agora inbox for [email protected]  

help / color / mirror / Atom feed

Re: [HACKERS] 7.0.3 branded
265+ messages / 2 participants
[nested] [flat]

* Re: [HACKERS] 7.0.3 branded
@ 2000-11-03 20:48  Ian Lance Taylor <[email protected]>
  0 siblings, 0 replies; 265+ messages in thread

From: Ian Lance Taylor @ 2000-11-03 20:48 UTC (permalink / raw)
  To: [email protected]

   From: Bruce Momjian <[email protected]>
   Date: Fri, 3 Nov 2000 15:35:01 -0500 (EST)

   > The natural way to fix this ought to be
   >     cvs co -rREL7_0_PATCHES pgsql
   >     cvs log .
   > Unfortunately, I tried it, and cvs log, I believe erroneously, seems
   > to pick up all files in the directory, even if they have not been
   > checked out.
   > 
   > I can tell you a hideous kludge to avoid this, but I can't claim that
   > it is the way to operate.  Check out the branch using the -r option as
   > above.  Then do this:
   >     find . -name CVS -type d -exec touch '{}/Entries.Static' \;
   > After that, in the same directory, do
   >     cvs log -rREL7_0_PATCHES
   > 
   > I'd hate to have to explain why that works.

   Does this cause any other problems, or does it just affect log?

The main effect is that a cvs update in that directory will not pick
up any newly added files.  That will catch you by surprise after a
while, so I wouldn't recommend leaving the Entries.Static files around
forever.

   > Which version of CVS are you running on the server?  When I find some
   > time I'll see about fixing cvs log.

       Concurrent Versions System (CVS) 1.10.3 (client/server)

   I couldn't imagine cvs was so broken as to do what it is doing, so I
   concluded I was doing something wrong.  Can I share this email with the
   hackers list?

CVS is a long aggregation of hacks.  Heck, the first version was a
bunch of shell scripts.  Since there is no theory underlying CVS, it's
easy to get the corner cases wrong unless you test them.  I would
guess that the author of the current cvs log implementation didn't
test this sort of thing.  (The author in question was, um, me,
although I think I might be able to blame John Gilmore for this
particular feature.)

Yes, please go ahead and share these e-mail messages if you like.

By the way, I gather you spoke with Nathan Meyers at the free database
summit.  I'm co-founder and CTO of Zembu, where he works.

Ian


-- 
  Bruce Momjian                        |  http://candle.pha.pa.us
  [email protected]               |  (610) 853-3000
  +  If your life is a hard drive,     |  830 Blythe Avenue
  +  Christ can be your backup.        |  Drexel Hill, Pennsylvania 19026

--ELM973373840-2705-0_
Content-Transfer-Encoding: 7bit
Content-Type: text/plain
Content-Disposition: inline; filename="/tmp/log"


                                                                 momjian
/doc/FAQ

 Update FAQ.

---
                                                                 momjian
/doc/FAQ_BSDI

 Update bsdi faq.

---
                                                                 momjian
/doc/FAQ_DEV

 update developers faq

---
                                                                 momjian
/src/interfaces/jdbc/postgresql/jdbc1/DatabaseMetaData.java

 Brand 7.1 release.  Also update jdbc version in release branch.

---
                                                                 momjian
/src/interfaces/jdbc/postgresql/jdbc2/DatabaseMetaData.java

 Brand 7.1 release.  Also update jdbc version in release branch.

---
                                                                  thomas
/doc/src/sgml/inherit.sgml
/doc/src/sgml/query.sgml
/doc/src/sgml/release.sgml

 Fix markup to allow doc building.

---
                                                                  thomas
/doc/src/sgml/sql.sgml

 Fix markup to allow doc building.

---
                                                                 momjian
/src/interfaces/jdbc/CHANGELOG
/src/interfaces/jdbc/Makefile
/src/interfaces/jdbc/example/basic.java
/src/interfaces/jdbc/org/postgresql/Connection.java
/src/interfaces/jdbc/org/postgresql/ResultSet.java
/src/interfaces/jdbc/org/postgresql/jdbc1/Connection.java
/src/interfaces/jdbc/org/postgresql/jdbc1/ResultSet.java
/src/interfaces/jdbc/org/postgresql/jdbc2/Connection.java
/src/interfaces/jdbc/org/postgresql/jdbc2/ResultSet.java

 Backpatch jdbc fixes into 7.0.X.

---
                                                                 momjian
/src/interfaces/jdbc/org/postgresql/jdbc2/Statement.java

 Backpatch jdbc fixes into 7.0.X.

---
                                                                     tgl
/src/backend/storage/large_object/inv_api.c

 Back-patch large-object fix.

---
                                                                 momjian
/doc/FAQ

 Update FAQ.

---
                                                                     tgl
/src/backend/commands/copy.c

 Back-patch COPY WITH OIDS leak fix.

---
                                                                     tgl
/src/backend/utils/adt/like.c
/src/backend/utils/adt/regexp.c
/src/backend/utils/adt/varchar.c

 Back-patch StrNCpy fix.

---
                                                                     tgl
/src/backend/optimizer/path/indxpath.c

 Backpatch backwards-index-scan fix.

---
                                                                   ishii
/src/backend/utils/time/tqual.c

 SELECT ... FOR UPDATE neglects duplicate key checking.
 patches submitted by Hiroshi Inoue.

---
                                                                     tgl
/src/backend/optimizer/plan/planner.c

 Back-patch primary fix for planner recursion bug.

---
                                                                 scrappy
/src/configure
/src/configure.in

 
 backpatch the --enable-syslog functionality to REL7_0 branch

---
                                                                 scrappy
/src/configure
/src/configure.in

 
 oops, in v7.x its USE_SYSLOG, not ENABLE_SYSLOG
 modify config.h.in so that it gets set by configure properly

---
                                                                 scrappy
/src/include/config.h.in

 
 oops, in v7.x its USE_SYSLOG, not ENABLE_SYSLOG
 modify config.h.in so that it gets set by configure properly

---
                                                                     tgl
/src/backend/tcop/postgres.c

 Back-patch fix to ensure we abort any open transaction at backend exit.

---
                                                                   ishii
/src/bin/psql/describe.c

 Fix psql crash. If MULTIBYTE is enabled, \l+ dumps core due to
 SQL buffer in listAllDbs is just too small.

---
                                                                     tgl
/src/include/executor/nodeMaterial.h
/src/backend/executor/execAmi.c
/src/backend/executor/nodeMaterial.c

 Back-patch fix for bogus plans involving non-mark/restorable plan
 as inner plan of a mergejoin.

---
                                                                     tgl
/src/backend/optimizer/plan/createplan.c

 Back-patch fix for bogus plans involving non-mark/restorable plan
 as inner plan of a mergejoin.

---
                                                                   ishii
/src/pl/plpgsql/src/scan.l

 Allow PL/pgSQL accept non ascii identifiers

---
                                                                     tgl
/src/backend/commands/vacuum.c

 Back-patch fix to ensure that VACUUM always calls FlushRelationBuffers.

---
                                                                   inoue
/src/backend/storage/lmgr/proc.c

 Cancel request while waiting for a lock should try to wake
 up sleeping processes.

---
                                                                     tgl
/src/bin/psql/help.c

 Back-patch fix for erroneous free() of getpwuid() result.

---
                                                                     tgl
/src/interfaces/odbc/info.c

 Back-patch fix to remove bogus use of int4out().

---
                                                                     tgl
/src/backend/optimizer/plan/subselect.c

 Back-patch fix to copy sub-Query nodes before planning them.  This
 fixes problems with subselects appearing in contexts like COALESCE or
 BETWEEN, where parser will make multiple links to same subexpression.

---
                                                                     tgl
/src/backend/utils/adt/ri_triggers.c

 Apply Jeroen van Vianen's patch for failure to check heap_openr failure
 in RI triggers.  This is fixed in another way in current sources, but
 might as well apply this patch to REL7_0 branch so that 7.0.3 need not
 suffer this crash.

---
                                                                     tgl
/src/backend/utils/adt/selfuncs.c

 Back-patch fix for erroneous selectivity of not-equals.

---
                                                                     tgl
/src/backend/utils/adt/ruleutils.c

 Back-patch fix for erroneous use of strcmp().

---
                                                                     tgl
/src/backend/storage/smgr/md.c

 Back-patch fix for 'Sorcerer's Apprentice' syndrome wherein md.c would
 create a vast quantity of zero-length files if asked to access a block
 number far beyond the actual end of a relation.

---
                                                                     tgl
/src/backend/storage/smgr/smgr.c

 Back-patch fix to include kernel errno message in all smgr elog messages.

---
                                                                     tgl
/src/pl/tcl/Makefile
/src/bin/pgtclsh/Makefile

 Back-patch fix for '.' not in PATH at build time, per SL Baur.

---
                                                                     tgl
/src/backend/storage/file/fd.c

 Back-patch fix that allows AllocateFile() to return errno=ENFILE/EMFILE
 after we are no longer able to close any more VFDs.  This is needed to
 avoid postmaster crash under out-of-file-descriptors conditions.

---
                                                                     tgl
/src/bin/pg_dump/pg_dump.c
/src/bin/pg_dump/pg_dump.h

 Back-patch fix to make pg_dump dump 'iscachable' flag for functions.

---
                                                                     tgl
/src/backend/optimizer/plan/setrefs.c

 Back-patch fix for subselect in targetlist of Append node.

---
                                                                     tgl
/src/include/optimizer/paths.h
/src/include/optimizer/planmain.h
/src/backend/optimizer/path/pathkeys.c
/src/backend/optimizer/plan/initsplan.c
/src/backend/optimizer/plan/planmain.c

 Back-patch code to deduce implied equalities from transitivity of
 mergejoin clauses, and add these equalities to the given WHERE clauses.
 This is necessary to ensure that sort keys we think are equivalent
 really are equivalent as soon as their rels have been joined.  Without
 this, 7.0 may create an incorrect mergejoin plan.

---
                                                                     tgl
/src/backend/storage/buffer/bufmgr.c

 Back-patch fix to grab read lock on a buffer while it is written out.

---
                                                                     tgl
/src/backend/catalog/heap.c

 Back-patch fix for TRUNCATE failure on relations with indexes.

---
                                                                   inoue
/src/backend/storage/buffer/bufmgr.c

 avoid database-wide restart on write error

---
                                                                     tgl
/src/backend/executor/nodeMaterial.c

 Back-patch nodeMaterial to honor chgParam by recomputing its output.

---
                                                                     tgl
/src/backend/commands/vacuum.c

 Patch VACUUM problem with moving chain of update tuples when source
 and destination of a tuple lie on the same page.

---
                                                                     tgl
/src/backend/commands/user.c

 Back-patch CommandCounterIncrement fix.

---
                                                                     tgl
/src/backend/utils/adt/formatting.c

 Back-patch fix for AM/PM boundary problem in to_char().
 Fix from Karel Zak, 10/18/00.

---
                                                                     tgl
/src/backend/utils/adt/date.c

 Fix time_larger, time_smaller, timetz_larger, timetz_smaller to meet
 nodeAgg.c's expectation that aggregate transition functions never return
 pointers to their input values.  This is fixed in a much better way in
 current sources, but in 7.0.* it's gotta be done like this.

---
                                                                     tgl
/src/backend/utils/adt/formatting.c

 Fix to_char() to avoid coredump on NULL input.  Not needed in current
 sources due to fmgr rewrite, but 7.0.3 can use the patch...

---
                                                                     tgl
/src/backend/storage/buffer/bufmgr.c

 Back-patch fix for bogus clearing of BufferDirtiedByMe.

---
                                                                   ishii
/src/backend/utils/adt/varchar.c

 Fix for inserting/copying longer multibyte strings into bpchar data
 types.

---
                                                                   wieck
/src/bin/pg_dump/Makefile.in

 New dump utility script pg_dumpaccounts.
 
 Dumps pg_shadow and pg_group (derived from pg_dumpall).
 
 Jan

---
                                                                   wieck
/src/bin/pg_dump/Makefile.in

 Revoked changes for pg_dumpaccounts
 
 Script will go into the contrib directory.
 
 Jan

---
                                                                   wieck
/contrib/pg_dumpaccounts/Makefile
/contrib/pg_dumpaccounts/README

 Added pg_dumpaccounts utility script in contrib.
 
 Derived from pg_dumpall it just dumps the pg_shadow and
 pg_group contents.
 
 Jan

---
                                                                   wieck
/contrib/pg_dumpaccounts/pg_dumpaccounts

 Added pg_dumpaccounts utility script in contrib.
 
 Derived from pg_dumpall it just dumps the pg_shadow and
 pg_group contents.
 
 Jan

---
                                                                 momjian
/HISTORY
/INSTALL
/README
/register.txt
/doc/FAQ
/doc/TODO
/doc/bug.template
/doc/src/FAQ.html
/doc/src/sgml/install.sgml
/doc/src/sgml/release.sgml
/src/include/version.h.in
/src/interfaces/jdbc/postgresql/jdbc1/DatabaseMetaData.java
/src/interfaces/jdbc/postgresql/jdbc2/DatabaseMetaData.java
/src/interfaces/libpq/libpq.rc

 Brand 7.0.3.

---
                                                                 momjian
/HISTORY
/doc/src/sgml/release.sgml

 cleanup

---

--ELM973373840-2705-0_--



^ permalink  raw  reply  [nested|flat] 265+ messages in thread

* [PATCH v6] README for 64bit xid
@ 2022-01-10 19:20  Pavel Borisov <[email protected]>
  0 siblings, 0 replies; 265+ messages in thread

From: Pavel Borisov @ 2022-01-10 19:20 UTC (permalink / raw)

Authors:
- Pavel Borisov <[email protected]>
- Maxim Orlov <[email protected]>
- Yura Sokolov <[email protected]> <[email protected]>
---
 src/backend/access/heap/README.XID64 | 128 +++++++++++++++++++++++++++
 1 file changed, 128 insertions(+)
 create mode 100644 src/backend/access/heap/README.XID64

diff --git a/src/backend/access/heap/README.XID64 b/src/backend/access/heap/README.XID64
new file mode 100644
index 00000000000..457ba9b9ef5
--- /dev/null
+++ b/src/backend/access/heap/README.XID64
@@ -0,0 +1,128 @@
+src/backend/access/heap/README.XID64
+
+64-bit Transaction ID's (XID)
+=============================
+
+A limited number (N = 2^32) of XID's required to do vacuum freeze to prevent
+wraparound every N/2 transactions. This causes performance degradation due
+to the need to exclusively lock tables while being vacuumed. In each
+wraparound cycle, SLRU buffers are also being cut.
+
+With 64-bit XID's wraparound is effectively postponed to a very distant
+future. Even in highly loaded systems that had 2^32 transactions per day
+it will take huge 2^31 days before the first enforced "vacuum to prevent
+wraparound"). Buffers cutting and routine vacuum are not enforced, and DBA
+can plan them independently at the time with the least system load and least
+critical for database performance. Also, it can be done less frequently
+(several times a year vs every several days) on systems with transaction rates
+similar to those mentioned above.
+
+On-disk tuple and page format
+-----------------------------
+
+On-disk tuple format remains unchanged. 32-bit t_xmin and t_xmax store the
+lower parts of 64-bit XMIN and XMAX values. Each heap page has additional
+64-bit pd_xid_base and pd_multi_base which are common for all tuples on a page.
+They are placed into a pd_special area - 16 bytes in the end of a heap page.
+Actual XMIN/XMAX for a tuple are calculated upon reading a tuple from a page
+as follows:
+
+XMIN = t_xmin + pd_xid_base. 					(1)
+XMAX = t_xmax + pd_xid_base/pd_multi_base.		(2)
+
+"Double XMAX" page format
+---------------------------------
+
+At first read of a heap page after pg_upgrade from 32-bit XID PostgreSQL
+version pd_special area with a size of 16 bytes should be added to a page.
+Though a page may not have space for this. Then it can be converted to a
+temporary format called "double XMAX".
+
+All tuples after pg-upgrade would necessarily have xmin = FrozenTransactionId.
+So we don't need tuple header t_xmin field and we reuse t_xmin to store higher
+32 bits of its XMAX.
+
+Double XMAX format is only for full pages that don't have 16 bytes for
+pd_special. So it neither has a�place for a single tuple. Insert and HOT update
+for double XMAX pages is impossible and not supported. We can only read or
+delete tuples from it.
+
+When we are able to prune page double XMAX it will be converted from it to
+general 64-bit XID page format with all operations on its tuples supported.
+
+In-memory tuple format
+----------------------
+
+In-memory tuple representation consists of two parts:
+- HeapTupleHeader from disk page (contains all heap tuple contents, not only
+header)
+- HeapTuple with additional in-memory fields
+
+HeapTuple for each tuple in memory stores t_xid_base/t_multi_base - a copies of
+page's pd_xid_base/pd_multi_base. With tuple's 32-bit t_xmin and t_xmax from
+HeapTupleHeader they are used to calculate actual 64-bit XMIN and XMAX:
+
+XMIN = t_xmin + t_xid_base. 					(3)
+XMAX = t_xmax + t_xid_base/t_multi_base.		(4)
+
+The downside of this is that we can not use tuple's XMIN and XMAX right away.
+We often need to re-read t_xmin and t_xmax - which could actually be pointers
+into a page in shared buffers and therefore they could be updated by any other
+backend.
+
+Update/delete with 64-bit XIDs and 32-bit t_xmin/t_xmax
+--------------------------------------------------------------
+
+When we try to delete/update a tuple, we check that XMAX for a page fits (2).
+I.e. that t_xmax will not be over MaxShortTransactionId relative to
+pd_xid_base/pd_multi_base of a its page.
+
+If the current XID doesn't fit a range
+(pd_xid_base, pd_xid_base + MaxShortTransactionId) (5):
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base/pd_multi_base on
+a page and update all t_xmin/t_xmax of the other tuples on the page to
+correspond new pd_xid_base/pd_multi_base.
+
+- If it was impossible, it will try to prune and freeze tuples on a page.
+
+- If this is unsuccessful it will throw an error. Normally this is very
+unlikely but if there is a very old living transaction with an age of around
+2^32 this can arise. Basically, this is a behavior similar to one during the
+vacuum to prevent wraparound when XID was 32-bit. Dba should take care and
+avoid very-long-living transactions with an age close to 2^32. So long-living
+transactions often they are most likely defunct.
+
+Insert with 64-bit XIDs and 32-bit t_xmin/t_xmax
+------------------------------------------------
+
+On insert we check if current XID fits a range (5). Otherwise:
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base for t_xmin will
+not be over MaxShortTransactionId.
+
+- If it is impossible, then it will try to prune and freeze tuples on a page.
+
+Known issue: if pd_xid_base could not be shifted to accommodate a tuple being
+inserted due to a very long-running transaction, we just throw an error. We
+neither try to insert a�tuple into another page nor mark the current page as
+full. So, in this (unlikely) case we will get regular insert errors on the next
+tries to insert to the page 'locked' by this very long-running transaction.
+
+Upgrade from 32-bit XID versions
+--------------------------------
+
+pg_upgrade doesn't change pages format itself. It is done lazily after.
+
+1. At first heap page read, tuples on a page are repacked to free 16 bytes
+at the end of a page, possibly freeing space from dead tuples.
+
+2A. 16 bytes of pd_special is added if there is a place for it
+
+2B. Page is converted to "Double XMAX" format if there is no place for
+pd_special
+
+3. If a page is in double XMAX format after its first read, and vacuum (or
+micro-vacuum at select query) could prune some tuples and free space for
+pd_special, prune_page will add pd_special and convert page from double XMAX
+to general 64-bit XID page format.
-- 
2.24.3 (Apple Git-128)


--cpok4wp6gsarlzvp--





^ permalink  raw  reply  [nested|flat] 265+ messages in thread

* [PATCH v6] README for 64bit xid
@ 2022-01-10 19:20  Pavel Borisov <[email protected]>
  0 siblings, 0 replies; 265+ messages in thread

From: Pavel Borisov @ 2022-01-10 19:20 UTC (permalink / raw)

Authors:
- Pavel Borisov <[email protected]>
- Maxim Orlov <[email protected]>
- Yura Sokolov <[email protected]> <[email protected]>
---
 src/backend/access/heap/README.XID64 | 128 +++++++++++++++++++++++++++
 1 file changed, 128 insertions(+)
 create mode 100644 src/backend/access/heap/README.XID64

diff --git a/src/backend/access/heap/README.XID64 b/src/backend/access/heap/README.XID64
new file mode 100644
index 00000000000..457ba9b9ef5
--- /dev/null
+++ b/src/backend/access/heap/README.XID64
@@ -0,0 +1,128 @@
+src/backend/access/heap/README.XID64
+
+64-bit Transaction ID's (XID)
+=============================
+
+A limited number (N = 2^32) of XID's required to do vacuum freeze to prevent
+wraparound every N/2 transactions. This causes performance degradation due
+to the need to exclusively lock tables while being vacuumed. In each
+wraparound cycle, SLRU buffers are also being cut.
+
+With 64-bit XID's wraparound is effectively postponed to a very distant
+future. Even in highly loaded systems that had 2^32 transactions per day
+it will take huge 2^31 days before the first enforced "vacuum to prevent
+wraparound"). Buffers cutting and routine vacuum are not enforced, and DBA
+can plan them independently at the time with the least system load and least
+critical for database performance. Also, it can be done less frequently
+(several times a year vs every several days) on systems with transaction rates
+similar to those mentioned above.
+
+On-disk tuple and page format
+-----------------------------
+
+On-disk tuple format remains unchanged. 32-bit t_xmin and t_xmax store the
+lower parts of 64-bit XMIN and XMAX values. Each heap page has additional
+64-bit pd_xid_base and pd_multi_base which are common for all tuples on a page.
+They are placed into a pd_special area - 16 bytes in the end of a heap page.
+Actual XMIN/XMAX for a tuple are calculated upon reading a tuple from a page
+as follows:
+
+XMIN = t_xmin + pd_xid_base. 					(1)
+XMAX = t_xmax + pd_xid_base/pd_multi_base.		(2)
+
+"Double XMAX" page format
+---------------------------------
+
+At first read of a heap page after pg_upgrade from 32-bit XID PostgreSQL
+version pd_special area with a size of 16 bytes should be added to a page.
+Though a page may not have space for this. Then it can be converted to a
+temporary format called "double XMAX".
+
+All tuples after pg-upgrade would necessarily have xmin = FrozenTransactionId.
+So we don't need tuple header t_xmin field and we reuse t_xmin to store higher
+32 bits of its XMAX.
+
+Double XMAX format is only for full pages that don't have 16 bytes for
+pd_special. So it neither has a�place for a single tuple. Insert and HOT update
+for double XMAX pages is impossible and not supported. We can only read or
+delete tuples from it.
+
+When we are able to prune page double XMAX it will be converted from it to
+general 64-bit XID page format with all operations on its tuples supported.
+
+In-memory tuple format
+----------------------
+
+In-memory tuple representation consists of two parts:
+- HeapTupleHeader from disk page (contains all heap tuple contents, not only
+header)
+- HeapTuple with additional in-memory fields
+
+HeapTuple for each tuple in memory stores t_xid_base/t_multi_base - a copies of
+page's pd_xid_base/pd_multi_base. With tuple's 32-bit t_xmin and t_xmax from
+HeapTupleHeader they are used to calculate actual 64-bit XMIN and XMAX:
+
+XMIN = t_xmin + t_xid_base. 					(3)
+XMAX = t_xmax + t_xid_base/t_multi_base.		(4)
+
+The downside of this is that we can not use tuple's XMIN and XMAX right away.
+We often need to re-read t_xmin and t_xmax - which could actually be pointers
+into a page in shared buffers and therefore they could be updated by any other
+backend.
+
+Update/delete with 64-bit XIDs and 32-bit t_xmin/t_xmax
+--------------------------------------------------------------
+
+When we try to delete/update a tuple, we check that XMAX for a page fits (2).
+I.e. that t_xmax will not be over MaxShortTransactionId relative to
+pd_xid_base/pd_multi_base of a its page.
+
+If the current XID doesn't fit a range
+(pd_xid_base, pd_xid_base + MaxShortTransactionId) (5):
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base/pd_multi_base on
+a page and update all t_xmin/t_xmax of the other tuples on the page to
+correspond new pd_xid_base/pd_multi_base.
+
+- If it was impossible, it will try to prune and freeze tuples on a page.
+
+- If this is unsuccessful it will throw an error. Normally this is very
+unlikely but if there is a very old living transaction with an age of around
+2^32 this can arise. Basically, this is a behavior similar to one during the
+vacuum to prevent wraparound when XID was 32-bit. Dba should take care and
+avoid very-long-living transactions with an age close to 2^32. So long-living
+transactions often they are most likely defunct.
+
+Insert with 64-bit XIDs and 32-bit t_xmin/t_xmax
+------------------------------------------------
+
+On insert we check if current XID fits a range (5). Otherwise:
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base for t_xmin will
+not be over MaxShortTransactionId.
+
+- If it is impossible, then it will try to prune and freeze tuples on a page.
+
+Known issue: if pd_xid_base could not be shifted to accommodate a tuple being
+inserted due to a very long-running transaction, we just throw an error. We
+neither try to insert a�tuple into another page nor mark the current page as
+full. So, in this (unlikely) case we will get regular insert errors on the next
+tries to insert to the page 'locked' by this very long-running transaction.
+
+Upgrade from 32-bit XID versions
+--------------------------------
+
+pg_upgrade doesn't change pages format itself. It is done lazily after.
+
+1. At first heap page read, tuples on a page are repacked to free 16 bytes
+at the end of a page, possibly freeing space from dead tuples.
+
+2A. 16 bytes of pd_special is added if there is a place for it
+
+2B. Page is converted to "Double XMAX" format if there is no place for
+pd_special
+
+3. If a page is in double XMAX format after its first read, and vacuum (or
+micro-vacuum at select query) could prune some tuples and free space for
+pd_special, prune_page will add pd_special and convert page from double XMAX
+to general 64-bit XID page format.
-- 
2.24.3 (Apple Git-128)


--cpok4wp6gsarlzvp--





^ permalink  raw  reply  [nested|flat] 265+ messages in thread

* [PATCH v6] README for 64bit xid
@ 2022-01-10 19:20  Pavel Borisov <[email protected]>
  0 siblings, 0 replies; 265+ messages in thread

From: Pavel Borisov @ 2022-01-10 19:20 UTC (permalink / raw)

Authors:
- Pavel Borisov <[email protected]>
- Maxim Orlov <[email protected]>
- Yura Sokolov <[email protected]> <[email protected]>
---
 src/backend/access/heap/README.XID64 | 128 +++++++++++++++++++++++++++
 1 file changed, 128 insertions(+)
 create mode 100644 src/backend/access/heap/README.XID64

diff --git a/src/backend/access/heap/README.XID64 b/src/backend/access/heap/README.XID64
new file mode 100644
index 00000000000..457ba9b9ef5
--- /dev/null
+++ b/src/backend/access/heap/README.XID64
@@ -0,0 +1,128 @@
+src/backend/access/heap/README.XID64
+
+64-bit Transaction ID's (XID)
+=============================
+
+A limited number (N = 2^32) of XID's required to do vacuum freeze to prevent
+wraparound every N/2 transactions. This causes performance degradation due
+to the need to exclusively lock tables while being vacuumed. In each
+wraparound cycle, SLRU buffers are also being cut.
+
+With 64-bit XID's wraparound is effectively postponed to a very distant
+future. Even in highly loaded systems that had 2^32 transactions per day
+it will take huge 2^31 days before the first enforced "vacuum to prevent
+wraparound"). Buffers cutting and routine vacuum are not enforced, and DBA
+can plan them independently at the time with the least system load and least
+critical for database performance. Also, it can be done less frequently
+(several times a year vs every several days) on systems with transaction rates
+similar to those mentioned above.
+
+On-disk tuple and page format
+-----------------------------
+
+On-disk tuple format remains unchanged. 32-bit t_xmin and t_xmax store the
+lower parts of 64-bit XMIN and XMAX values. Each heap page has additional
+64-bit pd_xid_base and pd_multi_base which are common for all tuples on a page.
+They are placed into a pd_special area - 16 bytes in the end of a heap page.
+Actual XMIN/XMAX for a tuple are calculated upon reading a tuple from a page
+as follows:
+
+XMIN = t_xmin + pd_xid_base. 					(1)
+XMAX = t_xmax + pd_xid_base/pd_multi_base.		(2)
+
+"Double XMAX" page format
+---------------------------------
+
+At first read of a heap page after pg_upgrade from 32-bit XID PostgreSQL
+version pd_special area with a size of 16 bytes should be added to a page.
+Though a page may not have space for this. Then it can be converted to a
+temporary format called "double XMAX".
+
+All tuples after pg-upgrade would necessarily have xmin = FrozenTransactionId.
+So we don't need tuple header t_xmin field and we reuse t_xmin to store higher
+32 bits of its XMAX.
+
+Double XMAX format is only for full pages that don't have 16 bytes for
+pd_special. So it neither has a�place for a single tuple. Insert and HOT update
+for double XMAX pages is impossible and not supported. We can only read or
+delete tuples from it.
+
+When we are able to prune page double XMAX it will be converted from it to
+general 64-bit XID page format with all operations on its tuples supported.
+
+In-memory tuple format
+----------------------
+
+In-memory tuple representation consists of two parts:
+- HeapTupleHeader from disk page (contains all heap tuple contents, not only
+header)
+- HeapTuple with additional in-memory fields
+
+HeapTuple for each tuple in memory stores t_xid_base/t_multi_base - a copies of
+page's pd_xid_base/pd_multi_base. With tuple's 32-bit t_xmin and t_xmax from
+HeapTupleHeader they are used to calculate actual 64-bit XMIN and XMAX:
+
+XMIN = t_xmin + t_xid_base. 					(3)
+XMAX = t_xmax + t_xid_base/t_multi_base.		(4)
+
+The downside of this is that we can not use tuple's XMIN and XMAX right away.
+We often need to re-read t_xmin and t_xmax - which could actually be pointers
+into a page in shared buffers and therefore they could be updated by any other
+backend.
+
+Update/delete with 64-bit XIDs and 32-bit t_xmin/t_xmax
+--------------------------------------------------------------
+
+When we try to delete/update a tuple, we check that XMAX for a page fits (2).
+I.e. that t_xmax will not be over MaxShortTransactionId relative to
+pd_xid_base/pd_multi_base of a its page.
+
+If the current XID doesn't fit a range
+(pd_xid_base, pd_xid_base + MaxShortTransactionId) (5):
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base/pd_multi_base on
+a page and update all t_xmin/t_xmax of the other tuples on the page to
+correspond new pd_xid_base/pd_multi_base.
+
+- If it was impossible, it will try to prune and freeze tuples on a page.
+
+- If this is unsuccessful it will throw an error. Normally this is very
+unlikely but if there is a very old living transaction with an age of around
+2^32 this can arise. Basically, this is a behavior similar to one during the
+vacuum to prevent wraparound when XID was 32-bit. Dba should take care and
+avoid very-long-living transactions with an age close to 2^32. So long-living
+transactions often they are most likely defunct.
+
+Insert with 64-bit XIDs and 32-bit t_xmin/t_xmax
+------------------------------------------------
+
+On insert we check if current XID fits a range (5). Otherwise:
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base for t_xmin will
+not be over MaxShortTransactionId.
+
+- If it is impossible, then it will try to prune and freeze tuples on a page.
+
+Known issue: if pd_xid_base could not be shifted to accommodate a tuple being
+inserted due to a very long-running transaction, we just throw an error. We
+neither try to insert a�tuple into another page nor mark the current page as
+full. So, in this (unlikely) case we will get regular insert errors on the next
+tries to insert to the page 'locked' by this very long-running transaction.
+
+Upgrade from 32-bit XID versions
+--------------------------------
+
+pg_upgrade doesn't change pages format itself. It is done lazily after.
+
+1. At first heap page read, tuples on a page are repacked to free 16 bytes
+at the end of a page, possibly freeing space from dead tuples.
+
+2A. 16 bytes of pd_special is added if there is a place for it
+
+2B. Page is converted to "Double XMAX" format if there is no place for
+pd_special
+
+3. If a page is in double XMAX format after its first read, and vacuum (or
+micro-vacuum at select query) could prune some tuples and free space for
+pd_special, prune_page will add pd_special and convert page from double XMAX
+to general 64-bit XID page format.
-- 
2.24.3 (Apple Git-128)


--cpok4wp6gsarlzvp--





^ permalink  raw  reply  [nested|flat] 265+ messages in thread

* [PATCH v6] README for 64bit xid
@ 2022-01-10 19:20  Pavel Borisov <[email protected]>
  0 siblings, 0 replies; 265+ messages in thread

From: Pavel Borisov @ 2022-01-10 19:20 UTC (permalink / raw)

Authors:
- Pavel Borisov <[email protected]>
- Maxim Orlov <[email protected]>
- Yura Sokolov <[email protected]> <[email protected]>
---
 src/backend/access/heap/README.XID64 | 128 +++++++++++++++++++++++++++
 1 file changed, 128 insertions(+)
 create mode 100644 src/backend/access/heap/README.XID64

diff --git a/src/backend/access/heap/README.XID64 b/src/backend/access/heap/README.XID64
new file mode 100644
index 00000000000..457ba9b9ef5
--- /dev/null
+++ b/src/backend/access/heap/README.XID64
@@ -0,0 +1,128 @@
+src/backend/access/heap/README.XID64
+
+64-bit Transaction ID's (XID)
+=============================
+
+A limited number (N = 2^32) of XID's required to do vacuum freeze to prevent
+wraparound every N/2 transactions. This causes performance degradation due
+to the need to exclusively lock tables while being vacuumed. In each
+wraparound cycle, SLRU buffers are also being cut.
+
+With 64-bit XID's wraparound is effectively postponed to a very distant
+future. Even in highly loaded systems that had 2^32 transactions per day
+it will take huge 2^31 days before the first enforced "vacuum to prevent
+wraparound"). Buffers cutting and routine vacuum are not enforced, and DBA
+can plan them independently at the time with the least system load and least
+critical for database performance. Also, it can be done less frequently
+(several times a year vs every several days) on systems with transaction rates
+similar to those mentioned above.
+
+On-disk tuple and page format
+-----------------------------
+
+On-disk tuple format remains unchanged. 32-bit t_xmin and t_xmax store the
+lower parts of 64-bit XMIN and XMAX values. Each heap page has additional
+64-bit pd_xid_base and pd_multi_base which are common for all tuples on a page.
+They are placed into a pd_special area - 16 bytes in the end of a heap page.
+Actual XMIN/XMAX for a tuple are calculated upon reading a tuple from a page
+as follows:
+
+XMIN = t_xmin + pd_xid_base. 					(1)
+XMAX = t_xmax + pd_xid_base/pd_multi_base.		(2)
+
+"Double XMAX" page format
+---------------------------------
+
+At first read of a heap page after pg_upgrade from 32-bit XID PostgreSQL
+version pd_special area with a size of 16 bytes should be added to a page.
+Though a page may not have space for this. Then it can be converted to a
+temporary format called "double XMAX".
+
+All tuples after pg-upgrade would necessarily have xmin = FrozenTransactionId.
+So we don't need tuple header t_xmin field and we reuse t_xmin to store higher
+32 bits of its XMAX.
+
+Double XMAX format is only for full pages that don't have 16 bytes for
+pd_special. So it neither has a�place for a single tuple. Insert and HOT update
+for double XMAX pages is impossible and not supported. We can only read or
+delete tuples from it.
+
+When we are able to prune page double XMAX it will be converted from it to
+general 64-bit XID page format with all operations on its tuples supported.
+
+In-memory tuple format
+----------------------
+
+In-memory tuple representation consists of two parts:
+- HeapTupleHeader from disk page (contains all heap tuple contents, not only
+header)
+- HeapTuple with additional in-memory fields
+
+HeapTuple for each tuple in memory stores t_xid_base/t_multi_base - a copies of
+page's pd_xid_base/pd_multi_base. With tuple's 32-bit t_xmin and t_xmax from
+HeapTupleHeader they are used to calculate actual 64-bit XMIN and XMAX:
+
+XMIN = t_xmin + t_xid_base. 					(3)
+XMAX = t_xmax + t_xid_base/t_multi_base.		(4)
+
+The downside of this is that we can not use tuple's XMIN and XMAX right away.
+We often need to re-read t_xmin and t_xmax - which could actually be pointers
+into a page in shared buffers and therefore they could be updated by any other
+backend.
+
+Update/delete with 64-bit XIDs and 32-bit t_xmin/t_xmax
+--------------------------------------------------------------
+
+When we try to delete/update a tuple, we check that XMAX for a page fits (2).
+I.e. that t_xmax will not be over MaxShortTransactionId relative to
+pd_xid_base/pd_multi_base of a its page.
+
+If the current XID doesn't fit a range
+(pd_xid_base, pd_xid_base + MaxShortTransactionId) (5):
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base/pd_multi_base on
+a page and update all t_xmin/t_xmax of the other tuples on the page to
+correspond new pd_xid_base/pd_multi_base.
+
+- If it was impossible, it will try to prune and freeze tuples on a page.
+
+- If this is unsuccessful it will throw an error. Normally this is very
+unlikely but if there is a very old living transaction with an age of around
+2^32 this can arise. Basically, this is a behavior similar to one during the
+vacuum to prevent wraparound when XID was 32-bit. Dba should take care and
+avoid very-long-living transactions with an age close to 2^32. So long-living
+transactions often they are most likely defunct.
+
+Insert with 64-bit XIDs and 32-bit t_xmin/t_xmax
+------------------------------------------------
+
+On insert we check if current XID fits a range (5). Otherwise:
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base for t_xmin will
+not be over MaxShortTransactionId.
+
+- If it is impossible, then it will try to prune and freeze tuples on a page.
+
+Known issue: if pd_xid_base could not be shifted to accommodate a tuple being
+inserted due to a very long-running transaction, we just throw an error. We
+neither try to insert a�tuple into another page nor mark the current page as
+full. So, in this (unlikely) case we will get regular insert errors on the next
+tries to insert to the page 'locked' by this very long-running transaction.
+
+Upgrade from 32-bit XID versions
+--------------------------------
+
+pg_upgrade doesn't change pages format itself. It is done lazily after.
+
+1. At first heap page read, tuples on a page are repacked to free 16 bytes
+at the end of a page, possibly freeing space from dead tuples.
+
+2A. 16 bytes of pd_special is added if there is a place for it
+
+2B. Page is converted to "Double XMAX" format if there is no place for
+pd_special
+
+3. If a page is in double XMAX format after its first read, and vacuum (or
+micro-vacuum at select query) could prune some tuples and free space for
+pd_special, prune_page will add pd_special and convert page from double XMAX
+to general 64-bit XID page format.
-- 
2.24.3 (Apple Git-128)


--cpok4wp6gsarlzvp--





^ permalink  raw  reply  [nested|flat] 265+ messages in thread

* [PATCH v6] README for 64bit xid
@ 2022-01-10 19:20  Pavel Borisov <[email protected]>
  0 siblings, 0 replies; 265+ messages in thread

From: Pavel Borisov @ 2022-01-10 19:20 UTC (permalink / raw)

