73 AS xml)::text = xmlcast(42.73 AS xml BY REF)::text, + xmlcast(42.73 AS xml)::text = xmlcast(42.73 AS xml BY VALUE)::text, + xmlcast('42.73'::xml AS numeric) = xmlcast('42.73'::xml AS numeric BY REF), + xmlcast('42.73'::xml AS numeric) = xmlcast('42.73'::xml AS numeric BY VALUE); + ?column? | ?column? | ?column? | ?column? +----------+----------+----------+---------- + t | t | t | t +(1 row) + +SELECT + xmlcast('2024-08-14'::date AS xml)::text = xmlcast('2024-08-14'::date AS xml BY REF)::text, + xmlcast('2024-08-14'::date AS xml)::text = xmlcast('2024-08-14'::date AS xml BY VALUE)::text, + xmlcast('2024-08-14'::xml AS date) = xmlcast('2024-08-14'::xml AS date BY REF), + xmlcast('2024-08-14'::xml AS date) = xmlcast('2024-08-14'::xml AS date BY VALUE); + ?column? | ?column? | ?column? | ?column? +----------+----------+----------+---------- + t | t | t | t +(1 row) + +SELECT + xmlcast('12:30:45'::time without time zone AS xml)::text = xmlcast('12:30:45'::time without time zone AS xml BY REF)::text, + xmlcast('12:30:45'::time without time zone AS xml)::text = xmlcast('12:30:45'::time without time zone AS xml BY VALUE)::text, + xmlcast('12:30:45'::xml AS time without time zone) = xmlcast('12:30:45'::xml AS time without time zone BY REF), + xmlcast('12:30:45'::xml AS time without time zone) = xmlcast('12:30:45'::xml AS time without time zone BY VALUE); + ?column? | ?column? | ?column? | ?column? +----------+----------+----------+---------- + t | t | t | t +(1 row) + +SELECT + xmlcast('09:30:10+06:00'::time with time zone AS xml)::text = xmlcast('09:30:10+06:00'::time with time zone AS xml BY REF)::text, + xmlcast('09:30:10+06:00'::time with time zone AS xml)::text = xmlcast('09:30:10+06:00'::time with time zone AS xml BY VALUE)::text, + xmlcast('09:30:10+06:00'::xml AS time with time zone) = xmlcast('09:30:10+06:00'::xml AS time with time zone BY REF), + xmlcast('09:30:10+06:00'::xml AS time with time zone) = xmlcast('09:30:10+06:00'::xml AS time with time zone BY VALUE); + ?column? | ?column? | ?column? | ?column? +----------+----------+----------+---------- + t | t | t | t +(1 row) + +SELECT + xmlcast('2002-05-30 09:30:10+06:00'::timestamp with time zone AS xml)::text = xmlcast('2002-05-30 09:30:10+06:00'::timestamp with time zone AS xml BY REF)::text, + xmlcast('2002-05-30 09:30:10+06:00'::timestamp with time zone AS xml)::text = xmlcast('2002-05-30 09:30:10+06:00'::timestamp with time zone AS xml BY VALUE)::text, + xmlcast('2002-05-30T09:30:10+06:00'::xml AS timestamp with time zone) = xmlcast('2002-05-30T09:30:10+06:00'::xml AS timestamp with time zone BY REF), + xmlcast('2002-05-30T09:30:10+06:00'::xml AS timestamp with time zone) = xmlcast('2002-05-30T09:30:10+06:00'::xml AS timestamp with time zone BY VALUE); + ?column? | ?column? | ?column? | ?column? +----------+----------+----------+---------- + t | t | t | t +(1 row) + +SELECT + xmlcast('1 year 2 mons'::interval AS xml)::text = xmlcast('1 year 2 mons'::interval AS xml BY REF)::text, + xmlcast('1 year 2 mons'::interval AS xml)::text = xmlcast('1 year 2 mons'::interval AS xml BY VALUE)::text, + xmlcast('1 year 2 mons'::xml AS interval) = xmlcast('1 year 2 mons'::xml AS interval BY REF), + xmlcast('1 year 2 mons'::xml AS interval) = xmlcast('1 year 2 mons'::xml AS interval BY VALUE); + ?column? | ?column? | ?column? | ?column? +----------+----------+----------+---------- + t | t | t | t +(1 row) + +CREATE VIEW view_xmlcast_to_xml AS +SELECT + xmlcast(NULL AS xml) AS c1, + xmlcast('foo' AS xml) AS c2, + xmlcast(''::text AS xml) AS c3, + xmlcast(NULL::text AS xml) AS c4, + xmlcast(''::xml AS text) AS c5, + xmlcast(NULL::xml AS text) c6, + xmlcast('foo & <"bar">'::text AS xml) AS c7, + xmlcast('foo & <"bar">'::varchar AS xml) AS c8, + xmlcast('foo & <"bar">'::name AS xml) AS c9, + xmlcast(xmltext(E'foo & <"bar">\r') AS text) AS c10, + xmlcast(xmlcast(E'foo & <"bar">\r' AS xml) AS text) AS c11, + xmlcast(to_date('29/05/2024','dd/mm/yyyy') AS xml) AS