Authors:
- Pavel Borisov <[email protected]>
- Maxim Orlov <[email protected]>
- Yura Sokolov <[email protected]> <[email protected]>
---
 src/backend/access/heap/README.XID64 | 128 +++++++++++++++++++++++++++
 1 file changed, 128 insertions(+)
 create mode 100644 src/backend/access/heap/README.XID64

diff --git a/src/backend/access/heap/README.XID64 b/src/backend/access/heap/README.XID64
new file mode 100644
index 00000000000..457ba9b9ef5
--- /dev/null
+++ b/src/backend/access/heap/README.XID64
@@ -0,0 +1,128 @@
+src/backend/access/heap/README.XID64
+
+64-bit Transaction ID's (XID)
+=============================
+
+A limited number (N = 2^32) of XID's required to do vacuum freeze to prevent
+wraparound every N/2 transactions. This causes performance degradation due
+to the need to exclusively lock tables while being vacuumed. In each
+wraparound cycle, SLRU buffers are also being cut.
+
+With 64-bit XID's wraparound is effectively postponed to a very distant
+future. Even in highly loaded systems that had 2^32 transactions per day
+it will take huge 2^31 days before the first enforced "vacuum to prevent
+wraparound"). Buffers cutting and routine vacuum are not enforced, and DBA
+can plan them independently at the time with the least system load and least
+critical for database performance. Also, it can be done less frequently
+(several times a year vs every several days) on systems with transaction rates
+similar to those mentioned above.
+
+On-disk tuple and page format
+-----------------------------
+
+On-disk tuple format remains unchanged. 32-bit t_xmin and t_xmax store the
+lower parts of 64-bit XMIN and XMAX values. Each heap page has additional
+64-bit pd_xid_base and pd_multi_base which are common for all tuples on a page.
+They are placed into a pd_special area - 16 bytes in the end of a heap page.
+Actual XMIN/XMAX for a tuple are calculated upon reading a tuple from a page
+as follows:
+
+XMIN = t_xmin + pd_xid_base. 					(1)
+XMAX = t_xmax + pd_xid_base/pd_multi_base.		(2)
+
+"Double XMAX" page format
+---------------------------------
+
+At first read of a heap page after pg_upgrade from 32-bit XID PostgreSQL
+version pd_special area with a size of 16 bytes should be added to a page.
+Though a page may not have space for this. Then it can be converted to a
+temporary format called "double XMAX".
+
+All tuples after pg-upgrade would necessarily have xmin = FrozenTransactionId.
+So we don't need tuple header t_xmin field and we reuse t_xmin to store higher
+32 bits of its XMAX.
+
+Double XMAX format is only for full pages that don't have 16 bytes for
+pd_special. So it neither has a�place for a single tuple. Insert and HOT update
+for double XMAX pages is impossible and not supported. We can only read or
+delete tuples from it.
+
+When we are able to prune page double XMAX it will be converted from it to
+general 64-bit XID page format with all operations on its tuples supported.
+
+In-memory tuple format
+----------------------
+
+In-memory tuple representation consists of two parts:
+- HeapTupleHeader from disk page (contains all heap tuple contents, not only
+header)
+- HeapTuple with additional in-memory fields
+
+HeapTuple for each tuple in memory stores t_xid_base/t_multi_base - a copies of
+page's pd_xid_base/pd_multi_base. With tuple's 32-bit t_xmin and t_xmax from
+HeapTupleHeader they are used to calculate actual 64-bit XMIN and XMAX:
+
+XMIN = t_xmin + t_xid_base. 					(3)
+XMAX = t_xmax + t_xid_base/t_multi_base.		(4)
+
+The downside of this is that we can not use tuple's XMIN and XMAX right away.
+We often need to re-read t_xmin and t_xmax - which could actually be pointers
+into a page in shared buffers and therefore they could be updated by any other
+backend.
+
+Update/delete with 64-bit XIDs and 32-bit t_xmin/t_xmax
+--------------------------------------------------------------
+
+When we try to delete/update a tuple, we check that XMAX for a page fits (2).
+I.e. that t_xmax will not be over MaxShortTransactionId relative to
+pd_xid_base/pd_multi_base of a its page.
+
+If the current XID doesn't fit a range
+(pd_xid_base, pd_xid_base + MaxShortTransactionId) (5):
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base/pd_multi_base on
+a page and update all t_xmin/t_xmax of the other tuples on the page to
+correspond new pd_xid_base/pd_multi_base.
+
+- If it was impossible, it will try to prune and freeze tuples on a page.
+
+- If this is unsuccessful it will throw an error. Normally this is very
+unlikely but if there is a very old living transaction with an age of around
+2^32 this can arise. Basically, this is a behavior similar to one during the
+vacuum to prevent wraparound when XID was 32-bit. Dba should take care and
+avoid very-long-living transactions with an age close to 2^32. So long-living
+transactions often they are most likely defunct.
+
+Insert with 64-bit XIDs and 32-bit t_xmin/t_xmax
+------------------------------------------------
+
+On insert we check if current XID fits a range (5). Otherwise:
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base for t_xmin will
+not be over MaxShortTransactionId.
+
+- If it is impossible, then it will try to prune and freeze tuples on a page.
+
+Known issue: if pd_xid_base could not be shifted to accommodate a tuple being
+inserted due to a very long-running transaction, we just throw an error. We
+neither try to insert a�tuple into another page nor mark the current page as
+full. So, in this (unlikely) case we will get regular insert errors on the next
+tries to insert to the page 'locked' by this very long-running transaction.
+
+Upgrade from 32-bit XID versions
+--------------------------------
+
+pg_upgrade doesn't change pages format itself. It is done lazily after.
+
+1. At first heap page read, tuples on a page are repacked to free 16 bytes
+at the end of a page, possibly freeing space from dead tuples.
+
+2A. 16 bytes of pd_special is added if there is a place for it
+
+2B. Page is converted to "Double XMAX" format if there is no place for
+pd_special
+
+3. If a page is in double XMAX format after its first read, and vacuum (or
+micro-vacuum at select query) could prune some tuples and free space for
+pd_special, prune_page will add pd_special and convert page from double XMAX
+to general 64-bit XID page format.
-- 
2.24.3 (Apple Git-128)


--cpok4wp6gsarlzvp--





^ permalink  raw  reply  [nested|flat] 265+ messages in thread

* [PATCH v6] README for 64bit xid
@ 2022-01-10 19:20  Pavel Borisov <[email protected]>
  0 siblings, 0 replies; 265+ messages in thread

From: Pavel Borisov @ 2022-01-10 19:20 UTC (permalink / raw)

Authors:
- Pavel Borisov <[email protected]>
- Maxim Orlov <[email protected]>
- Yura Sokolov <[email protected]> <[email protected]>
---
 src/backend/access/heap/README.XID64 | 128 +++++++++++++++++++++++++++
 1 file changed, 128 insertions(+)
 create mode 100644 src/backend/access/heap/README.XID64

diff --git a/src/backend/access/heap/README.XID64 b/src/backend/access/heap/README.XID64
new file mode 100644
index 00000000000..457ba9b9ef5
--- /dev/null
+++ b/src/backend/access/heap/README.XID64
@@ -0,0 +1,128 @@
+src/backend/access/heap/README.XID64
+
+64-bit Transaction ID's (XID)
+=============================
+
+A limited number (N = 2^32) of XID's required to do vacuum freeze to prevent
+wraparound every N/2 transactions. This causes performance degradation due
+to the need to exclusively lock tables while being vacuumed. In each
+wraparound cycle, SLRU buffers are also being cut.
+
+With 64-bit XID's wraparound is effectively postponed to a very distant
+future. Even in highly loaded systems that had 2^32 transactions per day
+it will take huge 2^31 days before the first enforced "vacuum to prevent
+wraparound"). Buffers cutting and routine vacuum are not enforced, and DBA
+can plan them independently at the time with the least system load and least
+critical for database performance. Also, it can be done less frequently
+(several times a year vs every several days) on systems with transaction rates
+similar to those mentioned above.
+
+On-disk tuple and page format
+-----------------------------
+
+On-disk tuple format remains unchanged. 32-bit t_xmin and t_xmax store the
+lower parts of 64-bit XMIN and XMAX values. Each heap page has additional
+64-bit pd_xid_base and pd_multi_base which are common for all tuples on a page.
+They are placed into a pd_special area - 16 bytes in the end of a heap page.
+Actual XMIN/XMAX for a tuple are calculated upon reading a tuple from a page
+as follows:
+
+XMIN = t_xmin + pd_xid_base. 					(1)
+XMAX = t_xmax + pd_xid_base/pd_multi_base.		(2)
+
+"Double XMAX" page format
+---------------------------------
+
+At first read of a heap page after pg_upgrade from 32-bit XID PostgreSQL
+version pd_special area with a size of 16 bytes should be added to a page.
+Though a page may not have space for this. Then it can be converted to a
+temporary format called "double XMAX".
+
+All tuples after pg-upgrade would necessarily have xmin = FrozenTransactionId.
+So we don't need tuple header t_xmin field and we reuse t_xmin to store higher
+32 bits of its XMAX.
+
+Double XMAX format is only for full pages that don't have 16 bytes for
+pd_special. So it neither has a�place for a single tuple. Insert and HOT update
+for double XMAX pages is impossible and not supported. We can only read or
+delete tuples from it.
+
+When we are able to prune page double XMAX it will be converted from it to
+general 64-bit XID page format with all operations on its tuples supported.
+
+In-memory tuple format
+----------------------
+
+In-memory tuple representation consists of two parts:
+- HeapTupleHeader from disk page (contains all heap tuple contents, not only
+header)
+- HeapTuple with additional in-memory fields
+
+HeapTuple for each tuple in memory stores t_xid_base/t_multi_base - a copies of
+page's pd_xid_base/pd_multi_base. With tuple's 32-bit t_xmin and t_xmax from
+HeapTupleHeader they are used to calculate actual 64-bit XMIN and XMAX:
+
+XMIN = t_xmin + t_xid_base. 					(3)
+XMAX = t_xmax + t_xid_base/t_multi_base.		(4)
+
+The downside of this is that we can not use tuple's XMIN and XMAX right away.
+We often need to re-read t_xmin and t_xmax - which could actually be pointers
+into a page in shared buffers and therefore they could be updated by any other
+backend.
+
+Update/delete with 64-bit XIDs and 32-bit t_xmin/t_xmax
+--------------------------------------------------------------
+
+When we try to delete/update a tuple, we check that XMAX for a page fits (2).
+I.e. that t_xmax will not be over MaxShortTransactionId relative to
+pd_xid_base/pd_multi_base of a its page.
+
+If the current XID doesn't fit a range
+(pd_xid_base, pd_xid_base + MaxShortTransactionId) (5):
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base/pd_multi_base on
+a page and update all t_xmin/t_xmax of the other tuples on the page to
+correspond new pd_xid_base/pd_multi_base.
+
+- If it was impossible, it will try to prune and freeze tuples on a page.
+
+- If this is unsuccessful it will throw an error. Normally this is very
+unlikely but if there is a very old living transaction with an age of around
+2^32 this can arise. Basically, this is a behavior similar to one during the
+vacuum to prevent wraparound when XID was 32-bit. Dba should take care and
+avoid very-long-living transactions with an age close to 2^32. So long-living
+transactions often they are most likely defunct.
+
+Insert with 64-bit XIDs and 32-bit t_xmin/t_xmax
+------------------------------------------------
+
+On insert we check if current XID fits a range (5). Otherwise:
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base for t_xmin will
+not be over MaxShortTransactionId.
+
+- If it is impossible, then it will try to prune and freeze tuples on a page.
+
+Known issue: if pd_xid_base could not be shifted to accommodate a tuple being
+inserted due to a very long-running transaction, we just throw an error. We
+neither try to insert a�tuple into another page nor mark the current page as
+full. So, in this (unlikely) case we will get regular insert errors on the next
+tries to insert to the page 'locked' by this very long-running transaction.
+
+Upgrade from 32-bit XID versions
+--------------------------------
+
+pg_upgrade doesn't change pages format itself. It is done lazily after.
+
+1. At first heap page read, tuples on a page are repacked to free 16 bytes
+at the end of a page, possibly freeing space from dead tuples.
+
+2A. 16 bytes of pd_special is added if there is a place for it
+
+2B. Page is converted to "Double XMAX" format if there is no place for
+pd_special
+
+3. If a page is in double XMAX format after its first read, and vacuum (or
+micro-vacuum at select query) could prune some tuples and free space for
+pd_special, prune_page will add pd_special and convert page from double XMAX
+to general 64-bit XID page format.
-- 
2.24.3 (Apple Git-128)


--cpok4wp6gsarlzvp--





^ permalink  raw  reply  [nested|flat] 265+ messages in thread

* [PATCH v6] README for 64bit xid
@ 2022-01-10 19:20  Pavel Borisov <[email protected]>
  0 siblings, 0 replies; 265+ messages in thread

From: Pavel Borisov @ 2022-01-10 19:20 UTC (permalink / raw)

Authors:
- Pavel Borisov <[email protected]>
- Maxim Orlov <[email protected]>
- Yura Sokolov <[email protected]> <[email protected]>
---
 src/backend/access/heap/README.XID64 | 128 +++++++++++++++++++++++++++
 1 file changed, 128 insertions(+)
 create mode 100644 src/backend/access/heap/README.XID64

diff --git a/src/backend/access/heap/README.XID64 b/src/backend/access/heap/README.XID64
new file mode 100644
index 00000000000..457ba9b9ef5
--- /dev/null
+++ b/src/backend/access/heap/README.XID64
@@ -0,0 +1,128 @@
+src/backend/access/heap/README.XID64
+
+64-bit Transaction ID's (XID)
+=============================
+
+A limited number (N = 2^32) of XID's required to do vacuum freeze to prevent
+wraparound every N/2 transactions. This causes performance degradation due
+to the need to exclusively lock tables while being vacuumed. In each
+wraparound cycle, SLRU buffers are also being cut.
+
+With 64-bit XID's wraparound is effectively postponed to a very distant
+future. Even in highly loaded systems that had 2^32 transactions per day
+it will take huge 2^31 days before the first enforced "vacuum to prevent
+wraparound"). Buffers cutting and routine vacuum are not enforced, and DBA
+can plan them independently at the time with the least system load and least
+critical for database performance. Also, it can be done less frequently
+(several times a year vs every several days) on systems with transaction rates
+similar to those mentioned above.
+
+On-disk tuple and page format
+-----------------------------
+
+On-disk tuple format remains unchanged. 32-bit t_xmin and t_xmax store the
+lower parts of 64-bit XMIN and XMAX values. Each heap page has additional
+64-bit pd_xid_base and pd_multi_base which are common for all tuples on a page.
+They are placed into a pd_special area - 16 bytes in the end of a heap page.
+Actual XMIN/XMAX for a tuple are calculated upon reading a tuple from a page
+as follows:
+
+XMIN = t_xmin + pd_xid_base. 					(1)
+XMAX = t_xmax + pd_xid_base/pd_multi_base.		(2)
+
+"Double XMAX" page format
+---------------------------------
+
+At first read of a heap page after pg_upgrade from 32-bit XID PostgreSQL
+version pd_special area with a size of 16 bytes should be added to a page.
+Though a page may not have space for this. Then it can be converted to a
+temporary format called "double XMAX".
+
+All tuples after pg-upgrade would necessarily have xmin = FrozenTransactionId.
+So we don't need tuple header t_xmin field and we reuse t_xmin to store higher
+32 bits of its XMAX.
+
+Double XMAX format is only for full pages that don't have 16 bytes for
+pd_special. So it neither has a�place for a single tuple. Insert and HOT update
+for double XMAX pages is impossible and not supported. We can only read or
+delete tuples from it.
+
+When we are able to prune page double XMAX it will be converted from it to
+general 64-bit XID page format with all operations on its tuples supported.
+
+In-memory tuple format
+----------------------
+
+In-memory tuple representation consists of two parts:
+- HeapTupleHeader from disk page (contains all heap tuple contents, not only
+header)
+- HeapTuple with additional in-memory fields
+
+HeapTuple for each tuple in memory stores t_xid_base/t_multi_base - a copies of
+page's pd_xid_base/pd_multi_base. With tuple's 32-bit t_xmin and t_xmax from
+HeapTupleHeader they are used to calculate actual 64-bit XMIN and XMAX:
+
+XMIN = t_xmin + t_xid_base. 					(3)
+XMAX = t_xmax + t_xid_base/t_multi_base.		(4)
+
+The downside of this is that we can not use tuple's XMIN and XMAX right away.
+We often need to re-read t_xmin and t_xmax - which could actually be pointers
+into a page in shared buffers and therefore they could be updated by any other
+backend.
+
+Update/delete with 64-bit XIDs and 32-bit t_xmin/t_xmax
+--------------------------------------------------------------
+
+When we try to delete/update a tuple, we check that XMAX for a page fits (2).
+I.e. that t_xmax will not be over MaxShortTransactionId relative to
+pd_xid_base/pd_multi_base of a its page.
+
+If the current XID doesn't fit a range
+(pd_xid_base, pd_xid_base + MaxShortTransactionId) (5):
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base/pd_multi_base on
+a page and update all t_xmin/t_xmax of the other tuples on the page to
+correspond new pd_xid_base/pd_multi_base.
+
+- If it was impossible, it will try to prune and freeze tuples on a page.
+
+- If this is unsuccessful it will throw an error. Normally this is very
+unlikely but if there is a very old living transaction with an age of around
+2^32 this can arise. Basically, this is a behavior similar to one during the
+vacuum to prevent wraparound when XID was 32-bit. Dba should take care and
+avoid very-long-living transactions with an age close to 2^32. So long-living
+transactions often they are most likely defunct.
+
+Insert with 64-bit XIDs and 32-bit t_xmin/t_xmax
+------------------------------------------------
+
+On insert we check if current XID fits a range (5). Otherwise:
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base for t_xmin will
+not be over MaxShortTransactionId.
+
+- If it is impossible, then it will try to prune and freeze tuples on a page.
+
+Known issue: if pd_xid_base could not be shifted to accommodate a tuple being
+inserted due to a very long-running transaction, we just throw an error. We
+neither try to insert a�tuple into another page nor mark the current page as
+full. So, in this (unlikely) case we will get regular insert errors on the next
+tries to insert to the page 'locked' by this very long-running transaction.
+
+Upgrade from 32-bit XID versions
+--------------------------------
+
+pg_upgrade doesn't change pages format itself. It is done lazily after.
+
+1. At first heap page read, tuples on a page are repacked to free 16 bytes
+at the end of a page, possibly freeing space from dead tuples.
+
+2A. 16 bytes of pd_special is added if there is a place for it
+
+2B. Page is converted to "Double XMAX" format if there is no place for
+pd_special
+
+3. If a page is in double XMAX format after its first read, and vacuum (or
+micro-vacuum at select query) could prune some tuples and free space for
+pd_special, prune_page will add pd_special and convert page from double XMAX
+to general 64-bit XID page format.
-- 
2.24.3 (Apple Git-128)


--cpok4wp6gsarlzvp--





^ permalink  raw  reply  [nested|flat] 265+ messages in thread

* [PATCH v6] README for 64bit xid
@ 2022-01-10 19:20  Pavel Borisov <[email protected]>
  0 siblings, 0 replies; 265+ messages in thread

From: Pavel Borisov @ 2022-01-10 19:20 UTC (permalink / raw)

Authors:
- Pavel Borisov <[email protected]>
- Maxim Orlov <[email protected]>
- Yura Sokolov <[email protected]> <[email protected]>
---
 src/backend/access/heap/README.XID64 | 128 +++++++++++++++++++++++++++
 1 file changed, 128 insertions(+)
 create mode 100644 src/backend/access/heap/README.XID64

diff --git a/src/backend/access/heap/README.XID64 b/src/backend/access/heap/README.XID64
new file mode 100644
index 00000000000..457ba9b9ef5
--- /dev/null
+++ b/src/backend/access/heap/README.XID64
@@ -0,0 +1,128 @@
+src/backend/access/heap/README.XID64
+
+64-bit Transaction ID's (XID)
+=============================
+
+A limited number (N = 2^32) of XID's required to do vacuum freeze to prevent
+wraparound every N/2 transactions. This causes performance degradation due
+to the need to exclusively lock tables while being vacuumed. In each
+wraparound cycle, SLRU buffers are also being cut.
+
+With 64-bit XID's wraparound is effectively postponed to a very distant
+future. Even in highly loaded systems that had 2^32 transactions per day
+it will take huge 2^31 days before the first enforced "vacuum to prevent
+wraparound"). Buffers cutting and routine vacuum are not enforced, and DBA
+can plan them independently at the time with the least system load and least
+critical for database performance. Also, it can be done less frequently
+(several times a year vs every several days) on systems with transaction rates
+similar to those mentioned above.
+
+On-disk tuple and page format
+-----------------------------
+
+On-disk tuple format remains unchanged. 32-bit t_xmin and t_xmax store the
+lower parts of 64-bit XMIN and XMAX values. Each heap page has additional
+64-bit pd_xid_base and pd_multi_base which are common for all tuples on a page.
+They are placed into a pd_special area - 16 bytes in the end of a heap page.
+Actual XMIN/XMAX for a tuple are calculated upon reading a tuple from a page
+as follows:
+
+XMIN = t_xmin + pd_xid_base. 					(1)
+XMAX = t_xmax + pd_xid_base/pd_multi_base.		(2)
+
+"Double XMAX" page format
+---------------------------------
+
+At first read of a heap page after pg_upgrade from 32-bit XID PostgreSQL
+version pd_special area with a size of 16 bytes should be added to a page.
+Though a page may not have space for this. Then it can be converted to a
+temporary format called "double XMAX".
+
+All tuples after pg-upgrade would necessarily have xmin = FrozenTransactionId.
+So we don't need tuple header t_xmin field and we reuse t_xmin to store higher
+32 bits of its XMAX.
+
+Double XMAX format is only for full pages that don't have 16 bytes for
+pd_special. So it neither has a�place for a single tuple. Insert and HOT update
+for double XMAX pages is impossible and not supported. We can only read or
+delete tuples from it.
+
+When we are able to prune page double XMAX it will be converted from it to
+general 64-bit XID page format with all operations on its tuples supported.
+
+In-memory tuple format
+----------------------
+
+In-memory tuple representation consists of two parts:
+- HeapTupleHeader from disk page (contains all heap tuple contents, not only
+header)
+- HeapTuple with additional in-memory fields
+
+HeapTuple for each tuple in memory stores t_xid_base/t_multi_base - a copies of
+page's pd_xid_base/pd_multi_base. With tuple's 32-bit t_xmin and t_xmax from
+HeapTupleHeader they are used to calculate actual 64-bit XMIN and XMAX:
+
+XMIN = t_xmin + t_xid_base. 					(3)
+XMAX = t_xmax + t_xid_base/t_multi_base.		(4)
+
+The downside of this is that we can not use tuple's XMIN and XMAX right away.
+We often need to re-read t_xmin and t_xmax - which could actually be pointers
+into a page in shared buffers and therefore they could be updated by any other
+backend.
+
+Update/delete with 64-bit XIDs and 32-bit t_xmin/t_xmax
+--------------------------------------------------------------
+
+When we try to delete/update a tuple, we check that XMAX for a page fits (2).
+I.e. that t_xmax will not be over MaxShortTransactionId relative to
+pd_xid_base/pd_multi_base of a its page.
+
+If the current XID doesn't fit a range
+(pd_xid_base, pd_xid_base + MaxShortTransactionId) (5):
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base/pd_multi_base on
+a page and update all t_xmin/t_xmax of the other tuples on the page to
+correspond new pd_xid_base/pd_multi_base.
+
+- If it was impossible, it will try to prune and freeze tuples on a page.
+
+- If this is unsuccessful it will throw an error. Normally this is very
+unlikely but if there is a very old living transaction with an age of around
+2^32 this can arise. Basically, this is a behavior similar to one during the
+vacuum to prevent wraparound when XID was 32-bit. Dba should take care and
+avoid very-long-living transactions with an age close to 2^32. So long-living
+transactions often they are most likely defunct.
+
+Insert with 64-bit XIDs and 32-bit t_xmin/t_xmax
+------------------------------------------------
+
+On insert we check if current XID fits a range (5). Otherwise:
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base for t_xmin will
+not be over MaxShortTransactionId.
+
+- If it is impossible, then it will try to prune and freeze tuples on a page.
+
+Known issue: if pd_xid_base could not be shifted to accommodate a tuple being
+inserted due to a very long-running transaction, we just throw an error. We
+neither try to insert a�tuple into another page nor mark the current page as
+full. So, in this (unlikely) case we will get regular insert errors on the next
+tries to insert to the page 'locked' by this very long-running transaction.
+
+Upgrade from 32-bit XID versions
+--------------------------------
+
+pg_upgrade doesn't change pages format itself. It is done lazily after.
+
+1. At first heap page read, tuples on a page are repacked to free 16 bytes
+at the end of a page, possibly freeing space from dead tuples.
+
+2A. 16 bytes of pd_special is added if there is a place for it
+
+2B. Page is converted to "Double XMAX" format if there is no place for
+pd_special
+
+3. If a page is in double XMAX format after its first read, and vacuum (or
+micro-vacuum at select query) could prune some tuples and free space for
+pd_special, prune_page will add pd_special and convert page from double XMAX
+to general 64-bit XID page format.
-- 
2.24.3 (Apple Git-128)


--cpok4wp6gsarlzvp--





^ permalink  raw  reply  [nested|flat] 265+ messages in thread

* [PATCH v6] README for 64bit xid
@ 2022-01-10 19:20  Pavel Borisov <[email protected]>
  0 siblings, 0 replies; 265+ messages in thread

From: Pavel Borisov @ 2022-01-10 19:20 UTC (permalink / raw)

Authors:
- Pavel Borisov <[email protected]>
- Maxim Orlov <[email protected]>
- Yura Sokolov <[email protected]> <[email protected]>
---
 src/backend/access/heap/README.XID64 | 128 +++++++++++++++++++++++++++
 1 file changed, 128 insertions(+)
 create mode 100644 src/backend/access/heap/README.XID64

diff --git a/src/backend/access/heap/README.XID64 b/src/backend/access/heap/README.XID64
new file mode 100644
index 00000000000..457ba9b9ef5
--- /dev/null
+++ b/src/backend/access/heap/README.XID64
@@ -0,0 +1,128 @@
+src/backend/access/heap/README.XID64
+
+64-bit Transaction ID's (XID)
+=============================
+
+A limited number (N = 2^32) of XID's required to do vacuum freeze to prevent
+wraparound every N/2 transactions. This causes performance degradation due
+to the need to exclusively lock tables while being vacuumed. In each
+wraparound cycle, SLRU buffers are also being cut.
+
+With 64-bit XID's wraparound is effectively postponed to a very distant
+future. Even in highly loaded systems that had 2^32 transactions per day
+it will take huge 2^31 days before the first enforced "vacuum to prevent
+wraparound"). Buffers cutting and routine vacuum are not enforced, and DBA
+can plan them independently at the time with the least system load and least
+critical for database performance. Also, it can be done less frequently
+(several times a year vs every several days) on systems with transaction rates
+similar to those mentioned above.
+
+On-disk tuple and page format
+-----------------------------
+
+On-disk tuple format remains unchanged. 32-bit t_xmin and t_xmax store the
+lower parts of 64-bit XMIN and XMAX values. Each heap page has additional
+64-bit pd_xid_base and pd_multi_base which are common for all tuples on a page.
+They are placed into a pd_special area - 16 bytes in the end of a heap page.
+Actual XMIN/XMAX for a tuple are calculated upon reading a tuple from a page
+as follows:
+
+XMIN = t_xmin + pd_xid_base. 					(1)
+XMAX = t_xmax + pd_xid_base/pd_multi_base.		(2)
+
+"Double XMAX" page format
+---------------------------------
+
+At first read of a heap page after pg_upgrade from 32-bit XID PostgreSQL
+version pd_special area with a size of 16 bytes should be added to a page.
+Though a page may not have space for this. Then it can be converted to a
+temporary format called "double XMAX".
+
+All tuples after pg-upgrade would necessarily have xmin = FrozenTransactionId.
+So we don't need tuple header t_xmin field and we reuse t_xmin to store higher
+32 bits of its XMAX.
+
+Double XMAX format is only for full pages that don't have 16 bytes for
+pd_special. So it neither has a�place for a single tuple. Insert and HOT update
+for double XMAX pages is impossible and not supported. We can only read or
+delete tuples from it.
+
+When we are able to prune page double XMAX it will be converted from it to
+general 64-bit XID page format with all operations on its tuples supported.
+
+In-memory tuple format
+----------------------
+
+In-memory tuple representation consists of two parts:
+- HeapTupleHeader from disk page (contains all heap tuple contents, not only
+header)
+- HeapTuple with additional in-memory fields
+
+HeapTuple for each tuple in memory stores t_xid_base/t_multi_base - a copies of
+page's pd_xid_base/pd_multi_base. With tuple's 32-bit t_xmin and t_xmax from
+HeapTupleHeader they are used to calculate actual 64-bit XMIN and XMAX:
+
+XMIN = t_xmin + t_xid_base. 					(3)
+XMAX = t_xmax + t_xid_base/t_multi_base.		(4)
+
+The downside of this is that we can not use tuple's XMIN and XMAX right away.
+We often need to re-read t_xmin and t_xmax - which could actually be pointers
+into a page in shared buffers and therefore they could be updated by any other
+backend.
+
+Update/delete with 64-bit XIDs and 32-bit t_xmin/t_xmax
+--------------------------------------------------------------
+
+When we try to delete/update a tuple, we check that XMAX for a page fits (2).
+I.e. that t_xmax will not be over MaxShortTransactionId relative to
+pd_xid_base/pd_multi_base of a its page.
+
+If the current XID doesn't fit a range
+(pd_xid_base, pd_xid_base + MaxShortTransactionId) (5):
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base/pd_multi_base on
+a page and update all t_xmin/t_xmax of the other tuples on the page to
+correspond new pd_xid_base/pd_multi_base.
+
+- If it was impossible, it will try to prune and freeze tuples on a page.
+
+- If this is unsuccessful it will throw an error. Normally this is very
+unlikely but if there is a very old living transaction with an age of around
+2^32 this can arise. Basically, this is a behavior similar to one during the
+vacuum to prevent wraparound when XID was 32-bit. Dba should take care and
+avoid very-long-living transactions with an age close to 2^32. So long-living
+transactions often they are most likely defunct.
+
+Insert with 64-bit XIDs and 32-bit t_xmin/t_xmax
+------------------------------------------------
+
+On insert we check if current XID fits a range (5). Otherwise:
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base for t_xmin will
+not be over MaxShortTransactionId.
+
+- If it is impossible, then it will try to prune and freeze tuples on a page.
+
+Known issue: if pd_xid_base could not be shifted to accommodate a tuple being
+inserted due to a very long-running transaction, we just throw an error. We
+neither try to insert a�tuple into another page nor mark the current page as
+full. So, in this (unlikely) case we will get regular insert errors on the next
+tries to insert to the page 'locked' by this very long-running transaction.
+
+Upgrade from 32-bit XID versions
+--------------------------------
+
+pg_upgrade doesn't change pages format itself. It is done lazily after.
+
+1. At first heap page read, tuples on a page are repacked to free 16 bytes
+at the end of a page, possibly freeing space from dead tuples.
+
+2A. 16 bytes of pd_special is added if there is a place for it
+
+2B. Page is converted to "Double XMAX" format if there is no place for
+pd_special
+
+3. If a page is in double XMAX format after its first read, and vacuum (or
+micro-vacuum at select query) could prune some tuples and free space for
+pd_special, prune_page will add pd_special and convert page from double XMAX
+to general 64-bit XID page format.
-- 
2.24.3 (Apple Git-128)


--cpok4wp6gsarlzvp--





^ permalink  raw  reply  [nested|flat] 265+ messages in thread

* [PATCH v6] README for 64bit xid
@ 2022-01-10 19:20  Pavel Borisov <[email protected]>
  0 siblings, 0 replies; 265+ messages in thread

From: Pavel Borisov @ 2022-01-10 19:20 UTC (permalink / raw)

Authors:
- Pavel Borisov <[email protected]>
- Maxim Orlov <[email protected]>
- Yura Sokolov <[email protected]> <[email protected]>
---
 src/backend/access/heap/README.XID64 | 128 +++++++++++++++++++++++++++
 1 file changed, 128 insertions(+)
 create mode 100644 src/backend/access/heap/README.XID64

diff --git a/src/backend/access/heap/README.XID64 b/src/backend/access/heap/README.XID64
new file mode 100644
index 00000000000..457ba9b9ef5
--- /dev/null
+++ b/src/backend/access/heap/README.XID64
@@ -0,0 +1,128 @@
+src/backend/access/heap/README.XID64
+
+64-bit Transaction ID's (XID)
+=============================
+
+A limited number (N = 2^32) of XID's required to do vacuum freeze to prevent
+wraparound every N/2 transactions. This causes performance degradation due
+to the need to exclusively lock tables while being vacuumed. In each
+wraparound cycle, SLRU buffers are also being cut.
+
+With 64-bit XID's wraparound is effectively postponed to a very distant
+future. Even in highly loaded systems that had 2^32 transactions per day
+it will take huge 2^31 days before the first enforced "vacuum to prevent
+wraparound"). Buffers cutting and routine vacuum are not enforced, and DBA
+can plan them independently at the time with the least system load and least
+critical for database performance. Also, it can be done less frequently
+(several times a year vs every several days) on systems with transaction rates
+similar to those mentioned above.
+
+On-disk tuple and page format
+-----------------------------
+
+On-disk tuple format remains unchanged. 32-bit t_xmin and t_xmax store the
+lower parts of 64-bit XMIN and XMAX values. Each heap page has additional
+64-bit pd_xid_base and pd_multi_base which are common for all tuples on a page.
+They are placed into a pd_special area - 16 bytes in the end of a heap page.
+Actual XMIN/XMAX for a tuple are calculated upon reading a tuple from a page
+as follows:
+
+XMIN = t_xmin + pd_xid_base. 					(1)
+XMAX = t_xmax + pd_xid_base/pd_multi_base.		(2)
+
+"Double XMAX" page format
+---------------------------------
+
+At first read of a heap page after pg_upgrade from 32-bit XID PostgreSQL
+version pd_special area with a size of 16 bytes should be added to a page.
+Though a page may not have space for this. Then it can be converted to a
+temporary format called "double XMAX".
+
+All tuples after pg-upgrade would necessarily have xmin = FrozenTransactionId.
+So we don't need tuple header t_xmin field and we reuse t_xmin to store higher
+32 bits of its XMAX.
+
+Double XMAX format is only for full pages that don't have 16 bytes for
+pd_special. So it neither has a�place for a single tuple. Insert and HOT update
+for double XMAX pages is impossible and not supported. We can only read or
+delete tuples from it.
+
+When we are able to prune page double XMAX it will be converted from it to
+general 64-bit XID page format with all operations on its tuples supported.
+
+In-memory tuple format
+----------------------
+
+In-memory tuple representation consists of two parts:
+- HeapTupleHeader from disk page (contains all heap tuple contents, not only
+header)
+- HeapTuple with additional in-memory fields
+
+HeapTuple for each tuple in memory stores t_xid_base/t_multi_base - a copies of
+page's pd_xid_base/pd_multi_base. With tuple's 32-bit t_xmin and t_xmax from
+HeapTupleHeader they are used to calculate actual 64-bit XMIN and XMAX:
+
+XMIN = t_xmin + t_xid_base. 					(3)
+XMAX = t_xmax + t_xid_base/t_multi_base.		(4)
+
+The downside of this is that we can not use tuple's XMIN and XMAX right away.
+We often need to re-read t_xmin and t_xmax - which could actually be pointers
+into a page in shared buffers and therefore they could be updated by any other
+backend.
+
+Update/delete with 64-bit XIDs and 32-bit t_xmin/t_xmax
+--------------------------------------------------------------
+
+When we try to delete/update a tuple, we check that XMAX for a page fits (2).
+I.e. that t_xmax will not be over MaxShortTransactionId relative to
+pd_xid_base/pd_multi_base of a its page.
+
+If the current XID doesn't fit a range
+(pd_xid_base, pd_xid_base + MaxShortTransactionId) (5):
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base/pd_multi_base on
+a page and update all t_xmin/t_xmax of the other tuples on the page to
+correspond new pd_xid_base/pd_multi_base.
+
+- If it was impossible, it will try to prune and freeze tuples on a page.
+
+- If this is unsuccessful it will throw an error. Normally this is very
+unlikely but if there is a very old living transaction with an age of around
+2^32 this can arise. Basically, this is a behavior similar to one during the
+vacuum to prevent wraparound when XID was 32-bit. Dba should take care and
+avoid very-long-living transactions with an age close to 2^32. So long-living
+transactions often they are most likely defunct.
+
+Insert with 64-bit XIDs and 32-bit t_xmin/t_xmax
+------------------------------------------------
+
+On insert we check if current XID fits a range (5). Otherwise:
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base for t_xmin will
+not be over MaxShortTransactionId.
+
+- If it is impossible, then it will try to prune and freeze tuples on a page.
+
+Known issue: if pd_xid_base could not be shifted to accommodate a tuple being
+inserted due to a very long-running transaction, we just throw an error. We
+neither try to insert a�tuple into another page nor mark the current page as
+full. So, in this (unlikely) case we will get regular insert errors on the next
+tries to insert to the page 'locked' by this very long-running transaction.
+
+Upgrade from 32-bit XID versions
+--------------------------------
+
+pg_upgrade doesn't change pages format itself. It is done lazily after.
+
+1. At first heap page read, tuples on a page are repacked to free 16 bytes
+at the end of a page, possibly freeing space from dead tuples.
+
+2A. 16 bytes of pd_special is added if there is a place for it
+
+2B. Page is converted to "Double XMAX" format if there is no place for
+pd_special
+
+3. If a page is in double XMAX format after its first read, and vacuum (or
+micro-vacuum at select query) could prune some tuples and free space for
+pd_special, prune_page will add pd_special and convert page from double XMAX
+to general 64-bit XID page format.
-- 
2.24.3 (Apple Git-128)


--cpok4wp6gsarlzvp--





^ permalink  raw  reply  [nested|flat] 265+ messages in thread

* [PATCH v6] README for 64bit xid
@ 2022-01-10 19:20  Pavel Borisov <[email protected]>
  0 siblings, 0 replies; 265+ messages in thread

From: Pavel Borisov @ 2022-01-10 19:20 UTC (permalink / raw)

Authors:
- Pavel Borisov <[email protected]>
- Maxim Orlov <[email protected]>
- Yura Sokolov <[email protected]> <[email protected]>
---
 src/backend/access/heap/README.XID64 | 128 +++++++++++++++++++++++++++
 1 file changed, 128 insertions(+)
 create mode 100644 src/backend/access/heap/README.XID64

diff --git a/src/backend/access/heap/README.XID64 b/src/backend/access/heap/README.XID64
new file mode 100644
index 00000000000..457ba9b9ef5
--- /dev/null
+++ b/src/backend/access/heap/README.XID64
@@ -0,0 +1,128 @@
+src/backend/access/heap/README.XID64
+
+64-bit Transaction ID's (XID)
+=============================
+
+A limited number (N = 2^32) of XID's required to do vacuum freeze to prevent
+wraparound every N/2 transactions. This causes performance degradation due
+to the need to exclusively lock tables while being vacuumed. In each
+wraparound cycle, SLRU buffers are also being cut.
+
+With 64-bit XID's wraparound is effectively postponed to a very distant
+future. Even in highly loaded systems that had 2^32 transactions per day
+it will take huge 2^31 days before the first enforced "vacuum to prevent
+wraparound"). Buffers cutting and routine vacuum are not enforced, and DBA
+can plan them independently at the time with the least system load and least
+critical for database performance. Also, it can be done less frequently
+(several times a year vs every several days) on systems with transaction rates
+similar to those mentioned above.
+
+On-disk tuple and page format
+-----------------------------
+
+On-disk tuple format remains unchanged. 32-bit t_xmin and t_xmax store the
+lower parts of 64-bit XMIN and XMAX values. Each heap page has additional
+64-bit pd_xid_base and pd_multi_base which are common for all tuples on a page.
+They are placed into a pd_special area - 16 bytes in the end of a heap page.
+Actual XMIN/XMAX for a tuple are calculated upon reading a tuple from a page
+as follows:
+
+XMIN = t_xmin + pd_xid_base. 					(1)
+XMAX = t_xmax + pd_xid_base/pd_multi_base.		(2)
+
+"Double XMAX" page format
+---------------------------------
+
+At first read of a heap page after pg_upgrade from 32-bit XID PostgreSQL
+version pd_special area with a size of 16 bytes should be added to a page.
+Though a page may not have space for this. Then it can be converted to a
+temporary format called "double XMAX".
+
+All tuples after pg-upgrade would necessarily have xmin = FrozenTransactionId.
+So we don't need tuple header t_xmin field and we reuse t_xmin to store higher
+32 bits of its XMAX.
+
+Double XMAX format is only for full pages that don't have 16 bytes for
+pd_special. So it neither has a�place for a single tuple. Insert and HOT update
+for double XMAX pages is impossible and not supported. We can only read or
+delete tuples from it.
+
+When we are able to prune page double XMAX it will be converted from it to
+general 64-bit XID page format with all operations on its tuples supported.
+
+In-memory tuple format
+----------------------
+
+In-memory tuple representation consists of two parts:
+- HeapTupleHeader from disk page (contains all heap tuple contents, not only
+header)
+- HeapTuple with additional in-memory fields
+
+HeapTuple for each tuple in memory stores t_xid_base/t_multi_base - a copies of
+page's pd_xid_base/pd_multi_base. With tuple's 32-bit t_xmin and t_xmax from
+HeapTupleHeader they are used to calculate actual 64-bit XMIN and XMAX:
+
+XMIN = t_xmin + t_xid_base. 					(3)
+XMAX = t_xmax + t_xid_base/t_multi_base.		(4)
+
+The downside of this is that we can not use tuple's XMIN and XMAX right away.
+We often need to re-read t_xmin and t_xmax - which could actually be pointers
+into a page in shared buffers and therefore they could be updated by any other
+backend.
+
+Update/delete with 64-bit XIDs and 32-bit t_xmin/t_xmax
+--------------------------------------------------------------
+
+When we try to delete/update a tuple, we check that XMAX for a page fits (2).
+I.e. that t_xmax will not be over MaxShortTransactionId relative to
+pd_xid_base/pd_multi_base of a its page.
+
+If the current XID doesn't fit a range
+(pd_xid_base, pd_xid_base + MaxShortTransactionId) (5):
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base/pd_multi_base on
+a page and update all t_xmin/t_xmax of the other tuples on the page to
+correspond new pd_xid_base/pd_multi_base.
+
+- If it was impossible, it will try to prune and freeze tuples on a page.
+
+- If this is unsuccessful it will throw an error. Normally this is very
+unlikely but if there is a very old living transaction with an age of around
+2^32 this can arise. Basically, this is a behavior similar to one during the
+vacuum to prevent wraparound when XID was 32-bit. Dba should take care and
+avoid very-long-living transactions with an age close to 2^32. So long-living
+transactions often they are most likely defunct.
+
+Insert with 64-bit XIDs and 32-bit t_xmin/t_xmax
+------------------------------------------------
+
+On insert we check if current XID fits a range (5). Otherwise:
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base for t_xmin will
+not be over MaxShortTransactionId.
+
+- If it is impossible, then it will try to prune and freeze tuples on a page.
+
+Known issue: if pd_xid_base could not be shifted to accommodate a tuple being
+inserted due to a very long-running transaction, we just throw an error. We
+neither try to insert a�tuple into another page nor mark the current page as
+full. So, in this (unlikely) case we will get regular insert errors on the next
+tries to insert to the page 'locked' by this very long-running transaction.
+
+Upgrade from 32-bit XID versions
+--------------------------------
+
+pg_upgrade doesn't change pages format itself. It is done lazily after.
+
+1. At first heap page read, tuples on a page are repacked to free 16 bytes
+at the end of a page, possibly freeing space from dead tuples.
+
+2A. 16 bytes of pd_special is added if there is a place for it
+
+2B. Page is converted to "Double XMAX" format if there is no place for
+pd_special
+
+3. If a page is in double XMAX format after its first read, and vacuum (or
+micro-vacuum at select query) could prune some tuples and free space for
+pd_special, prune_page will add pd_special and convert page from double XMAX
+to general 64-bit XID page format.
-- 
2.24.3 (Apple Git-128)


--cpok4wp6gsarlzvp--





^ permalink  raw  reply  [nested|flat] 265+ messages in thread

* [PATCH v6] README for 64bit xid
@ 2022-01-10 19:20  Pavel Borisov <[email protected]>
  0 siblings, 0 replies; 265+ messages in thread

From: Pavel Borisov @ 2022-01-10 19:20 UTC (permalink / raw)

Authors:
- Pavel Borisov <[email protected]>
- Maxim Orlov <[email protected]>
- Yura Sokolov <[email protected]> <[email protected]>
---
 src/backend/access/heap/README.XID64 | 128 +++++++++++++++++++++++++++
 1 file changed, 128 insertions(+)
 create mode 100644 src/backend/access/heap/README.XID64

diff --git a/src/backend/access/heap/README.XID64 b/src/backend/access/heap/README.XID64
new file mode 100644
index 00000000000..457ba9b9ef5
--- /dev/null
+++ b/src/backend/access/heap/README.XID64
@@ -0,0 +1,128 @@
+src/backend/access/heap/README.XID64
+
+64-bit Transaction ID's (XID)
+=============================
+
+A limited number (N = 2^32) of XID's required to do vacuum freeze to prevent
+wraparound every N/2 transactions. This causes performance degradation due
+to the need to exclusively lock tables while being vacuumed. In each
+wraparound cycle, SLRU buffers are also being cut.
+
+With 64-bit XID's wraparound is effectively postponed to a very distant
+future. Even in highly loaded systems that had 2^32 transactions per day
+it will take huge 2^31 days before the first enforced "vacuum to prevent
+wraparound"). Buffers cutting and routine vacuum are not enforced, and DBA
+can plan them independently at the time with the least system load and least
+critical for database performance. Also, it can be done less frequently
+(several times a year vs every several days) on systems with transaction rates
+similar to those mentioned above.
+
+On-disk tuple and page format
+-----------------------------
+
+On-disk tuple format remains unchanged. 32-bit t_xmin and t_xmax store the
+lower parts of 64-bit XMIN and XMAX values. Each heap page has additional
+64-bit pd_xid_base and pd_multi_base which are common for all tuples on a page.
+They are placed into a pd_special area - 16 bytes in the end of a heap page.
+Actual XMIN/XMAX for a tuple are calculated upon reading a tuple from a page
+as follows:
+
+XMIN = t_xmin + pd_xid_base. 					(1)
+XMAX = t_xmax + pd_xid_base/pd_multi_base.		(2)
+
+"Double XMAX" page format
+---------------------------------
+
+At first read of a heap page after pg_upgrade from 32-bit XID PostgreSQL
+version pd_special area with a size of 16 bytes should be added to a page.
+Though a page may not have space for this. Then it can be converted to a
+temporary format called "double XMAX".
+
+All tuples after pg-upgrade would necessarily have xmin = FrozenTransactionId.
+So we don't need tuple header t_xmin field and we reuse t_xmin to store higher
+32 bits of its XMAX.
+
+Double XMAX format is only for full pages that don't have 16 bytes for
+pd_special. So it neither has a�place for a single tuple. Insert and HOT update
+for double XMAX pages is impossible and not supported. We can only read or
+delete tuples from it.
+
+When we are able to prune page double XMAX it will be converted from it to
+general 64-bit XID page format with all operations on its tuples supported.
+
+In-memory tuple format
+----------------------
+
+In-memory tuple representation consists of two parts:
+- HeapTupleHeader from disk page (contains all heap tuple contents, not only
+header)
+- HeapTuple with additional in-memory fields
+
+HeapTuple for each tuple in memory stores t_xid_base/t_multi_base - a copies of
+page's pd_xid_base/pd_multi_base. With tuple's 32-bit t_xmin and t_xmax from
+HeapTupleHeader they are used to calculate actual 64-bit XMIN and XMAX:
+
+XMIN = t_xmin + t_xid_base. 					(3)
+XMAX = t_xmax + t_xid_base/t_multi_base.		(4)
+
+The downside of this is that we can not use tuple's XMIN and XMAX right away.
+We often need to re-read t_xmin and t_xmax - which could actually be pointers
+into a page in shared buffers and therefore they could be updated by any other
+backend.
+
+Update/delete with 64-bit XIDs and 32-bit t_xmin/t_xmax
+--------------------------------------------------------------
+
+When we try to delete/update a tuple, we check that XMAX for a page fits (2).
+I.e. that t_xmax will not be over MaxShortTransactionId relative to
+pd_xid_base/pd_multi_base of a its page.
+
+If the current XID doesn't fit a range
+(pd_xid_base, pd_xid_base + MaxShortTransactionId) (5):
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base/pd_multi_base on
+a page and update all t_xmin/t_xmax of the other tuples on the page to
+correspond new pd_xid_base/pd_multi_base.
+
+- If it was impossible, it will try to prune and freeze tuples on a page.
+
+- If this is unsuccessful it will throw an error. Normally this is very
+unlikely but if there is a very old living transaction with an age of around
+2^32 this can arise. Basically, this is a behavior similar to one during the
+vacuum to prevent wraparound when XID was 32-bit. Dba should take care and
+avoid very-long-living transactions with an age close to 2^32. So long-living
+transactions often they are most likely defunct.
+
+Insert with 64-bit XIDs and 32-bit t_xmin/t_xmax
+------------------------------------------------
+
+On insert we check if current XID fits a range (5). Otherwise:
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base for t_xmin will
+not be over MaxShortTransactionId.
+
+- If it is impossible, then it will try to prune and freeze tuples on a page.
+
+Known issue: if pd_xid_base could not be shifted to accommodate a tuple being
+inserted due to a very long-running transaction, we just throw an error. We
+neither try to insert a�tuple into another page nor mark the current page as
+full. So, in this (unlikely) case we will get regular insert errors on the next
+tries to insert to the page 'locked' by this very long-running transaction.
+
+Upgrade from 32-bit XID versions
+--------------------------------
+
+pg_upgrade doesn't change pages format itself. It is done lazily after.
+
+1. At first heap page read, tuples on a page are repacked to free 16 bytes
+at the end of a page, possibly freeing space from dead tuples.
+
+2A. 16 bytes of pd_special is added if there is a place for it
+
+2B. Page is converted to "Double XMAX" format if there is no place for
+pd_special
+
+3. If a page is in double XMAX format after its first read, and vacuum (or
+micro-vacuum at select query) could prune some tuples and free space for
+pd_special, prune_page will add pd_special and convert page from double XMAX
+to general 64-bit XID page format.
-- 
2.24.3 (Apple Git-128)


--cpok4wp6gsarlzvp--





^ permalink  raw  reply  [nested|flat] 265+ messages in thread

* [PATCH v6] README for 64bit xid
@ 2022-01-10 19:20  Pavel Borisov <[email protected]>
  0 siblings, 0 replies; 265+ messages in thread

From: Pavel Borisov @ 2022-01-10 19:20 UTC (permalink / raw)

Authors:
- Pavel Borisov <[email protected]>
- Maxim Orlov <[email protected]>
- Yura Sokolov <[email protected]> <[email protected]>
---
 src/backend/access/heap/README.XID64 | 128 +++++++++++++++++++++++++++
 1 file changed, 128 insertions(+)
 create mode 100644 src/backend/access/heap/README.XID64

diff --git a/src/backend/access/heap/README.XID64 b/src/backend/access/heap/README.XID64
new file mode 100644
index 00000000000..457ba9b9ef5
--- /dev/null
+++ b/src/backend/access/heap/README.XID64
@@ -0,0 +1,128 @@
+src/backend/access/heap/README.XID64
+
+64-bit Transaction ID's (XID)
+=============================
+
+A limited number (N = 2^32) of XID's required to do vacuum freeze to prevent
+wraparound every N/2 transactions. This causes performance degradation due
+to the need to exclusively lock tables while being vacuumed. In each
+wraparound cycle, SLRU buffers are also being cut.
+
+With 64-bit XID's wraparound is effectively postponed to a very distant
+future. Even in highly loaded systems that had 2^32 transactions per day
+it will take huge 2^31 days before the first enforced "vacuum to prevent
+wraparound"). Buffers cutting and routine vacuum are not enforced, and DBA
+can plan them independently at the time with the least system load and least
+critical for database performance. Also, it can be done less frequently
+(several times a year vs every several days) on systems with transaction rates
+similar to those mentioned above.
+
+On-disk tuple and page format
+-----------------------------
+
+On-disk tuple format remains unchanged. 32-bit t_xmin and t_xmax store the
+lower parts of 64-bit XMIN and XMAX values. Each heap page has additional
+64-bit pd_xid_base and pd_multi_base which are common for all tuples on a page.
+They are placed into a pd_special area - 16 bytes in the end of a heap page.
+Actual XMIN/XMAX for a tuple are calculated upon reading a tuple from a page
+as follows:
+
+XMIN = t_xmin + pd_xid_base. 					(1)
+XMAX = t_xmax + pd_xid_base/pd_multi_base.		(2)
+
+"Double XMAX" page format
+---------------------------------
+
+At first read of a heap page after pg_upgrade from 32-bit XID PostgreSQL
+version pd_special area with a size of 16 bytes should be added to a page.
+Though a page may not have space for this. Then it can be converted to a
+temporary format called "double XMAX".
+
+All tuples after pg-upgrade would necessarily have xmin = FrozenTransactionId.
+So we don't need tuple header t_xmin field and we reuse t_xmin to store higher
+32 bits of its XMAX.
+
+Double XMAX format is only for full pages that don't have 16 bytes for
+pd_special. So it neither has a�place for a single tuple. Insert and HOT update
+for double XMAX pages is impossible and not supported. We can only read or
+delete tuples from it.
+
+When we are able to prune page double XMAX it will be converted from it to
+general 64-bit XID page format with all operations on its tuples supported.
+
+In-memory tuple format
+----------------------
+
+In-memory tuple representation consists of two parts:
+- HeapTupleHeader from disk page (contains all heap tuple contents, not only
+header)
+- HeapTuple with additional in-memory fields
+
+HeapTuple for each tuple in memory stores t_xid_base/t_multi_base - a copies of
+page's pd_xid_base/pd_multi_base. With tuple's 32-bit t_xmin and t_xmax from
+HeapTupleHeader they are used to calculate actual 64-bit XMIN and XMAX:
+
+XMIN = t_xmin + t_xid_base. 					(3)
+XMAX = t_xmax + t_xid_base/t_multi_base.		(4)
+
+The downside of this is that we can not use tuple's XMIN and XMAX right away.
+We often need to re-read t_xmin and t_xmax - which could actually be pointers
+into a page in shared buffers and therefore they could be updated by any other
+backend.
+
+Update/delete with 64-bit XIDs and 32-bit t_xmin/t_xmax
+--------------------------------------------------------------
+
+When we try to delete/update a tuple, we check that XMAX for a page fits (2).
+I.e. that t_xmax will not be over MaxShortTransactionId relative to
+pd_xid_base/pd_multi_base of a its page.
+
+If the current XID doesn't fit a range
+(pd_xid_base, pd_xid_base + MaxShortTransactionId) (5):
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base/pd_multi_base on
+a page and update all t_xmin/t_xmax of the other tuples on the page to
+correspond new pd_xid_base/pd_multi_base.
+
+- If it was impossible, it will try to prune and freeze tuples on a page.
+
+- If this is unsuccessful it will throw an error. Normally this is very
+unlikely but if there is a very old living transaction with an age of around
+2^32 this can arise. Basically, this is a behavior similar to one during the
+vacuum to prevent wraparound when XID was 32-bit. Dba should take care and
+avoid very-long-living transactions with an age close to 2^32. So long-living
+transactions often they are most likely defunct.
+
+Insert with 64-bit XIDs and 32-bit t_xmin/t_xmax
+------------------------------------------------
+
+On insert we check if current XID fits a range (5). Otherwise:
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base for t_xmin will
+not be over MaxShortTransactionId.
+
+- If it is impossible, then it will try to prune and freeze tuples on a page.
+
+Known issue: if pd_xid_base could not be shifted to accommodate a tuple being
+inserted due to a very long-running transaction, we just throw an error. We
+neither try to insert a�tuple into another page nor mark the current page as
+full. So, in this (unlikely) case we will get regular insert errors on the next
+tries to insert to the page 'locked' by this very long-running transaction.
+
+Upgrade from 32-bit XID versions
+--------------------------------
+
+pg_upgrade doesn't change pages format itself. It is done lazily after.
+
+1. At first heap page read, tuples on a page are repacked to free 16 bytes
+at the end of a page, possibly freeing space from dead tuples.
+
+2A. 16 bytes of pd_special is added if there is a place for it
+
+2B. Page is converted to "Double XMAX" format if there is no place for
+pd_special
+
+3. If a page is in double XMAX format after its first read, and vacuum (or
+micro-vacuum at select query) could prune some tuples and free space for
+pd_special, prune_page will add pd_special and convert page from double XMAX
+to general 64-bit XID page format.
-- 
2.24.3 (Apple Git-128)


--cpok4wp6gsarlzvp--





^ permalink  raw  reply  [nested|flat] 265+ messages in thread

* [PATCH v6] README for 64bit xid
@ 2022-01-10 19:20  Pavel Borisov <[email protected]>
  0 siblings, 0 replies; 265+ messages in thread

From: Pavel Borisov @ 2022-01-10 19:20 UTC (permalink / raw)

Authors:
- Pavel Borisov <[email protected]>
- Maxim Orlov <[email protected]>
- Yura Sokolov <[email protected]> <[email protected]>
---
 src/backend/access/heap/README.XID64 | 128 +++++++++++++++++++++++++++
 1 file changed, 128 insertions(+)
 create mode 100644 src/backend/access/heap/README.XID64

diff --git a/src/backend/access/heap/README.XID64 b/src/backend/access/heap/README.XID64
new file mode 100644
index 00000000000..457ba9b9ef5
--- /dev/null
+++ b/src/backend/access/heap/README.XID64
@@ -0,0 +1,128 @@
+src/backend/access/heap/README.XID64
+
+64-bit Transaction ID's (XID)
+=============================
+
+A limited number (N = 2^32) of XID's required to do vacuum freeze to prevent
+wraparound every N/2 transactions. This causes performance degradation due
+to the need to exclusively lock tables while being vacuumed. In each
+wraparound cycle, SLRU buffers are also being cut.
+
+With 64-bit XID's wraparound is effectively postponed to a very distant
+future. Even in highly loaded systems that had 2^32 transactions per day
+it will take huge 2^31 days before the first enforced "vacuum to prevent
+wraparound"). Buffers cutting and routine vacuum are not enforced, and DBA
+can plan them independently at the time with the least system load and least
+critical for database performance. Also, it can be done less frequently
+(several times a year vs every several days) on systems with transaction rates
+similar to those mentioned above.
+
+On-disk tuple and page format
+-----------------------------
+
+On-disk tuple format remains unchanged. 32-bit t_xmin and t_xmax store the
+lower parts of 64-bit XMIN and XMAX values. Each heap page has additional
+64-bit pd_xid_base and pd_multi_base which are common for all tuples on a page.
+They are placed into a pd_special area - 16 bytes in the end of a heap page.
+Actual XMIN/XMAX for a tuple are calculated upon reading a tuple from a page
+as follows:
+
+XMIN = t_xmin + pd_xid_base. 					(1)
+XMAX = t_xmax + pd_xid_base/pd_multi_base.		(2)
+
+"Double XMAX" page format
+---------------------------------
+
+At first read of a heap page after pg_upgrade from 32-bit XID PostgreSQL
+version pd_special area with a size of 16 bytes should be added to a page.
+Though a page may not have space for this. Then it can be converted to a
+temporary format called "double XMAX".
+
+All tuples after pg-upgrade would necessarily have xmin = FrozenTransactionId.
+So we don't need tuple header t_xmin field and we reuse t_xmin to store higher
+32 bits of its XMAX.
+
+Double XMAX format is only for full pages that don't have 16 bytes for
+pd_special. So it neither has a�place for a single tuple. Insert and HOT update
+for double XMAX pages is impossible and not supported. We can only read or
+delete tuples from it.
+
+When we are able to prune page double XMAX it will be converted from it to
+general 64-bit XID page format with all operations on its tuples supported.
+
+In-memory tuple format
+----------------------
+
+In-memory tuple representation consists of two parts:
+- HeapTupleHeader from disk page (contains all heap tuple contents, not only
+header)
+- HeapTuple with additional in-memory fields
+
+HeapTuple for each tuple in memory stores t_xid_base/t_multi_base - a copies of
+page's pd_xid_base/pd_multi_base. With tuple's 32-bit t_xmin and t_xmax from
+HeapTupleHeader they are used to calculate actual 64-bit XMIN and XMAX:
+
+XMIN = t_xmin + t_xid_base. 					(3)
+XMAX = t_xmax + t_xid_base/t_multi_base.		(4)
+
+The downside of this is that we can not use tuple's XMIN and XMAX right away.
+We often need to re-read t_xmin and t_xmax - which could actually be pointers
+into a page in shared buffers and therefore they could be updated by any other
+backend.
+
+Update/delete with 64-bit XIDs and 32-bit t_xmin/t_xmax
+--------------------------------------------------------------
+
+When we try to delete/update a tuple, we check that XMAX for a page fits (2).
+I.e. that t_xmax will not be over MaxShortTransactionId relative to
+pd_xid_base/pd_multi_base of a its page.
+
+If the current XID doesn't fit a range
+(pd_xid_base, pd_xid_base + MaxShortTransactionId) (5):
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base/pd_multi_base on
+a page and update all t_xmin/t_xmax of the other tuples on the page to
+correspond new pd_xid_base/pd_multi_base.
+
+- If it was impossible, it will try to prune and freeze tuples on a page.
+
+- If this is unsuccessful it will throw an error. Normally this is very
+unlikely but if there is a very old living transaction with an age of around
+2^32 this can arise. Basically, this is a behavior similar to one during the
+vacuum to prevent wraparound when XID was 32-bit. Dba should take care and
+avoid very-long-living transactions with an age close to 2^32. So long-living
+transactions often they are most likely defunct.
+
+Insert with 64-bit XIDs and 32-bit t_xmin/t_xmax
+------------------------------------------------
+
+On insert we check if current XID fits a range (5). Otherwise:
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base for t_xmin will
+not be over MaxShortTransactionId.
+
+- If it is impossible, then it will try to prune and freeze tuples on a page.
+
+Known issue: if pd_xid_base could not be shifted to accommodate a tuple being
+inserted due to a very long-running transaction, we just throw an error. We
+neither try to insert a�tuple into another page nor mark the current page as
+full. So, in this (unlikely) case we will get regular insert errors on the next
+tries to insert to the page 'locked' by this very long-running transaction.
+
+Upgrade from 32-bit XID versions
+--------------------------------
+
+pg_upgrade doesn't change pages format itself. It is done lazily after.
+
+1. At first heap page read, tuples on a page are repacked to free 16 bytes
+at the end of a page, possibly freeing space from dead tuples.
+
+2A. 16 bytes of pd_special is added if there is a place for it
+
+2B. Page is converted to "Double XMAX" format if there is no place for
+pd_special
+
+3. If a page is in double XMAX format after its first read, and vacuum (or
+micro-vacuum at select query) could prune some tuples and free space for
+pd_special, prune_page will add pd_special and convert page from double XMAX
+to general 64-bit XID page format.
-- 
2.24.3 (Apple Git-128)


--cpok4wp6gsarlzvp--





^ permalink  raw  reply  [nested|flat] 265+ messages in thread

* [PATCH v6] README for 64bit xid
@ 2022-01-10 19:20  Pavel Borisov <[email protected]>
  0 siblings, 0 replies; 265+ messages in thread

From: Pavel Borisov @ 2022-01-10 19:20 UTC (permalink / raw)

Authors:
- Pavel Borisov <[email protected]>
- Maxim Orlov <[email protected]>
- Yura Sokolov <[email protected]> <[email protected]>
---
 src/backend/access/heap/README.XID64 | 128 +++++++++++++++++++++++++++
 1 file changed, 128 insertions(+)
 create mode 100644 src/backend/access/heap/README.XID64

diff --git a/src/backend/access/heap/README.XID64 b/src/backend/access/heap/README.XID64
new file mode 100644
index 00000000000..457ba9b9ef5
--- /dev/null
+++ b/src/backend/access/heap/README.XID64
@@ -0,0 +1,128 @@
+src/backend/access/heap/README.XID64
+
+64-bit Transaction ID's (XID)
+=============================
+
+A limited number (N = 2^32) of XID's required to do vacuum freeze to prevent
+wraparound every N/2 transactions. This causes performance degradation due
+to the need to exclusively lock tables while being vacuumed. In each
+wraparound cycle, SLRU buffers are also being cut.
+
+With 64-bit XID's wraparound is effectively postponed to a very distant
+future. Even in highly loaded systems that had 2^32 transactions per day
+it will take huge 2^31 days before the first enforced "vacuum to prevent
+wraparound"). Buffers cutting and routine vacuum are not enforced, and DBA
+can plan them independently at the time with the least system load and least
+critical for database performance. Also, it can be done less frequently
+(several times a year vs every several days) on systems with transaction rates
+similar to those mentioned above.
+
+On-disk tuple and page format
+-----------------------------
+
+On-disk tuple format remains unchanged. 32-bit t_xmin and t_xmax store the
+lower parts of 64-bit XMIN and XMAX values. Each heap page has additional
+64-bit pd_xid_base and pd_multi_base which are common for all tuples on a page.
+They are placed into a pd_special area - 16 bytes in the end of a heap page.
+Actual XMIN/XMAX for a tuple are calculated upon reading a tuple from a page
+as follows:
+
+XMIN = t_xmin + pd_xid_base. 					(1)
+XMAX = t_xmax + pd_xid_base/pd_multi_base.		(2)
+
+"Double XMAX" page format
+---------------------------------
+
+At first read of a heap page after pg_upgrade from 32-bit XID PostgreSQL
+version pd_special area with a size of 16 bytes should be added to a page.
+Though a page may not have space for this. Then it can be converted to a
+temporary format called "double XMAX".
+
+All tuples after pg-upgrade would necessarily have xmin = FrozenTransactionId.
+So we don't need tuple header t_xmin field and we reuse t_xmin to store higher
+32 bits of its XMAX.
+
+Double XMAX format is only for full pages that don't have 16 bytes for
+pd_special. So it neither has a�place for a single tuple. Insert and HOT update
+for double XMAX pages is impossible and not supported. We can only read or
+delete tuples from it.
+
+When we are able to prune page double XMAX it will be converted from it to
+general 64-bit XID page format with all operations on its tuples supported.
+
+In-memory tuple format
+----------------------
+
+In-memory tuple representation consists of two parts:
+- HeapTupleHeader from disk page (contains all heap tuple contents, not only
+header)
+- HeapTuple with additional in-memory fields
+
+HeapTuple for each tuple in memory stores t_xid_base/t_multi_base - a copies of
+page's pd_xid_base/pd_multi_base. With tuple's 32-bit t_xmin and t_xmax from
+HeapTupleHeader they are used to calculate actual 64-bit XMIN and XMAX:
+
+XMIN = t_xmin + t_xid_base. 					(3)
+XMAX = t_xmax + t_xid_base/t_multi_base.		(4)
+
+The downside of this is that we can not use tuple's XMIN and XMAX right away.
+We often need to re-read t_xmin and t_xmax - which could actually be pointers
+into a page in shared buffers and therefore they could be updated by any other
+backend.
+
+Update/delete with 64-bit XIDs and 32-bit t_xmin/t_xmax
+--------------------------------------------------------------
+
+When we try to delete/update a tuple, we check that XMAX for a page fits (2).
+I.e. that t_xmax will not be over MaxShortTransactionId relative to
+pd_xid_base/pd_multi_base of a its page.
+
+If the current XID doesn't fit a range
+(pd_xid_base, pd_xid_base + MaxShortTransactionId) (5):
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base/pd_multi_base on
+a page and update all t_xmin/t_xmax of the other tuples on the page to
+correspond new pd_xid_base/pd_multi_base.
+
+- If it was impossible, it will try to prune and freeze tuples on a page.
+
+- If this is unsuccessful it will throw an error. Normally this is very
+unlikely but if there is a very old living transaction with an age of around
+2^32 this can arise. Basically, this is a behavior similar to one during the
+vacuum to prevent wraparound when XID was 32-bit. Dba should take care and
+avoid very-long-living transactions with an age close to 2^32. So long-living
+transactions often they are most likely defunct.
+
+Insert with 64-bit XIDs and 32-bit t_xmin/t_xmax
+------------------------------------------------
+
+On insert we check if current XID fits a range (5). Otherwise:
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base for t_xmin will
+not be over MaxShortTransactionId.
+
+- If it is impossible, then it will try to prune and freeze tuples on a page.
+
+Known issue: if pd_xid_base could not be shifted to accommodate a tuple being
+inserted due to a very long-running transaction, we just throw an error. We
+neither try to insert a�tuple into another page nor mark the current page as
+full. So, in this (unlikely) case we will get regular insert errors on the next
+tries to insert to the page 'locked' by this very long-running transaction.
+
+Upgrade from 32-bit XID versions
+--------------------------------
+
+pg_upgrade doesn't change pages format itself. It is done lazily after.
+
+1. At first heap page read, tuples on a page are repacked to free 16 bytes
+at the end of a page, possibly freeing space from dead tuples.
+
+2A. 16 bytes of pd_special is added if there is a place for it
+
+2B. Page is converted to "Double XMAX" format if there is no place for
+pd_special
+
+3. If a page is in double XMAX format after its first read, and vacuum (or
+micro-vacuum at select query) could prune some tuples and free space for
+pd_special, prune_page will add pd_special and convert page from double XMAX
+to general 64-bit XID page format.
-- 
2.24.3 (Apple Git-128)


--cpok4wp6gsarlzvp--





^ permalink  raw  reply  [nested|flat] 265+ messages in thread

* [PATCH v6] README for 64bit xid
@ 2022-01-10 19:20  Pavel Borisov <[email protected]>
  0 siblings, 0 replies; 265+ messages in thread

From: Pavel Borisov @ 2022-01-10 19:20 UTC (permalink / raw)

Authors:
- Pavel Borisov <[email protected]>
- Maxim Orlov <[email protected]>
- Yura Sokolov <[email protected]> <[email protected]>
---
 src/backend/access/heap/README.XID64 | 128 +++++++++++++++++++++++++++
 1 file changed, 128 insertions(+)
 create mode 100644 src/backend/access/heap/README.XID64

diff --git a/src/backend/access/heap/README.XID64 b/src/backend/access/heap/README.XID64
new file mode 100644
index 00000000000..457ba9b9ef5
--- /dev/null
+++ b/src/backend/access/heap/README.XID64
@@ -0,0 +1,128 @@
+src/backend/access/heap/README.XID64
+
+64-bit Transaction ID's (XID)
+=============================
+
+A limited number (N = 2^32) of XID's required to do vacuum freeze to prevent
+wraparound every N/2 transactions. This causes performance degradation due
+to the need to exclusively lock tables while being vacuumed. In each
+wraparound cycle, SLRU buffers are also being cut.
+
+With 64-bit XID's wraparound is effectively postponed to a very distant
+future. Even in highly loaded systems that had 2^32 transactions per day
+it will take huge 2^31 days before the first enforced "vacuum to prevent
+wraparound"). Buffers cutting and routine vacuum are not enforced, and DBA
+can plan them independently at the time with the least system load and least
+critical for database performance. Also, it can be done less frequently
+(several times a year vs every several days) on systems with transaction rates
+similar to those mentioned above.
+
+On-disk tuple and page format
+-----------------------------
+
+On-disk tuple format remains unchanged. 32-bit t_xmin and t_xmax store the
+lower parts of 64-bit XMIN and XMAX values. Each heap page has additional
+64-bit pd_xid_base and pd_multi_base which are common for all tuples on a page.
+They are placed into a pd_special area - 16 bytes in the end of a heap page.
+Actual XMIN/XMAX for a tuple are calculated upon reading a tuple from a page
+as follows:
+
+XMIN = t_xmin + pd_xid_base. 					(1)
+XMAX = t_xmax + pd_xid_base/pd_multi_base.		(2)
+
+"Double XMAX" page format
+---------------------------------
+
+At first read of a heap page after pg_upgrade from 32-bit XID PostgreSQL
+version pd_special area with a size of 16 bytes should be added to a page.
+Though a page may not have space for this. Then it can be converted to a
+temporary format called "double XMAX".
+
+All tuples after pg-upgrade would necessarily have xmin = FrozenTransactionId.
+So we don't need tuple header t_xmin field and we reuse t_xmin to store higher
+32 bits of its XMAX.
+
+Double XMAX format is only for full pages that don't have 16 bytes for
+pd_special. So it neither has a�place for a single tuple. Insert and HOT update
+for double XMAX pages is impossible and not supported. We can only read or
+delete tuples from it.
+
+When we are able to prune page double XMAX it will be converted from it to
+general 64-bit XID page format with all operations on its tuples supported.
+
+In-memory tuple format
+----------------------
+
+In-memory tuple representation consists of two parts:
+- HeapTupleHeader from disk page (contains all heap tuple contents, not only
+header)
+- HeapTuple with additional in-memory fields
+
+HeapTuple for each tuple in memory stores t_xid_base/t_multi_base - a copies of
+page's pd_xid_base/pd_multi_base. With tuple's 32-bit t_xmin and t_xmax from
+HeapTupleHeader they are used to calculate actual 64-bit XMIN and XMAX:
+
+XMIN = t_xmin + t_xid_base. 					(3)
+XMAX = t_xmax + t_xid_base/t_multi_base.		(4)
+
+The downside of this is that we can not use tuple's XMIN and XMAX right away.
+We often need to re-read t_xmin and t_xmax - which could actually be pointers
+into a page in shared buffers and therefore they could be updated by any other
+backend.
+
+Update/delete with 64-bit XIDs and 32-bit t_xmin/t_xmax
+--------------------------------------------------------------
+
+When we try to delete/update a tuple, we check that XMAX for a page fits (2).
+I.e. that t_xmax will not be over MaxShortTransactionId relative to
+pd_xid_base/pd_multi_base of a its page.
+
+If the current XID doesn't fit a range
+(pd_xid_base, pd_xid_base + MaxShortTransactionId) (5):
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base/pd_multi_base on
+a page and update all t_xmin/t_xmax of the other tuples on the page to
+correspond new pd_xid_base/pd_multi_base.
+
+- If it was impossible, it will try to prune and freeze tuples on a page.
+
+- If this is unsuccessful it will throw an error. Normally this is very
+unlikely but if there is a very old living transaction with an age of around
+2^32 this can arise. Basically, this is a behavior similar to one during the
+vacuum to prevent wraparound when XID was 32-bit. Dba should take care and
+avoid very-long-living transactions with an age close to 2^32. So long-living
+transactions often they are most likely defunct.
+
+Insert with 64-bit XIDs and 32-bit t_xmin/t_xmax
+------------------------------------------------
+
+On insert we check if current XID fits a range (5). Otherwise:
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base for t_xmin will
+not be over MaxShortTransactionId.
+
+- If it is impossible, then it will try to prune and freeze tuples on a page.
+
+Known issue: if pd_xid_base could not be shifted to accommodate a tuple being
+inserted due to a very long-running transaction, we just throw an error. We
+neither try to insert a�tuple into another page nor mark the current page as
+full. So, in this (unlikely) case we will get regular insert errors on the next
+tries to insert to the page 'locked' by this very long-running transaction.
+
+Upgrade from 32-bit XID versions
+--------------------------------
+
+pg_upgrade doesn't change pages format itself. It is done lazily after.
+
+1. At first heap page read, tuples on a page are repacked to free 16 bytes
+at the end of a page, possibly freeing space from dead tuples.
+
+2A. 16 bytes of pd_special is added if there is a place for it
+
+2B. Page is converted to "Double XMAX" format if there is no place for
+pd_special
+
+3. If a page is in double XMAX format after its first read, and vacuum (or
+micro-vacuum at select query) could prune some tuples and free space for
+pd_special, prune_page will add pd_special and convert page from double XMAX
+to general 64-bit XID page format.
-- 
2.24.3 (Apple Git-128)


--cpok4wp6gsarlzvp--





^ permalink  raw  reply  [nested|flat] 265+ messages in thread

* [PATCH v6] README for 64bit xid
@ 2022-01-10 19:20  Pavel Borisov <[email protected]>
  0 siblings, 0 replies; 265+ messages in thread

From: Pavel Borisov @ 2022-01-10 19:20 UTC (permalink / raw)

Authors:
- Pavel Borisov <[email protected]>
- Maxim Orlov <[email protected]>
- Yura Sokolov <[email protected]> <[email protected]>
---
 src/backend/access/heap/README.XID64 | 128 +++++++++++++++++++++++++++
 1 file changed, 128 insertions(+)
 create mode 100644 src/backend/access/heap/README.XID64

diff --git a/src/backend/access/heap/README.XID64 b/src/backend/access/heap/README.XID64
new file mode 100644
index 00000000000..457ba9b9ef5
--- /dev/null
+++ b/src/backend/access/heap/README.XID64
@@ -0,0 +1,128 @@
+src/backend/access/heap/README.XID64
+
+64-bit Transaction ID's (XID)
+=============================
+
+A limited number (N = 2^32) of XID's required to do vacuum freeze to prevent
+wraparound every N/2 transactions. This causes performance degradation due
+to the need to exclusively lock tables while being vacuumed. In each
+wraparound cycle, SLRU buffers are also being cut.
+
+With 64-bit XID's wraparound is effectively postponed to a very distant
+future. Even in highly loaded systems that had 2^32 transactions per day
+it will take huge 2^31 days before the first enforced "vacuum to prevent
+wraparound"). Buffers cutting and routine vacuum are not enforced, and DBA
+can plan them independently at the time with the least system load and least
+critical for database performance. Also, it can be done less frequently
+(several times a year vs every several days) on systems with transaction rates
+similar to those mentioned above.
+
+On-disk tuple and page format
+-----------------------------
+
+On-disk tuple format remains unchanged. 32-bit t_xmin and t_xmax store the
+lower parts of 64-bit XMIN and XMAX values. Each heap page has additional
+64-bit pd_xid_base and pd_multi_base which are common for all tuples on a page.
+They are placed into a pd_special area - 16 bytes in the end of a heap page.
+Actual XMIN/XMAX for a tuple are calculated upon reading a tuple from a page
+as follows:
+
+XMIN = t_xmin + pd_xid_base. 					(1)
+XMAX = t_xmax + pd_xid_base/pd_multi_base.		(2)
+
+"Double XMAX" page format
+---------------------------------
+
+At first read of a heap page after pg_upgrade from 32-bit XID PostgreSQL
+version pd_special area with a size of 16 bytes should be added to a page.
+Though a page may not have space for this. Then it can be converted to a
+temporary format called "double XMAX".
+
+All tuples after pg-upgrade would necessarily have xmin = FrozenTransactionId.
+So we don't need tuple header t_xmin field and we reuse t_xmin to store higher
+32 bits of its XMAX.
+
+Double XMAX format is only for full pages that don't have 16 bytes for
+pd_special. So it neither has a�place for a single tuple. Insert and HOT update
+for double XMAX pages is impossible and not supported. We can only read or
+delete tuples from it.
+
+When we are able to prune page double XMAX it will be converted from it to
+general 64-bit XID page format with all operations on its tuples supported.
+
+In-memory tuple format
+----------------------
+
+In-memory tuple representation consists of two parts:
+- HeapTupleHeader from disk page (contains all heap tuple contents, not only
+header)
+- HeapTuple with additional in-memory fields
+
+HeapTuple for each tuple in memory stores t_xid_base/t_multi_base - a copies of
+page's pd_xid_base/pd_multi_base. With tuple's 32-bit t_xmin and t_xmax from
+HeapTupleHeader they are used to calculate actual 64-bit XMIN and XMAX:
+
+XMIN = t_xmin + t_xid_base. 					(3)
+XMAX = t_xmax + t_xid_base/t_multi_base.		(4)
+
+The downside of this is that we can not use tuple's XMIN and XMAX right away.
+We often need to re-read t_xmin and t_xmax - which could actually be pointers
+into a page in shared buffers and therefore they could be updated by any other
+backend.
+
+Update/delete with 64-bit XIDs and 32-bit t_xmin/t_xmax
+--------------------------------------------------------------
+
+When we try to delete/update a tuple, we check that XMAX for a page fits (2).
+I.e. that t_xmax will not be over MaxShortTransactionId relative to
+pd_xid_base/pd_multi_base of a its page.
+
+If the current XID doesn't fit a range
+(pd_xid_base, pd_xid_base + MaxShortTransactionId) (5):
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base/pd_multi_base on
+a page and update all t_xmin/t_xmax of the other tuples on the page to
+correspond new pd_xid_base/pd_multi_base.
+
+- If it was impossible, it will try to prune and freeze tuples on a page.
+
+- If this is unsuccessful it will throw an error. Normally this is very
+unlikely but if there is a very old living transaction with an age of around
+2^32 this can arise. Basically, this is a behavior similar to one during the
+vacuum to prevent wraparound when XID was 32-bit. Dba should take care and
+avoid very-long-living transactions with an age close to 2^32. So long-living
+transactions often they are most likely defunct.
+
+Insert with 64-bit XIDs and 32-bit t_xmin/t_xmax
+------------------------------------------------
+
+On insert we check if current XID fits a range (5). Otherwise:
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base for t_xmin will
+not be over MaxShortTransactionId.
+
+- If it is impossible, then it will try to prune and freeze tuples on a page.
+
+Known issue: if pd_xid_base could not be shifted to accommodate a tuple being
+inserted due to a very long-running transaction, we just throw an error. We
+neither try to insert a�tuple into another page nor mark the current page as
+full. So, in this (unlikely) case we will get regular insert errors on the next
+tries to insert to the page 'locked' by this very long-running transaction.
+
+Upgrade from 32-bit XID versions
+--------------------------------
+
+pg_upgrade doesn't change pages format itself. It is done lazily after.
+
+1. At first heap page read, tuples on a page are repacked to free 16 bytes
+at the end of a page, possibly freeing space from dead tuples.
+
+2A. 16 bytes of pd_special is added if there is a place for it
+
+2B. Page is converted to "Double XMAX" format if there is no place for
+pd_special
+
+3. If a page is in double XMAX format after its first read, and vacuum (or
+micro-vacuum at select query) could prune some tuples and free space for
+pd_special, prune_page will add pd_special and convert page from double XMAX
+to general 64-bit XID page format.
-- 
2.24.3 (Apple Git-128)