c12, + xmlcast('2024-05-29 12:04:10.703585+02'::timestamp with time zone at time zone 'Europe/Berlin' AS xml) AS c13, + xmlcast('2024-05-29 12:04:10.703585+02'::timestamp without time zone AS xml) AS c14, + xmlcast('1 year 2 months 3 days 4 hours 5 minutes 6 seconds'::interval AS xml) AS c15, + xmlcast(427353542 AS xml) AS c16, + xmlcast(4273535420162021 AS xml) AS c17, + xmlcast(42.007312345678910 AS xml) AS c18, + xmlcast(42.007312345678910::double precision AS xml) AS c19, + xmlcast(true AS xml) AS c20, + xmlcast(false AS xml) AS c21, + xmlcast(42 = 73 AS xml) AS c22, + xmlcast(42 <> 73 AS xml) AS c23, + xmlcast('11:11:11.5'::time AS xml) AS c24, + xmlcast('11:11:11.5'::time with time zone at time zone 'Europe/Berlin' AS xml) AS c25; +\sv view_xmlcast_to_xml +CREATE OR REPLACE VIEW public.view_xmlcast_to_xml AS + SELECT XMLCAST(NULL::text AS xml)::xml AS c1, + XMLCAST('foo'::text AS xml)::xml AS c2, + XMLCAST(''::text AS xml)::xml AS c3, + XMLCAST(NULL::text AS xml)::xml AS c4, + XMLCAST(''::xml AS text)::text AS c5, + XMLCAST(NULL::xml AS text)::text AS c6, + XMLCAST('foo & <"bar">'::text AS xml)::xml AS c7, + XMLCAST('foo & <"bar">'::character varying AS xml)::xml AS c8, + XMLCAST('foo & <"bar">'::name::text AS xml)::xml AS c9, + XMLCAST(xmltext('foo & <"bar"> +'::text) AS text)::text AS c10, + XMLCAST(XMLCAST('foo & <"bar"> +'::text AS xml)::xml AS text)::text AS c11, + XMLCAST(to_date('29/05/2024'::text, 'dd/mm/yyyy'::text) AS xml)::xml AS c12, + XMLCAST(('Wed May 29 03:04:10.703585 2024 PDT'::timestamp with time zone AT TIME ZONE 'Europe/Berlin'::text) AS xml)::xml AS c13, + XMLCAST('Wed May 29 12:04:10.703585 2024'::timestamp without time zone AS xml)::xml AS c14, + XMLCAST('@ 1 year 2 mons 3 days 4 hours 5 mins 6 secs'::interval AS xml)::xml AS c15, + XMLCAST(427353542::text AS xml)::xml AS c16, + XMLCAST('4273535420162021'::bigint::text AS xml)::xml AS c17, + XMLCAST(42.007312345678910 AS xml)::xml AS c18, + XMLCAST(42.007312345678910::double precision AS xml)::xml AS c19, + XMLCAST(true AS xml)::xml AS c20, + XMLCAST(false AS xml)::xml AS c21, + XMLCAST(42 = 73 AS xml)::xml AS c22, + XMLCAST(42 <> 73 AS xml)::xml AS c23, + XMLCAST('11:11:11.5'::time without time zone::text AS xml)::xml AS c24, + XMLCAST(('11:11:11.5-08'::time with time zone AT TIME ZONE 'Europe/Berlin'::text)::text AS xml)::xml AS c25 +SELECT * FROM view_xmlcast_to_xml; + c1 | c2 | c3 | c4 | c5 | c6 | c7 | c8 | c9 | c10 | c11 | c12 | c13 | c14 | c15 | c16 | c17 | c18 | c19 | c20 | c21 | c22 | c23 | c24 | c25 +----+-----+----+----+----+----+-----------------------------------+-----------------------------------+-----------------------------------+-----------------+-----------------+------------+----------------------------+----------------------------+----------------+-----------+------------------+--------------------+-------------------+------+-------+-------+------+------------+--------------- + | foo | | | | | foo & <"bar"> | foo & <"bar"> | foo & <"bar"> | foo & <"bar">\r | foo & <"bar">\r | 2024-05-29 | 2024-05-29T12:04:10.703585 | 2024-05-29T12:04:10.703585 | P1Y2M3DT4H5M6S | 427353542 | 4273535420162021 | 42.007312345678910 | 42.00731234567891 | true | false | false | true | 11:11:11.5 | 20:11:11.5+01 +(1 row) + +CREATE VIEW view_xmlcast_from_xml AS +SELECT + xmlcast('P1Y2M3DT4H5M6S'::xml AS interval) AS c1, + xmlcast('1 year 2 mons 3 days 4 hours 5 minutes 6 seconds'::xml AS interval) AS c2, + xmlcast('2002-09-24'::xml AS date) AS c3, + xmlcast('2002-09-24+06:00'::xml AS date) AS c4, + xmlcast('09:30:10Z'::xml AS time with time zone) at time zone 'Europe/Berlin' AS c5, + xmlcast('09:30:10-06:00'::xml AS time with time zone) at time zone 'Europe/Berlin' AS c6, + xmlcast('09:30:10+06:00'::xml AS time with time zone) at time zone 'Europe/Berlin' AS c7, + xmlcast('2002-05-30T09:30:10Z'::xml AS timestamp with time zone) at time zone 'Europe/Berlin' AS c8, + xmlcast('2002-05-30T09:30:10-06:00'::xml AS timestamp with time zone) at time zone 'Europe/Berlin' AS c9, + xmlcast('2002-05-30T09:30:10+06:00'::xml AS timestamp with time zone) at time zone 'Europe/Berlin' AS c10, + xmlcast('foo bar'::xml AS text) AS c11, + xmlcast(' foo bar '::xml AS varchar) AS c12, + xmlcast('foo & <"bar">'::xml AS text) AS c13, + xmlcast('42.7312345678910'::xml AS numeric) AS c14, + xmlcast('+42.7312345678910'::xml AS numeric) AS c15, + xmlcast('-42.7312345678910'::xml AS numeric) AS c16, + xmlcast('42'::xml AS integer) AS c17, + xmlcast('+42'::xml AS integer) AS c18, + xmlcast('-42'::xml AS integer) AS c19, + xmlcast('4273535420162021'::xml AS bigint) AS c20, + xmlcast('+4273535420162021'::xml AS bigint) AS c21, + xmlcast('-4273535420162021'::xml AS bigint) AS c22, + xmlcast('true'::xml AS boolean) AS c23, + xmlcast('false'::xml AS boolean) AS c24, + xmlcast(''::xml AS character varying) AS c25, + xmlcast(NULL::xml AS character varying) AS c26; +\sv view_xmlcast_from_xml +CREATE OR REPLACE VIEW public.view_xmlcast_from_xml AS + SELECT XMLCAST('P1Y2M3DT4H5M6S'::xml AS interval)::interval AS c1, + XMLCAST('1 year 2 mons 3 days 4 hours 5 minutes 6 seconds'::xml AS interval)::interval AS c2, + XMLCAST('2002-09-24'::xml AS date)::date AS c3, + XMLCAST('2002-09-24+06:00'::xml AS date)::date AS c4, + (XMLCAST('09:30:10Z'::xml AS time with time zone)::time with time zone AT TIME ZONE 'Europe/Berlin'::text) AS c5, + (XMLCAST('09:30:10-06:00'::xml AS time with time zone)::time with time zone AT TIME ZONE 'Europe/Berlin'::text) AS c6, + (XMLCAST('09:30:10+06:00'::xml AS time with time zone)::time with time zone AT TIME ZONE 'Europe/Berlin'::text) AS c7, + (XMLCAST('2002-05-30T09:30:10Z'::xml AS timestamp with time zone)::timestamp with time zone AT TIME ZONE 'Europe/Berlin'::text) AS c8, + (XMLCAST('2002-05-30T09:30:10-06:00'::xml AS timestamp with time zone)::timestamp with time zone AT TIME ZONE 'Europe/Berlin'::text) AS c9, + (XMLCAST('2002-05-30T09:30:10+06:00'::xml AS timestamp with time zone)::timestamp with time zone AT TIME ZONE 'Europe/Berlin'::text) AS c10, + XMLCAST('foo bar'::xml AS text)::text AS c11, + XMLCAST(' foo bar '::xml AS character varying)::character varying AS c12, + XMLCAST('foo & <"bar">'::xml AS text)::text AS c13, + XMLCAST('42.7312345678910'::xml AS numeric)::numeric AS c14, + XMLCAST('+42.7312345678910'::xml AS numeric)::numeric AS c15, + XMLCAST('-42.7312345678910'::xml AS numeric)::numeric AS c16, + XMLCAST('42'::xml AS integer)::integer AS c17, + XMLCAST('+42'::xml AS integer)::integer AS c18, + XMLCAST('-42'::xml AS integer)::integer AS c19, + XMLCAST('4273535420162021'::xml AS bigint)::bigint AS c20, + XMLCAST('+4273535420162021'::xml AS bigint)::bigint AS c21, + XMLCAST('-4273535420162021'::xml AS bigint)::bigint AS c22, + XMLCAST('true'::xml AS boolean)::boolean AS c23, + XMLCAST('false'::xml AS boolean)::boolean AS c24, + XMLCAST(''::xml AS character varying)::character varying AS c25, + XMLCAST(NULL::xml AS character varying)::character varying AS c26 +SELECT * FROM view_xmlcast_from_xml; + c1 | c2 | c3 | c4 | c5 | c6 | c7 | c8 | c9 | c10 | c11 | c12 | c13 | c14 | c15 | c16 | c17 | c18 | c19 | c20 | c21 | c22 | c23 | c24 | c25 | c26 +----------------------------------------------+----------------------------------------------+------------+------------+-------------+-------------+-------------+--------------------------+--------------------------+--------------------------+---------+---------------------+---------------+------------------+------------------+-------------------+-----+-----+-----+------------------+------------------+-------------------+-----+-----+-----+----- + @ 1 year 2 mons 3 days 4 hours 5 mins 6 secs | @ 1 year 2 mons 3 days 4 hours 5 mins 6 secs | 09-24-2002 | 09-24-2002 | 10:30:10+01 | 16:30:10+01 | 04:30:10+01 | Thu May 30 11:30:10 2002 | Thu May 30 17:30:10 2002 | Thu May 30 05:30:10 2002 | foo bar | foo bar | foo & <"bar"> | 42.7312345678910 | 42.7312345678910 | -42.7312345678910 | 42 | 42 | -42 | 4273535420162021 | 4273535420162021 | -4273535420162021 | t | f | | +(1 row) + diff --git a/src/test/regress/sql/xml.sql b/src/test/regress/sql/xml.sql index 0ea4f50883..648751d212 100644 --- a/src/test/regress/sql/xml.sql +++ b/src/test/regress/sql/xml.sql @@ -679,3 +679,293 @@ SELECT xmltext(' '); SELECT xmltext('foo `$_-+?