--cpok4wp6gsarlzvp--





^ permalink  raw  reply  [nested|flat] 265+ messages in thread

* [PATCH v6] README for 64bit xid
@ 2022-01-10 19:20  Pavel Borisov <[email protected]>
  0 siblings, 0 replies; 265+ messages in thread

From: Pavel Borisov @ 2022-01-10 19:20 UTC (permalink / raw)

Authors:
- Pavel Borisov <[email protected]>
- Maxim Orlov <[email protected]>
- Yura Sokolov <[email protected]> <[email protected]>
---
 src/backend/access/heap/README.XID64 | 128 +++++++++++++++++++++++++++
 1 file changed, 128 insertions(+)
 create mode 100644 src/backend/access/heap/README.XID64

diff --git a/src/backend/access/heap/README.XID64 b/src/backend/access/heap/README.XID64
new file mode 100644
index 00000000000..457ba9b9ef5
--- /dev/null
+++ b/src/backend/access/heap/README.XID64
@@ -0,0 +1,128 @@
+src/backend/access/heap/README.XID64
+
+64-bit Transaction ID's (XID)
+=============================
+
+A limited number (N = 2^32) of XID's required to do vacuum freeze to prevent
+wraparound every N/2 transactions. This causes performance degradation due
+to the need to exclusively lock tables while being vacuumed. In each
+wraparound cycle, SLRU buffers are also being cut.
+
+With 64-bit XID's wraparound is effectively postponed to a very distant
+future. Even in highly loaded systems that had 2^32 transactions per day
+it will take huge 2^31 days before the first enforced "vacuum to prevent
+wraparound"). Buffers cutting and routine vacuum are not enforced, and DBA
+can plan them independently at the time with the least system load and least
+critical for database performance. Also, it can be done less frequently
+(several times a year vs every several days) on systems with transaction rates
+similar to those mentioned above.
+
+On-disk tuple and page format
+-----------------------------
+
+On-disk tuple format remains unchanged. 32-bit t_xmin and t_xmax store the
+lower parts of 64-bit XMIN and XMAX values. Each heap page has additional
+64-bit pd_xid_base and pd_multi_base which are common for all tuples on a page.
+They are placed into a pd_special area - 16 bytes in the end of a heap page.
+Actual XMIN/XMAX for a tuple are calculated upon reading a tuple from a page
+as follows:
+
+XMIN = t_xmin + pd_xid_base. 					(1)
+XMAX = t_xmax + pd_xid_base/pd_multi_base.		(2)
+
+"Double XMAX" page format
+---------------------------------
+
+At first read of a heap page after pg_upgrade from 32-bit XID PostgreSQL
+version pd_special area with a size of 16 bytes should be added to a page.
+Though a page may not have space for this. Then it can be converted to a
+temporary format called "double XMAX".
+
+All tuples after pg-upgrade would necessarily have xmin = FrozenTransactionId.
+So we don't need tuple header t_xmin field and we reuse t_xmin to store higher
+32 bits of its XMAX.
+
+Double XMAX format is only for full pages that don't have 16 bytes for
+pd_special. So it neither has a�place for a single tuple. Insert and HOT update
+for double XMAX pages is impossible and not supported. We can only read or
+delete tuples from it.
+
+When we are able to prune page double XMAX it will be converted from it to
+general 64-bit XID page format with all operations on its tuples supported.
+
+In-memory tuple format
+----------------------
+
+In-memory tuple representation consists of two parts:
+- HeapTupleHeader from disk page (contains all heap tuple contents, not only
+header)
+- HeapTuple with additional in-memory fields
+
+HeapTuple for each tuple in memory stores t_xid_base/t_multi_base - a copies of
+page's pd_xid_base/pd_multi_base. With tuple's 32-bit t_xmin and t_xmax from
+HeapTupleHeader they are used to calculate actual 64-bit XMIN and XMAX:
+
+XMIN = t_xmin + t_xid_base. 					(3)
+XMAX = t_xmax + t_xid_base/t_multi_base.		(4)
+
+The downside of this is that we can not use tuple's XMIN and XMAX right away.
+We often need to re-read t_xmin and t_xmax - which could actually be pointers
+into a page in shared buffers and therefore they could be updated by any other
+backend.
+
+Update/delete with 64-bit XIDs and 32-bit t_xmin/t_xmax
+--------------------------------------------------------------
+
+When we try to delete/update a tuple, we check that XMAX for a page fits (2).
+I.e. that t_xmax will not be over MaxShortTransactionId relative to
+pd_xid_base/pd_multi_base of a its page.
+
+If the current XID doesn't fit a range
+(pd_xid_base, pd_xid_base + MaxShortTransactionId) (5):
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base/pd_multi_base on
+a page and update all t_xmin/t_xmax of the other tuples on the page to
+correspond new pd_xid_base/pd_multi_base.
+
+- If it was impossible, it will try to prune and freeze tuples on a page.
+
+- If this is unsuccessful it will throw an error. Normally this is very
+unlikely but if there is a very old living transaction with an age of around
+2^32 this can arise. Basically, this is a behavior similar to one during the
+vacuum to prevent wraparound when XID was 32-bit. Dba should take care and
+avoid very-long-living transactions with an age close to 2^32. So long-living
+transactions often they are most likely defunct.
+
+Insert with 64-bit XIDs and 32-bit t_xmin/t_xmax
+------------------------------------------------
+
+On insert we check if current XID fits a range (5). Otherwise:
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base for t_xmin will
+not be over MaxShortTransactionId.
+
+- If it is impossible, then it will try to prune and freeze tuples on a page.
+
+Known issue: if pd_xid_base could not be shifted to accommodate a tuple being
+inserted due to a very long-running transaction, we just throw an error. We
+neither try to insert a�tuple into another page nor mark the current page as
+full. So, in this (unlikely) case we will get regular insert errors on the next
+tries to insert to the page 'locked' by this very long-running transaction.
+
+Upgrade from 32-bit XID versions
+--------------------------------
+
+pg_upgrade doesn't change pages format itself. It is done lazily after.
+
+1. At first heap page read, tuples on a page are repacked to free 16 bytes
+at the end of a page, possibly freeing space from dead tuples.
+
+2A. 16 bytes of pd_special is added if there is a place for it
+
+2B. Page is converted to "Double XMAX" format if there is no place for
+pd_special
+
+3. If a page is in double XMAX format after its first read, and vacuum (or
+micro-vacuum at select query) could prune some tuples and free space for
+pd_special, prune_page will add pd_special and convert page from double XMAX
+to general 64-bit XID page format.
-- 
2.24.3 (Apple Git-128)


--cpok4wp6gsarlzvp--





^ permalink  raw  reply  [nested|flat] 265+ messages in thread

* [PATCH v6] README for 64bit xid
@ 2022-01-10 19:20  Pavel Borisov <[email protected]>
  0 siblings, 0 replies; 265+ messages in thread

From: Pavel Borisov @ 2022-01-10 19:20 UTC (permalink / raw)

Authors:
- Pavel Borisov <[email protected]>
- Maxim Orlov <[email protected]>
- Yura Sokolov <[email protected]> <[email protected]>
---
 src/backend/access/heap/README.XID64 | 128 +++++++++++++++++++++++++++
 1 file changed, 128 insertions(+)
 create mode 100644 src/backend/access/heap/README.XID64

diff --git a/src/backend/access/heap/README.XID64 b/src/backend/access/heap/README.XID64
new file mode 100644
index 00000000000..457ba9b9ef5
--- /dev/null
+++ b/src/backend/access/heap/README.XID64
@@ -0,0 +1,128 @@
+src/backend/access/heap/README.XID64
+
+64-bit Transaction ID's (XID)
+=============================
+
+A limited number (N = 2^32) of XID's required to do vacuum freeze to prevent
+wraparound every N/2 transactions. This causes performance degradation due
+to the need to exclusively lock tables while being vacuumed. In each
+wraparound cycle, SLRU buffers are also being cut.
+
+With 64-bit XID's wraparound is effectively postponed to a very distant
+future. Even in highly loaded systems that had 2^32 transactions per day
+it will take huge 2^31 days before the first enforced "vacuum to prevent
+wraparound"). Buffers cutting and routine vacuum are not enforced, and DBA
+can plan them independently at the time with the least system load and least
+critical for database performance. Also, it can be done less frequently
+(several times a year vs every several days) on systems with transaction rates
+similar to those mentioned above.
+
+On-disk tuple and page format
+-----------------------------
+
+On-disk tuple format remains unchanged. 32-bit t_xmin and t_xmax store the
+lower parts of 64-bit XMIN and XMAX values. Each heap page has additional
+64-bit pd_xid_base and pd_multi_base which are common for all tuples on a page.
+They are placed into a pd_special area - 16 bytes in the end of a heap page.
+Actual XMIN/XMAX for a tuple are calculated upon reading a tuple from a page
+as follows:
+
+XMIN = t_xmin + pd_xid_base. 					(1)
+XMAX = t_xmax + pd_xid_base/pd_multi_base.		(2)
+
+"Double XMAX" page format
+---------------------------------
+
+At first read of a heap page after pg_upgrade from 32-bit XID PostgreSQL
+version pd_special area with a size of 16 bytes should be added to a page.
+Though a page may not have space for this. Then it can be converted to a
+temporary format called "double XMAX".
+
+All tuples after pg-upgrade would necessarily have xmin = FrozenTransactionId.
+So we don't need tuple header t_xmin field and we reuse t_xmin to store higher
+32 bits of its XMAX.
+
+Double XMAX format is only for full pages that don't have 16 bytes for
+pd_special. So it neither has a�place for a single tuple. Insert and HOT update
+for double XMAX pages is impossible and not supported. We can only read or
+delete tuples from it.
+
+When we are able to prune page double XMAX it will be converted from it to
+general 64-bit XID page format with all operations on its tuples supported.
+
+In-memory tuple format
+----------------------
+
+In-memory tuple representation consists of two parts:
+- HeapTupleHeader from disk page (contains all heap tuple contents, not only
+header)
+- HeapTuple with additional in-memory fields
+
+HeapTuple for each tuple in memory stores t_xid_base/t_multi_base - a copies of
+page's pd_xid_base/pd_multi_base. With tuple's 32-bit t_xmin and t_xmax from
+HeapTupleHeader they are used to calculate actual 64-bit XMIN and XMAX:
+
+XMIN = t_xmin + t_xid_base. 					(3)
+XMAX = t_xmax + t_xid_base/t_multi_base.		(4)
+
+The downside of this is that we can not use tuple's XMIN and XMAX right away.
+We often need to re-read t_xmin and t_xmax - which could actually be pointers
+into a page in shared buffers and therefore they could be updated by any other
+backend.
+
+Update/delete with 64-bit XIDs and 32-bit t_xmin/t_xmax
+--------------------------------------------------------------
+
+When we try to delete/update a tuple, we check that XMAX for a page fits (2).
+I.e. that t_xmax will not be over MaxShortTransactionId relative to
+pd_xid_base/pd_multi_base of a its page.
+
+If the current XID doesn't fit a range
+(pd_xid_base, pd_xid_base + MaxShortTransactionId) (5):
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base/pd_multi_base on
+a page and update all t_xmin/t_xmax of the other tuples on the page to
+correspond new pd_xid_base/pd_multi_base.
+
+- If it was impossible, it will try to prune and freeze tuples on a page.
+
+- If this is unsuccessful it will throw an error. Normally this is very
+unlikely but if there is a very old living transaction with an age of around
+2^32 this can arise. Basically, this is a behavior similar to one during the
+vacuum to prevent wraparound when XID was 32-bit. Dba should take care and
+avoid very-long-living transactions with an age close to 2^32. So long-living
+transactions often they are most likely defunct.
+
+Insert with 64-bit XIDs and 32-bit t_xmin/t_xmax
+------------------------------------------------
+
+On insert we check if current XID fits a range (5). Otherwise:
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base for t_xmin will
+not be over MaxShortTransactionId.
+
+- If it is impossible, then it will try to prune and freeze tuples on a page.
+
+Known issue: if pd_xid_base could not be shifted to accommodate a tuple being
+inserted due to a very long-running transaction, we just throw an error. We
+neither try to insert a�tuple into another page nor mark the current page as
+full. So, in this (unlikely) case we will get regular insert errors on the next
+tries to insert to the page 'locked' by this very long-running transaction.
+
+Upgrade from 32-bit XID versions
+--------------------------------
+
+pg_upgrade doesn't change pages format itself. It is done lazily after.
+
+1. At first heap page read, tuples on a page are repacked to free 16 bytes
+at the end of a page, possibly freeing space from dead tuples.
+
+2A. 16 bytes of pd_special is added if there is a place for it
+
+2B. Page is converted to "Double XMAX" format if there is no place for
+pd_special
+
+3. If a page is in double XMAX format after its first read, and vacuum (or
+micro-vacuum at select query) could prune some tuples and free space for
+pd_special, prune_page will add pd_special and convert page from double XMAX
+to general 64-bit XID page format.
-- 
2.24.3 (Apple Git-128)


--cpok4wp6gsarlzvp--





^ permalink  raw  reply  [nested|flat] 265+ messages in thread

* [PATCH v6] README for 64bit xid
@ 2022-01-10 19:20  Pavel Borisov <[email protected]>
  0 siblings, 0 replies; 265+ messages in thread

From: Pavel Borisov @ 2022-01-10 19:20 UTC (permalink / raw)

Authors:
- Pavel Borisov <[email protected]>
- Maxim Orlov <[email protected]>
- Yura Sokolov <[email protected]> <[email protected]>
---
 src/backend/access/heap/README.XID64 | 128 +++++++++++++++++++++++++++
 1 file changed, 128 insertions(+)
 create mode 100644 src/backend/access/heap/README.XID64

diff --git a/src/backend/access/heap/README.XID64 b/src/backend/access/heap/README.XID64
new file mode 100644
index 00000000000..457ba9b9ef5
--- /dev/null
+++ b/src/backend/access/heap/README.XID64
@@ -0,0 +1,128 @@
+src/backend/access/heap/README.XID64
+
+64-bit Transaction ID's (XID)
+=============================
+
+A limited number (N = 2^32) of XID's required to do vacuum freeze to prevent
+wraparound every N/2 transactions. This causes performance degradation due
+to the need to exclusively lock tables while being vacuumed. In each
+wraparound cycle, SLRU buffers are also being cut.
+
+With 64-bit XID's wraparound is effectively postponed to a very distant
+future. Even in highly loaded systems that had 2^32 transactions per day
+it will take huge 2^31 days before the first enforced "vacuum to prevent
+wraparound"). Buffers cutting and routine vacuum are not enforced, and DBA
+can plan them independently at the time with the least system load and least
+critical for database performance. Also, it can be done less frequently
+(several times a year vs every several days) on systems with transaction rates
+similar to those mentioned above.
+
+On-disk tuple and page format
+-----------------------------
+
+On-disk tuple format remains unchanged. 32-bit t_xmin and t_xmax store the
+lower parts of 64-bit XMIN and XMAX values. Each heap page has additional
+64-bit pd_xid_base and pd_multi_base which are common for all tuples on a page.
+They are placed into a pd_special area - 16 bytes in the end of a heap page.
+Actual XMIN/XMAX for a tuple are calculated upon reading a tuple from a page
+as follows:
+
+XMIN = t_xmin + pd_xid_base. 					(1)
+XMAX = t_xmax + pd_xid_base/pd_multi_base.		(2)
+
+"Double XMAX" page format
+---------------------------------
+
+At first read of a heap page after pg_upgrade from 32-bit XID PostgreSQL
+version pd_special area with a size of 16 bytes should be added to a page.
+Though a page may not have space for this. Then it can be converted to a
+temporary format called "double XMAX".
+
+All tuples after pg-upgrade would necessarily have xmin = FrozenTransactionId.
+So we don't need tuple header t_xmin field and we reuse t_xmin to store higher
+32 bits of its XMAX.
+
+Double XMAX format is only for full pages that don't have 16 bytes for
+pd_special. So it neither has a�place for a single tuple. Insert and HOT update
+for double XMAX pages is impossible and not supported. We can only read or
+delete tuples from it.
+
+When we are able to prune page double XMAX it will be converted from it to
+general 64-bit XID page format with all operations on its tuples supported.
+
+In-memory tuple format
+----------------------
+
+In-memory tuple representation consists of two parts:
+- HeapTupleHeader from disk page (contains all heap tuple contents, not only
+header)
+- HeapTuple with additional in-memory fields
+
+HeapTuple for each tuple in memory stores t_xid_base/t_multi_base - a copies of
+page's pd_xid_base/pd_multi_base. With tuple's 32-bit t_xmin and t_xmax from
+HeapTupleHeader they are used to calculate actual 64-bit XMIN and XMAX:
+
+XMIN = t_xmin + t_xid_base. 					(3)
+XMAX = t_xmax + t_xid_base/t_multi_base.		(4)
+
+The downside of this is that we can not use tuple's XMIN and XMAX right away.
+We often need to re-read t_xmin and t_xmax - which could actually be pointers
+into a page in shared buffers and therefore they could be updated by any other
+backend.
+
+Update/delete with 64-bit XIDs and 32-bit t_xmin/t_xmax
+--------------------------------------------------------------
+
+When we try to delete/update a tuple, we check that XMAX for a page fits (2).
+I.e. that t_xmax will not be over MaxShortTransactionId relative to
+pd_xid_base/pd_multi_base of a its page.
+
+If the current XID doesn't fit a range
+(pd_xid_base, pd_xid_base + MaxShortTransactionId) (5):
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base/pd_multi_base on
+a page and update all t_xmin/t_xmax of the other tuples on the page to
+correspond new pd_xid_base/pd_multi_base.
+
+- If it was impossible, it will try to prune and freeze tuples on a page.
+
+- If this is unsuccessful it will throw an error. Normally this is very
+unlikely but if there is a very old living transaction with an age of around
+2^32 this can arise. Basically, this is a behavior similar to one during the
+vacuum to prevent wraparound when XID was 32-bit. Dba should take care and
+avoid very-long-living transactions with an age close to 2^32. So long-living
+transactions often they are most likely defunct.
+
+Insert with 64-bit XIDs and 32-bit t_xmin/t_xmax
+------------------------------------------------
+
+On insert we check if current XID fits a range (5). Otherwise:
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base for t_xmin will
+not be over MaxShortTransactionId.
+
+- If it is impossible, then it will try to prune and freeze tuples on a page.
+
+Known issue: if pd_xid_base could not be shifted to accommodate a tuple being
+inserted due to a very long-running transaction, we just throw an error. We
+neither try to insert a�tuple into another page nor mark the current page as
+full. So, in this (unlikely) case we will get regular insert errors on the next
+tries to insert to the page 'locked' by this very long-running transaction.
+
+Upgrade from 32-bit XID versions
+--------------------------------
+
+pg_upgrade doesn't change pages format itself. It is done lazily after.
+
+1. At first heap page read, tuples on a page are repacked to free 16 bytes
+at the end of a page, possibly freeing space from dead tuples.
+
+2A. 16 bytes of pd_special is added if there is a place for it
+
+2B. Page is converted to "Double XMAX" format if there is no place for
+pd_special
+
+3. If a page is in double XMAX format after its first read, and vacuum (or
+micro-vacuum at select query) could prune some tuples and free space for
+pd_special, prune_page will add pd_special and convert page from double XMAX
+to general 64-bit XID page format.
-- 
2.24.3 (Apple Git-128)


--cpok4wp6gsarlzvp--





^ permalink  raw  reply  [nested|flat] 265+ messages in thread

* [PATCH v6] README for 64bit xid
@ 2022-01-10 19:20  Pavel Borisov <[email protected]>
  0 siblings, 0 replies; 265+ messages in thread

From: Pavel Borisov @ 2022-01-10 19:20 UTC (permalink / raw)

Authors:
- Pavel Borisov <[email protected]>
- Maxim Orlov <[email protected]>
- Yura Sokolov <[email protected]> <[email protected]>
---
 src/backend/access/heap/README.XID64 | 128 +++++++++++++++++++++++++++
 1 file changed, 128 insertions(+)
 create mode 100644 src/backend/access/heap/README.XID64

diff --git a/src/backend/access/heap/README.XID64 b/src/backend/access/heap/README.XID64
new file mode 100644
index 00000000000..457ba9b9ef5
--- /dev/null
+++ b/src/backend/access/heap/README.XID64
@@ -0,0 +1,128 @@
+src/backend/access/heap/README.XID64
+
+64-bit Transaction ID's (XID)
+=============================
+
+A limited number (N = 2^32) of XID's required to do vacuum freeze to prevent
+wraparound every N/2 transactions. This causes performance degradation due
+to the need to exclusively lock tables while being vacuumed. In each
+wraparound cycle, SLRU buffers are also being cut.
+
+With 64-bit XID's wraparound is effectively postponed to a very distant
+future. Even in highly loaded systems that had 2^32 transactions per day
+it will take huge 2^31 days before the first enforced "vacuum to prevent
+wraparound"). Buffers cutting and routine vacuum are not enforced, and DBA
+can plan them independently at the time with the least system load and least
+critical for database performance. Also, it can be done less frequently
+(several times a year vs every several days) on systems with transaction rates
+similar to those mentioned above.
+
+On-disk tuple and page format
+-----------------------------
+
+On-disk tuple format remains unchanged. 32-bit t_xmin and t_xmax store the
+lower parts of 64-bit XMIN and XMAX values. Each heap page has additional
+64-bit pd_xid_base and pd_multi_base which are common for all tuples on a page.
+They are placed into a pd_special area - 16 bytes in the end of a heap page.
+Actual XMIN/XMAX for a tuple are calculated upon reading a tuple from a page
+as follows:
+
+XMIN = t_xmin + pd_xid_base. 					(1)
+XMAX = t_xmax + pd_xid_base/pd_multi_base.		(2)
+
+"Double XMAX" page format
+---------------------------------
+
+At first read of a heap page after pg_upgrade from 32-bit XID PostgreSQL
+version pd_special area with a size of 16 bytes should be added to a page.
+Though a page may not have space for this. Then it can be converted to a
+temporary format called "double XMAX".
+
+All tuples after pg-upgrade would necessarily have xmin = FrozenTransactionId.
+So we don't need tuple header t_xmin field and we reuse t_xmin to store higher
+32 bits of its XMAX.
+
+Double XMAX format is only for full pages that don't have 16 bytes for
+pd_special. So it neither has a�place for a single tuple. Insert and HOT update
+for double XMAX pages is impossible and not supported. We can only read or
+delete tuples from it.
+
+When we are able to prune page double XMAX it will be converted from it to
+general 64-bit XID page format with all operations on its tuples supported.
+
+In-memory tuple format
+----------------------
+
+In-memory tuple representation consists of two parts:
+- HeapTupleHeader from disk page (contains all heap tuple contents, not only
+header)
+- HeapTuple with additional in-memory fields
+
+HeapTuple for each tuple in memory stores t_xid_base/t_multi_base - a copies of
+page's pd_xid_base/pd_multi_base. With tuple's 32-bit t_xmin and t_xmax from
+HeapTupleHeader they are used to calculate actual 64-bit XMIN and XMAX:
+
+XMIN = t_xmin + t_xid_base. 					(3)
+XMAX = t_xmax + t_xid_base/t_multi_base.		(4)
+
+The downside of this is that we can not use tuple's XMIN and XMAX right away.
+We often need to re-read t_xmin and t_xmax - which could actually be pointers
+into a page in shared buffers and therefore they could be updated by any other
+backend.
+
+Update/delete with 64-bit XIDs and 32-bit t_xmin/t_xmax
+--------------------------------------------------------------
+
+When we try to delete/update a tuple, we check that XMAX for a page fits (2).
+I.e. that t_xmax will not be over MaxShortTransactionId relative to
+pd_xid_base/pd_multi_base of a its page.
+
+If the current XID doesn't fit a range
+(pd_xid_base, pd_xid_base + MaxShortTransactionId) (5):
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base/pd_multi_base on
+a page and update all t_xmin/t_xmax of the other tuples on the page to
+correspond new pd_xid_base/pd_multi_base.
+
+- If it was impossible, it will try to prune and freeze tuples on a page.
+
+- If this is unsuccessful it will throw an error. Normally this is very
+unlikely but if there is a very old living transaction with an age of around
+2^32 this can arise. Basically, this is a behavior similar to one during the
+vacuum to prevent wraparound when XID was 32-bit. Dba should take care and
+avoid very-long-living transactions with an age close to 2^32. So long-living
+transactions often they are most likely defunct.
+
+Insert with 64-bit XIDs and 32-bit t_xmin/t_xmax
+------------------------------------------------
+
+On insert we check if current XID fits a range (5). Otherwise:
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base for t_xmin will
+not be over MaxShortTransactionId.
+
+- If it is impossible, then it will try to prune and freeze tuples on a page.
+
+Known issue: if pd_xid_base could not be shifted to accommodate a tuple being
+inserted due to a very long-running transaction, we just throw an error. We
+neither try to insert a�tuple into another page nor mark the current page as
+full. So, in this (unlikely) case we will get regular insert errors on the next
+tries to insert to the page 'locked' by this very long-running transaction.
+
+Upgrade from 32-bit XID versions
+--------------------------------
+
+pg_upgrade doesn't change pages format itself. It is done lazily after.
+
+1. At first heap page read, tuples on a page are repacked to free 16 bytes
+at the end of a page, possibly freeing space from dead tuples.
+
+2A. 16 bytes of pd_special is added if there is a place for it
+
+2B. Page is converted to "Double XMAX" format if there is no place for
+pd_special
+
+3. If a page is in double XMAX format after its first read, and vacuum (or
+micro-vacuum at select query) could prune some tuples and free space for
+pd_special, prune_page will add pd_special and convert page from double XMAX
+to general 64-bit XID page format.
-- 
2.24.3 (Apple Git-128)


--cpok4wp6gsarlzvp--





^ permalink  raw  reply  [nested|flat] 265+ messages in thread

* [PATCH v6] README for 64bit xid
@ 2022-01-10 19:20  Pavel Borisov <[email protected]>
  0 siblings, 0 replies; 265+ messages in thread

From: Pavel Borisov @ 2022-01-10 19:20 UTC (permalink / raw)

Authors:
- Pavel Borisov <[email protected]>
- Maxim Orlov <[email protected]>
- Yura Sokolov <[email protected]> <[email protected]>
---
 src/backend/access/heap/README.XID64 | 128 +++++++++++++++++++++++++++
 1 file changed, 128 insertions(+)
 create mode 100644 src/backend/access/heap/README.XID64

diff --git a/src/backend/access/heap/README.XID64 b/src/backend/access/heap/README.XID64
new file mode 100644
index 00000000000..457ba9b9ef5
--- /dev/null
+++ b/src/backend/access/heap/README.XID64
@@ -0,0 +1,128 @@
+src/backend/access/heap/README.XID64
+
+64-bit Transaction ID's (XID)
+=============================
+
+A limited number (N = 2^32) of XID's required to do vacuum freeze to prevent
+wraparound every N/2 transactions. This causes performance degradation due
+to the need to exclusively lock tables while being vacuumed. In each
+wraparound cycle, SLRU buffers are also being cut.
+
+With 64-bit XID's wraparound is effectively postponed to a very distant
+future. Even in highly loaded systems that had 2^32 transactions per day
+it will take huge 2^31 days before the first enforced "vacuum to prevent
+wraparound"). Buffers cutting and routine vacuum are not enforced, and DBA
+can plan them independently at the time with the least system load and least
+critical for database performance. Also, it can be done less frequently
+(several times a year vs every several days) on systems with transaction rates
+similar to those mentioned above.
+
+On-disk tuple and page format
+-----------------------------
+
+On-disk tuple format remains unchanged. 32-bit t_xmin and t_xmax store the
+lower parts of 64-bit XMIN and XMAX values. Each heap page has additional
+64-bit pd_xid_base and pd_multi_base which are common for all tuples on a page.
+They are placed into a pd_special area - 16 bytes in the end of a heap page.
+Actual XMIN/XMAX for a tuple are calculated upon reading a tuple from a page
+as follows:
+
+XMIN = t_xmin + pd_xid_base. 					(1)
+XMAX = t_xmax + pd_xid_base/pd_multi_base.		(2)
+
+"Double XMAX" page format
+---------------------------------
+
+At first read of a heap page after pg_upgrade from 32-bit XID PostgreSQL
+version pd_special area with a size of 16 bytes should be added to a page.
+Though a page may not have space for this. Then it can be converted to a
+temporary format called "double XMAX".
+
+All tuples after pg-upgrade would necessarily have xmin = FrozenTransactionId.
+So we don't need tuple header t_xmin field and we reuse t_xmin to store higher
+32 bits of its XMAX.
+
+Double XMAX format is only for full pages that don't have 16 bytes for
+pd_special. So it neither has a�place for a single tuple. Insert and HOT update
+for double XMAX pages is impossible and not supported. We can only read or
+delete tuples from it.
+
+When we are able to prune page double XMAX it will be converted from it to
+general 64-bit XID page format with all operations on its tuples supported.
+
+In-memory tuple format
+----------------------
+
+In-memory tuple representation consists of two parts:
+- HeapTupleHeader from disk page (contains all heap tuple contents, not only
+header)
+- HeapTuple with additional in-memory fields
+
+HeapTuple for each tuple in memory stores t_xid_base/t_multi_base - a copies of
+page's pd_xid_base/pd_multi_base. With tuple's 32-bit t_xmin and t_xmax from
+HeapTupleHeader they are used to calculate actual 64-bit XMIN and XMAX:
+
+XMIN = t_xmin + t_xid_base. 					(3)
+XMAX = t_xmax + t_xid_base/t_multi_base.		(4)
+
+The downside of this is that we can not use tuple's XMIN and XMAX right away.
+We often need to re-read t_xmin and t_xmax - which could actually be pointers
+into a page in shared buffers and therefore they could be updated by any other
+backend.
+
+Update/delete with 64-bit XIDs and 32-bit t_xmin/t_xmax
+--------------------------------------------------------------
+
+When we try to delete/update a tuple, we check that XMAX for a page fits (2).
+I.e. that t_xmax will not be over MaxShortTransactionId relative to
+pd_xid_base/pd_multi_base of a its page.
+
+If the current XID doesn't fit a range
+(pd_xid_base, pd_xid_base + MaxShortTransactionId) (5):
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base/pd_multi_base on
+a page and update all t_xmin/t_xmax of the other tuples on the page to
+correspond new pd_xid_base/pd_multi_base.
+
+- If it was impossible, it will try to prune and freeze tuples on a page.
+
+- If this is unsuccessful it will throw an error. Normally this is very
+unlikely but if there is a very old living transaction with an age of around
+2^32 this can arise. Basically, this is a behavior similar to one during the
+vacuum to prevent wraparound when XID was 32-bit. Dba should take care and
+avoid very-long-living transactions with an age close to 2^32. So long-living
+transactions often they are most likely defunct.
+
+Insert with 64-bit XIDs and 32-bit t_xmin/t_xmax
+------------------------------------------------
+
+On insert we check if current XID fits a range (5). Otherwise:
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base for t_xmin will
+not be over MaxShortTransactionId.
+
+- If it is impossible, then it will try to prune and freeze tuples on a page.
+
+Known issue: if pd_xid_base could not be shifted to accommodate a tuple being
+inserted due to a very long-running transaction, we just throw an error. We
+neither try to insert a�tuple into another page nor mark the current page as
+full. So, in this (unlikely) case we will get regular insert errors on the next
+tries to insert to the page 'locked' by this very long-running transaction.
+
+Upgrade from 32-bit XID versions
+--------------------------------
+
+pg_upgrade doesn't change pages format itself. It is done lazily after.
+
+1. At first heap page read, tuples on a page are repacked to free 16 bytes
+at the end of a page, possibly freeing space from dead tuples.
+
+2A. 16 bytes of pd_special is added if there is a place for it
+
+2B. Page is converted to "Double XMAX" format if there is no place for
+pd_special
+
+3. If a page is in double XMAX format after its first read, and vacuum (or
+micro-vacuum at select query) could prune some tuples and free space for
+pd_special, prune_page will add pd_special and convert page from double XMAX
+to general 64-bit XID page format.
-- 
2.24.3 (Apple Git-128)


--cpok4wp6gsarlzvp--





^ permalink  raw  reply  [nested|flat] 265+ messages in thread

* [PATCH v6] README for 64bit xid
@ 2022-01-10 19:20  Pavel Borisov <[email protected]>
  0 siblings, 0 replies; 265+ messages in thread

From: Pavel Borisov @ 2022-01-10 19:20 UTC (permalink / raw)

Authors:
- Pavel Borisov <[email protected]>
- Maxim Orlov <[email protected]>
- Yura Sokolov <[email protected]> <[email protected]>
---
 src/backend/access/heap/README.XID64 | 128 +++++++++++++++++++++++++++
 1 file changed, 128 insertions(+)
 create mode 100644 src/backend/access/heap/README.XID64

diff --git a/src/backend/access/heap/README.XID64 b/src/backend/access/heap/README.XID64
new file mode 100644
index 00000000000..457ba9b9ef5
--- /dev/null
+++ b/src/backend/access/heap/README.XID64
@@ -0,0 +1,128 @@
+src/backend/access/heap/README.XID64
+
+64-bit Transaction ID's (XID)
+=============================
+
+A limited number (N = 2^32) of XID's required to do vacuum freeze to prevent
+wraparound every N/2 transactions. This causes performance degradation due
+to the need to exclusively lock tables while being vacuumed. In each
+wraparound cycle, SLRU buffers are also being cut.
+
+With 64-bit XID's wraparound is effectively postponed to a very distant
+future. Even in highly loaded systems that had 2^32 transactions per day
+it will take huge 2^31 days before the first enforced "vacuum to prevent
+wraparound"). Buffers cutting and routine vacuum are not enforced, and DBA
+can plan them independently at the time with the least system load and least
+critical for database performance. Also, it can be done less frequently
+(several times a year vs every several days) on systems with transaction rates
+similar to those mentioned above.
+
+On-disk tuple and page format
+-----------------------------
+
+On-disk tuple format remains unchanged. 32-bit t_xmin and t_xmax store the
+lower parts of 64-bit XMIN and XMAX values. Each heap page has additional
+64-bit pd_xid_base and pd_multi_base which are common for all tuples on a page.
+They are placed into a pd_special area - 16 bytes in the end of a heap page.
+Actual XMIN/XMAX for a tuple are calculated upon reading a tuple from a page
+as follows:
+
+XMIN = t_xmin + pd_xid_base. 					(1)
+XMAX = t_xmax + pd_xid_base/pd_multi_base.		(2)
+
+"Double XMAX" page format
+---------------------------------
+
+At first read of a heap page after pg_upgrade from 32-bit XID PostgreSQL
+version pd_special area with a size of 16 bytes should be added to a page.
+Though a page may not have space for this. Then it can be converted to a
+temporary format called "double XMAX".
+
+All tuples after pg-upgrade would necessarily have xmin = FrozenTransactionId.
+So we don't need tuple header t_xmin field and we reuse t_xmin to store higher
+32 bits of its XMAX.
+
+Double XMAX format is only for full pages that don't have 16 bytes for
+pd_special. So it neither has a�place for a single tuple. Insert and HOT update
+for double XMAX pages is impossible and not supported. We can only read or
+delete tuples from it.
+
+When we are able to prune page double XMAX it will be converted from it to
+general 64-bit XID page format with all operations on its tuples supported.
+
+In-memory tuple format
+----------------------
+
+In-memory tuple representation consists of two parts:
+- HeapTupleHeader from disk page (contains all heap tuple contents, not only
+header)
+- HeapTuple with additional in-memory fields
+
+HeapTuple for each tuple in memory stores t_xid_base/t_multi_base - a copies of
+page's pd_xid_base/pd_multi_base. With tuple's 32-bit t_xmin and t_xmax from
+HeapTupleHeader they are used to calculate actual 64-bit XMIN and XMAX:
+
+XMIN = t_xmin + t_xid_base. 					(3)
+XMAX = t_xmax + t_xid_base/t_multi_base.		(4)
+
+The downside of this is that we can not use tuple's XMIN and XMAX right away.
+We often need to re-read t_xmin and t_xmax - which could actually be pointers
+into a page in shared buffers and therefore they could be updated by any other
+backend.
+
+Update/delete with 64-bit XIDs and 32-bit t_xmin/t_xmax
+--------------------------------------------------------------
+
+When we try to delete/update a tuple, we check that XMAX for a page fits (2).
+I.e. that t_xmax will not be over MaxShortTransactionId relative to
+pd_xid_base/pd_multi_base of a its page.
+
+If the current XID doesn't fit a range
+(pd_xid_base, pd_xid_base + MaxShortTransactionId) (5):
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base/pd_multi_base on
+a page and update all t_xmin/t_xmax of the other tuples on the page to
+correspond new pd_xid_base/pd_multi_base.
+
+- If it was impossible, it will try to prune and freeze tuples on a page.
+
+- If this is unsuccessful it will throw an error. Normally this is very
+unlikely but if there is a very old living transaction with an age of around
+2^32 this can arise. Basically, this is a behavior similar to one during the
+vacuum to prevent wraparound when XID was 32-bit. Dba should take care and
+avoid very-long-living transactions with an age close to 2^32. So long-living
+transactions often they are most likely defunct.
+
+Insert with 64-bit XIDs and 32-bit t_xmin/t_xmax
+------------------------------------------------
+
+On insert we check if current XID fits a range (5). Otherwise:
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base for t_xmin will
+not be over MaxShortTransactionId.
+
+- If it is impossible, then it will try to prune and freeze tuples on a page.
+
+Known issue: if pd_xid_base could not be shifted to accommodate a tuple being
+inserted due to a very long-running transaction, we just throw an error. We
+neither try to insert a�tuple into another page nor mark the current page as
+full. So, in this (unlikely) case we will get regular insert errors on the next
+tries to insert to the page 'locked' by this very long-running transaction.
+
+Upgrade from 32-bit XID versions
+--------------------------------
+
+pg_upgrade doesn't change pages format itself. It is done lazily after.
+
+1. At first heap page read, tuples on a page are repacked to free 16 bytes
+at the end of a page, possibly freeing space from dead tuples.
+
+2A. 16 bytes of pd_special is added if there is a place for it
+
+2B. Page is converted to "Double XMAX" format if there is no place for
+pd_special
+
+3. If a page is in double XMAX format after its first read, and vacuum (or
+micro-vacuum at select query) could prune some tuples and free space for
+pd_special, prune_page will add pd_special and convert page from double XMAX
+to general 64-bit XID page format.
-- 
2.24.3 (Apple Git-128)


--cpok4wp6gsarlzvp--





^ permalink  raw  reply  [nested|flat] 265+ messages in thread

* [PATCH v6] README for 64bit xid
@ 2022-01-10 19:20  Pavel Borisov <[email protected]>
  0 siblings, 0 replies; 265+ messages in thread