=*^%!|/\()[]{}'); SELECT xmltext('foo & <"bar">'); SELECT xmltext('x'|| '<P>73</P>'::xml || .42 || true || 'j'::char); + +-- for xmlcast() tests +INSERT INTO xmltest + VALUES (42, +'<?xml version="1.0" encoding="utf-8"?> + <xmlcast> + <period1>P1Y2M3DT4H5M6S</period1> + <period2>1 year 2 mons 3 days 4 hours 5 minutes 6 seconds</period2> + <date1>2002-09-24</date1> + <date2>2002-09-24+06:00</date2> + <time>09:30:10.5</time> + <time_tz1>09:30:10Z</time_tz1> + <time_tz2>09:30:10-06:00</time_tz2> + <time_tz3>09:30:10+06:00</time_tz3> + <timestamp1>2002-05-30T09:00:00</timestamp1> + <timestamp2>2002-05-30T09:30:10.5</timestamp2> + <timestamp_tz1>2002-05-30T09:30:10Z</timestamp_tz1> + <timestamp_tz2>2002-05-30T09:30:10-06:00</timestamp_tz2> + <timestamp_tz3>2002-05-30T09:30:10+06:00</timestamp_tz3> + <text1>foo bar</text1> + <text2> foo bar </text2> + <text3>foo & <"bar"></text3> + <decimal1>42.7312345678910</decimal1> + <decimal2>+42.7312345678910</decimal2> + <decimal3>-42.7312345678910</decimal3> + <decimal4>INF</decimal4> + <decimal5>-INF</decimal5> + <integer1>42</integer1> + <integer2>+42</integer2> + <integer3>-42</integer3> + <long1>4273535420162021</long1> + <long2>+4273535420162021</long2> + <long3>-4273535420162021</long3> + <bool1 att="true">42</bool1> + <bool2 att="false">73</bool2> + <empty></empty> + </xmlcast>'::xml +); + +-- This prevents the xmlcast regression tests from failing if the system's timezone has been changed. +SET timezone TO 'America/Los_Angeles'; + +-- xmlcast exceptions +\set VERBOSITY terse +SELECT xmlcast((xpath('//text1/text()', data))[1] AS text[]) FROM xmltest WHERE id = 42; +SELECT xmlcast((xpath('//text1/integer1()', data))[1] AS int[]) FROM xmltest WHERE id = 42; +SELECT xmlcast(NULL AS text); +SELECT xmlcast('foo'::text AS varchar); +SELECT xmlcast(42 AS text); +SELECT xmlcast(array['foo','bar'] AS xml); +\set VERBOSITY default + +-- xmlcast tests for "XML to non-XML" expressions +SELECT + xmlcast((xpath('//date1/text()', data))[1] AS date), pg_typeof(xmlcast((xpath('//date1/text()', data))[1] AS date)), + xmlcast((xpath('//date2/text()', data))[1] AS date), pg_typeof(xmlcast((xpath('//date2/text()', data))[1] AS date)) +FROM xmltest WHERE id = 42; + +SELECT + xmlcast((xpath('//period1/text()', data))[1] AS interval), pg_typeof(xmlcast((xpath('//period1/text()', data))[1] AS interval)), + xmlcast((xpath('//period2/text()', data))[1] AS interval), pg_typeof(xmlcast((xpath('//period2/text()', data))[1] AS interval)) +FROM xmltest WHERE id = 42; + +SELECT + xmlcast((xpath('//time/text()', data))[1] AS time), pg_typeof(xmlcast((xpath('//time/text()', data))[1] AS time)), + xmlcast((xpath('//time_tz1/text()', data))[1] AS time with time zone), pg_typeof(xmlcast((xpath('//time_tz1/text()', data))[1] AS time with time zone)), + xmlcast((xpath('//time_tz2/text()', data))[1] AS time with time zone), pg_typeof(xmlcast((xpath('//time_tz2/text()', data))[1] AS time with time zone)), + xmlcast((xpath('//time_tz3/text()', data))[1] AS time with time zone), pg_typeof(xmlcast((xpath('//time_tz3/text()', data))[1] AS time with time zone)) +FROM xmltest WHERE id = 42; + +SELECT + xmlcast((xpath('//text1/text()', data))[1] AS text), pg_typeof(xmlcast((xpath('//text1/text()', data))[1] AS text)), + xmlcast((xpath('//text2/text()', data))[1] AS text), pg_typeof(xmlcast((xpath('//text2/text()', data))[1] AS text)), + xmlcast((xpath('//text3/text()', data))[1] AS text), pg_typeof(xmlcast((xpath('//text3/text()', data))[1] AS text)) +FROM xmltest WHERE id = 42; + +SELECT + xmlcast((xpath('//text1/text()', data))[1] AS varchar), pg_typeof(xmlcast((xpath('//text1/text()', data))[1] AS varchar)), + xmlcast((xpath('//text2/text()', data))[1] AS varchar), pg_typeof(xmlcast((xpath('//text2/text()', data))[1] AS varchar)), + xmlcast((xpath('//text3/text()', data))[1] AS varchar), pg_typeof(xmlcast((xpath('//text3/text()', data))[1] AS varchar)) +FROM xmltest WHERE id = 42; + +SELECT + xmlcast((xpath('//text1/text()', data))[1] AS name), pg_typeof(xmlcast((xpath('//text1/text()', data))[1] AS name)), + xmlcast((xpath('//text2/text()', data))[1] AS name), pg_typeof(xmlcast((xpath('//text2/text()', data))[1] AS name)), + xmlcast((xpath('//text3/text()', data))[1] AS name), pg_typeof(xmlcast((xpath('//text3/text()', data))[1] AS name)) +FROM xmltest WHERE id = 42; + +SELECT + xmlcast((xpath('//text1/text()', data))[1] AS bpchar), pg_typeof(xmlcast((xpath('//text1/text()', data))[1] AS bpchar)), + xmlcast((xpath('//text2/text()', data))[1] AS bpchar), pg_typeof(xmlcast((xpath('//text2/text()', data))[1] AS bpchar)), + xmlcast((xpath('//text3/text()', data))[1] AS bpchar), pg_typeof(xmlcast((xpath('//text3/text()', data))[1] AS bpchar)) +FROM xmltest WHERE id = 42; + +SELECT + xmlcast((xpath('//decimal1/text()', data))[1] AS numeric), pg_typeof(xmlcast((xpath('//decimal1/text()', data))[1] AS numeric)), + xmlcast((xpath('//decimal2/text()', data))[1] AS numeric), pg_typeof(xmlcast((xpath('//decimal2/text()', data))[1] AS numeric)), + xmlcast((xpath('//decimal3/text()', data))[1] AS numeric), pg_typeof(xmlcast((xpath('//decimal3/text()', data))[1] AS numeric)), + xmlcast((xpath('//decimal4/text()', data))[1] AS numeric), pg_typeof(xmlcast((xpath('//decimal4/text()', data))[1] AS numeric)), + xmlcast((xpath('//decimal5/text()', data))[1] AS numeric), pg_typeof(xmlcast((xpath('//decimal5/text()', data))[1] AS numeric)) +FROM xmltest WHERE id = 42; + +SELECT + xmlcast((xpath('//decimal1/text()', data))[1] AS double precision), pg_typeof(xmlcast((xpath('//decimal1/text()', data))[1] AS double precision)), + xmlcast((xpath('//decimal2/text()', data))[1] AS double precision), pg_typeof(xmlcast((xpath('//decimal2/text()', data))[1] AS double precision)), + xmlcast((xpath('//decimal3/text()', data))[1] AS double precision), pg_typeof(xmlcast((xpath('//decimal3/text()', data))[1] AS double precision)), + xmlcast((xpath('//decimal4/text()', data))[1] AS double precision), pg_typeof(xmlcast((xpath('//decimal4/text()', data))[1] AS double precision)), + xmlcast((xpath('//decimal5/text()', data))[1] AS double precision), pg_typeof(xmlcast((xpath('//decimal5/text()', data))[1] AS double precision)) +FROM xmltest WHERE id = 42; + +SELECT + xmlcast((xpath('//integer1/text()', data))[1] AS int), pg_typeof(xmlcast((xpath('//integer1/text()', data))[1] AS int)), + xmlcast((xpath('//integer2/text()', data))[1] AS int), pg_typeof(xmlcast((xpath('//integer2/text()', data))[1] AS int)), + xmlcast((xpath('//integer3/text()', data))[1] AS int), pg_typeof(xmlcast((xpath('//integer3/text()', data))[1] AS int)) +FROM xmltest WHERE id = 42; + +SELECT + xmlcast((xpath('//long1/text()', data))[1] AS bigint), pg_typeof(xmlcast((xpath('//long1/text()', data))[1] AS bigint)), + xmlcast((xpath('//long2/text()', data))[1] AS bigint), pg_typeof(xmlcast((xpath('//long2/text()', data))[1] AS bigint)), + xmlcast((xpath('//long3/text()', data))[1] AS bigint), pg_typeof(xmlcast((xpath('//long3/text()', data))[1] AS bigint)) +FROM xmltest WHERE id = 42; + +SELECT + xmlcast((xpath('//bool1/@att', data))[1] AS boolean), pg_typeof(xmlcast((xpath('//bool1/@att', data))[1] AS boolean)), + xmlcast((xpath('//bool2/@att', data))[1] AS boolean), pg_typeof(xmlcast((xpath('//bool1/@att', data))[1] AS boolean)) +FROM xmltest WHERE id = 42; + +SELECT xmlcast((xpath('//empty/text()', data))[1] AS text), pg_typeof(xmlcast((xpath('//empty/text()', data))[1] AS text)) +FROM xmltest WHERE id = 42; + +-- xmlcast tests for "XML to XML" expressions +SELECT + xmlcast((xpath('//text1/text()', data))[1] AS xml), pg_typeof(xmlcast((xpath('//text1/text()', data))[1] AS