From: Pavel Borisov @ 2022-01-10 19:20 UTC (permalink / raw)

Authors:
- Pavel Borisov <[email protected]>
- Maxim Orlov <[email protected]>
- Yura Sokolov <[email protected]> <[email protected]>
---
 src/backend/access/heap/README.XID64 | 128 +++++++++++++++++++++++++++
 1 file changed, 128 insertions(+)
 create mode 100644 src/backend/access/heap/README.XID64

diff --git a/src/backend/access/heap/README.XID64 b/src/backend/access/heap/README.XID64
new file mode 100644
index 00000000000..457ba9b9ef5
--- /dev/null
+++ b/src/backend/access/heap/README.XID64
@@ -0,0 +1,128 @@
+src/backend/access/heap/README.XID64
+
+64-bit Transaction ID's (XID)
+=============================
+
+A limited number (N = 2^32) of XID's required to do vacuum freeze to prevent
+wraparound every N/2 transactions. This causes performance degradation due
+to the need to exclusively lock tables while being vacuumed. In each
+wraparound cycle, SLRU buffers are also being cut.
+
+With 64-bit XID's wraparound is effectively postponed to a very distant
+future. Even in highly loaded systems that had 2^32 transactions per day
+it will take huge 2^31 days before the first enforced "vacuum to prevent
+wraparound"). Buffers cutting and routine vacuum are not enforced, and DBA
+can plan them independently at the time with the least system load and least
+critical for database performance. Also, it can be done less frequently
+(several times a year vs every several days) on systems with transaction rates
+similar to those mentioned above.
+
+On-disk tuple and page format
+-----------------------------
+
+On-disk tuple format remains unchanged. 32-bit t_xmin and t_xmax store the
+lower parts of 64-bit XMIN and XMAX values. Each heap page has additional
+64-bit pd_xid_base and pd_multi_base which are common for all tuples on a page.
+They are placed into a pd_special area - 16 bytes in the end of a heap page.
+Actual XMIN/XMAX for a tuple are calculated upon reading a tuple from a page
+as follows:
+
+XMIN = t_xmin + pd_xid_base. 					(1)
+XMAX = t_xmax + pd_xid_base/pd_multi_base.		(2)
+
+"Double XMAX" page format
+---------------------------------
+
+At first read of a heap page after pg_upgrade from 32-bit XID PostgreSQL
+version pd_special area with a size of 16 bytes should be added to a page.
+Though a page may not have space for this. Then it can be converted to a
+temporary format called "double XMAX".
+
+All tuples after pg-upgrade would necessarily have xmin = FrozenTransactionId.
+So we don't need tuple header t_xmin field and we reuse t_xmin to store higher
+32 bits of its XMAX.
+
+Double XMAX format is only for full pages that don't have 16 bytes for
+pd_special. So it neither has a�place for a single tuple. Insert and HOT update
+for double XMAX pages is impossible and not supported. We can only read or
+delete tuples from it.
+
+When we are able to prune page double XMAX it will be converted from it to
+general 64-bit XID page format with all operations on its tuples supported.
+
+In-memory tuple format
+----------------------
+
+In-memory tuple representation consists of two parts:
+- HeapTupleHeader from disk page (contains all heap tuple contents, not only
+header)
+- HeapTuple with additional in-memory fields
+
+HeapTuple for each tuple in memory stores t_xid_base/t_multi_base - a copies of
+page's pd_xid_base/pd_multi_base. With tuple's 32-bit t_xmin and t_xmax from
+HeapTupleHeader they are used to calculate actual 64-bit XMIN and XMAX:
+
+XMIN = t_xmin + t_xid_base. 					(3)
+XMAX = t_xmax + t_xid_base/t_multi_base.		(4)
+
+The downside of this is that we can not use tuple's XMIN and XMAX right away.
+We often need to re-read t_xmin and t_xmax - which could actually be pointers
+into a page in shared buffers and therefore they could be updated by any other
+backend.
+
+Update/delete with 64-bit XIDs and 32-bit t_xmin/t_xmax
+--------------------------------------------------------------
+
+When we try to delete/update a tuple, we check that XMAX for a page fits (2).
+I.e. that t_xmax will not be over MaxShortTransactionId relative to
+pd_xid_base/pd_multi_base of a its page.
+
+If the current XID doesn't fit a range
+(pd_xid_base, pd_xid_base + MaxShortTransactionId) (5):
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base/pd_multi_base on
+a page and update all t_xmin/t_xmax of the other tuples on the page to
+correspond new pd_xid_base/pd_multi_base.
+
+- If it was impossible, it will try to prune and freeze tuples on a page.
+
+- If this is unsuccessful it will throw an error. Normally this is very
+unlikely but if there is a very old living transaction with an age of around
+2^32 this can arise. Basically, this is a behavior similar to one during the
+vacuum to prevent wraparound when XID was 32-bit. Dba should take care and
+avoid very-long-living transactions with an age close to 2^32. So long-living
+transactions often they are most likely defunct.
+
+Insert with 64-bit XIDs and 32-bit t_xmin/t_xmax
+------------------------------------------------
+
+On insert we check if current XID fits a range (5). Otherwise:
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base for t_xmin will
+not be over MaxShortTransactionId.
+
+- If it is impossible, then it will try to prune and freeze tuples on a page.
+
+Known issue: if pd_xid_base could not be shifted to accommodate a tuple being
+inserted due to a very long-running transaction, we just throw an error. We
+neither try to insert a�tuple into another page nor mark the current page as
+full. So, in this (unlikely) case we will get regular insert errors on the next
+tries to insert to the page 'locked' by this very long-running transaction.
+
+Upgrade from 32-bit XID versions
+--------------------------------
+
+pg_upgrade doesn't change pages format itself. It is done lazily after.
+
+1. At first heap page read, tuples on a page are repacked to free 16 bytes
+at the end of a page, possibly freeing space from dead tuples.
+
+2A. 16 bytes of pd_special is added if there is a place for it
+
+2B. Page is converted to "Double XMAX" format if there is no place for
+pd_special
+
+3. If a page is in double XMAX format after its first read, and vacuum (or
+micro-vacuum at select query) could prune some tuples and free space for
+pd_special, prune_page will add pd_special and convert page from double XMAX
+to general 64-bit XID page format.
-- 
2.24.3 (Apple Git-128)


--cpok4wp6gsarlzvp--





^ permalink  raw  reply  [nested|flat] 265+ messages in thread

* [PATCH v6] README for 64bit xid
@ 2022-01-10 19:20  Pavel Borisov <[email protected]>
  0 siblings, 0 replies; 265+ messages in thread

From: Pavel Borisov @ 2022-01-10 19:20 UTC (permalink / raw)

Authors:
- Pavel Borisov <[email protected]>
- Maxim Orlov <[email protected]>
- Yura Sokolov <[email protected]> <[email protected]>
---
 src/backend/access/heap/README.XID64 | 128 +++++++++++++++++++++++++++
 1 file changed, 128 insertions(+)
 create mode 100644 src/backend/access/heap/README.XID64

diff --git a/src/backend/access/heap/README.XID64 b/src/backend/access/heap/README.XID64
new file mode 100644
index 00000000000..457ba9b9ef5
--- /dev/null
+++ b/src/backend/access/heap/README.XID64
@@ -0,0 +1,128 @@
+src/backend/access/heap/README.XID64
+
+64-bit Transaction ID's (XID)
+=============================
+
+A limited number (N = 2^32) of XID's required to do vacuum freeze to prevent
+wraparound every N/2 transactions. This causes performance degradation due
+to the need to exclusively lock tables while being vacuumed. In each
+wraparound cycle, SLRU buffers are also being cut.
+
+With 64-bit XID's wraparound is effectively postponed to a very distant
+future. Even in highly loaded systems that had 2^32 transactions per day
+it will take huge 2^31 days before the first enforced "vacuum to prevent
+wraparound"). Buffers cutting and routine vacuum are not enforced, and DBA
+can plan them independently at the time with the least system load and least
+critical for database performance. Also, it can be done less frequently
+(several times a year vs every several days) on systems with transaction rates
+similar to those mentioned above.
+
+On-disk tuple and page format
+-----------------------------
+
+On-disk tuple format remains unchanged. 32-bit t_xmin and t_xmax store the
+lower parts of 64-bit XMIN and XMAX values. Each heap page has additional
+64-bit pd_xid_base and pd_multi_base which are common for all tuples on a page.
+They are placed into a pd_special area - 16 bytes in the end of a heap page.
+Actual XMIN/XMAX for a tuple are calculated upon reading a tuple from a page
+as follows:
+
+XMIN = t_xmin + pd_xid_base. 					(1)
+XMAX = t_xmax + pd_xid_base/pd_multi_base.		(2)
+
+"Double XMAX" page format
+---------------------------------
+
+At first read of a heap page after pg_upgrade from 32-bit XID PostgreSQL
+version pd_special area with a size of 16 bytes should be added to a page.
+Though a page may not have space for this. Then it can be converted to a
+temporary format called "double XMAX".
+
+All tuples after pg-upgrade would necessarily have xmin = FrozenTransactionId.
+So we don't need tuple header t_xmin field and we reuse t_xmin to store higher
+32 bits of its XMAX.
+
+Double XMAX format is only for full pages that don't have 16 bytes for
+pd_special. So it neither has a�place for a single tuple. Insert and HOT update
+for double XMAX pages is impossible and not supported. We can only read or
+delete tuples from it.
+
+When we are able to prune page double XMAX it will be converted from it to
+general 64-bit XID page format with all operations on its tuples supported.
+
+In-memory tuple format
+----------------------
+
+In-memory tuple representation consists of two parts:
+- HeapTupleHeader from disk page (contains all heap tuple contents, not only
+header)
+- HeapTuple with additional in-memory fields
+
+HeapTuple for each tuple in memory stores t_xid_base/t_multi_base - a copies of
+page's pd_xid_base/pd_multi_base. With tuple's 32-bit t_xmin and t_xmax from
+HeapTupleHeader they are used to calculate actual 64-bit XMIN and XMAX:
+
+XMIN = t_xmin + t_xid_base. 					(3)
+XMAX = t_xmax + t_xid_base/t_multi_base.		(4)
+
+The downside of this is that we can not use tuple's XMIN and XMAX right away.
+We often need to re-read t_xmin and t_xmax - which could actually be pointers
+into a page in shared buffers and therefore they could be updated by any other
+backend.
+
+Update/delete with 64-bit XIDs and 32-bit t_xmin/t_xmax
+--------------------------------------------------------------
+
+When we try to delete/update a tuple, we check that XMAX for a page fits (2).
+I.e. that t_xmax will not be over MaxShortTransactionId relative to
+pd_xid_base/pd_multi_base of a its page.
+
+If the current XID doesn't fit a range
+(pd_xid_base, pd_xid_base + MaxShortTransactionId) (5):
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base/pd_multi_base on
+a page and update all t_xmin/t_xmax of the other tuples on the page to
+correspond new pd_xid_base/pd_multi_base.
+
+- If it was impossible, it will try to prune and freeze tuples on a page.
+
+- If this is unsuccessful it will throw an error. Normally this is very
+unlikely but if there is a very old living transaction with an age of around
+2^32 this can arise. Basically, this is a behavior similar to one during the
+vacuum to prevent wraparound when XID was 32-bit. Dba should take care and
+avoid very-long-living transactions with an age close to 2^32. So long-living
+transactions often they are most likely defunct.
+
+Insert with 64-bit XIDs and 32-bit t_xmin/t_xmax
+------------------------------------------------
+
+On insert we check if current XID fits a range (5). Otherwise:
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base for t_xmin will
+not be over MaxShortTransactionId.
+
+- If it is impossible, then it will try to prune and freeze tuples on a page.
+
+Known issue: if pd_xid_base could not be shifted to accommodate a tuple being
+inserted due to a very long-running transaction, we just throw an error. We
+neither try to insert a�tuple into another page nor mark the current page as
+full. So, in this (unlikely) case we will get regular insert errors on the next
+tries to insert to the page 'locked' by this very long-running transaction.
+
+Upgrade from 32-bit XID versions
+--------------------------------
+
+pg_upgrade doesn't change pages format itself. It is done lazily after.
+
+1. At first heap page read, tuples on a page are repacked to free 16 bytes
+at the end of a page, possibly freeing space from dead tuples.
+
+2A. 16 bytes of pd_special is added if there is a place for it
+
+2B. Page is converted to "Double XMAX" format if there is no place for
+pd_special
+
+3. If a page is in double XMAX format after its first read, and vacuum (or
+micro-vacuum at select query) could prune some tuples and free space for
+pd_special, prune_page will add pd_special and convert page from double XMAX
+to general 64-bit XID page format.
-- 
2.24.3 (Apple Git-128)


--cpok4wp6gsarlzvp--





^ permalink  raw  reply  [nested|flat] 265+ messages in thread

* [PATCH v6] README for 64bit xid
@ 2022-01-10 19:20  Pavel Borisov <[email protected]>
  0 siblings, 0 replies; 265+ messages in thread

From: Pavel Borisov @ 2022-01-10 19:20 UTC (permalink / raw)

Authors:
- Pavel Borisov <[email protected]>
- Maxim Orlov <[email protected]>
- Yura Sokolov <[email protected]> <[email protected]>
---
 src/backend/access/heap/README.XID64 | 128 +++++++++++++++++++++++++++
 1 file changed, 128 insertions(+)
 create mode 100644 src/backend/access/heap/README.XID64

diff --git a/src/backend/access/heap/README.XID64 b/src/backend/access/heap/README.XID64
new file mode 100644
index 00000000000..457ba9b9ef5
--- /dev/null
+++ b/src/backend/access/heap/README.XID64
@@ -0,0 +1,128 @@
+src/backend/access/heap/README.XID64
+
+64-bit Transaction ID's (XID)
+=============================
+
+A limited number (N = 2^32) of XID's required to do vacuum freeze to prevent
+wraparound every N/2 transactions. This causes performance degradation due
+to the need to exclusively lock tables while being vacuumed. In each
+wraparound cycle, SLRU buffers are also being cut.
+
+With 64-bit XID's wraparound is effectively postponed to a very distant
+future. Even in highly loaded systems that had 2^32 transactions per day
+it will take huge 2^31 days before the first enforced "vacuum to prevent
+wraparound"). Buffers cutting and routine vacuum are not enforced, and DBA
+can plan them independently at the time with the least system load and least
+critical for database performance. Also, it can be done less frequently
+(several times a year vs every several days) on systems with transaction rates
+similar to those mentioned above.
+
+On-disk tuple and page format
+-----------------------------
+
+On-disk tuple format remains unchanged. 32-bit t_xmin and t_xmax store the
+lower parts of 64-bit XMIN and XMAX values. Each heap page has additional
+64-bit pd_xid_base and pd_multi_base which are common for all tuples on a page.
+They are placed into a pd_special area - 16 bytes in the end of a heap page.
+Actual XMIN/XMAX for a tuple are calculated upon reading a tuple from a page
+as follows:
+
+XMIN = t_xmin + pd_xid_base. 					(1)
+XMAX = t_xmax + pd_xid_base/pd_multi_base.		(2)
+
+"Double XMAX" page format
+---------------------------------
+
+At first read of a heap page after pg_upgrade from 32-bit XID PostgreSQL
+version pd_special area with a size of 16 bytes should be added to a page.
+Though a page may not have space for this. Then it can be converted to a
+temporary format called "double XMAX".
+
+All tuples after pg-upgrade would necessarily have xmin = FrozenTransactionId.
+So we don't need tuple header t_xmin field and we reuse t_xmin to store higher
+32 bits of its XMAX.
+
+Double XMAX format is only for full pages that don't have 16 bytes for
+pd_special. So it neither has a�place for a single tuple. Insert and HOT update
+for double XMAX pages is impossible and not supported. We can only read or
+delete tuples from it.
+
+When we are able to prune page double XMAX it will be converted from it to
+general 64-bit XID page format with all operations on its tuples supported.
+
+In-memory tuple format
+----------------------
+
+In-memory tuple representation consists of two parts:
+- HeapTupleHeader from disk page (contains all heap tuple contents, not only
+header)
+- HeapTuple with additional in-memory fields
+
+HeapTuple for each tuple in memory stores t_xid_base/t_multi_base - a copies of
+page's pd_xid_base/pd_multi_base. With tuple's 32-bit t_xmin and t_xmax from
+HeapTupleHeader they are used to calculate actual 64-bit XMIN and XMAX:
+
+XMIN = t_xmin + t_xid_base. 					(3)
+XMAX = t_xmax + t_xid_base/t_multi_base.		(4)
+
+The downside of this is that we can not use tuple's XMIN and XMAX right away.
+We often need to re-read t_xmin and t_xmax - which could actually be pointers
+into a page in shared buffers and therefore they could be updated by any other
+backend.
+
+Update/delete with 64-bit XIDs and 32-bit t_xmin/t_xmax
+--------------------------------------------------------------
+
+When we try to delete/update a tuple, we check that XMAX for a page fits (2).
+I.e. that t_xmax will not be over MaxShortTransactionId relative to
+pd_xid_base/pd_multi_base of a its page.
+
+If the current XID doesn't fit a range
+(pd_xid_base, pd_xid_base + MaxShortTransactionId) (5):
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base/pd_multi_base on
+a page and update all t_xmin/t_xmax of the other tuples on the page to
+correspond new pd_xid_base/pd_multi_base.
+
+- If it was impossible, it will try to prune and freeze tuples on a page.
+
+- If this is unsuccessful it will throw an error. Normally this is very
+unlikely but if there is a very old living transaction with an age of around
+2^32 this can arise. Basically, this is a behavior similar to one during the
+vacuum to prevent wraparound when XID was 32-bit. Dba should take care and
+avoid very-long-living transactions with an age close to 2^32. So long-living
+transactions often they are most likely defunct.
+
+Insert with 64-bit XIDs and 32-bit t_xmin/t_xmax
+------------------------------------------------
+
+On insert we check if current XID fits a range (5). Otherwise:
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base for t_xmin will
+not be over MaxShortTransactionId.
+
+- If it is impossible, then it will try to prune and freeze tuples on a page.
+
+Known issue: if pd_xid_base could not be shifted to accommodate a tuple being
+inserted due to a very long-running transaction, we just throw an error. We
+neither try to insert a�tuple into another page nor mark the current page as
+full. So, in this (unlikely) case we will get regular insert errors on the next
+tries to insert to the page 'locked' by this very long-running transaction.
+
+Upgrade from 32-bit XID versions
+--------------------------------
+
+pg_upgrade doesn't change pages format itself. It is done lazily after.
+
+1. At first heap page read, tuples on a page are repacked to free 16 bytes
+at the end of a page, possibly freeing space from dead tuples.
+
+2A. 16 bytes of pd_special is added if there is a place for it
+
+2B. Page is converted to "Double XMAX" format if there is no place for
+pd_special
+
+3. If a page is in double XMAX format after its first read, and vacuum (or
+micro-vacuum at select query) could prune some tuples and free space for
+pd_special, prune_page will add pd_special and convert page from double XMAX
+to general 64-bit XID page format.
-- 
2.24.3 (Apple Git-128)


--cpok4wp6gsarlzvp--





^ permalink  raw  reply  [nested|flat] 265+ messages in thread

* [PATCH v6] README for 64bit xid
@ 2022-01-10 19:20  Pavel Borisov <[email protected]>
  0 siblings, 0 replies; 265+ messages in thread

From: Pavel Borisov @ 2022-01-10 19:20 UTC (permalink / raw)

Authors:
- Pavel Borisov <[email protected]>
- Maxim Orlov <[email protected]>
- Yura Sokolov <[email protected]> <[email protected]>
---
 src/backend/access/heap/README.XID64 | 128 +++++++++++++++++++++++++++
 1 file changed, 128 insertions(+)
 create mode 100644 src/backend/access/heap/README.XID64

diff --git a/src/backend/access/heap/README.XID64 b/src/backend/access/heap/README.XID64
new file mode 100644
index 00000000000..457ba9b9ef5
--- /dev/null
+++ b/src/backend/access/heap/README.XID64
@@ -0,0 +1,128 @@
+src/backend/access/heap/README.XID64
+
+64-bit Transaction ID's (XID)
+=============================
+
+A limited number (N = 2^32) of XID's required to do vacuum freeze to prevent
+wraparound every N/2 transactions. This causes performance degradation due
+to the need to exclusively lock tables while being vacuumed. In each
+wraparound cycle, SLRU buffers are also being cut.
+
+With 64-bit XID's wraparound is effectively postponed to a very distant
+future. Even in highly loaded systems that had 2^32 transactions per day
+it will take huge 2^31 days before the first enforced "vacuum to prevent
+wraparound"). Buffers cutting and routine vacuum are not enforced, and DBA
+can plan them independently at the time with the least system load and least
+critical for database performance. Also, it can be done less frequently
+(several times a year vs every several days) on systems with transaction rates
+similar to those mentioned above.
+
+On-disk tuple and page format
+-----------------------------
+
+On-disk tuple format remains unchanged. 32-bit t_xmin and t_xmax store the
+lower parts of 64-bit XMIN and XMAX values. Each heap page has additional
+64-bit pd_xid_base and pd_multi_base which are common for all tuples on a page.
+They are placed into a pd_special area - 16 bytes in the end of a heap page.
+Actual XMIN/XMAX for a tuple are calculated upon reading a tuple from a page
+as follows:
+
+XMIN = t_xmin + pd_xid_base. 					(1)
+XMAX = t_xmax + pd_xid_base/pd_multi_base.		(2)
+
+"Double XMAX" page format
+---------------------------------
+
+At first read of a heap page after pg_upgrade from 32-bit XID PostgreSQL
+version pd_special area with a size of 16 bytes should be added to a page.
+Though a page may not have space for this. Then it can be converted to a
+temporary format called "double XMAX".
+
+All tuples after pg-upgrade would necessarily have xmin = FrozenTransactionId.
+So we don't need tuple header t_xmin field and we reuse t_xmin to store higher
+32 bits of its XMAX.
+
+Double XMAX format is only for full pages that don't have 16 bytes for
+pd_special. So it neither has a�place for a single tuple. Insert and HOT update
+for double XMAX pages is impossible and not supported. We can only read or
+delete tuples from it.
+
+When we are able to prune page double XMAX it will be converted from it to
+general 64-bit XID page format with all operations on its tuples supported.
+
+In-memory tuple format
+----------------------
+
+In-memory tuple representation consists of two parts:
+- HeapTupleHeader from disk page (contains all heap tuple contents, not only
+header)
+- HeapTuple with additional in-memory fields
+
+HeapTuple for each tuple in memory stores t_xid_base/t_multi_base - a copies of
+page's pd_xid_base/pd_multi_base. With tuple's 32-bit t_xmin and t_xmax from
+HeapTupleHeader they are used to calculate actual 64-bit XMIN and XMAX:
+
+XMIN = t_xmin + t_xid_base. 					(3)
+XMAX = t_xmax + t_xid_base/t_multi_base.		(4)
+
+The downside of this is that we can not use tuple's XMIN and XMAX right away.
+We often need to re-read t_xmin and t_xmax - which could actually be pointers
+into a page in shared buffers and therefore they could be updated by any other
+backend.
+
+Update/delete with 64-bit XIDs and 32-bit t_xmin/t_xmax
+--------------------------------------------------------------
+
+When we try to delete/update a tuple, we check that XMAX for a page fits (2).
+I.e. that t_xmax will not be over MaxShortTransactionId relative to
+pd_xid_base/pd_multi_base of a its page.
+
+If the current XID doesn't fit a range
+(pd_xid_base, pd_xid_base + MaxShortTransactionId) (5):
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base/pd_multi_base on
+a page and update all t_xmin/t_xmax of the other tuples on the page to
+correspond new pd_xid_base/pd_multi_base.
+
+- If it was impossible, it will try to prune and freeze tuples on a page.
+
+- If this is unsuccessful it will throw an error. Normally this is very
+unlikely but if there is a very old living transaction with an age of around
+2^32 this can arise. Basically, this is a behavior similar to one during the
+vacuum to prevent wraparound when XID was 32-bit. Dba should take care and
+avoid very-long-living transactions with an age close to 2^32. So long-living
+transactions often they are most likely defunct.
+
+Insert with 64-bit XIDs and 32-bit t_xmin/t_xmax
+------------------------------------------------
+
+On insert we check if current XID fits a range (5). Otherwise:
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base for t_xmin will
+not be over MaxShortTransactionId.
+
+- If it is impossible, then it will try to prune and freeze tuples on a page.
+
+Known issue: if pd_xid_base could not be shifted to accommodate a tuple being
+inserted due to a very long-running transaction, we just throw an error. We
+neither try to insert a�tuple into another page nor mark the current page as
+full. So, in this (unlikely) case we will get regular insert errors on the next
+tries to insert to the page 'locked' by this very long-running transaction.
+
+Upgrade from 32-bit XID versions
+--------------------------------
+
+pg_upgrade doesn't change pages format itself. It is done lazily after.
+
+1. At first heap page read, tuples on a page are repacked to free 16 bytes
+at the end of a page, possibly freeing space from dead tuples.
+
+2A. 16 bytes of pd_special is added if there is a place for it
+
+2B. Page is converted to "Double XMAX" format if there is no place for
+pd_special
+
+3. If a page is in double XMAX format after its first read, and vacuum (or
+micro-vacuum at select query) could prune some tuples and free space for
+pd_special, prune_page will add pd_special and convert page from double XMAX
+to general 64-bit XID page format.
-- 
2.24.3 (Apple Git-128)


--cpok4wp6gsarlzvp--





^ permalink  raw  reply  [nested|flat] 265+ messages in thread

* [PATCH v6] README for 64bit xid
@ 2022-01-10 19:20  Pavel Borisov <[email protected]>
  0 siblings, 0 replies; 265+ messages in thread

From: Pavel Borisov @ 2022-01-10 19:20 UTC (permalink / raw)

Authors:
- Pavel Borisov <[email protected]>
- Maxim Orlov <[email protected]>
- Yura Sokolov <[email protected]> <[email protected]>
---
 src/backend/access/heap/README.XID64 | 128 +++++++++++++++++++++++++++
 1 file changed, 128 insertions(+)
 create mode 100644 src/backend/access/heap/README.XID64

diff --git a/src/backend/access/heap/README.XID64 b/src/backend/access/heap/README.XID64
new file mode 100644
index 00000000000..457ba9b9ef5
--- /dev/null
+++ b/src/backend/access/heap/README.XID64
@@ -0,0 +1,128 @@
+src/backend/access/heap/README.XID64
+
+64-bit Transaction ID's (XID)
+=============================
+
+A limited number (N = 2^32) of XID's required to do vacuum freeze to prevent
+wraparound every N/2 transactions. This causes performance degradation due
+to the need to exclusively lock tables while being vacuumed. In each
+wraparound cycle, SLRU buffers are also being cut.
+
+With 64-bit XID's wraparound is effectively postponed to a very distant
+future. Even in highly loaded systems that had 2^32 transactions per day
+it will take huge 2^31 days before the first enforced "vacuum to prevent
+wraparound"). Buffers cutting and routine vacuum are not enforced, and DBA
+can plan them independently at the time with the least system load and least
+critical for database performance. Also, it can be done less frequently
+(several times a year vs every several days) on systems with transaction rates
+similar to those mentioned above.
+
+On-disk tuple and page format
+-----------------------------
+
+On-disk tuple format remains unchanged. 32-bit t_xmin and t_xmax store the
+lower parts of 64-bit XMIN and XMAX values. Each heap page has additional
+64-bit pd_xid_base and pd_multi_base which are common for all tuples on a page.
+They are placed into a pd_special area - 16 bytes in the end of a heap page.
+Actual XMIN/XMAX for a tuple are calculated upon reading a tuple from a page
+as follows:
+
+XMIN = t_xmin + pd_xid_base. 					(1)
+XMAX = t_xmax + pd_xid_base/pd_multi_base.		(2)
+
+"Double XMAX" page format
+---------------------------------
+
+At first read of a heap page after pg_upgrade from 32-bit XID PostgreSQL
+version pd_special area with a size of 16 bytes should be added to a page.
+Though a page may not have space for this. Then it can be converted to a
+temporary format called "double XMAX".
+
+All tuples after pg-upgrade would necessarily have xmin = FrozenTransactionId.
+So we don't need tuple header t_xmin field and we reuse t_xmin to store higher
+32 bits of its XMAX.
+
+Double XMAX format is only for full pages that don't have 16 bytes for
+pd_special. So it neither has a�place for a single tuple. Insert and HOT update
+for double XMAX pages is impossible and not supported. We can only read or
+delete tuples from it.
+
+When we are able to prune page double XMAX it will be converted from it to
+general 64-bit XID page format with all operations on its tuples supported.
+
+In-memory tuple format
+----------------------
+
+In-memory tuple representation consists of two parts:
+- HeapTupleHeader from disk page (contains all heap tuple contents, not only
+header)
+- HeapTuple with additional in-memory fields
+
+HeapTuple for each tuple in memory stores t_xid_base/t_multi_base - a copies of
+page's pd_xid_base/pd_multi_base. With tuple's 32-bit t_xmin and t_xmax from
+HeapTupleHeader they are used to calculate actual 64-bit XMIN and XMAX:
+
+XMIN = t_xmin + t_xid_base. 					(3)
+XMAX = t_xmax + t_xid_base/t_multi_base.		(4)
+
+The downside of this is that we can not use tuple's XMIN and XMAX right away.
+We often need to re-read t_xmin and t_xmax - which could actually be pointers
+into a page in shared buffers and therefore they could be updated by any other
+backend.
+
+Update/delete with 64-bit XIDs and 32-bit t_xmin/t_xmax
+--------------------------------------------------------------
+
+When we try to delete/update a tuple, we check that XMAX for a page fits (2).
+I.e. that t_xmax will not be over MaxShortTransactionId relative to
+pd_xid_base/pd_multi_base of a its page.
+
+If the current XID doesn't fit a range
+(pd_xid_base, pd_xid_base + MaxShortTransactionId) (5):
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base/pd_multi_base on
+a page and update all t_xmin/t_xmax of the other tuples on the page to
+correspond new pd_xid_base/pd_multi_base.
+
+- If it was impossible, it will try to prune and freeze tuples on a page.
+
+- If this is unsuccessful it will throw an error. Normally this is very
+unlikely but if there is a very old living transaction with an age of around
+2^32 this can arise. Basically, this is a behavior similar to one during the
+vacuum to prevent wraparound when XID was 32-bit. Dba should take care and
+avoid very-long-living transactions with an age close to 2^32. So long-living
+transactions often they are most likely defunct.
+
+Insert with 64-bit XIDs and 32-bit t_xmin/t_xmax
+------------------------------------------------
+
+On insert we check if current XID fits a range (5). Otherwise:
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base for t_xmin will
+not be over MaxShortTransactionId.
+
+- If it is impossible, then it will try to prune and freeze tuples on a page.
+
+Known issue: if pd_xid_base could not be shifted to accommodate a tuple being
+inserted due to a very long-running transaction, we just throw an error. We
+neither try to insert a�tuple into another page nor mark the current page as
+full. So, in this (unlikely) case we will get regular insert errors on the next
+tries to insert to the page 'locked' by this very long-running transaction.
+
+Upgrade from 32-bit XID versions
+--------------------------------
+
+pg_upgrade doesn't change pages format itself. It is done lazily after.
+
+1. At first heap page read, tuples on a page are repacked to free 16 bytes
+at the end of a page, possibly freeing space from dead tuples.
+
+2A. 16 bytes of pd_special is added if there is a place for it
+
+2B. Page is converted to "Double XMAX" format if there is no place for
+pd_special
+
+3. If a page is in double XMAX format after its first read, and vacuum (or
+micro-vacuum at select query) could prune some tuples and free space for
+pd_special, prune_page will add pd_special and convert page from double XMAX
+to general 64-bit XID page format.
-- 
2.24.3 (Apple Git-128)


--cpok4wp6gsarlzvp--





^ permalink  raw  reply  [nested|flat] 265+ messages in thread

* [PATCH v6] README for 64bit xid
@ 2022-01-10 19:20  Pavel Borisov <[email protected]>
  0 siblings, 0 replies; 265+ messages in thread

From: Pavel Borisov @ 2022-01-10 19:20 UTC (permalink / raw)

Authors:
- Pavel Borisov <[email protected]>
- Maxim Orlov <[email protected]>
- Yura Sokolov <[email protected]> <[email protected]>
---
 src/backend/access/heap/README.XID64 | 128 +++++++++++++++++++++++++++
 1 file changed, 128 insertions(+)
 create mode 100644 src/backend/access/heap/README.XID64

diff --git a/src/backend/access/heap/README.XID64 b/src/backend/access/heap/README.XID64
new file mode 100644
index 00000000000..457ba9b9ef5
--- /dev/null
+++ b/src/backend/access/heap/README.XID64
@@ -0,0 +1,128 @@
+src/backend/access/heap/README.XID64
+
+64-bit Transaction ID's (XID)
+=============================
+
+A limited number (N = 2^32) of XID's required to do vacuum freeze to prevent
+wraparound every N/2 transactions. This causes performance degradation due
+to the need to exclusively lock tables while being vacuumed. In each
+wraparound cycle, SLRU buffers are also being cut.
+
+With 64-bit XID's wraparound is effectively postponed to a very distant
+future. Even in highly loaded systems that had 2^32 transactions per day
+it will take huge 2^31 days before the first enforced "vacuum to prevent
+wraparound"). Buffers cutting and routine vacuum are not enforced, and DBA
+can plan them independently at the time with the least system load and least
+critical for database performance. Also, it can be done less frequently
+(several times a year vs every several days) on systems with transaction rates
+similar to those mentioned above.
+
+On-disk tuple and page format
+-----------------------------
+
+On-disk tuple format remains unchanged. 32-bit t_xmin and t_xmax store the
+lower parts of 64-bit XMIN and XMAX values. Each heap page has additional
+64-bit pd_xid_base and pd_multi_base which are common for all tuples on a page.
+They are placed into a pd_special area - 16 bytes in the end of a heap page.
+Actual XMIN/XMAX for a tuple are calculated upon reading a tuple from a page
+as follows:
+
+XMIN = t_xmin + pd_xid_base. 					(1)
+XMAX = t_xmax + pd_xid_base/pd_multi_base.		(2)
+
+"Double XMAX" page format
+---------------------------------
+
+At first read of a heap page after pg_upgrade from 32-bit XID PostgreSQL
+version pd_special area with a size of 16 bytes should be added to a page.
+Though a page may not have space for this. Then it can be converted to a
+temporary format called "double XMAX".
+
+All tuples after pg-upgrade would necessarily have xmin = FrozenTransactionId.
+So we don't need tuple header t_xmin field and we reuse t_xmin to store higher
+32 bits of its XMAX.
+
+Double XMAX format is only for full pages that don't have 16 bytes for
+pd_special. So it neither has a�place for a single tuple. Insert and HOT update
+for double XMAX pages is impossible and not supported. We can only read or
+delete tuples from it.
+
+When we are able to prune page double XMAX it will be converted from it to
+general 64-bit XID page format with all operations on its tuples supported.
+
+In-memory tuple format
+----------------------
+
+In-memory tuple representation consists of two parts:
+- HeapTupleHeader from disk page (contains all heap tuple contents, not only
+header)
+- HeapTuple with additional in-memory fields
+
+HeapTuple for each tuple in memory stores t_xid_base/t_multi_base - a copies of
+page's pd_xid_base/pd_multi_base. With tuple's 32-bit t_xmin and t_xmax from
+HeapTupleHeader they are used to calculate actual 64-bit XMIN and XMAX:
+
+XMIN = t_xmin + t_xid_base. 					(3)
+XMAX = t_xmax + t_xid_base/t_multi_base.		(4)
+
+The downside of this is that we can not use tuple's XMIN and XMAX right away.
+We often need to re-read t_xmin and t_xmax - which could actually be pointers
+into a page in shared buffers and therefore they could be updated by any other
+backend.
+
+Update/delete with 64-bit XIDs and 32-bit t_xmin/t_xmax
+--------------------------------------------------------------
+
+When we try to delete/update a tuple, we check that XMAX for a page fits (2).
+I.e. that t_xmax will not be over MaxShortTransactionId relative to
+pd_xid_base/pd_multi_base of a its page.
+
+If the current XID doesn't fit a range
+(pd_xid_base, pd_xid_base + MaxShortTransactionId) (5):
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base/pd_multi_base on
+a page and update all t_xmin/t_xmax of the other tuples on the page to
+correspond new pd_xid_base/pd_multi_base.
+
+- If it was impossible, it will try to prune and freeze tuples on a page.
+
+- If this is unsuccessful it will throw an error. Normally this is very
+unlikely but if there is a very old living transaction with an age of around
+2^32 this can arise. Basically, this is a behavior similar to one during the
+vacuum to prevent wraparound when XID was 32-bit. Dba should take care and
+avoid very-long-living transactions with an age close to 2^32. So long-living
+transactions often they are most likely defunct.
+
+Insert with 64-bit XIDs and 32-bit t_xmin/t_xmax
+------------------------------------------------
+
+On insert we check if current XID fits a range (5). Otherwise:
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base for t_xmin will
+not be over MaxShortTransactionId.
+
+- If it is impossible, then it will try to prune and freeze tuples on a page.
+
+Known issue: if pd_xid_base could not be shifted to accommodate a tuple being
+inserted due to a very long-running transaction, we just throw an error. We
+neither try to insert a�tuple into another page nor mark the current page as
+full. So, in this (unlikely) case we will get regular insert errors on the next
+tries to insert to the page 'locked' by this very long-running transaction.
+
+Upgrade from 32-bit XID versions
+--------------------------------
+
+pg_upgrade doesn't change pages format itself. It is done lazily after.
+
+1. At first heap page read, tuples on a page are repacked to free 16 bytes
+at the end of a page, possibly freeing space from dead tuples.
+
+2A. 16 bytes of pd_special is added if there is a place for it
+
+2B. Page is converted to "Double XMAX" format if there is no place for
+pd_special
+
+3. If a page is in double XMAX format after its first read, and vacuum (or
+micro-vacuum at select query) could prune some tuples and free space for
+pd_special, prune_page will add pd_special and convert page from double XMAX
+to general 64-bit XID page format.
-- 
2.24.3 (Apple Git-128)


--cpok4wp6gsarlzvp--





^ permalink  raw  reply  [nested|flat] 265+ messages in thread

* [PATCH v6] README for 64bit xid
@ 2022-01-10 19:20  Pavel Borisov <[email protected]>
  0 siblings, 0 replies; 265+ messages in thread

From: Pavel Borisov @ 2022-01-10 19:20 UTC (permalink / raw)

Authors:
- Pavel Borisov <[email protected]>
- Maxim Orlov <[email protected]>
- Yura Sokolov <[email protected]> <[email protected]>
---
 src/backend/access/heap/README.XID64 | 128 +++++++++++++++++++++++++++
 1 file changed, 128 insertions(+)
 create mode 100644 src/backend/access/heap/README.XID64

diff --git a/src/backend/access/heap/README.XID64 b/src/backend/access/heap/README.XID64
new file mode 100644
index 00000000000..457ba9b9ef5
--- /dev/null
+++ b/src/backend/access/heap/README.XID64
@@ -0,0 +1,128 @@
+src/backend/access/heap/README.XID64
+
+64-bit Transaction ID's (XID)
+=============================
+
+A limited number (N = 2^32) of XID's required to do vacuum freeze to prevent
+wraparound every N/2 transactions. This causes performance degradation due
+to the need to exclusively lock tables while being vacuumed. In each
+wraparound cycle, SLRU buffers are also being cut.
+
+With 64-bit XID's wraparound is effectively postponed to a very distant
+future. Even in highly loaded systems that had 2^32 transactions per day
+it will take huge 2^31 days before the first enforced "vacuum to prevent
+wraparound"). Buffers cutting and routine vacuum are not enforced, and DBA
+can plan them independently at the time with the least system load and least
+critical for database performance. Also, it can be done less frequently
+(several times a year vs every several days) on systems with transaction rates
+similar to those mentioned above.
+
+On-disk tuple and page format
+-----------------------------
+
+On-disk tuple format remains unchanged. 32-bit t_xmin and t_xmax store the
+lower parts of 64-bit XMIN and XMAX values. Each heap page has additional
+64-bit pd_xid_base and pd_multi_base which are common for all tuples on a page.
+They are placed into a pd_special area - 16 bytes in the end of a heap page.
+Actual XMIN/XMAX for a tuple are calculated upon reading a tuple from a page
+as follows:
+
+XMIN = t_xmin + pd_xid_base. 					(1)
+XMAX = t_xmax + pd_xid_base/pd_multi_base.		(2)
+
+"Double XMAX" page format
+---------------------------------
+
+At first read of a heap page after pg_upgrade from 32-bit XID PostgreSQL
+version pd_special area with a size of 16 bytes should be added to a page.
+Though a page may not have space for this. Then it can be converted to a
+temporary format called "double XMAX".
+
+All tuples after pg-upgrade would necessarily have xmin = FrozenTransactionId.
+So we don't need tuple header t_xmin field and we reuse t_xmin to store higher
+32 bits of its XMAX.
+
+Double XMAX format is only for full pages that don't have 16 bytes for
+pd_special. So it neither has a�place for a single tuple. Insert and HOT update
+for double XMAX pages is impossible and not supported. We can only read or
+delete tuples from it.
+
+When we are able to prune page double XMAX it will be converted from it to
+general 64-bit XID page format with all operations on its tuples supported.
+
+In-memory tuple format
+----------------------
+
+In-memory tuple representation consists of two parts:
+- HeapTupleHeader from disk page (contains all heap tuple contents, not only
+header)
+- HeapTuple with additional in-memory fields
+
+HeapTuple for each tuple in memory stores t_xid_base/t_multi_base - a copies of
+page's pd_xid_base/pd_multi_base. With tuple's 32-bit t_xmin and t_xmax from
+HeapTupleHeader they are used to calculate actual 64-bit XMIN and XMAX:
+
+XMIN = t_xmin + t_xid_base. 					(3)
+XMAX = t_xmax + t_xid_base/t_multi_base.		(4)
+
+The downside of this is that we can not use tuple's XMIN and XMAX right away.
+We often need to re-read t_xmin and t_xmax - which could actually be pointers
+into a page in shared buffers and therefore they could be updated by any other
+backend.
+
+Update/delete with 64-bit XIDs and 32-bit t_xmin/t_xmax
+--------------------------------------------------------------
+
+When we try to delete/update a tuple, we check that XMAX for a page fits (2).
+I.e. that t_xmax will not be over MaxShortTransactionId relative to
+pd_xid_base/pd_multi_base of a its page.
+
+If the current XID doesn't fit a range
+(pd_xid_base, pd_xid_base + MaxShortTransactionId) (5):
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base/pd_multi_base on
+a page and update all t_xmin/t_xmax of the other tuples on the page to
+correspond new pd_xid_base/pd_multi_base.
+
+- If it was impossible, it will try to prune and freeze tuples on a page.
+
+- If this is unsuccessful it will throw an error. Normally this is very
+unlikely but if there is a very old living transaction with an age of around
+2^32 this can arise. Basically, this is a behavior similar to one during the
+vacuum to prevent wraparound when XID was 32-bit. Dba should take care and
+avoid very-long-living transactions with an age close to 2^32. So long-living
+transactions often they are most likely defunct.
+
+Insert with 64-bit XIDs and 32-bit t_xmin/t_xmax
+------------------------------------------------
+
+On insert we check if current XID fits a range (5). Otherwise:
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base for t_xmin will
+not be over MaxShortTransactionId.
+
+- If it is impossible, then it will try to prune and freeze tuples on a page.
+
+Known issue: if pd_xid_base could not be shifted to accommodate a tuple being
+inserted due to a very long-running transaction, we just throw an error. We
+neither try to insert a�tuple into another page nor mark the current page as
+full. So, in this (unlikely) case we will get regular insert errors on the next
+tries to insert to the page 'locked' by this very long-running transaction.
+
+Upgrade from 32-bit XID versions
+--------------------------------
+
+pg_upgrade doesn't change pages format itself. It is done lazily after.
+
+1. At first heap page read, tuples on a page are repacked to free 16 bytes
+at the end of a page, possibly freeing space from dead tuples.
+
+2A. 16 bytes of pd_special is added if there is a place for it
+
+2B. Page is converted to "Double XMAX" format if there is no place for
+pd_special
+
+3. If a page is in double XMAX format after its first read, and vacuum (or
+micro-vacuum at select query) could prune some tuples and free space for
+pd_special, prune_page will add pd_special and convert page from double XMAX
+to general 64-bit XID page format.
-- 
2.24.3 (Apple Git-128)


--cpok4wp6gsarlzvp--





^ permalink  raw  reply  [nested|flat] 265+ messages in thread

* [PATCH v6] README for 64bit xid
@ 2022-01-10 19:20  Pavel Borisov <[email protected]>
  0 siblings, 0 replies; 265+ messages in thread

From: Pavel Borisov @ 2022-01-10 19:20 UTC (permalink / raw)

Authors:
- Pavel Borisov <[email protected]>
- Maxim Orlov <[email protected]>
- Yura Sokolov <[email protected]> <[email protected]>
---
 src/backend/access/heap/README.XID64 | 128 +++++++++++++++++++++++++++
 1 file changed, 128 insertions(+)
 create mode 100644 src/backend/access/heap/README.XID64

diff --git a/src/backend/access/heap/README.XID64 b/src/backend/access/heap/README.XID64
new file mode 100644
index 00000000000..457ba9b9ef5
--- /dev/null
+++ b/src/backend/access/heap/README.XID64
@@ -0,0 +1,128 @@
+src/backend/access/heap/README.XID64
+
+64-bit Transaction ID's (XID)
+=============================
+
+A limited number (N = 2^32) of XID's required to do vacuum freeze to prevent
+wraparound every N/2 transactions. This causes performance degradation due
+to the need to exclusively lock tables while being vacuumed. In each
+wraparound cycle, SLRU buffers are also being cut.
+
+With 64-bit XID's wraparound is effectively postponed to a very distant
+future. Even in highly loaded systems that had 2^32 transactions per day
+it will take huge 2^31 days before the first enforced "vacuum to prevent
+wraparound"). Buffers cutting and routine vacuum are not enforced, and DBA
+can plan them independently at the time with the least system load and least
+critical for database performance. Also, it can be done less frequently
+(several times a year vs every several days) on systems with transaction rates
+similar to those mentioned above.
+
+On-disk tuple and page format
+-----------------------------
+
+On-disk tuple format remains unchanged. 32-bit t_xmin and t_xmax store the
+lower parts of 64-bit XMIN and XMAX values. Each heap page has additional
+64-bit pd_xid_base and pd_multi_base which are common for all tuples on a page.
+They are placed into a pd_special area - 16 bytes in the end of a heap page.
+Actual XMIN/XMAX for a tuple are calculated upon reading a tuple from a page
+as follows:
+
+XMIN = t_xmin + pd_xid_base. 					(1)
+XMAX = t_xmax + pd_xid_base/pd_multi_base.		(2)
+
+"Double XMAX" page format
+---------------------------------
+
+At first read of a heap page after pg_upgrade from 32-bit XID PostgreSQL
+version pd_special area with a size of 16 bytes should be added to a page.
+Though a page may not have space for this. Then it can be converted to a
+temporary format called "double XMAX".
+
+All tuples after pg-upgrade would necessarily have xmin = FrozenTransactionId.
+So we don't need tuple header t_xmin field and we reuse t_xmin to store higher
+32 bits of its XMAX.
+
+Double XMAX format is only for full pages that don't have 16 bytes for
+pd_special. So it neither has a�place for a single tuple. Insert and HOT update
+for double XMAX pages is impossible and not supported. We can only read or
+delete tuples from it.
+
+When we are able to prune page double XMAX it will be converted from it to
+general 64-bit XID page format with all operations on its tuples supported.
+
+In-memory tuple format
+----------------------
+
+In-memory tuple representation consists of two parts:
+- HeapTupleHeader from disk page (contains all heap tuple contents, not only
+header)
+- HeapTuple with additional in-memory fields
+
+HeapTuple for each tuple in memory stores t_xid_base/t_multi_base - a copies of
+page's pd_xid_base/pd_multi_base. With tuple's 32-bit t_xmin and t_xmax from
+HeapTupleHeader they are used to calculate actual 64-bit XMIN and XMAX:
+
+XMIN = t_xmin + t_xid_base. 					(3)
+XMAX = t_xmax + t_xid_base/t_multi_base.		(4)
+
+The downside of this is that we can not use tuple's XMIN and XMAX right away.
+We often need to re-read t_xmin and t_xmax - which could actually be pointers
+into a page in shared buffers and therefore they could be updated by any other
+backend.
+
+Update/delete with 64-bit XIDs and 32-bit t_xmin/t_xmax
+--------------------------------------------------------------
+
+When we try to delete/update a tuple, we check that XMAX for a page fits (2).
+I.e. that t_xmax will not be over MaxShortTransactionId relative to
+pd_xid_base/pd_multi_base of a its page.
+
+If the current XID doesn't fit a range
+(pd_xid_base, pd_xid_base + MaxShortTransactionId) (5):
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base/pd_multi_base on
+a page and update all t_xmin/t_xmax of the other tuples on the page to
+correspond new pd_xid_base/pd_multi_base.
+
+- If it was impossible, it will try to prune and freeze tuples on a page.
+
+- If this is unsuccessful it will throw an error. Normally this is very
+unlikely but if there is a very old living transaction with an age of around
+2^32 this can arise. Basically, this is a behavior similar to one during the
+vacuum to prevent wraparound when XID was 32-bit. Dba should take care and
+avoid very-long-living transactions with an age close to 2^32. So long-living
+transactions often they are most likely defunct.
+
+Insert with 64-bit XIDs and 32-bit t_xmin/t_xmax
+------------------------------------------------
+
+On insert we check if current XID fits a range (5). Otherwise:
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base for t_xmin will
+not be over MaxShortTransactionId.
+
+- If it is impossible, then it will try to prune and freeze tuples on a page.
+
+Known issue: if pd_xid_base could not be shifted to accommodate a tuple being
+inserted due to a very long-running transaction, we just throw an error. We
+neither try to insert a�tuple into another page nor mark the current page as
+full. So, in this (unlikely) case we will get regular insert errors on the next
+tries to insert to the page 'locked' by this very long-running transaction.
+
+Upgrade from 32-bit XID versions
+--------------------------------
+
+pg_upgrade doesn't change pages format itself. It is done lazily after.
+
+1. At first heap page read, tuples on a page are repacked to free 16 bytes
+at the end of a page, possibly freeing space from dead tuples.
+
+2A. 16 bytes of pd_special is added if there is a place for it
+
+2B. Page is converted to "Double XMAX" format if there is no place for
+pd_special
+
+3. If a page is in double XMAX format after its first read, and vacuum (or
+micro-vacuum at select query) could prune some tuples and free space for
+pd_special, prune_page will add pd_special and convert page from double XMAX
+to general 64-bit XID page format.
-- 
2.24.3 (Apple Git-128)


--cpok4wp6gsarlzvp--





^ permalink  raw  reply  [nested|flat] 265+ messages in thread

* [PATCH v6] README for 64bit xid
@ 2022-01-10 19:20  Pavel Borisov <[email protected]>
  0 siblings, 0 replies; 265+ messages in thread

From: Pavel Borisov @ 2022-01-10 19:20 UTC (permalink / raw)

Authors:
- Pavel Borisov <[email protected]>
- Maxim Orlov <[email protected]>
- Yura Sokolov <[email protected]> <[email protected]>
---
 src/backend/access/heap/README.XID64 | 128 +++++++++++++++++++++++++++
 1 file changed, 128 insertions(+)
 create mode 100644 src/backend/access/heap/README.XID64

diff --git a/src/backend/access/heap/README.XID64 b/src/backend/access/heap/README.XID64
new file mode 100644
index 00000000000..457ba9b9ef5
--- /dev/null
+++ b/src/backend/access/heap/README.XID64
@@ -0,0 +1,128 @@
+src/backend/access/heap/README.XID64
+
+64-bit Transaction ID's (XID)
+=============================
+
+A limited number (N = 2^32) of XID's required to do vacuum freeze to prevent
+wraparound every N/2 transactions. This causes performance degradation due
+to the need to exclusively lock tables while being vacuumed. In each
+wraparound cycle, SLRU buffers are also being cut.
+
+With 64-bit XID's wraparound is effectively postponed to a very distant
+future. Even in highly loaded systems that had 2^32 transactions per day
+it will take huge 2^31 days before the first enforced "vacuum to prevent
+wraparound"). Buffers cutting and routine vacuum are not enforced, and DBA
+can plan them independently at the time with the least system load and least
+critical for database performance. Also, it can be done less frequently
+(several times a year vs every several days) on systems with transaction rates
+similar to those mentioned above.
+
+On-disk tuple and page format
+-----------------------------
+
+On-disk tuple format remains unchanged. 32-bit t_xmin and t_xmax store the
+lower parts of 64-bit XMIN and XMAX values. Each heap page has additional
+64-bit pd_xid_base and pd_multi_base which are common for all tuples on a page.
+They are placed into a pd_special area - 16 bytes in the end of a heap page.
+Actual XMIN/XMAX for a tuple are calculated upon reading a tuple from a page
+as follows:
+
+XMIN = t_xmin + pd_xid_base. 					(1)
+XMAX = t_xmax + pd_xid_base/pd_multi_base.		(2)
+
+"Double XMAX" page format
+---------------------------------
+
+At first read of a heap page after pg_upgrade from 32-bit XID PostgreSQL
+version pd_special area with a size of 16 bytes should be added to a page.
+Though a page may not have space for this. Then it can be converted to a
+temporary format called "double XMAX".
+
+All tuples after pg-upgrade would necessarily have xmin = FrozenTransactionId.
+So we don't need tuple header t_xmin field and we reuse t_xmin to store higher
+32 bits of its XMAX.
+
+Double XMAX format is only for full pages that don't have 16 bytes for
+pd_special. So it neither has a�place for a single tuple. Insert and HOT update
+for double XMAX pages is impossible and not supported. We can only read or
+delete tuples from it.
+
+When we are able to prune page double XMAX it will be converted from it to
+general 64-bit XID page format with all operations on its tuples supported.
+
+In-memory tuple format
+----------------------
+
+In-memory tuple representation consists of two parts:
+- HeapTupleHeader from disk page (contains all heap tuple contents, not only
+header)
+- HeapTuple with additional in-memory fields
+
+HeapTuple for each tuple in memory stores t_xid_base/t_multi_base - a copies of
+page's pd_xid_base/pd_multi_base. With tuple's 32-bit t_xmin and t_xmax from
+HeapTupleHeader they are used to calculate actual 64-bit XMIN and XMAX:
+
+XMIN = t_xmin + t_xid_base. 					(3)
+XMAX = t_xmax + t_xid_base/t_multi_base.		(4)
+
+The downside of this is that we can not use tuple's XMIN and XMAX right away.
+We often need to re-read t_xmin and t_xmax - which could actually be pointers
+into a page in shared buffers and therefore they could be updated by any other
+backend.
+
+Update/delete with 64-bit XIDs and 32-bit t_xmin/t_xmax
+--------------------------------------------------------------
+
+When we try to delete/update a tuple, we check that XMAX for a page fits (2).
+I.e. that t_xmax will not be over MaxShortTransactionId relative to
+pd_xid_base/pd_multi_base of a its page.
+
+If the current XID doesn't fit a range
+(pd_xid_base, pd_xid_base + MaxShortTransactionId) (5):
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base/pd_multi_base on
+a page and update all t_xmin/t_xmax of the other tuples on the page to
+correspond new pd_xid_base/pd_multi_base.
+
+- If it was impossible, it will try to prune and freeze tuples on a page.
+
+- If this is unsuccessful it will throw an error. Normally this is very
+unlikely but if there is a very old living transaction with an age of around
+2^32 this can arise. Basically, this is a behavior similar to one during the
+vacuum to prevent wraparound when XID was 32-bit. Dba should take care and
+avoid very-long-living transactions with an age close to 2^32. So long-living
+transactions often they are most likely defunct.
+
+Insert with 64-bit XIDs and 32-bit t_xmin/t_xmax
+------------------------------------------------
+
+On insert we check if current XID fits a range (5). Otherwise:
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base for t_xmin will
+not be over MaxShortTransactionId.
+
+- If it is impossible, then it will try to prune and freeze tuples on a page.
+
+Known issue: if pd_xid_base could not be shifted to accommodate a tuple being
+inserted due to a very long-running transaction, we just throw an error. We
+neither try to insert a�tuple into another page nor mark the current page as
+full. So, in this (unlikely) case we will get regular insert errors on the next
+tries to insert to the page 'locked' by this very long-running transaction.
+
+Upgrade from 32-bit XID versions
+--------------------------------
+
+pg_upgrade doesn't change pages format itself. It is done lazily after.
+
+1. At first heap page read, tuples on a page are repacked to free 16 bytes
+at the end of a page, possibly freeing space from dead tuples.
+
+2A. 16 bytes of pd_special is added if there is a place for it
+
+2B. Page is converted to "Double XMAX" format if there is no place for
+pd_special
+
+3. If a page is in double XMAX format after its first read, and vacuum (or
+micro-vacuum at select query) could prune some tuples and free space for
+pd_special, prune_page will add pd_special and convert page from double XMAX
+to general 64-bit XID page format.
-- 
2.24.3 (Apple Git-128)


--cpok4wp6gsarlzvp--





^ permalink  raw  reply  [nested|flat] 265+ messages in thread

* [PATCH v6] README for 64bit xid
@ 2022-01-10 19:20  Pavel Borisov <[email protected]>
  0 siblings, 0 replies; 265+ messages in thread

From: Pavel Borisov @ 2022-01-10 19:20 UTC (permalink / raw)

Authors:
- Pavel Borisov <[email protected]>
- Maxim Orlov <[email protected]>
- Yura Sokolov <[email protected]> <[email protected]>
---
 src/backend/access/heap/README.XID64 | 128 +++++++++++++++++++++++++++
 1 file changed, 128 insertions(+)
 create mode 100644 src/backend/access/heap/README.XID64

diff --git a/src/backend/access/heap/README.XID64 b/src/backend/access/heap/README.XID64
new file mode 100644
index 00000000000..457ba9b9ef5
--- /dev/null
+++ b/src/backend/access/heap/README.XID64
@@ -0,0 +1,128 @@
+src/backend/access/heap/README.XID64
+
+64-bit Transaction ID's (XID)
+=============================
+
+A limited number (N = 2^32) of XID's required to do vacuum freeze to prevent
+wraparound every N/2 transactions. This causes performance degradation due
+to the need to exclusively lock tables while being vacuumed. In each
+wraparound cycle, SLRU buffers are also being cut.
+
+With 64-bit XID's wraparound is effectively postponed to a very distant
+future. Even in highly loaded systems that had 2^32 transactions per day
+it will take huge 2^31 days before the first enforced "vacuum to prevent
+wraparound"). Buffers cutting and routine vacuum are not enforced, and DBA
+can plan them independently at the time with the least system load and least
+critical for database performance. Also, it can be done less frequently
+(several times a year vs every several days) on systems with transaction rates
+similar to those mentioned above.
+
+On-disk tuple and page format
+-----------------------------
+
+On-disk tuple format remains unchanged. 32-bit t_xmin and t_xmax store the
+lower parts of 64-bit XMIN and XMAX values. Each heap page has additional
+64-bit pd_xid_base and pd_multi_base which are common for all tuples on a page.
+They are placed into a pd_special area - 16 bytes in the end of a heap page.
+Actual XMIN/XMAX for a tuple are calculated upon reading a tuple from a page
+as follows:
+
+XMIN = t_xmin + pd_xid_base. 					(1)
+XMAX = t_xmax + pd_xid_base/pd_multi_base.		(2)
+
+"Double XMAX" page format
+---------------------------------
+
+At first read of a heap page after pg_upgrade from 32-bit XID PostgreSQL
+version pd_special area with a size of 16 bytes should be added to a page.
+Though a page may not have space for this. Then it can be converted to a
+temporary format called "double XMAX".
+
+All tuples after pg-upgrade would necessarily have xmin = FrozenTransactionId.
+So we don't need tuple header t_xmin field and we reuse t_xmin to store higher
+32 bits of its XMAX.
+
+Double XMAX format is only for full pages that don't have 16 bytes for
+pd_special. So it neither has a�place for a single tuple. Insert and HOT update
+for double XMAX pages is impossible and not supported. We can only read or
+delete tuples from it.
+
+When we are able to prune page double XMAX it will be converted from it to
+general 64-bit XID page format with all operations on its tuples supported.
+
+In-memory tuple format
+----------------------
+
+In-memory tuple representation consists of two parts:
+- HeapTupleHeader from disk page (contains all heap tuple contents, not only
+header)
+- HeapTuple with additional in-memory fields
+
+HeapTuple for each tuple in memory stores t_xid_base/t_multi_base - a copies of
+page's pd_xid_base/pd_multi_base. With tuple's 32-bit t_xmin and t_xmax from
+HeapTupleHeader they are used to calculate actual 64-bit XMIN and XMAX:
+
+XMIN = t_xmin + t_xid_base. 					(3)
+XMAX = t_xmax + t_xid_base/t_multi_base.		(4)
+
+The downside of this is that we can not use tuple's XMIN and XMAX right away.
+We often need to re-read t_xmin and t_xmax - which could actually be pointers
+into a page in shared buffers and therefore they could be updated by any other
+backend.
+
+Update/delete with 64-bit XIDs and 32-bit t_xmin/t_xmax
+--------------------------------------------------------------
+
+When we try to delete/update a tuple, we check that XMAX for a page fits (2).
+I.e. that t_xmax will not be over MaxShortTransactionId relative to
+pd_xid_base/pd_multi_base of a its page.
+
+If the current XID doesn't fit a range
+(pd_xid_base, pd_xid_base + MaxShortTransactionId) (5):
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base/pd_multi_base on
+a page and update all t_xmin/t_xmax of the other tuples on the page to
+correspond new pd_xid_base/pd_multi_base.
+
+- If it was impossible, it will try to prune and freeze tuples on a page.
+
+- If this is unsuccessful it will throw an error. Normally this is very
+unlikely but if there is a very old living transaction with an age of around
+2^32 this can arise. Basically, this is a behavior similar to one during the
+vacuum to prevent wraparound when XID was 32-bit. Dba should take care and
+avoid very-long-living transactions with an age close to 2^32. So long-living
+transactions often they are most likely defunct.
+
+Insert with 64-bit XIDs and 32-bit t_xmin/t_xmax
+------------------------------------------------
+
+On insert we check if current XID fits a range (5). Otherwise:
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base for t_xmin will
+not be over MaxShortTransactionId.
+
+- If it is impossible, then it will try to prune and freeze tuples on a page.
+
+Known issue: if pd_xid_base could not be shifted to accommodate a tuple being
+inserted due to a very long-running transaction, we just throw an error. We
+neither try to insert a�tuple into another page nor mark the current page as
+full. So, in this (unlikely) case we will get regular insert errors on the next
+tries to insert to the page 'locked' by this very long-running transaction.
+
+Upgrade from 32-bit XID versions
+--------------------------------
+
+pg_upgrade doesn't change pages format itself. It is done lazily after.
+
+1. At first heap page read, tuples on a page are repacked to free 16 bytes
+at the end of a page, possibly freeing space from dead tuples.
+
+2A. 16 bytes of pd_special is added if there is a place for it
+
+2B. Page is converted to "Double XMAX" format if there is no place for
+pd_special
+
+3. If a page is in double XMAX format after its first read, and vacuum (or
+micro-vacuum at select query) could prune some tuples and free space for
+pd_special, prune_page will add pd_special and convert page from double XMAX
+to general 64-bit XID page format.
-- 
2.24.3 (Apple Git-128)


--cpok4wp6gsarlzvp--





^ permalink  raw  reply  [nested|flat] 265+ messages in thread

* [PATCH v6] README for 64bit xid
@ 2022-01-10 19:20  Pavel Borisov <[email protected]>
  0 siblings, 0 replies; 265+ messages in thread

From: Pavel Borisov @ 2022-01-10 19:20 UTC (permalink / raw)

Authors:
- Pavel Borisov <[email protected]>
- Maxim Orlov <[email protected]>
- Yura Sokolov <[email protected]> <[email protected]>
---
 src/backend/access/heap/README.XID64 | 128 +++++++++++++++++++++++++++
 1 file changed, 128 insertions(+)
 create mode 100644 src/backend/access/heap/README.XID64

diff --git a/src/backend/access/heap/README.XID64 b/src/backend/access/heap/README.XID64
new file mode 100644
index 00000000000..457ba9b9ef5
--- /dev/null
+++ b/src/backend/access/heap/README.XID64
@@ -0,0 +1,128 @@
+src/backend/access/heap/README.XID64
+
+64-bit Transaction ID's (XID)
+=============================
+
+A limited number (N = 2^32) of XID's required to do vacuum freeze to prevent
+wraparound every N/2 transactions. This causes performance degradation due
+to the need to exclusively lock tables while being vacuumed. In each
+wraparound cycle, SLRU buffers are also being cut.
+
+With 64-bit XID's wraparound is effectively postponed to a very distant
+future. Even in highly loaded systems that had 2^32 transactions per day
+it will take huge 2^31 days before the first enforced "vacuum to prevent
+wraparound"). Buffers cutting and routine vacuum are not enforced, and DBA
+can plan them independently at the time with the least system load and least
+critical for database performance. Also, it can be done less frequently
+(several times a year vs every several days) on systems with transaction rates
+similar to those mentioned above.
+
+On-disk tuple and page format
+-----------------------------
+
+On-disk tuple format remains unchanged. 32-bit t_xmin and t_xmax store the
+lower parts of 64-bit XMIN and XMAX values. Each heap page has additional
+64-bit pd_xid_base and pd_multi_base which are common for all tuples on a page.
+They are placed into a pd_special area - 16 bytes in the end of a heap page.
+Actual XMIN/XMAX for a tuple are calculated upon reading a tuple from a page
+as follows:
+
+XMIN = t_xmin + pd_xid_base. 					(1)
+XMAX = t_xmax + pd_xid_base/pd_multi_base.		(2)
+
+"Double XMAX" page format
+---------------------------------
+
+At first read of a heap page after pg_upgrade from 32-bit XID PostgreSQL
+version pd_special area with a size of 16 bytes should be added to a page.
+Though a page may not have space for this. Then it can be converted to a
+temporary format called "double XMAX".
+
+All tuples after pg-upgrade would necessarily have xmin = FrozenTransactionId.
+So we don't need tuple header t_xmin field and we reuse t_xmin to store higher
+32 bits of its XMAX.
+
+Double XMAX format is only for full pages that don't have 16 bytes for
+pd_special. So it neither has a�place for a single tuple. Insert and HOT update
+for double XMAX pages is impossible and not supported. We can only read or
+delete tuples from it.
+
+When we are able to prune page double XMAX it will be converted from it to
+general 64-bit XID page format with all operations on its tuples supported.
+
+In-memory tuple format
+----------------------
+
+In-memory tuple representation consists of two parts:
+- HeapTupleHeader from disk page (contains all heap tuple contents, not only
+header)
+- HeapTuple with additional in-memory fields
+
+HeapTuple for each tuple in memory stores t_xid_base/t_multi_base - a copies of
+page's pd_xid_base/pd_multi_base. With tuple's 32-bit t_xmin and t_xmax from
+HeapTupleHeader they are used to calculate actual 64-bit XMIN and XMAX:
+
+XMIN = t_xmin + t_xid_base. 					(3)
+XMAX = t_xmax + t_xid_base/t_multi_base.		(4)
+
+The downside of this is that we can not use tuple's XMIN and XMAX right away.
+We often need to re-read t_xmin and t_xmax - which could actually be pointers
+into a page in shared buffers and therefore they could be updated by any other
+backend.
+
+Update/delete with 64-bit XIDs and 32-bit t_xmin/t_xmax
+--------------------------------------------------------------
+
+When we try to delete/update a tuple, we check that XMAX for a page fits (2).
+I.e. that t_xmax will not be over MaxShortTransactionId relative to
+pd_xid_base/pd_multi_base of a its page.
+
+If the current XID doesn't fit a range
+(pd_xid_base, pd_xid_base + MaxShortTransactionId) (5):
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base/pd_multi_base on
+a page and update all t_xmin/t_xmax of the other tuples on the page to
+correspond new pd_xid_base/pd_multi_base.
+
+- If it was impossible, it will try to prune and freeze tuples on a page.
+
+- If this is unsuccessful it will throw an error. Normally this is very
+unlikely but if there is a very old living transaction with an age of around
+2^32 this can arise. Basically, this is a behavior similar to one during the
+vacuum to prevent wraparound when XID was 32-bit. Dba should take care and
+avoid very-long-living transactions with an age close to 2^32. So long-living
+transactions often they are most likely defunct.
+
+Insert with 64-bit XIDs and 32-bit t_xmin/t_xmax
+------------------------------------------------
+
+On insert we check if current XID fits a range (5). Otherwise:
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base for t_xmin will
+not be over MaxShortTransactionId.
+
+- If it is impossible, then it will try to prune and freeze tuples on a page.
+
+Known issue: if pd_xid_base could not be shifted to accommodate a tuple being
+inserted due to a very long-running transaction, we just throw an error. We
+neither try to insert a�tuple into another page nor mark the current page as
+full. So, in this (unlikely) case we will get regular insert errors on the next
+tries to insert to the page 'locked' by this very long-running transaction.
+
+Upgrade from 32-bit XID versions
+--------------------------------
+
+pg_upgrade doesn't change pages format itself. It is done lazily after.
+
+1. At first heap page read, tuples on a page are repacked to free 16 bytes
+at the end of a page, possibly freeing space from dead tuples.
+
+2A. 16 bytes of pd_special is added if there is a place for it
+
+2B. Page is converted to "Double XMAX" format if there is no place for
+pd_special
+
+3. If a page is in double XMAX format after its first read, and vacuum (or
+micro-vacuum at select query) could prune some tuples and free space for
+pd_special, prune_page will add pd_special and convert page from double XMAX
+to general 64-bit XID page format.
-- 
2.24.3 (Apple Git-128)


--cpok4wp6gsarlzvp--





^ permalink  raw  reply  [nested|flat] 265+ messages in thread

* [PATCH v6] README for 64bit xid
@ 2022-01-10 19:20  Pavel Borisov <[email protected]>
  0 siblings, 0 replies; 265+ messages in thread

From: Pavel Borisov @ 2022-01-10 19:20 UTC (permalink / raw)

Authors:
- Pavel Borisov <[email protected]>
- Maxim Orlov <[email protected]>
- Yura Sokolov <[email protected]> <[email protected]>
---
 src/backend/access/heap/README.XID64 | 128 +++++++++++++++++++++++++++
 1 file changed, 128 insertions(+)
 create mode 100644 src/backend/access/heap/README.XID64

diff --git a/src/backend/access/heap/README.XID64 b/src/backend/access/heap/README.XID64
new file mode 100644
index 00000000000..457ba9b9ef5
--- /dev/null
+++ b/src/backend/access/heap/README.XID64
@@ -0,0 +1,128 @@
+src/backend/access/heap/README.XID64
+
+64-bit Transaction ID's (XID)
+=============================
+
+A limited number (N = 2^32) of XID's required to do vacuum freeze to prevent
+wraparound every N/2 transactions. This causes performance degradation due
+to the need to exclusively lock tables while being vacuumed. In each
+wraparound cycle, SLRU buffers are also being cut.
+
+With 64-bit XID's wraparound is effectively postponed to a very distant
+future. Even in highly loaded systems that had 2^32 transactions per day
+it will take huge 2^31 days before the first enforced "vacuum to prevent
+wraparound"). Buffers cutting and routine vacuum are not enforced, and DBA
+can plan them independently at the time with the least system load and least
+critical for database performance. Also, it can be done less frequently
+(several times a year vs every several days) on systems with transaction rates
+similar to those mentioned above.
+
+On-disk tuple and page format
+-----------------------------
+
+On-disk tuple format remains unchanged. 32-bit t_xmin and t_xmax store the
+lower parts of 64-bit XMIN and XMAX values. Each heap page has additional
+64-bit pd_xid_base and pd_multi_base which are common for all tuples on a page.
+They are placed into a pd_special area - 16 bytes in the end of a heap page.
+Actual XMIN/XMAX for a tuple are calculated upon reading a tuple from a page
+as follows:
+
+XMIN = t_xmin + pd_xid_base. 					(1)
+XMAX = t_xmax + pd_xid_base/pd_multi_base.		(2)
+
+"Double XMAX" page format
+---------------------------------
+
+At first read of a heap page after pg_upgrade from 32-bit XID PostgreSQL
+version pd_special area with a size of 16 bytes should be added to a page.
+Though a page may not have space for this. Then it can be converted to a
+temporary format called "double XMAX".
+
+All tuples after pg-upgrade would necessarily have xmin = FrozenTransactionId.
+So we don't need tuple header t_xmin field and we reuse t_xmin to store higher
+32 bits of its XMAX.
+
+Double XMAX format is only for full pages that don't have 16 bytes for
+pd_special. So it neither has a�place for a single tuple. Insert and HOT update
+for double XMAX pages is impossible and not supported. We can only read or
+delete tuples from it.
+
+When we are able to prune page double XMAX it will be converted from it to
+general 64-bit XID page format with all operations on its tuples supported.
+
+In-memory tuple format
+----------------------
+
+In-memory tuple representation consists of two parts:
+- HeapTupleHeader from disk page (contains all heap tuple contents, not only
+header)
+- HeapTuple with additional in-memory fields
+
+HeapTuple for each tuple in memory stores t_xid_base/t_multi_base - a copies of
+page's pd_xid_base/pd_multi_base. With tuple's 32-bit t_xmin and t_xmax from
+HeapTupleHeader they are used to calculate actual 64-bit XMIN and XMAX:
+
+XMIN = t_xmin + t_xid_base. 					(3)
+XMAX = t_xmax + t_xid_base/t_multi_base.		(4)
+
+The downside of this is that we can not use tuple's XMIN and XMAX right away.
+We often need to re-read t_xmin and t_xmax - which could actually be pointers
+into a page in shared buffers and therefore they could be updated by any other
+backend.
+
+Update/delete with 64-bit XIDs and 32-bit t_xmin/t_xmax
+--------------------------------------------------------------
+
+When we try to delete/update a tuple, we check that XMAX for a page fits (2).
+I.e. that t_xmax will not be over MaxShortTransactionId relative to
+pd_xid_base/pd_multi_base of a its page.
+
+If the current XID doesn't fit a range
+(pd_xid_base, pd_xid_base + MaxShortTransactionId) (5):
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base/pd_multi_base on
+a page and update all t_xmin/t_xmax of the other tuples on the page to
+correspond new pd_xid_base/pd_multi_base.
+
+- If it was impossible, it will try to prune and freeze tuples on a page.
+
+- If this is unsuccessful it will throw an error. Normally this is very
+unlikely but if there is a very old living transaction with an age of around
+2^32 this can arise. Basically, this is a behavior similar to one during the
+vacuum to prevent wraparound when XID was 32-bit. Dba should take care and
+avoid very-long-living transactions with an age close to 2^32. So long-living
+transactions often they are most likely defunct.
+
+Insert with 64-bit XIDs and 32-bit t_xmin/t_xmax
+------------------------------------------------
+
+On insert we check if current XID fits a range (5). Otherwise:
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base for t_xmin will
+not be over MaxShortTransactionId.
+
+- If it is impossible, then it will try to prune and freeze tuples on a page.
+
+Known issue: if pd_xid_base could not be shifted to accommodate a tuple being
+inserted due to a very long-running transaction, we just throw an error. We
+neither try to insert a�tuple into another page nor mark the current page as
+full. So, in this (unlikely) case we will get regular insert errors on the next
+tries to insert to the page 'locked' by this very long-running transaction.
+
+Upgrade from 32-bit XID versions
+--------------------------------
+
+pg_upgrade doesn't change pages format itself. It is done lazily after.
+
+1. At first heap page read, tuples on a page are repacked to free 16 bytes
+at the end of a page, possibly freeing space from dead tuples.
+
+2A. 16 bytes of pd_special is added if there is a place for it
+
+2B. Page is converted to "Double XMAX" format if there is no place for
+pd_special
+
+3. If a page is in double XMAX format after its first read, and vacuum (or
+micro-vacuum at select query) could prune some tuples and free space for
+pd_special, prune_page will add pd_special and convert page from double XMAX
+to general 64-bit XID page format.
-- 
2.24.3 (Apple Git-128)


--cpok4wp6gsarlzvp--





^ permalink  raw  reply  [nested|flat] 265+ messages in thread

* [PATCH v6] README for 64bit xid
@ 2022-01-10 19:20  Pavel Borisov <[email protected]>
  0 siblings, 0 replies; 265+ messages in thread

From: Pavel Borisov @ 2022-01-10 19:20 UTC (permalink / raw)

Authors:
- Pavel Borisov <[email protected]>
- Maxim Orlov <[email protected]>
- Yura Sokolov <[email protected]> <[email protected]>
---
 src/backend/access/heap/README.XID64 | 128 +++++++++++++++++++++++++++
 1 file changed, 128 insertions(+)
 create mode 100644 src/backend/access/heap/README.XID64

diff --git a/src/backend/access/heap/README.XID64 b/src/backend/access/heap/README.XID64
new file mode 100644
index 00000000000..457ba9b9ef5
--- /dev/null
+++ b/src/backend/access/heap/README.XID64
@@ -0,0 +1,128 @@
+src/backend/access/heap/README.XID64
+
+64-bit Transaction ID's (XID)
+=============================
+
+A limited number (N = 2^32) of XID's required to do vacuum freeze to prevent
+wraparound every N/2 transactions. This causes performance degradation due
+to the need to exclusively lock tables while being vacuumed. In each
+wraparound cycle, SLRU buffers are also being cut.
+
+With 64-bit XID's wraparound is effectively postponed to a very distant
+future. Even in highly loaded systems that had 2^32 transactions per day
+it will take huge 2^31 days before the first enforced "vacuum to prevent
+wraparound"). Buffers cutting and routine vacuum are not enforced, and DBA
+can plan them independently at the time with the least system load and least
+critical for database performance. Also, it can be done less frequently
+(several times a year vs every several days) on systems with transaction rates
+similar to those mentioned above.
+
+On-disk tuple and page format
+-----------------------------
+
+On-disk tuple format remains unchanged. 32-bit t_xmin and t_xmax store the
+lower parts of 64-bit XMIN and XMAX values. Each heap page has additional
+64-bit pd_xid_base and pd_multi_base which are common for all tuples on a page.
+They are placed into a pd_special area - 16 bytes in the end of a heap page.
+Actual XMIN/XMAX for a tuple are calculated upon reading a tuple from a page
+as follows:
+
+XMIN = t_xmin + pd_xid_base. 					(1)
+XMAX = t_xmax + pd_xid_base/pd_multi_base.		(2)
+
+"Double XMAX" page format
+---------------------------------
+
+At first read of a heap page after pg_upgrade from 32-bit XID PostgreSQL
+version pd_special area with a size of 16 bytes should be added to a page.
+Though a page may not have space for this. Then it can be converted to a
+temporary format called "double XMAX".
+
+All tuples after pg-upgrade would necessarily have xmin = FrozenTransactionId.
+So we don't need tuple header t_xmin field and we reuse t_xmin to store higher
+32 bits of its XMAX.
+
+Double XMAX format is only for full pages that don't have 16 bytes for
+pd_special. So it neither has a�place for a single tuple. Insert and HOT update
+for double XMAX pages is impossible and not supported. We can only read or
+delete tuples from it.
+
+When we are able to prune page double XMAX it will be converted from it to
+general 64-bit XID page format with all operations on its tuples supported.
+
+In-memory tuple format
+----------------------
+
+In-memory tuple representation consists of two parts:
+- HeapTupleHeader from disk page (contains all heap tuple contents, not only
+header)
+- HeapTuple with additional in-memory fields
+
+HeapTuple for each tuple in memory stores t_xid_base/t_multi_base - a copies of
+page's pd_xid_base/pd_multi_base. With tuple's 32-bit t_xmin and t_xmax from
+HeapTupleHeader they are used to calculate actual 64-bit XMIN and XMAX:
+
+XMIN = t_xmin + t_xid_base. 					(3)
+XMAX = t_xmax + t_xid_base/t_multi_base.		(4)
+
+The downside of this is that we can not use tuple's XMIN and XMAX right away.
+We often need to re-read t_xmin and t_xmax - which could actually be pointers
+into a page in shared buffers and therefore they could be updated by any other
+backend.
+
+Update/delete with 64-bit XIDs and 32-bit t_xmin/t_xmax
+--------------------------------------------------------------
+
+When we try to delete/update a tuple, we check that XMAX for a page fits (2).
+I.e. that t_xmax will not be over MaxShortTransactionId relative to
+pd_xid_base/pd_multi_base of a its page.
+
+If the current XID doesn't fit a range
+(pd_xid_base, pd_xid_base + MaxShortTransactionId) (5):
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base/pd_multi_base on
+a page and update all t_xmin/t_xmax of the other tuples on the page to
+correspond new pd_xid_base/pd_multi_base.
+
+- If it was impossible, it will try to prune and freeze tuples on a page.
+
+- If this is unsuccessful it will throw an error. Normally this is very
+unlikely but if there is a very old living transaction with an age of around
+2^32 this can arise. Basically, this is a behavior similar to one during the
+vacuum to prevent wraparound when XID was 32-bit. Dba should take care and
+avoid very-long-living transactions with an age close to 2^32. So long-living
+transactions often they are most likely defunct.
+
+Insert with 64-bit XIDs and 32-bit t_xmin/t_xmax
+------------------------------------------------
+
+On insert we check if current XID fits a range (5). Otherwise:
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base for t_xmin will
+not be over MaxShortTransactionId.
+
+- If it is impossible, then it will try to prune and freeze tuples on a page.
+
+Known issue: if pd_xid_base could not be shifted to accommodate a tuple being
+inserted due to a very long-running transaction, we just throw an error. We
+neither try to insert a�tuple into another page nor mark the current page as
+full. So, in this (unlikely) case we will get regular insert errors on the next
+tries to insert to the page 'locked' by this very long-running transaction.
+
+Upgrade from 32-bit XID versions
+--------------------------------
+
+pg_upgrade doesn't change pages format itself. It is done lazily after.
+
+1. At first heap page read, tuples on a page are repacked to free 16 bytes
+at the end of a page, possibly freeing space from dead tuples.
+
+2A. 16 bytes of pd_special is added if there is a place for it
+
+2B. Page is converted to "Double XMAX" format if there is no place for
+pd_special
+
+3. If a page is in double XMAX format after its first read, and vacuum (or
+micro-vacuum at select query) could prune some tuples and free space for
+pd_special, prune_page will add pd_special and convert page from double XMAX
+to general 64-bit XID page format.
-- 
2.24.3 (Apple Git-128)