xml)), + xmlcast((xpath('//text2/text()', data))[1] AS xml), pg_typeof(xmlcast((xpath('//text2/text()', data))[1] AS xml)), + xmlcast((xpath('//text3/text()', data))[1] AS xml), pg_typeof(xmlcast((xpath('//text3/text()', data))[1] AS xml)) +FROM xmltest WHERE id = 42; + +-- xmlcast tests for "non-XML to XML" expressions +SELECT j, pg_typeof(j) FROM xmlcast(NULL AS xml) t(j); +SELECT j, pg_typeof(j) FROM xmlcast('foo' AS xml) t(j); +SELECT j, pg_typeof(j) FROM xmlcast(''::text AS xml) t(j); +SELECT j, pg_typeof(j) FROM xmlcast(NULL::text AS xml) t(j); +SELECT j, pg_typeof(j) FROM xmlcast(''::xml AS text) t(j); +SELECT j, pg_typeof(j) FROM xmlcast(NULL::xml AS text) t(j); +SELECT j, pg_typeof(j) FROM xmlcast('foo & <"bar">'::text AS xml) t(j); +SELECT j, pg_typeof(j) FROM xmlcast('foo & <"bar">'::varchar AS xml) t(j); +SELECT j, pg_typeof(j) FROM xmlcast('foo & <"bar">'::name AS xml) t(j); +SELECT j, pg_typeof(j) FROM xmlcast(xmltext(E'foo & <"bar">\r') AS text) t(j); +SELECT j, pg_typeof(j) FROM xmlcast(xmlcast(E'foo & <"bar">\r' AS xml) AS text) t(j); +SELECT j, pg_typeof(j) FROM xmlcast(to_date('29/05/2024','dd/mm/yyyy') AS xml) t(j); +SELECT j, pg_typeof(j) FROM xmlcast('2024-05-29 12:04:10.703585+02'::timestamp with time zone at time zone 'Europe/Berlin' AS xml) t(j); +SELECT j, pg_typeof(j) FROM xmlcast('2024-05-29 12:04:10.703585+02'::timestamp without time zone AS xml) t(j); +SELECT j, pg_typeof(j) FROM xmlcast('1 year 2 months 3 days 4 hours 5 minutes 6 seconds'::interval AS xml) t(j); +SELECT j, pg_typeof(j) FROM xmlcast(427353542 AS xml) t(j); +SELECT j, pg_typeof(j) FROM xmlcast(4273535420162021 AS xml) t(j); +SELECT j, pg_typeof(j) FROM xmlcast(42.007312345678910 AS xml) t(j); +SELECT j, pg_typeof(j) FROM xmlcast(42.007312345678910::double precision AS xml) t(j); +SELECT j, pg_typeof(j) FROM xmlcast('infinity'::double precision AS xml) t(j); +SELECT j, pg_typeof(j) FROM xmlcast('-infinity'::double precision AS xml) t(j); +SELECT j, pg_typeof(j) FROM xmlcast('nan'::double precision AS xml) t(j); +SELECT j, pg_typeof(j) FROM xmlcast('infinity'::numeric AS xml) t(j); +SELECT j, pg_typeof(j) FROM xmlcast('-infinity'::numeric AS xml) t(j); +SELECT j, pg_typeof(j) FROM xmlcast('nan'::numeric AS xml) t(j); +SELECT j, pg_typeof(j) FROM xmlcast(true AS xml) t(j); +SELECT j, pg_typeof(j) FROM xmlcast(false AS xml) t(j); +SELECT j, pg_typeof(j) FROM xmlcast(42 = 73 AS xml) t(j); +SELECT j, pg_typeof(j) FROM xmlcast(42 <> 73 AS xml) t(j); +SELECT j, pg_typeof(j) FROM xmlcast('11:11:11.5'::time AS xml) t(j); +SELECT j, pg_typeof(j) FROM xmlcast('11:11:11.5'::time with time zone at time zone 'Europe/Berlin' AS xml) t(j); + +-- Convert an XML string to bytea and back to xml +SELECT xmlcast(convert_from(xmlcast('<"foo&bar">'::xml AS bytea),'UTF8')::xml AS text); + +SET xmlbinary TO hex; +SELECT xmlcast(E'\\xDEADBEEF'::bytea AS xml); +SET xmlbinary TO base64; +SELECT xmlcast(E'\\xDEADBEEF'::bytea AS xml); + +-- The BY REF and BY VALUE clauses are accepted but ignored. +-- This checks if the results are indeed the same as without the clauses. +SELECT + xmlcast('foo' AS xml)::text = xmlcast('foo' AS xml BY REF)::text, + xmlcast('foo' AS xml)::text = xmlcast('foo' AS xml BY VALUE)::text, + xmlcast('foo'::xml AS text) = xmlcast('foo'::xml AS text BY REF), + xmlcast('foo'::xml AS text) = xmlcast('foo'::xml AS text BY VALUE); + +SELECT + xmlcast(42 AS xml)::text = xmlcast(42 AS xml BY REF)::text, + xmlcast(42 AS xml)::text = xmlcast(42 AS xml BY VALUE)::text, + xmlcast('42'::xml AS int) = xmlcast('42'::xml AS int BY REF), + xmlcast('42'::xml AS int) = xmlcast('42'::xml AS int BY VALUE); + +SELECT + xmlcast(42.73 AS xml)::text = xmlcast(42.73 AS xml BY REF)::text, + xmlcast(42.73 AS xml)::text = xmlcast(42.73 AS xml BY VALUE)::text, + xmlcast('42.73'::xml AS numeric) = xmlcast('42.73'::xml AS numeric BY REF), + xmlcast('42.73'::xml AS numeric) = xmlcast('42.73'::xml AS numeric BY VALUE); + +SELECT + xmlcast('2024-08-14'::date AS xml)::text = xmlcast('2024-08-14'::date AS xml BY REF)::text, + xmlcast('2024-08-14'::date AS xml)::text = xmlcast('2024-08-14'::date AS xml BY VALUE)::text, + xmlcast('2024-08-14'::xml AS date) = xmlcast('2024-08-14'::xml AS date BY REF), + xmlcast('2024-08-14'::xml AS date) = xmlcast('2024-08-14'::xml AS date BY VALUE); + +SELECT + xmlcast('12:30:45'::time without time zone AS xml)::text = xmlcast('12:30:45'::time without time zone AS xml BY REF)::text, + xmlcast('12:30:45'::time without time zone AS xml)::text = xmlcast('12:30:45'::time without time zone AS xml BY VALUE)::text, + xmlcast('12:30:45'::xml AS time without time zone) = xmlcast('12:30:45'::xml AS time without time zone BY REF), + xmlcast('12:30:45'::xml AS time without time zone) = xmlcast('12:30:45'::xml AS time without time zone BY VALUE); + +SELECT + xmlcast('09:30:10+06:00'::time with time zone AS xml)::text = xmlcast('09:30:10+06:00'::time with time zone AS xml BY REF)::text, + xmlcast('09:30:10+06:00'::time with time zone AS xml)::text = xmlcast('09:30:10+06:00'::time with time zone AS xml BY VALUE)::text, + xmlcast('09:30:10+06:00'::xml AS time with time zone) = xmlcast('09:30:10+06:00'::xml AS time with time zone BY REF), + xmlcast('09:30:10+06:00'::xml AS time with time zone) = xmlcast('09:30:10+06:00'::xml AS time with time zone BY VALUE); + +SELECT + xmlcast('2002-05-30 09:30:10+06:00'::timestamp with time zone AS xml)::text = xmlcast('2002-05-30 09:30:10+06:00'::timestamp with time zone AS xml BY REF)::text, + xmlcast('2002-05-30 09:30:10+06:00'::timestamp with time zone AS xml)::text = xmlcast('2002-05-30 09:30:10+06:00'::timestamp with time zone AS xml BY VALUE)::text, + xmlcast('2002-05-30T09:30:10+06:00'::xml AS timestamp with time zone) = xmlcast('2002-05-30T09:30:10+06:00'::xml AS timestamp with time zone BY REF), + xmlcast('2002-05-30T09:30:10+06:00'::xml AS timestamp with time zone) = xmlcast('2002-05-30T09:30:10+06:00'::xml AS timestamp with time zone BY VALUE); + +SELECT + xmlcast('1 year 2 mons'::interval AS xml)::text = xmlcast('1 year 2 mons'::interval AS xml BY REF)::text, + xmlcast('1 year 2 mons'::interval AS xml)::text = xmlcast('1 year 2 mons'::interval AS xml BY VALUE)::text, + xmlcast('1 year 2 mons'::xml AS interval) = xmlcast('1 year 2 mons'::xml AS interval BY REF), + xmlcast('1 year 2 mons'::xml AS interval) = xmlcast('1 year 2 mons'::xml AS interval BY VALUE); + +CREATE VIEW view_xmlcast_to_xml AS +SELECT + xmlcast(NULL AS xml) AS c1, + xmlcast('foo' AS xml) AS c2, + xmlcast(''::text AS xml) AS c3, + xmlcast(NULL::text AS xml) AS c4, + xmlcast(''::xml AS text) AS c5, + xmlcast(NULL::xml AS text) c6, + xmlcast('foo & <"bar">'::text AS xml) AS c7, + xmlcast('foo & <"bar">'::varchar AS xml) AS c8, + xmlcast('foo & <"bar">'::name AS xml) AS c9, + xmlcast(xmltext(E'foo & <"bar">\r') AS text) AS c10, + xmlcast(xmlcast(E'foo & <"bar">\r' AS xml) AS text) AS c11, + xmlcast(to_date('29/05/2024','dd/mm/yyyy') AS xml) AS c12, + xmlcast('2024-05-29 12:04:10.703585+02'::timestamp with time zone at time zone 'Europe/Berlin' AS xml) AS c13, + xmlcast('2024-05-29 12:04:10.703585+02'::timestamp without time zone AS xml) AS c14, + xmlcast('1 year 2 months 3 days 4 hours 5 minutes 6 seconds'::interval AS xml) AS c15, + xmlcast(427353542 AS xml) AS c16, + xmlcast(4273535420162021 AS xml) AS c17, + xmlcast(42.007312345678910 AS xml) AS c18, + xmlcast(42.007312345678910::double precision AS xml) AS c19, + xmlcast(true AS xml) AS c20, + xmlcast(false AS xml) AS c21, + xmlcast(42 = 73 AS xml) AS c22, + xmlcast(42 <> 73 AS xml) AS c23, + xmlcast('11:11:11.5'::time AS xml) AS c24, + xmlcast('11:11:11.5'::time with time zone at time zone 'Europe/Berlin' AS xml) AS c25; + +\sv view_xmlcast_to_xml +SELECT * FROM view_xmlcast_to_xml; + +CREATE VIEW view_xmlcast_from_xml AS +SELECT + xmlcast('P1Y2M3DT4H5M6S'::xml AS interval) AS c1, + xmlcast('1 year 2 mons 3 days 4 hours 5 minutes 6 seconds'::xml AS interval) AS c2, + xmlcast('2002-09-24'::xml AS date) AS c3, + xmlcast('2002-09-24+06:00'::xml AS date) AS c4, + xmlcast('09:30:10Z'::xml AS time with time zone) at time zone 'Europe/Berlin' AS c5, + xmlcast('09:30:10-06:00'::xml AS time with time zone) at time zone 'Europe/Berlin' AS c6, + xmlcast('09:30:10+06:00'::xml AS time with time zone) at time zone 'Europe/Berlin' AS c7, + xmlcast('2002-05-30T09:30:10Z'::xml AS timestamp with time zone) at time zone 'Europe/Berlin' AS c8, + xmlcast('2002-05-30T09:30:10-06:00'::xml AS timestamp with time zone) at time zone 'Europe/Berlin' AS c9, + xmlcast('2002-05-30T09:30:10+06:00'::xml AS timestamp with time zone) at time zone 'Europe/Berlin' AS c10, + xmlcast('foo bar'::xml AS text) AS c11, + xmlcast(' foo bar '::xml AS varchar) AS c12, + xmlcast('foo & <"bar">'::xml AS text) AS c13, + xmlcast('42.7312345678910'::xml AS numeric) AS c14, + xmlcast('+42.7312345678910'::xml AS numeric) AS c15, + xmlcast('-42.7312345678910'::xml AS numeric) AS c16, + xmlcast('42'::xml AS integer) AS c17, + xmlcast('+42'::xml AS integer) AS c18, + xmlcast('-42'::xml AS integer) AS c19, + xmlcast('4273535420162021'::xml AS bigint) AS c20, + xmlcast('+4273535420162021'::xml AS bigint) AS c21, + xmlcast('-4273535420162021'::xml AS bigint) AS c22, + xmlcast('true'::xml AS boolean) AS c23, + xmlcast('false'::xml AS boolean) AS c24, + xmlcast(''::xml AS character varying) AS c25, + xmlcast(NULL::xml AS character varying) AS c26; + +\sv view_xmlcast_from_xml +SELECT * FROM view_xmlcast_from_xml; \ No newline at end of file diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 14dec2d49c..1258b7baef 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -3426,6 +3426,7 @@ XidStatus XmlExpr XmlExprOp XmlOptionType +XmlCast XmlSerialize XmlTableBuilderData YYLTYPE -- 2.43.0 ^ permalink raw reply [nested|flat] 102+ messages in thread
end of thread, other threads:[~2026-01-13 09:31 UTC | newest] Thread overview: 102+ messages (download: mbox mbox.gz follow: Atom feed) -- links below jump to the message on this page -- 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 3/8] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 3/8] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 3/8] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 3/8] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 3/8] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 3/8] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH v7 1/9] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 3/8] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 3/8] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH v9 1/9] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 3/8] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 3/8] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 3/8] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 3/8] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 3/8] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 3/8] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 3/8] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 3/8] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 3/8] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 3/8] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 3/8] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 3/8] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 3/8] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 3/8] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/12] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH v8 1/9] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/12] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 3/8] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 3/8] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2021-02-27 04:03 [PATCH 01/10] Allow alternate compression methods for wal_compression Andrey Borodin <[email protected]> 2026-01-13 09:31 Re: [PoC] XMLCast (SQL/XML X025) Jim Jones <[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