--cpok4wp6gsarlzvp--





^ permalink  raw  reply  [nested|flat] 265+ messages in thread

* [PATCH v6] README for 64bit xid
@ 2022-01-10 19:20  Pavel Borisov <[email protected]>
  0 siblings, 0 replies; 265+ messages in thread

From: Pavel Borisov @ 2022-01-10 19:20 UTC (permalink / raw)

Authors:
- Pavel Borisov <[email protected]>
- Maxim Orlov <[email protected]>
- Yura Sokolov <[email protected]> <[email protected]>
---
 src/backend/access/heap/README.XID64 | 128 +++++++++++++++++++++++++++
 1 file changed, 128 insertions(+)
 create mode 100644 src/backend/access/heap/README.XID64

diff --git a/src/backend/access/heap/README.XID64 b/src/backend/access/heap/README.XID64
new file mode 100644
index 00000000000..457ba9b9ef5
--- /dev/null
+++ b/src/backend/access/heap/README.XID64
@@ -0,0 +1,128 @@
+src/backend/access/heap/README.XID64
+
+64-bit Transaction ID's (XID)
+=============================
+
+A limited number (N = 2^32) of XID's required to do vacuum freeze to prevent
+wraparound every N/2 transactions. This causes performance degradation due
+to the need to exclusively lock tables while being vacuumed. In each
+wraparound cycle, SLRU buffers are also being cut.
+
+With 64-bit XID's wraparound is effectively postponed to a very distant
+future. Even in highly loaded systems that had 2^32 transactions per day
+it will take huge 2^31 days before the first enforced "vacuum to prevent
+wraparound"). Buffers cutting and routine vacuum are not enforced, and DBA
+can plan them independently at the time with the least system load and least
+critical for database performance. Also, it can be done less frequently
+(several times a year vs every several days) on systems with transaction rates
+similar to those mentioned above.
+
+On-disk tuple and page format
+-----------------------------
+
+On-disk tuple format remains unchanged. 32-bit t_xmin and t_xmax store the
+lower parts of 64-bit XMIN and XMAX values. Each heap page has additional
+64-bit pd_xid_base and pd_multi_base which are common for all tuples on a page.
+They are placed into a pd_special area - 16 bytes in the end of a heap page.
+Actual XMIN/XMAX for a tuple are calculated upon reading a tuple from a page
+as follows:
+
+XMIN = t_xmin + pd_xid_base. 					(1)
+XMAX = t_xmax + pd_xid_base/pd_multi_base.		(2)
+
+"Double XMAX" page format
+---------------------------------
+
+At first read of a heap page after pg_upgrade from 32-bit XID PostgreSQL
+version pd_special area with a size of 16 bytes should be added to a page.
+Though a page may not have space for this. Then it can be converted to a
+temporary format called "double XMAX".
+
+All tuples after pg-upgrade would necessarily have xmin = FrozenTransactionId.
+So we don't need tuple header t_xmin field and we reuse t_xmin to store higher
+32 bits of its XMAX.
+
+Double XMAX format is only for full pages that don't have 16 bytes for
+pd_special. So it neither has a�place for a single tuple. Insert and HOT update
+for double XMAX pages is impossible and not supported. We can only read or
+delete tuples from it.
+
+When we are able to prune page double XMAX it will be converted from it to
+general 64-bit XID page format with all operations on its tuples supported.
+
+In-memory tuple format
+----------------------
+
+In-memory tuple representation consists of two parts:
+- HeapTupleHeader from disk page (contains all heap tuple contents, not only
+header)
+- HeapTuple with additional in-memory fields
+
+HeapTuple for each tuple in memory stores t_xid_base/t_multi_base - a copies of
+page's pd_xid_base/pd_multi_base. With tuple's 32-bit t_xmin and t_xmax from
+HeapTupleHeader they are used to calculate actual 64-bit XMIN and XMAX:
+
+XMIN = t_xmin + t_xid_base. 					(3)
+XMAX = t_xmax + t_xid_base/t_multi_base.		(4)
+
+The downside of this is that we can not use tuple's XMIN and XMAX right away.
+We often need to re-read t_xmin and t_xmax - which could actually be pointers
+into a page in shared buffers and therefore they could be updated by any other
+backend.
+
+Update/delete with 64-bit XIDs and 32-bit t_xmin/t_xmax
+--------------------------------------------------------------
+
+When we try to delete/update a tuple, we check that XMAX for a page fits (2).
+I.e. that t_xmax will not be over MaxShortTransactionId relative to
+pd_xid_base/pd_multi_base of a its page.
+
+If the current XID doesn't fit a range
+(pd_xid_base, pd_xid_base + MaxShortTransactionId) (5):
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base/pd_multi_base on
+a page and update all t_xmin/t_xmax of the other tuples on the page to
+correspond new pd_xid_base/pd_multi_base.
+
+- If it was impossible, it will try to prune and freeze tuples on a page.
+
+- If this is unsuccessful it will throw an error. Normally this is very
+unlikely but if there is a very old living transaction with an age of around
+2^32 this can arise. Basically, this is a behavior similar to one during the
+vacuum to prevent wraparound when XID was 32-bit. Dba should take care and
+avoid very-long-living transactions with an age close to 2^32. So long-living
+transactions often they are most likely defunct.
+
+Insert with 64-bit XIDs and 32-bit t_xmin/t_xmax
+------------------------------------------------
+
+On insert we check if current XID fits a range (5). Otherwise:
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base for t_xmin will
+not be over MaxShortTransactionId.
+
+- If it is impossible, then it will try to prune and freeze tuples on a page.
+
+Known issue: if pd_xid_base could not be shifted to accommodate a tuple being
+inserted due to a very long-running transaction, we just throw an error. We
+neither try to insert a�tuple into another page nor mark the current page as
+full. So, in this (unlikely) case we will get regular insert errors on the next
+tries to insert to the page 'locked' by this very long-running transaction.
+
+Upgrade from 32-bit XID versions
+--------------------------------
+
+pg_upgrade doesn't change pages format itself. It is done lazily after.
+
+1. At first heap page read, tuples on a page are repacked to free 16 bytes
+at the end of a page, possibly freeing space from dead tuples.
+
+2A. 16 bytes of pd_special is added if there is a place for it
+
+2B. Page is converted to "Double XMAX" format if there is no place for
+pd_special
+
+3. If a page is in double XMAX format after its first read, and vacuum (or
+micro-vacuum at select query) could prune some tuples and free space for
+pd_special, prune_page will add pd_special and convert page from double XMAX
+to general 64-bit XID page format.
-- 
2.24.3 (Apple Git-128)


--cpok4wp6gsarlzvp--





^ permalink  raw  reply  [nested|flat] 265+ messages in thread

* [PATCH v6] README for 64bit xid
@ 2022-01-10 19:20  Pavel Borisov <[email protected]>
  0 siblings, 0 replies; 265+ messages in thread

From: Pavel Borisov @ 2022-01-10 19:20 UTC (permalink / raw)

Authors:
- Pavel Borisov <[email protected]>
- Maxim Orlov <[email protected]>
- Yura Sokolov <[email protected]> <[email protected]>
---
 src/backend/access/heap/README.XID64 | 128 +++++++++++++++++++++++++++
 1 file changed, 128 insertions(+)
 create mode 100644 src/backend/access/heap/README.XID64

diff --git a/src/backend/access/heap/README.XID64 b/src/backend/access/heap/README.XID64
new file mode 100644
index 00000000000..457ba9b9ef5
--- /dev/null
+++ b/src/backend/access/heap/README.XID64
@@ -0,0 +1,128 @@
+src/backend/access/heap/README.XID64
+
+64-bit Transaction ID's (XID)
+=============================
+
+A limited number (N = 2^32) of XID's required to do vacuum freeze to prevent
+wraparound every N/2 transactions. This causes performance degradation due
+to the need to exclusively lock tables while being vacuumed. In each
+wraparound cycle, SLRU buffers are also being cut.
+
+With 64-bit XID's wraparound is effectively postponed to a very distant
+future. Even in highly loaded systems that had 2^32 transactions per day
+it will take huge 2^31 days before the first enforced "vacuum to prevent
+wraparound"). Buffers cutting and routine vacuum are not enforced, and DBA
+can plan them independently at the time with the least system load and least
+critical for database performance. Also, it can be done less frequently
+(several times a year vs every several days) on systems with transaction rates
+similar to those mentioned above.
+
+On-disk tuple and page format
+-----------------------------
+
+On-disk tuple format remains unchanged. 32-bit t_xmin and t_xmax store the
+lower parts of 64-bit XMIN and XMAX values. Each heap page has additional
+64-bit pd_xid_base and pd_multi_base which are common for all tuples on a page.
+They are placed into a pd_special area - 16 bytes in the end of a heap page.
+Actual XMIN/XMAX for a tuple are calculated upon reading a tuple from a page
+as follows:
+
+XMIN = t_xmin + pd_xid_base. 					(1)
+XMAX = t_xmax + pd_xid_base/pd_multi_base.		(2)
+
+"Double XMAX" page format
+---------------------------------
+
+At first read of a heap page after pg_upgrade from 32-bit XID PostgreSQL
+version pd_special area with a size of 16 bytes should be added to a page.
+Though a page may not have space for this. Then it can be converted to a
+temporary format called "double XMAX".
+
+All tuples after pg-upgrade would necessarily have xmin = FrozenTransactionId.
+So we don't need tuple header t_xmin field and we reuse t_xmin to store higher
+32 bits of its XMAX.
+
+Double XMAX format is only for full pages that don't have 16 bytes for
+pd_special. So it neither has a�place for a single tuple. Insert and HOT update
+for double XMAX pages is impossible and not supported. We can only read or
+delete tuples from it.
+
+When we are able to prune page double XMAX it will be converted from it to
+general 64-bit XID page format with all operations on its tuples supported.
+
+In-memory tuple format
+----------------------
+
+In-memory tuple representation consists of two parts:
+- HeapTupleHeader from disk page (contains all heap tuple contents, not only
+header)
+- HeapTuple with additional in-memory fields
+
+HeapTuple for each tuple in memory stores t_xid_base/t_multi_base - a copies of
+page's pd_xid_base/pd_multi_base. With tuple's 32-bit t_xmin and t_xmax from
+HeapTupleHeader they are used to calculate actual 64-bit XMIN and XMAX:
+
+XMIN = t_xmin + t_xid_base. 					(3)
+XMAX = t_xmax + t_xid_base/t_multi_base.		(4)
+
+The downside of this is that we can not use tuple's XMIN and XMAX right away.
+We often need to re-read t_xmin and t_xmax - which could actually be pointers
+into a page in shared buffers and therefore they could be updated by any other
+backend.
+
+Update/delete with 64-bit XIDs and 32-bit t_xmin/t_xmax
+--------------------------------------------------------------
+
+When we try to delete/update a tuple, we check that XMAX for a page fits (2).
+I.e. that t_xmax will not be over MaxShortTransactionId relative to
+pd_xid_base/pd_multi_base of a its page.
+
+If the current XID doesn't fit a range
+(pd_xid_base, pd_xid_base + MaxShortTransactionId) (5):
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base/pd_multi_base on
+a page and update all t_xmin/t_xmax of the other tuples on the page to
+correspond new pd_xid_base/pd_multi_base.
+
+- If it was impossible, it will try to prune and freeze tuples on a page.
+
+- If this is unsuccessful it will throw an error. Normally this is very
+unlikely but if there is a very old living transaction with an age of around
+2^32 this can arise. Basically, this is a behavior similar to one during the
+vacuum to prevent wraparound when XID was 32-bit. Dba should take care and
+avoid very-long-living transactions with an age close to 2^32. So long-living
+transactions often they are most likely defunct.
+
+Insert with 64-bit XIDs and 32-bit t_xmin/t_xmax
+------------------------------------------------
+
+On insert we check if current XID fits a range (5). Otherwise:
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base for t_xmin will
+not be over MaxShortTransactionId.
+
+- If it is impossible, then it will try to prune and freeze tuples on a page.
+
+Known issue: if pd_xid_base could not be shifted to accommodate a tuple being
+inserted due to a very long-running transaction, we just throw an error. We
+neither try to insert a�tuple into another page nor mark the current page as
+full. So, in this (unlikely) case we will get regular insert errors on the next
+tries to insert to the page 'locked' by this very long-running transaction.
+
+Upgrade from 32-bit XID versions
+--------------------------------
+
+pg_upgrade doesn't change pages format itself. It is done lazily after.
+
+1. At first heap page read, tuples on a page are repacked to free 16 bytes
+at the end of a page, possibly freeing space from dead tuples.
+
+2A. 16 bytes of pd_special is added if there is a place for it
+
+2B. Page is converted to "Double XMAX" format if there is no place for
+pd_special
+
+3. If a page is in double XMAX format after its first read, and vacuum (or
+micro-vacuum at select query) could prune some tuples and free space for
+pd_special, prune_page will add pd_special and convert page from double XMAX
+to general 64-bit XID page format.
-- 
2.24.3 (Apple Git-128)


--cpok4wp6gsarlzvp--





^ permalink  raw  reply  [nested|flat] 265+ messages in thread

* [PATCH v6] README for 64bit xid
@ 2022-01-10 19:20  Pavel Borisov <[email protected]>
  0 siblings, 0 replies; 265+ messages in thread

From: Pavel Borisov @ 2022-01-10 19:20 UTC (permalink / raw)

Authors:
- Pavel Borisov <[email protected]>
- Maxim Orlov <[email protected]>
- Yura Sokolov <[email protected]> <[email protected]>
---
 src/backend/access/heap/README.XID64 | 128 +++++++++++++++++++++++++++
 1 file changed, 128 insertions(+)
 create mode 100644 src/backend/access/heap/README.XID64

diff --git a/src/backend/access/heap/README.XID64 b/src/backend/access/heap/README.XID64
new file mode 100644
index 00000000000..457ba9b9ef5
--- /dev/null
+++ b/src/backend/access/heap/README.XID64
@@ -0,0 +1,128 @@
+src/backend/access/heap/README.XID64
+
+64-bit Transaction ID's (XID)
+=============================
+
+A limited number (N = 2^32) of XID's required to do vacuum freeze to prevent
+wraparound every N/2 transactions. This causes performance degradation due
+to the need to exclusively lock tables while being vacuumed. In each
+wraparound cycle, SLRU buffers are also being cut.
+
+With 64-bit XID's wraparound is effectively postponed to a very distant
+future. Even in highly loaded systems that had 2^32 transactions per day
+it will take huge 2^31 days before the first enforced "vacuum to prevent
+wraparound"). Buffers cutting and routine vacuum are not enforced, and DBA
+can plan them independently at the time with the least system load and least
+critical for database performance. Also, it can be done less frequently
+(several times a year vs every several days) on systems with transaction rates
+similar to those mentioned above.
+
+On-disk tuple and page format
+-----------------------------
+
+On-disk tuple format remains unchanged. 32-bit t_xmin and t_xmax store the
+lower parts of 64-bit XMIN and XMAX values. Each heap page has additional
+64-bit pd_xid_base and pd_multi_base which are common for all tuples on a page.
+They are placed into a pd_special area - 16 bytes in the end of a heap page.
+Actual XMIN/XMAX for a tuple are calculated upon reading a tuple from a page
+as follows:
+
+XMIN = t_xmin + pd_xid_base. 					(1)
+XMAX = t_xmax + pd_xid_base/pd_multi_base.		(2)
+
+"Double XMAX" page format
+---------------------------------
+
+At first read of a heap page after pg_upgrade from 32-bit XID PostgreSQL
+version pd_special area with a size of 16 bytes should be added to a page.
+Though a page may not have space for this. Then it can be converted to a
+temporary format called "double XMAX".
+
+All tuples after pg-upgrade would necessarily have xmin = FrozenTransactionId.
+So we don't need tuple header t_xmin field and we reuse t_xmin to store higher
+32 bits of its XMAX.
+
+Double XMAX format is only for full pages that don't have 16 bytes for
+pd_special. So it neither has a�place for a single tuple. Insert and HOT update
+for double XMAX pages is impossible and not supported. We can only read or
+delete tuples from it.
+
+When we are able to prune page double XMAX it will be converted from it to
+general 64-bit XID page format with all operations on its tuples supported.
+
+In-memory tuple format
+----------------------
+
+In-memory tuple representation consists of two parts:
+- HeapTupleHeader from disk page (contains all heap tuple contents, not only
+header)
+- HeapTuple with additional in-memory fields
+
+HeapTuple for each tuple in memory stores t_xid_base/t_multi_base - a copies of
+page's pd_xid_base/pd_multi_base. With tuple's 32-bit t_xmin and t_xmax from
+HeapTupleHeader they are used to calculate actual 64-bit XMIN and XMAX:
+
+XMIN = t_xmin + t_xid_base. 					(3)
+XMAX = t_xmax + t_xid_base/t_multi_base.		(4)
+
+The downside of this is that we can not use tuple's XMIN and XMAX right away.
+We often need to re-read t_xmin and t_xmax - which could actually be pointers
+into a page in shared buffers and therefore they could be updated by any other
+backend.
+
+Update/delete with 64-bit XIDs and 32-bit t_xmin/t_xmax
+--------------------------------------------------------------
+
+When we try to delete/update a tuple, we check that XMAX for a page fits (2).
+I.e. that t_xmax will not be over MaxShortTransactionId relative to
+pd_xid_base/pd_multi_base of a its page.
+
+If the current XID doesn't fit a range
+(pd_xid_base, pd_xid_base + MaxShortTransactionId) (5):
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base/pd_multi_base on
+a page and update all t_xmin/t_xmax of the other tuples on the page to
+correspond new pd_xid_base/pd_multi_base.
+
+- If it was impossible, it will try to prune and freeze tuples on a page.
+
+- If this is unsuccessful it will throw an error. Normally this is very
+unlikely but if there is a very old living transaction with an age of around
+2^32 this can arise. Basically, this is a behavior similar to one during the
+vacuum to prevent wraparound when XID was 32-bit. Dba should take care and
+avoid very-long-living transactions with an age close to 2^32. So long-living
+transactions often they are most likely defunct.
+
+Insert with 64-bit XIDs and 32-bit t_xmin/t_xmax
+------------------------------------------------
+
+On insert we check if current XID fits a range (5). Otherwise:
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base for t_xmin will
+not be over MaxShortTransactionId.
+
+- If it is impossible, then it will try to prune and freeze tuples on a page.
+
+Known issue: if pd_xid_base could not be shifted to accommodate a tuple being
+inserted due to a very long-running transaction, we just throw an error. We
+neither try to insert a�tuple into another page nor mark the current page as
+full. So, in this (unlikely) case we will get regular insert errors on the next
+tries to insert to the page 'locked' by this very long-running transaction.
+
+Upgrade from 32-bit XID versions
+--------------------------------
+
+pg_upgrade doesn't change pages format itself. It is done lazily after.
+
+1. At first heap page read, tuples on a page are repacked to free 16 bytes
+at the end of a page, possibly freeing space from dead tuples.
+
+2A. 16 bytes of pd_special is added if there is a place for it
+
+2B. Page is converted to "Double XMAX" format if there is no place for
+pd_special
+
+3. If a page is in double XMAX format after its first read, and vacuum (or
+micro-vacuum at select query) could prune some tuples and free space for
+pd_special, prune_page will add pd_special and convert page from double XMAX
+to general 64-bit XID page format.
-- 
2.24.3 (Apple Git-128)


--cpok4wp6gsarlzvp--





^ permalink  raw  reply  [nested|flat] 265+ messages in thread

* [PATCH v6] README for 64bit xid
@ 2022-01-10 19:20  Pavel Borisov <[email protected]>
  0 siblings, 0 replies; 265+ messages in thread

From: Pavel Borisov @ 2022-01-10 19:20 UTC (permalink / raw)

Authors:
- Pavel Borisov <[email protected]>
- Maxim Orlov <[email protected]>
- Yura Sokolov <[email protected]> <[email protected]>
---
 src/backend/access/heap/README.XID64 | 128 +++++++++++++++++++++++++++
 1 file changed, 128 insertions(+)
 create mode 100644 src/backend/access/heap/README.XID64

diff --git a/src/backend/access/heap/README.XID64 b/src/backend/access/heap/README.XID64
new file mode 100644
index 00000000000..457ba9b9ef5
--- /dev/null
+++ b/src/backend/access/heap/README.XID64
@@ -0,0 +1,128 @@
+src/backend/access/heap/README.XID64
+
+64-bit Transaction ID's (XID)
+=============================
+
+A limited number (N = 2^32) of XID's required to do vacuum freeze to prevent
+wraparound every N/2 transactions. This causes performance degradation due
+to the need to exclusively lock tables while being vacuumed. In each
+wraparound cycle, SLRU buffers are also being cut.
+
+With 64-bit XID's wraparound is effectively postponed to a very distant
+future. Even in highly loaded systems that had 2^32 transactions per day
+it will take huge 2^31 days before the first enforced "vacuum to prevent
+wraparound"). Buffers cutting and routine vacuum are not enforced, and DBA
+can plan them independently at the time with the least system load and least
+critical for database performance. Also, it can be done less frequently
+(several times a year vs every several days) on systems with transaction rates
+similar to those mentioned above.
+
+On-disk tuple and page format
+-----------------------------
+
+On-disk tuple format remains unchanged. 32-bit t_xmin and t_xmax store the
+lower parts of 64-bit XMIN and XMAX values. Each heap page has additional
+64-bit pd_xid_base and pd_multi_base which are common for all tuples on a page.
+They are placed into a pd_special area - 16 bytes in the end of a heap page.
+Actual XMIN/XMAX for a tuple are calculated upon reading a tuple from a page
+as follows:
+
+XMIN = t_xmin + pd_xid_base. 					(1)
+XMAX = t_xmax + pd_xid_base/pd_multi_base.		(2)
+
+"Double XMAX" page format
+---------------------------------
+
+At first read of a heap page after pg_upgrade from 32-bit XID PostgreSQL
+version pd_special area with a size of 16 bytes should be added to a page.
+Though a page may not have space for this. Then it can be converted to a
+temporary format called "double XMAX".
+
+All tuples after pg-upgrade would necessarily have xmin = FrozenTransactionId.
+So we don't need tuple header t_xmin field and we reuse t_xmin to store higher
+32 bits of its XMAX.
+
+Double XMAX format is only for full pages that don't have 16 bytes for
+pd_special. So it neither has a�place for a single tuple. Insert and HOT update
+for double XMAX pages is impossible and not supported. We can only read or
+delete tuples from it.
+
+When we are able to prune page double XMAX it will be converted from it to
+general 64-bit XID page format with all operations on its tuples supported.
+
+In-memory tuple format
+----------------------
+
+In-memory tuple representation consists of two parts:
+- HeapTupleHeader from disk page (contains all heap tuple contents, not only
+header)
+- HeapTuple with additional in-memory fields
+
+HeapTuple for each tuple in memory stores t_xid_base/t_multi_base - a copies of
+page's pd_xid_base/pd_multi_base. With tuple's 32-bit t_xmin and t_xmax from
+HeapTupleHeader they are used to calculate actual 64-bit XMIN and XMAX:
+
+XMIN = t_xmin + t_xid_base. 					(3)
+XMAX = t_xmax + t_xid_base/t_multi_base.		(4)
+
+The downside of this is that we can not use tuple's XMIN and XMAX right away.
+We often need to re-read t_xmin and t_xmax - which could actually be pointers
+into a page in shared buffers and therefore they could be updated by any other
+backend.
+
+Update/delete with 64-bit XIDs and 32-bit t_xmin/t_xmax
+--------------------------------------------------------------
+
+When we try to delete/update a tuple, we check that XMAX for a page fits (2).
+I.e. that t_xmax will not be over MaxShortTransactionId relative to
+pd_xid_base/pd_multi_base of a its page.
+
+If the current XID doesn't fit a range
+(pd_xid_base, pd_xid_base + MaxShortTransactionId) (5):
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base/pd_multi_base on
+a page and update all t_xmin/t_xmax of the other tuples on the page to
+correspond new pd_xid_base/pd_multi_base.
+
+- If it was impossible, it will try to prune and freeze tuples on a page.
+
+- If this is unsuccessful it will throw an error. Normally this is very
+unlikely but if there is a very old living transaction with an age of around
+2^32 this can arise. Basically, this is a behavior similar to one during the
+vacuum to prevent wraparound when XID was 32-bit. Dba should take care and
+avoid very-long-living transactions with an age close to 2^32. So long-living
+transactions often they are most likely defunct.
+
+Insert with 64-bit XIDs and 32-bit t_xmin/t_xmax
+------------------------------------------------
+
+On insert we check if current XID fits a range (5). Otherwise:
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base for t_xmin will
+not be over MaxShortTransactionId.
+
+- If it is impossible, then it will try to prune and freeze tuples on a page.
+
+Known issue: if pd_xid_base could not be shifted to accommodate a tuple being
+inserted due to a very long-running transaction, we just throw an error. We
+neither try to insert a�tuple into another page nor mark the current page as
+full. So, in this (unlikely) case we will get regular insert errors on the next
+tries to insert to the page 'locked' by this very long-running transaction.
+
+Upgrade from 32-bit XID versions
+--------------------------------
+
+pg_upgrade doesn't change pages format itself. It is done lazily after.
+
+1. At first heap page read, tuples on a page are repacked to free 16 bytes
+at the end of a page, possibly freeing space from dead tuples.
+
+2A. 16 bytes of pd_special is added if there is a place for it
+
+2B. Page is converted to "Double XMAX" format if there is no place for
+pd_special
+
+3. If a page is in double XMAX format after its first read, and vacuum (or
+micro-vacuum at select query) could prune some tuples and free space for
+pd_special, prune_page will add pd_special and convert page from double XMAX
+to general 64-bit XID page format.
-- 
2.24.3 (Apple Git-128)


--cpok4wp6gsarlzvp--





^ permalink  raw  reply  [nested|flat] 265+ messages in thread

* [PATCH v6] README for 64bit xid
@ 2022-01-10 19:20  Pavel Borisov <[email protected]>
  0 siblings, 0 replies; 265+ messages in thread

From: Pavel Borisov @ 2022-01-10 19:20 UTC (permalink / raw)

Authors:
- Pavel Borisov <[email protected]>
- Maxim Orlov <[email protected]>
- Yura Sokolov <[email protected]> <[email protected]>
---
 src/backend/access/heap/README.XID64 | 128 +++++++++++++++++++++++++++
 1 file changed, 128 insertions(+)
 create mode 100644 src/backend/access/heap/README.XID64

diff --git a/src/backend/access/heap/README.XID64 b/src/backend/access/heap/README.XID64
new file mode 100644
index 00000000000..457ba9b9ef5
--- /dev/null
+++ b/src/backend/access/heap/README.XID64
@@ -0,0 +1,128 @@
+src/backend/access/heap/README.XID64
+
+64-bit Transaction ID's (XID)
+=============================
+
+A limited number (N = 2^32) of XID's required to do vacuum freeze to prevent
+wraparound every N/2 transactions. This causes performance degradation due
+to the need to exclusively lock tables while being vacuumed. In each
+wraparound cycle, SLRU buffers are also being cut.
+
+With 64-bit XID's wraparound is effectively postponed to a very distant
+future. Even in highly loaded systems that had 2^32 transactions per day
+it will take huge 2^31 days before the first enforced "vacuum to prevent
+wraparound"). Buffers cutting and routine vacuum are not enforced, and DBA
+can plan them independently at the time with the least system load and least
+critical for database performance. Also, it can be done less frequently
+(several times a year vs every several days) on systems with transaction rates
+similar to those mentioned above.
+
+On-disk tuple and page format
+-----------------------------
+
+On-disk tuple format remains unchanged. 32-bit t_xmin and t_xmax store the
+lower parts of 64-bit XMIN and XMAX values. Each heap page has additional
+64-bit pd_xid_base and pd_multi_base which are common for all tuples on a page.
+They are placed into a pd_special area - 16 bytes in the end of a heap page.
+Actual XMIN/XMAX for a tuple are calculated upon reading a tuple from a page
+as follows:
+
+XMIN = t_xmin + pd_xid_base. 					(1)
+XMAX = t_xmax + pd_xid_base/pd_multi_base.		(2)
+
+"Double XMAX" page format
+---------------------------------
+
+At first read of a heap page after pg_upgrade from 32-bit XID PostgreSQL
+version pd_special area with a size of 16 bytes should be added to a page.
+Though a page may not have space for this. Then it can be converted to a
+temporary format called "double XMAX".
+
+All tuples after pg-upgrade would necessarily have xmin = FrozenTransactionId.
+So we don't need tuple header t_xmin field and we reuse t_xmin to store higher
+32 bits of its XMAX.
+
+Double XMAX format is only for full pages that don't have 16 bytes for
+pd_special. So it neither has a�place for a single tuple. Insert and HOT update
+for double XMAX pages is impossible and not supported. We can only read or
+delete tuples from it.
+
+When we are able to prune page double XMAX it will be converted from it to
+general 64-bit XID page format with all operations on its tuples supported.
+
+In-memory tuple format
+----------------------
+
+In-memory tuple representation consists of two parts:
+- HeapTupleHeader from disk page (contains all heap tuple contents, not only
+header)
+- HeapTuple with additional in-memory fields
+
+HeapTuple for each tuple in memory stores t_xid_base/t_multi_base - a copies of
+page's pd_xid_base/pd_multi_base. With tuple's 32-bit t_xmin and t_xmax from
+HeapTupleHeader they are used to calculate actual 64-bit XMIN and XMAX:
+
+XMIN = t_xmin + t_xid_base. 					(3)
+XMAX = t_xmax + t_xid_base/t_multi_base.		(4)
+
+The downside of this is that we can not use tuple's XMIN and XMAX right away.
+We often need to re-read t_xmin and t_xmax - which could actually be pointers
+into a page in shared buffers and therefore they could be updated by any other
+backend.
+
+Update/delete with 64-bit XIDs and 32-bit t_xmin/t_xmax
+--------------------------------------------------------------
+
+When we try to delete/update a tuple, we check that XMAX for a page fits (2).
+I.e. that t_xmax will not be over MaxShortTransactionId relative to
+pd_xid_base/pd_multi_base of a its page.
+
+If the current XID doesn't fit a range
+(pd_xid_base, pd_xid_base + MaxShortTransactionId) (5):
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base/pd_multi_base on
+a page and update all t_xmin/t_xmax of the other tuples on the page to
+correspond new pd_xid_base/pd_multi_base.
+
+- If it was impossible, it will try to prune and freeze tuples on a page.
+
+- If this is unsuccessful it will throw an error. Normally this is very
+unlikely but if there is a very old living transaction with an age of around
+2^32 this can arise. Basically, this is a behavior similar to one during the
+vacuum to prevent wraparound when XID was 32-bit. Dba should take care and
+avoid very-long-living transactions with an age close to 2^32. So long-living
+transactions often they are most likely defunct.
+
+Insert with 64-bit XIDs and 32-bit t_xmin/t_xmax
+------------------------------------------------
+
+On insert we check if current XID fits a range (5). Otherwise:
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base for t_xmin will
+not be over MaxShortTransactionId.
+
+- If it is impossible, then it will try to prune and freeze tuples on a page.
+
+Known issue: if pd_xid_base could not be shifted to accommodate a tuple being
+inserted due to a very long-running transaction, we just throw an error. We
+neither try to insert a�tuple into another page nor mark the current page as
+full. So, in this (unlikely) case we will get regular insert errors on the next
+tries to insert to the page 'locked' by this very long-running transaction.
+
+Upgrade from 32-bit XID versions
+--------------------------------
+
+pg_upgrade doesn't change pages format itself. It is done lazily after.
+
+1. At first heap page read, tuples on a page are repacked to free 16 bytes
+at the end of a page, possibly freeing space from dead tuples.
+
+2A. 16 bytes of pd_special is added if there is a place for it
+
+2B. Page is converted to "Double XMAX" format if there is no place for
+pd_special
+
+3. If a page is in double XMAX format after its first read, and vacuum (or
+micro-vacuum at select query) could prune some tuples and free space for
+pd_special, prune_page will add pd_special and convert page from double XMAX
+to general 64-bit XID page format.
-- 
2.24.3 (Apple Git-128)


--cpok4wp6gsarlzvp--





^ permalink  raw  reply  [nested|flat] 265+ messages in thread

* [PATCH v6] README for 64bit xid
@ 2022-01-10 19:20  Pavel Borisov <[email protected]>
  0 siblings, 0 replies; 265+ messages in thread

From: Pavel Borisov @ 2022-01-10 19:20 UTC (permalink / raw)

Authors:
- Pavel Borisov <[email protected]>
- Maxim Orlov <[email protected]>
- Yura Sokolov <[email protected]> <[email protected]>
---
 src/backend/access/heap/README.XID64 | 128 +++++++++++++++++++++++++++
 1 file changed, 128 insertions(+)
 create mode 100644 src/backend/access/heap/README.XID64

diff --git a/src/backend/access/heap/README.XID64 b/src/backend/access/heap/README.XID64
new file mode 100644
index 00000000000..457ba9b9ef5
--- /dev/null
+++ b/src/backend/access/heap/README.XID64
@@ -0,0 +1,128 @@
+src/backend/access/heap/README.XID64
+
+64-bit Transaction ID's (XID)
+=============================
+
+A limited number (N = 2^32) of XID's required to do vacuum freeze to prevent
+wraparound every N/2 transactions. This causes performance degradation due
+to the need to exclusively lock tables while being vacuumed. In each
+wraparound cycle, SLRU buffers are also being cut.
+
+With 64-bit XID's wraparound is effectively postponed to a very distant
+future. Even in highly loaded systems that had 2^32 transactions per day
+it will take huge 2^31 days before the first enforced "vacuum to prevent
+wraparound"). Buffers cutting and routine vacuum are not enforced, and DBA
+can plan them independently at the time with the least system load and least
+critical for database performance. Also, it can be done less frequently
+(several times a year vs every several days) on systems with transaction rates
+similar to those mentioned above.
+
+On-disk tuple and page format
+-----------------------------
+
+On-disk tuple format remains unchanged. 32-bit t_xmin and t_xmax store the
+lower parts of 64-bit XMIN and XMAX values. Each heap page has additional
+64-bit pd_xid_base and pd_multi_base which are common for all tuples on a page.
+They are placed into a pd_special area - 16 bytes in the end of a heap page.
+Actual XMIN/XMAX for a tuple are calculated upon reading a tuple from a page
+as follows:
+
+XMIN = t_xmin + pd_xid_base. 					(1)
+XMAX = t_xmax + pd_xid_base/pd_multi_base.		(2)
+
+"Double XMAX" page format
+---------------------------------
+
+At first read of a heap page after pg_upgrade from 32-bit XID PostgreSQL
+version pd_special area with a size of 16 bytes should be added to a page.
+Though a page may not have space for this. Then it can be converted to a
+temporary format called "double XMAX".
+
+All tuples after pg-upgrade would necessarily have xmin = FrozenTransactionId.
+So we don't need tuple header t_xmin field and we reuse t_xmin to store higher
+32 bits of its XMAX.
+
+Double XMAX format is only for full pages that don't have 16 bytes for
+pd_special. So it neither has a�place for a single tuple. Insert and HOT update
+for double XMAX pages is impossible and not supported. We can only read or
+delete tuples from it.
+
+When we are able to prune page double XMAX it will be converted from it to
+general 64-bit XID page format with all operations on its tuples supported.
+
+In-memory tuple format
+----------------------
+
+In-memory tuple representation consists of two parts:
+- HeapTupleHeader from disk page (contains all heap tuple contents, not only
+header)
+- HeapTuple with additional in-memory fields
+
+HeapTuple for each tuple in memory stores t_xid_base/t_multi_base - a copies of
+page's pd_xid_base/pd_multi_base. With tuple's 32-bit t_xmin and t_xmax from
+HeapTupleHeader they are used to calculate actual 64-bit XMIN and XMAX:
+
+XMIN = t_xmin + t_xid_base. 					(3)
+XMAX = t_xmax + t_xid_base/t_multi_base.		(4)
+
+The downside of this is that we can not use tuple's XMIN and XMAX right away.
+We often need to re-read t_xmin and t_xmax - which could actually be pointers
+into a page in shared buffers and therefore they could be updated by any other
+backend.
+
+Update/delete with 64-bit XIDs and 32-bit t_xmin/t_xmax
+--------------------------------------------------------------
+
+When we try to delete/update a tuple, we check that XMAX for a page fits (2).
+I.e. that t_xmax will not be over MaxShortTransactionId relative to
+pd_xid_base/pd_multi_base of a its page.
+
+If the current XID doesn't fit a range
+(pd_xid_base, pd_xid_base + MaxShortTransactionId) (5):
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base/pd_multi_base on
+a page and update all t_xmin/t_xmax of the other tuples on the page to
+correspond new pd_xid_base/pd_multi_base.
+
+- If it was impossible, it will try to prune and freeze tuples on a page.
+
+- If this is unsuccessful it will throw an error. Normally this is very
+unlikely but if there is a very old living transaction with an age of around
+2^32 this can arise. Basically, this is a behavior similar to one during the
+vacuum to prevent wraparound when XID was 32-bit. Dba should take care and
+avoid very-long-living transactions with an age close to 2^32. So long-living
+transactions often they are most likely defunct.
+
+Insert with 64-bit XIDs and 32-bit t_xmin/t_xmax
+------------------------------------------------
+
+On insert we check if current XID fits a range (5). Otherwise:
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base for t_xmin will
+not be over MaxShortTransactionId.
+
+- If it is impossible, then it will try to prune and freeze tuples on a page.
+
+Known issue: if pd_xid_base could not be shifted to accommodate a tuple being
+inserted due to a very long-running transaction, we just throw an error. We
+neither try to insert a�tuple into another page nor mark the current page as
+full. So, in this (unlikely) case we will get regular insert errors on the next
+tries to insert to the page 'locked' by this very long-running transaction.
+
+Upgrade from 32-bit XID versions
+--------------------------------
+
+pg_upgrade doesn't change pages format itself. It is done lazily after.
+
+1. At first heap page read, tuples on a page are repacked to free 16 bytes
+at the end of a page, possibly freeing space from dead tuples.
+
+2A. 16 bytes of pd_special is added if there is a place for it
+
+2B. Page is converted to "Double XMAX" format if there is no place for
+pd_special
+
+3. If a page is in double XMAX format after its first read, and vacuum (or
+micro-vacuum at select query) could prune some tuples and free space for
+pd_special, prune_page will add pd_special and convert page from double XMAX
+to general 64-bit XID page format.
-- 
2.24.3 (Apple Git-128)


--cpok4wp6gsarlzvp--





^ permalink  raw  reply  [nested|flat] 265+ messages in thread

* [PATCH v6] README for 64bit xid
@ 2022-01-10 19:20  Pavel Borisov <[email protected]>
  0 siblings, 0 replies; 265+ messages in thread

From: Pavel Borisov @ 2022-01-10 19:20 UTC (permalink / raw)

Authors:
- Pavel Borisov <[email protected]>
- Maxim Orlov <[email protected]>
- Yura Sokolov <[email protected]> <[email protected]>
---
 src/backend/access/heap/README.XID64 | 128 +++++++++++++++++++++++++++
 1 file changed, 128 insertions(+)
 create mode 100644 src/backend/access/heap/README.XID64

diff --git a/src/backend/access/heap/README.XID64 b/src/backend/access/heap/README.XID64
new file mode 100644
index 00000000000..457ba9b9ef5
--- /dev/null
+++ b/src/backend/access/heap/README.XID64
@@ -0,0 +1,128 @@
+src/backend/access/heap/README.XID64
+
+64-bit Transaction ID's (XID)
+=============================
+
+A limited number (N = 2^32) of XID's required to do vacuum freeze to prevent
+wraparound every N/2 transactions. This causes performance degradation due
+to the need to exclusively lock tables while being vacuumed. In each
+wraparound cycle, SLRU buffers are also being cut.
+
+With 64-bit XID's wraparound is effectively postponed to a very distant
+future. Even in highly loaded systems that had 2^32 transactions per day
+it will take huge 2^31 days before the first enforced "vacuum to prevent
+wraparound"). Buffers cutting and routine vacuum are not enforced, and DBA
+can plan them independently at the time with the least system load and least
+critical for database performance. Also, it can be done less frequently
+(several times a year vs every several days) on systems with transaction rates
+similar to those mentioned above.
+
+On-disk tuple and page format
+-----------------------------
+
+On-disk tuple format remains unchanged. 32-bit t_xmin and t_xmax store the
+lower parts of 64-bit XMIN and XMAX values. Each heap page has additional
+64-bit pd_xid_base and pd_multi_base which are common for all tuples on a page.
+They are placed into a pd_special area - 16 bytes in the end of a heap page.
+Actual XMIN/XMAX for a tuple are calculated upon reading a tuple from a page
+as follows:
+
+XMIN = t_xmin + pd_xid_base. 					(1)
+XMAX = t_xmax + pd_xid_base/pd_multi_base.		(2)
+
+"Double XMAX" page format
+---------------------------------
+
+At first read of a heap page after pg_upgrade from 32-bit XID PostgreSQL
+version pd_special area with a size of 16 bytes should be added to a page.
+Though a page may not have space for this. Then it can be converted to a
+temporary format called "double XMAX".
+
+All tuples after pg-upgrade would necessarily have xmin = FrozenTransactionId.
+So we don't need tuple header t_xmin field and we reuse t_xmin to store higher
+32 bits of its XMAX.
+
+Double XMAX format is only for full pages that don't have 16 bytes for
+pd_special. So it neither has a�place for a single tuple. Insert and HOT update
+for double XMAX pages is impossible and not supported. We can only read or
+delete tuples from it.
+
+When we are able to prune page double XMAX it will be converted from it to
+general 64-bit XID page format with all operations on its tuples supported.
+
+In-memory tuple format
+----------------------
+
+In-memory tuple representation consists of two parts:
+- HeapTupleHeader from disk page (contains all heap tuple contents, not only
+header)
+- HeapTuple with additional in-memory fields
+
+HeapTuple for each tuple in memory stores t_xid_base/t_multi_base - a copies of
+page's pd_xid_base/pd_multi_base. With tuple's 32-bit t_xmin and t_xmax from
+HeapTupleHeader they are used to calculate actual 64-bit XMIN and XMAX:
+
+XMIN = t_xmin + t_xid_base. 					(3)
+XMAX = t_xmax + t_xid_base/t_multi_base.		(4)
+
+The downside of this is that we can not use tuple's XMIN and XMAX right away.
+We often need to re-read t_xmin and t_xmax - which could actually be pointers
+into a page in shared buffers and therefore they could be updated by any other
+backend.
+
+Update/delete with 64-bit XIDs and 32-bit t_xmin/t_xmax
+--------------------------------------------------------------
+
+When we try to delete/update a tuple, we check that XMAX for a page fits (2).
+I.e. that t_xmax will not be over MaxShortTransactionId relative to
+pd_xid_base/pd_multi_base of a its page.
+
+If the current XID doesn't fit a range
+(pd_xid_base, pd_xid_base + MaxShortTransactionId) (5):
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base/pd_multi_base on
+a page and update all t_xmin/t_xmax of the other tuples on the page to
+correspond new pd_xid_base/pd_multi_base.
+
+- If it was impossible, it will try to prune and freeze tuples on a page.
+
+- If this is unsuccessful it will throw an error. Normally this is very
+unlikely but if there is a very old living transaction with an age of around
+2^32 this can arise. Basically, this is a behavior similar to one during the
+vacuum to prevent wraparound when XID was 32-bit. Dba should take care and
+avoid very-long-living transactions with an age close to 2^32. So long-living
+transactions often they are most likely defunct.
+
+Insert with 64-bit XIDs and 32-bit t_xmin/t_xmax
+------------------------------------------------
+
+On insert we check if current XID fits a range (5). Otherwise:
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base for t_xmin will
+not be over MaxShortTransactionId.
+
+- If it is impossible, then it will try to prune and freeze tuples on a page.
+
+Known issue: if pd_xid_base could not be shifted to accommodate a tuple being
+inserted due to a very long-running transaction, we just throw an error. We
+neither try to insert a�tuple into another page nor mark the current page as
+full. So, in this (unlikely) case we will get regular insert errors on the next
+tries to insert to the page 'locked' by this very long-running transaction.
+
+Upgrade from 32-bit XID versions
+--------------------------------
+
+pg_upgrade doesn't change pages format itself. It is done lazily after.
+
+1. At first heap page read, tuples on a page are repacked to free 16 bytes
+at the end of a page, possibly freeing space from dead tuples.
+
+2A. 16 bytes of pd_special is added if there is a place for it
+
+2B. Page is converted to "Double XMAX" format if there is no place for
+pd_special
+
+3. If a page is in double XMAX format after its first read, and vacuum (or
+micro-vacuum at select query) could prune some tuples and free space for
+pd_special, prune_page will add pd_special and convert page from double XMAX
+to general 64-bit XID page format.
-- 
2.24.3 (Apple Git-128)


--cpok4wp6gsarlzvp--





^ permalink  raw  reply  [nested|flat] 265+ messages in thread

* [PATCH v6] README for 64bit xid
@ 2022-01-10 19:20  Pavel Borisov <[email protected]>
  0 siblings, 0 replies; 265+ messages in thread

From: Pavel Borisov @ 2022-01-10 19:20 UTC (permalink / raw)

Authors:
- Pavel Borisov <[email protected]>
- Maxim Orlov <[email protected]>
- Yura Sokolov <[email protected]> <[email protected]>
---
 src/backend/access/heap/README.XID64 | 128 +++++++++++++++++++++++++++
 1 file changed, 128 insertions(+)
 create mode 100644 src/backend/access/heap/README.XID64

diff --git a/src/backend/access/heap/README.XID64 b/src/backend/access/heap/README.XID64
new file mode 100644
index 00000000000..457ba9b9ef5
--- /dev/null
+++ b/src/backend/access/heap/README.XID64
@@ -0,0 +1,128 @@
+src/backend/access/heap/README.XID64
+
+64-bit Transaction ID's (XID)
+=============================
+
+A limited number (N = 2^32) of XID's required to do vacuum freeze to prevent
+wraparound every N/2 transactions. This causes performance degradation due
+to the need to exclusively lock tables while being vacuumed. In each
+wraparound cycle, SLRU buffers are also being cut.
+
+With 64-bit XID's wraparound is effectively postponed to a very distant
+future. Even in highly loaded systems that had 2^32 transactions per day
+it will take huge 2^31 days before the first enforced "vacuum to prevent
+wraparound"). Buffers cutting and routine vacuum are not enforced, and DBA
+can plan them independently at the time with the least system load and least
+critical for database performance. Also, it can be done less frequently
+(several times a year vs every several days) on systems with transaction rates
+similar to those mentioned above.
+
+On-disk tuple and page format
+-----------------------------
+
+On-disk tuple format remains unchanged. 32-bit t_xmin and t_xmax store the
+lower parts of 64-bit XMIN and XMAX values. Each heap page has additional
+64-bit pd_xid_base and pd_multi_base which are common for all tuples on a page.
+They are placed into a pd_special area - 16 bytes in the end of a heap page.
+Actual XMIN/XMAX for a tuple are calculated upon reading a tuple from a page
+as follows:
+
+XMIN = t_xmin + pd_xid_base. 					(1)
+XMAX = t_xmax + pd_xid_base/pd_multi_base.		(2)
+
+"Double XMAX" page format
+---------------------------------
+
+At first read of a heap page after pg_upgrade from 32-bit XID PostgreSQL
+version pd_special area with a size of 16 bytes should be added to a page.
+Though a page may not have space for this. Then it can be converted to a
+temporary format called "double XMAX".
+
+All tuples after pg-upgrade would necessarily have xmin = FrozenTransactionId.
+So we don't need tuple header t_xmin field and we reuse t_xmin to store higher
+32 bits of its XMAX.
+
+Double XMAX format is only for full pages that don't have 16 bytes for
+pd_special. So it neither has a�place for a single tuple. Insert and HOT update
+for double XMAX pages is impossible and not supported. We can only read or
+delete tuples from it.
+
+When we are able to prune page double XMAX it will be converted from it to
+general 64-bit XID page format with all operations on its tuples supported.
+
+In-memory tuple format
+----------------------
+
+In-memory tuple representation consists of two parts:
+- HeapTupleHeader from disk page (contains all heap tuple contents, not only
+header)
+- HeapTuple with additional in-memory fields
+
+HeapTuple for each tuple in memory stores t_xid_base/t_multi_base - a copies of
+page's pd_xid_base/pd_multi_base. With tuple's 32-bit t_xmin and t_xmax from
+HeapTupleHeader they are used to calculate actual 64-bit XMIN and XMAX:
+
+XMIN = t_xmin + t_xid_base. 					(3)
+XMAX = t_xmax + t_xid_base/t_multi_base.		(4)
+
+The downside of this is that we can not use tuple's XMIN and XMAX right away.
+We often need to re-read t_xmin and t_xmax - which could actually be pointers
+into a page in shared buffers and therefore they could be updated by any other
+backend.
+
+Update/delete with 64-bit XIDs and 32-bit t_xmin/t_xmax
+--------------------------------------------------------------
+
+When we try to delete/update a tuple, we check that XMAX for a page fits (2).
+I.e. that t_xmax will not be over MaxShortTransactionId relative to
+pd_xid_base/pd_multi_base of a its page.
+
+If the current XID doesn't fit a range
+(pd_xid_base, pd_xid_base + MaxShortTransactionId) (5):
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base/pd_multi_base on
+a page and update all t_xmin/t_xmax of the other tuples on the page to
+correspond new pd_xid_base/pd_multi_base.
+
+- If it was impossible, it will try to prune and freeze tuples on a page.
+
+- If this is unsuccessful it will throw an error. Normally this is very
+unlikely but if there is a very old living transaction with an age of around
+2^32 this can arise. Basically, this is a behavior similar to one during the
+vacuum to prevent wraparound when XID was 32-bit. Dba should take care and
+avoid very-long-living transactions with an age close to 2^32. So long-living
+transactions often they are most likely defunct.
+
+Insert with 64-bit XIDs and 32-bit t_xmin/t_xmax
+------------------------------------------------
+
+On insert we check if current XID fits a range (5). Otherwise:
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base for t_xmin will
+not be over MaxShortTransactionId.
+
+- If it is impossible, then it will try to prune and freeze tuples on a page.
+
+Known issue: if pd_xid_base could not be shifted to accommodate a tuple being
+inserted due to a very long-running transaction, we just throw an error. We
+neither try to insert a�tuple into another page nor mark the current page as
+full. So, in this (unlikely) case we will get regular insert errors on the next
+tries to insert to the page 'locked' by this very long-running transaction.
+
+Upgrade from 32-bit XID versions
+--------------------------------
+
+pg_upgrade doesn't change pages format itself. It is done lazily after.
+
+1. At first heap page read, tuples on a page are repacked to free 16 bytes
+at the end of a page, possibly freeing space from dead tuples.
+
+2A. 16 bytes of pd_special is added if there is a place for it
+
+2B. Page is converted to "Double XMAX" format if there is no place for
+pd_special
+
+3. If a page is in double XMAX format after its first read, and vacuum (or
+micro-vacuum at select query) could prune some tuples and free space for
+pd_special, prune_page will add pd_special and convert page from double XMAX
+to general 64-bit XID page format.
-- 
2.24.3 (Apple Git-128)


--cpok4wp6gsarlzvp--





^ permalink  raw  reply  [nested|flat] 265+ messages in thread

* [PATCH v6] README for 64bit xid
@ 2022-01-10 19:20  Pavel Borisov <[email protected]>
  0 siblings, 0 replies; 265+ messages in thread

From: Pavel Borisov @ 2022-01-10 19:20 UTC (permalink / raw)

Authors:
- Pavel Borisov <[email protected]>
- Maxim Orlov <[email protected]>
- Yura Sokolov <[email protected]> <[email protected]>
---
 src/backend/access/heap/README.XID64 | 128 +++++++++++++++++++++++++++
 1 file changed, 128 insertions(+)
 create mode 100644 src/backend/access/heap/README.XID64

diff --git a/src/backend/access/heap/README.XID64 b/src/backend/access/heap/README.XID64
new file mode 100644
index 00000000000..457ba9b9ef5
--- /dev/null
+++ b/src/backend/access/heap/README.XID64
@@ -0,0 +1,128 @@
+src/backend/access/heap/README.XID64
+
+64-bit Transaction ID's (XID)
+=============================
+
+A limited number (N = 2^32) of XID's required to do vacuum freeze to prevent
+wraparound every N/2 transactions. This causes performance degradation due
+to the need to exclusively lock tables while being vacuumed. In each
+wraparound cycle, SLRU buffers are also being cut.
+
+With 64-bit XID's wraparound is effectively postponed to a very distant
+future. Even in highly loaded systems that had 2^32 transactions per day
+it will take huge 2^31 days before the first enforced "vacuum to prevent
+wraparound"). Buffers cutting and routine vacuum are not enforced, and DBA
+can plan them independently at the time with the least system load and least
+critical for database performance. Also, it can be done less frequently
+(several times a year vs every several days) on systems with transaction rates
+similar to those mentioned above.
+
+On-disk tuple and page format
+-----------------------------
+
+On-disk tuple format remains unchanged. 32-bit t_xmin and t_xmax store the
+lower parts of 64-bit XMIN and XMAX values. Each heap page has additional
+64-bit pd_xid_base and pd_multi_base which are common for all tuples on a page.
+They are placed into a pd_special area - 16 bytes in the end of a heap page.
+Actual XMIN/XMAX for a tuple are calculated upon reading a tuple from a page
+as follows:
+
+XMIN = t_xmin + pd_xid_base. 					(1)
+XMAX = t_xmax + pd_xid_base/pd_multi_base.		(2)
+
+"Double XMAX" page format
+---------------------------------
+
+At first read of a heap page after pg_upgrade from 32-bit XID PostgreSQL
+version pd_special area with a size of 16 bytes should be added to a page.
+Though a page may not have space for this. Then it can be converted to a
+temporary format called "double XMAX".
+
+All tuples after pg-upgrade would necessarily have xmin = FrozenTransactionId.
+So we don't need tuple header t_xmin field and we reuse t_xmin to store higher
+32 bits of its XMAX.
+
+Double XMAX format is only for full pages that don't have 16 bytes for
+pd_special. So it neither has a�place for a single tuple. Insert and HOT update
+for double XMAX pages is impossible and not supported. We can only read or
+delete tuples from it.
+
+When we are able to prune page double XMAX it will be converted from it to
+general 64-bit XID page format with all operations on its tuples supported.
+
+In-memory tuple format
+----------------------
+
+In-memory tuple representation consists of two parts:
+- HeapTupleHeader from disk page (contains all heap tuple contents, not only
+header)
+- HeapTuple with additional in-memory fields
+
+HeapTuple for each tuple in memory stores t_xid_base/t_multi_base - a copies of
+page's pd_xid_base/pd_multi_base. With tuple's 32-bit t_xmin and t_xmax from
+HeapTupleHeader they are used to calculate actual 64-bit XMIN and XMAX:
+
+XMIN = t_xmin + t_xid_base. 					(3)
+XMAX = t_xmax + t_xid_base/t_multi_base.		(4)
+
+The downside of this is that we can not use tuple's XMIN and XMAX right away.
+We often need to re-read t_xmin and t_xmax - which could actually be pointers
+into a page in shared buffers and therefore they could be updated by any other
+backend.
+
+Update/delete with 64-bit XIDs and 32-bit t_xmin/t_xmax
+--------------------------------------------------------------
+
+When we try to delete/update a tuple, we check that XMAX for a page fits (2).
+I.e. that t_xmax will not be over MaxShortTransactionId relative to
+pd_xid_base/pd_multi_base of a its page.
+
+If the current XID doesn't fit a range
+(pd_xid_base, pd_xid_base + MaxShortTransactionId) (5):
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base/pd_multi_base on
+a page and update all t_xmin/t_xmax of the other tuples on the page to
+correspond new pd_xid_base/pd_multi_base.
+
+- If it was impossible, it will try to prune and freeze tuples on a page.
+
+- If this is unsuccessful it will throw an error. Normally this is very
+unlikely but if there is a very old living transaction with an age of around
+2^32 this can arise. Basically, this is a behavior similar to one during the
+vacuum to prevent wraparound when XID was 32-bit. Dba should take care and
+avoid very-long-living transactions with an age close to 2^32. So long-living
+transactions often they are most likely defunct.
+
+Insert with 64-bit XIDs and 32-bit t_xmin/t_xmax
+------------------------------------------------
+
+On insert we check if current XID fits a range (5). Otherwise:
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base for t_xmin will
+not be over MaxShortTransactionId.
+
+- If it is impossible, then it will try to prune and freeze tuples on a page.
+
+Known issue: if pd_xid_base could not be shifted to accommodate a tuple being
+inserted due to a very long-running transaction, we just throw an error. We
+neither try to insert a�tuple into another page nor mark the current page as
+full. So, in this (unlikely) case we will get regular insert errors on the next
+tries to insert to the page 'locked' by this very long-running transaction.
+
+Upgrade from 32-bit XID versions
+--------------------------------
+
+pg_upgrade doesn't change pages format itself. It is done lazily after.
+
+1. At first heap page read, tuples on a page are repacked to free 16 bytes
+at the end of a page, possibly freeing space from dead tuples.
+
+2A. 16 bytes of pd_special is added if there is a place for it
+
+2B. Page is converted to "Double XMAX" format if there is no place for
+pd_special
+
+3. If a page is in double XMAX format after its first read, and vacuum (or
+micro-vacuum at select query) could prune some tuples and free space for
+pd_special, prune_page will add pd_special and convert page from double XMAX
+to general 64-bit XID page format.
-- 
2.24.3 (Apple Git-128)


--cpok4wp6gsarlzvp--





^ permalink  raw  reply  [nested|flat] 265+ messages in thread

* [PATCH v6] README for 64bit xid
@ 2022-01-10 19:20  Pavel Borisov <[email protected]>
  0 siblings, 0 replies; 265+ messages in thread

From: Pavel Borisov @ 2022-01-10 19:20 UTC (permalink / raw)

Authors:
- Pavel Borisov <[email protected]>
- Maxim Orlov <[email protected]>
- Yura Sokolov <[email protected]> <[email protected]>
---
 src/backend/access/heap/README.XID64 | 128 +++++++++++++++++++++++++++
 1 file changed, 128 insertions(+)
 create mode 100644 src/backend/access/heap/README.XID64

diff --git a/src/backend/access/heap/README.XID64 b/src/backend/access/heap/README.XID64
new file mode 100644
index 00000000000..457ba9b9ef5
--- /dev/null
+++ b/src/backend/access/heap/README.XID64
@@ -0,0 +1,128 @@
+src/backend/access/heap/README.XID64
+
+64-bit Transaction ID's (XID)
+=============================
+
+A limited number (N = 2^32) of XID's required to do vacuum freeze to prevent
+wraparound every N/2 transactions. This causes performance degradation due
+to the need to exclusively lock tables while being vacuumed. In each
+wraparound cycle, SLRU buffers are also being cut.
+
+With 64-bit XID's wraparound is effectively postponed to a very distant
+future. Even in highly loaded systems that had 2^32 transactions per day
+it will take huge 2^31 days before the first enforced "vacuum to prevent
+wraparound"). Buffers cutting and routine vacuum are not enforced, and DBA
+can plan them independently at the time with the least system load and least
+critical for database performance. Also, it can be done less frequently
+(several times a year vs every several days) on systems with transaction rates
+similar to those mentioned above.
+
+On-disk tuple and page format
+-----------------------------
+
+On-disk tuple format remains unchanged. 32-bit t_xmin and t_xmax store the
+lower parts of 64-bit XMIN and XMAX values. Each heap page has additional
+64-bit pd_xid_base and pd_multi_base which are common for all tuples on a page.
+They are placed into a pd_special area - 16 bytes in the end of a heap page.
+Actual XMIN/XMAX for a tuple are calculated upon reading a tuple from a page
+as follows:
+
+XMIN = t_xmin + pd_xid_base. 					(1)
+XMAX = t_xmax + pd_xid_base/pd_multi_base.		(2)
+
+"Double XMAX" page format
+---------------------------------
+
+At first read of a heap page after pg_upgrade from 32-bit XID PostgreSQL
+version pd_special area with a size of 16 bytes should be added to a page.
+Though a page may not have space for this. Then it can be converted to a
+temporary format called "double XMAX".
+
+All tuples after pg-upgrade would necessarily have xmin = FrozenTransactionId.
+So we don't need tuple header t_xmin field and we reuse t_xmin to store higher
+32 bits of its XMAX.
+
+Double XMAX format is only for full pages that don't have 16 bytes for
+pd_special. So it neither has a�place for a single tuple. Insert and HOT update
+for double XMAX pages is impossible and not supported. We can only read or
+delete tuples from it.
+
+When we are able to prune page double XMAX it will be converted from it to
+general 64-bit XID page format with all operations on its tuples supported.
+
+In-memory tuple format
+----------------------
+
+In-memory tuple representation consists of two parts:
+- HeapTupleHeader from disk page (contains all heap tuple contents, not only
+header)
+- HeapTuple with additional in-memory fields
+
+HeapTuple for each tuple in memory stores t_xid_base/t_multi_base - a copies of
+page's pd_xid_base/pd_multi_base. With tuple's 32-bit t_xmin and t_xmax from
+HeapTupleHeader they are used to calculate actual 64-bit XMIN and XMAX:
+
+XMIN = t_xmin + t_xid_base. 					(3)
+XMAX = t_xmax + t_xid_base/t_multi_base.		(4)
+
+The downside of this is that we can not use tuple's XMIN and XMAX right away.
+We often need to re-read t_xmin and t_xmax - which could actually be pointers
+into a page in shared buffers and therefore they could be updated by any other
+backend.
+
+Update/delete with 64-bit XIDs and 32-bit t_xmin/t_xmax
+--------------------------------------------------------------
+
+When we try to delete/update a tuple, we check that XMAX for a page fits (2).
+I.e. that t_xmax will not be over MaxShortTransactionId relative to
+pd_xid_base/pd_multi_base of a its page.
+
+If the current XID doesn't fit a range
+(pd_xid_base, pd_xid_base + MaxShortTransactionId) (5):
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base/pd_multi_base on
+a page and update all t_xmin/t_xmax of the other tuples on the page to
+correspond new pd_xid_base/pd_multi_base.
+
+- If it was impossible, it will try to prune and freeze tuples on a page.
+
+- If this is unsuccessful it will throw an error. Normally this is very
+unlikely but if there is a very old living transaction with an age of around
+2^32 this can arise. Basically, this is a behavior similar to one during the
+vacuum to prevent wraparound when XID was 32-bit. Dba should take care and
+avoid very-long-living transactions with an age close to 2^32. So long-living
+transactions often they are most likely defunct.
+
+Insert with 64-bit XIDs and 32-bit t_xmin/t_xmax
+------------------------------------------------
+
+On insert we check if current XID fits a range (5). Otherwise:
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base for t_xmin will
+not be over MaxShortTransactionId.
+
+- If it is impossible, then it will try to prune and freeze tuples on a page.
+
+Known issue: if pd_xid_base could not be shifted to accommodate a tuple being
+inserted due to a very long-running transaction, we just throw an error. We
+neither try to insert a�tuple into another page nor mark the current page as
+full. So, in this (unlikely) case we will get regular insert errors on the next
+tries to insert to the page 'locked' by this very long-running transaction.
+
+Upgrade from 32-bit XID versions
+--------------------------------
+
+pg_upgrade doesn't change pages format itself. It is done lazily after.
+
+1. At first heap page read, tuples on a page are repacked to free 16 bytes
+at the end of a page, possibly freeing space from dead tuples.
+
+2A. 16 bytes of pd_special is added if there is a place for it
+
+2B. Page is converted to "Double XMAX" format if there is no place for
+pd_special
+
+3. If a page is in double XMAX format after its first read, and vacuum (or
+micro-vacuum at select query) could prune some tuples and free space for
+pd_special, prune_page will add pd_special and convert page from double XMAX
+to general 64-bit XID page format.
-- 
2.24.3 (Apple Git-128)


--cpok4wp6gsarlzvp--





^ permalink  raw  reply  [nested|flat] 265+ messages in thread

* [PATCH v6] README for 64bit xid
@ 2022-01-10 19:20  Pavel Borisov <[email protected]>
  0 siblings, 0 replies; 265+ messages in thread

From: Pavel Borisov @ 2022-01-10 19:20 UTC (permalink / raw)

Authors:
- Pavel Borisov <[email protected]>
- Maxim Orlov <[email protected]>
- Yura Sokolov <[email protected]> <[email protected]>
---
 src/backend/access/heap/README.XID64 | 128 +++++++++++++++++++++++++++
 1 file changed, 128 insertions(+)
 create mode 100644 src/backend/access/heap/README.XID64

diff --git a/src/backend/access/heap/README.XID64 b/src/backend/access/heap/README.XID64
new file mode 100644
index 00000000000..457ba9b9ef5
--- /dev/null
+++ b/src/backend/access/heap/README.XID64
@@ -0,0 +1,128 @@
+src/backend/access/heap/README.XID64
+
+64-bit Transaction ID's (XID)
+=============================
+
+A limited number (N = 2^32) of XID's required to do vacuum freeze to prevent
+wraparound every N/2 transactions. This causes performance degradation due
+to the need to exclusively lock tables while being vacuumed. In each
+wraparound cycle, SLRU buffers are also being cut.
+
+With 64-bit XID's wraparound is effectively postponed to a very distant
+future. Even in highly loaded systems that had 2^32 transactions per day
+it will take huge 2^31 days before the first enforced "vacuum to prevent
+wraparound"). Buffers cutting and routine vacuum are not enforced, and DBA
+can plan them independently at the time with the least system load and least
+critical for database performance. Also, it can be done less frequently
+(several times a year vs every several days) on systems with transaction rates
+similar to those mentioned above.
+
+On-disk tuple and page format
+-----------------------------
+
+On-disk tuple format remains unchanged. 32-bit t_xmin and t_xmax store the
+lower parts of 64-bit XMIN and XMAX values. Each heap page has additional
+64-bit pd_xid_base and pd_multi_base which are common for all tuples on a page.
+They are placed into a pd_special area - 16 bytes in the end of a heap page.
+Actual XMIN/XMAX for a tuple are calculated upon reading a tuple from a page
+as follows:
+
+XMIN = t_xmin + pd_xid_base. 					(1)
+XMAX = t_xmax + pd_xid_base/pd_multi_base.		(2)
+
+"Double XMAX" page format
+---------------------------------
+
+At first read of a heap page after pg_upgrade from 32-bit XID PostgreSQL
+version pd_special area with a size of 16 bytes should be added to a page.
+Though a page may not have space for this. Then it can be converted to a
+temporary format called "double XMAX".
+
+All tuples after pg-upgrade would necessarily have xmin = FrozenTransactionId.
+So we don't need tuple header t_xmin field and we reuse t_xmin to store higher
+32 bits of its XMAX.
+
+Double XMAX format is only for full pages that don't have 16 bytes for
+pd_special. So it neither has a�place for a single tuple. Insert and HOT update
+for double XMAX pages is impossible and not supported. We can only read or
+delete tuples from it.
+
+When we are able to prune page double XMAX it will be converted from it to
+general 64-bit XID page format with all operations on its tuples supported.
+
+In-memory tuple format
+----------------------
+
+In-memory tuple representation consists of two parts:
+- HeapTupleHeader from disk page (contains all heap tuple contents, not only
+header)
+- HeapTuple with additional in-memory fields
+
+HeapTuple for each tuple in memory stores t_xid_base/t_multi_base - a copies of
+page's pd_xid_base/pd_multi_base. With tuple's 32-bit t_xmin and t_xmax from
+HeapTupleHeader they are used to calculate actual 64-bit XMIN and XMAX:
+
+XMIN = t_xmin + t_xid_base. 					(3)
+XMAX = t_xmax + t_xid_base/t_multi_base.		(4)
+
+The downside of this is that we can not use tuple's XMIN and XMAX right away.
+We often need to re-read t_xmin and t_xmax - which could actually be pointers
+into a page in shared buffers and therefore they could be updated by any other
+backend.
+
+Update/delete with 64-bit XIDs and 32-bit t_xmin/t_xmax
+--------------------------------------------------------------
+
+When we try to delete/update a tuple, we check that XMAX for a page fits (2).
+I.e. that t_xmax will not be over MaxShortTransactionId relative to
+pd_xid_base/pd_multi_base of a its page.
+
+If the current XID doesn't fit a range
+(pd_xid_base, pd_xid_base + MaxShortTransactionId) (5):
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base/pd_multi_base on
+a page and update all t_xmin/t_xmax of the other tuples on the page to
+correspond new pd_xid_base/pd_multi_base.
+
+- If it was impossible, it will try to prune and freeze tuples on a page.
+
+- If this is unsuccessful it will throw an error. Normally this is very
+unlikely but if there is a very old living transaction with an age of around
+2^32 this can arise. Basically, this is a behavior similar to one during the
+vacuum to prevent wraparound when XID was 32-bit. Dba should take care and
+avoid very-long-living transactions with an age close to 2^32. So long-living
+transactions often they are most likely defunct.
+
+Insert with 64-bit XIDs and 32-bit t_xmin/t_xmax
+------------------------------------------------
+
+On insert we check if current XID fits a range (5). Otherwise:
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base for t_xmin will
+not be over MaxShortTransactionId.
+
+- If it is impossible, then it will try to prune and freeze tuples on a page.
+
+Known issue: if pd_xid_base could not be shifted to accommodate a tuple being
+inserted due to a very long-running transaction, we just throw an error. We
+neither try to insert a�tuple into another page nor mark the current page as
+full. So, in this (unlikely) case we will get regular insert errors on the next
+tries to insert to the page 'locked' by this very long-running transaction.
+
+Upgrade from 32-bit XID versions
+--------------------------------
+
+pg_upgrade doesn't change pages format itself. It is done lazily after.
+
+1. At first heap page read, tuples on a page are repacked to free 16 bytes
+at the end of a page, possibly freeing space from dead tuples.
+
+2A. 16 bytes of pd_special is added if there is a place for it
+
+2B. Page is converted to "Double XMAX" format if there is no place for
+pd_special
+
+3. If a page is in double XMAX format after its first read, and vacuum (or
+micro-vacuum at select query) could prune some tuples and free space for
+pd_special, prune_page will add pd_special and convert page from double XMAX
+to general 64-bit XID page format.
-- 
2.24.3 (Apple Git-128)


--cpok4wp6gsarlzvp--





^ permalink  raw  reply  [nested|flat] 265+ messages in thread

* [PATCH v6] README for 64bit xid
@ 2022-01-10 19:20  Pavel Borisov <[email protected]>
  0 siblings, 0 replies; 265+ messages in thread

From: Pavel Borisov @ 2022-01-10 19:20 UTC (permalink / raw)

Authors:
- Pavel Borisov <[email protected]>
- Maxim Orlov <[email protected]>
- Yura Sokolov <[email protected]> <[email protected]>
---
 src/backend/access/heap/README.XID64 | 128 +++++++++++++++++++++++++++
 1 file changed, 128 insertions(+)
 create mode 100644 src/backend/access/heap/README.XID64

diff --git a/src/backend/access/heap/README.XID64 b/src/backend/access/heap/README.XID64
new file mode 100644
index 00000000000..457ba9b9ef5
--- /dev/null
+++ b/src/backend/access/heap/README.XID64
@@ -0,0 +1,128 @@
+src/backend/access/heap/README.XID64
+
+64-bit Transaction ID's (XID)
+=============================
+
+A limited number (N = 2^32) of XID's required to do vacuum freeze to prevent
+wraparound every N/2 transactions. This causes performance degradation due
+to the need to exclusively lock tables while being vacuumed. In each
+wraparound cycle, SLRU buffers are also being cut.
+
+With 64-bit XID's wraparound is effectively postponed to a very distant
+future. Even in highly loaded systems that had 2^32 transactions per day
+it will take huge 2^31 days before the first enforced "vacuum to prevent
+wraparound"). Buffers cutting and routine vacuum are not enforced, and DBA
+can plan them independently at the time with the least system load and least
+critical for database performance. Also, it can be done less frequently
+(several times a year vs every several days) on systems with transaction rates
+similar to those mentioned above.
+
+On-disk tuple and page format
+-----------------------------
+
+On-disk tuple format remains unchanged. 32-bit t_xmin and t_xmax store the
+lower parts of 64-bit XMIN and XMAX values. Each heap page has additional
+64-bit pd_xid_base and pd_multi_base which are common for all tuples on a page.
+They are placed into a pd_special area - 16 bytes in the end of a heap page.
+Actual XMIN/XMAX for a tuple are calculated upon reading a tuple from a page
+as follows:
+
+XMIN = t_xmin + pd_xid_base. 					(1)
+XMAX = t_xmax + pd_xid_base/pd_multi_base.		(2)
+
+"Double XMAX" page format
+---------------------------------
+
+At first read of a heap page after pg_upgrade from 32-bit XID PostgreSQL
+version pd_special area with a size of 16 bytes should be added to a page.
+Though a page may not have space for this. Then it can be converted to a
+temporary format called "double XMAX".
+
+All tuples after pg-upgrade would necessarily have xmin = FrozenTransactionId.
+So we don't need tuple header t_xmin field and we reuse t_xmin to store higher
+32 bits of its XMAX.
+
+Double XMAX format is only for full pages that don't have 16 bytes for
+pd_special. So it neither has a�place for a single tuple. Insert and HOT update
+for double XMAX pages is impossible and not supported. We can only read or
+delete tuples from it.
+
+When we are able to prune page double XMAX it will be converted from it to
+general 64-bit XID page format with all operations on its tuples supported.
+
+In-memory tuple format
+----------------------
+
+In-memory tuple representation consists of two parts:
+- HeapTupleHeader from disk page (contains all heap tuple contents, not only
+header)
+- HeapTuple with additional in-memory fields
+
+HeapTuple for each tuple in memory stores t_xid_base/t_multi_base - a copies of
+page's pd_xid_base/pd_multi_base. With tuple's 32-bit t_xmin and t_xmax from
+HeapTupleHeader they are used to calculate actual 64-bit XMIN and XMAX:
+
+XMIN = t_xmin + t_xid_base. 					(3)
+XMAX = t_xmax + t_xid_base/t_multi_base.		(4)
+
+The downside of this is that we can not use tuple's XMIN and XMAX right away.
+We often need to re-read t_xmin and t_xmax - which could actually be pointers
+into a page in shared buffers and therefore they could be updated by any other
+backend.
+
+Update/delete with 64-bit XIDs and 32-bit t_xmin/t_xmax
+--------------------------------------------------------------
+
+When we try to delete/update a tuple, we check that XMAX for a page fits (2).
+I.e. that t_xmax will not be over MaxShortTransactionId relative to
+pd_xid_base/pd_multi_base of a its page.
+
+If the current XID doesn't fit a range
+(pd_xid_base, pd_xid_base + MaxShortTransactionId) (5):
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base/pd_multi_base on
+a page and update all t_xmin/t_xmax of the other tuples on the page to
+correspond new pd_xid_base/pd_multi_base.
+
+- If it was impossible, it will try to prune and freeze tuples on a page.
+
+- If this is unsuccessful it will throw an error. Normally this is very
+unlikely but if there is a very old living transaction with an age of around
+2^32 this can arise. Basically, this is a behavior similar to one during the
+vacuum to prevent wraparound when XID was 32-bit. Dba should take care and
+avoid very-long-living transactions with an age close to 2^32. So long-living
+transactions often they are most likely defunct.
+
+Insert with 64-bit XIDs and 32-bit t_xmin/t_xmax
+------------------------------------------------
+
+On insert we check if current XID fits a range (5). Otherwise:
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base for t_xmin will
+not be over MaxShortTransactionId.
+
+- If it is impossible, then it will try to prune and freeze tuples on a page.
+
+Known issue: if pd_xid_base could not be shifted to accommodate a tuple being
+inserted due to a very long-running transaction, we just throw an error. We
+neither try to insert a�tuple into another page nor mark the current page as
+full. So, in this (unlikely) case we will get regular insert errors on the next
+tries to insert to the page 'locked' by this very long-running transaction.
+
+Upgrade from 32-bit XID versions
+--------------------------------
+
+pg_upgrade doesn't change pages format itself. It is done lazily after.
+
+1. At first heap page read, tuples on a page are repacked to free 16 bytes
+at the end of a page, possibly freeing space from dead tuples.
+
+2A. 16 bytes of pd_special is added if there is a place for it
+
+2B. Page is converted to "Double XMAX" format if there is no place for
+pd_special
+
+3. If a page is in double XMAX format after its first read, and vacuum (or
+micro-vacuum at select query) could prune some tuples and free space for
+pd_special, prune_page will add pd_special and convert page from double XMAX
+to general 64-bit XID page format.
-- 
2.24.3 (Apple Git-128)


--cpok4wp6gsarlzvp--





^ permalink  raw  reply  [nested|flat] 265+ messages in thread

* [PATCH v6] README for 64bit xid
@ 2022-01-10 19:20  Pavel Borisov <[email protected]>
  0 siblings, 0 replies; 265+ messages in thread

From: Pavel Borisov @ 2022-01-10 19:20 UTC (permalink / raw)

Authors:
- Pavel Borisov <[email protected]>
- Maxim Orlov <[email protected]>
- Yura Sokolov <[email protected]> <[email protected]>
---
 src/backend/access/heap/README.XID64 | 128 +++++++++++++++++++++++++++
 1 file changed, 128 insertions(+)
 create mode 100644 src/backend/access/heap/README.XID64

diff --git a/src/backend/access/heap/README.XID64 b/src/backend/access/heap/README.XID64
new file mode 100644
index 00000000000..457ba9b9ef5
--- /dev/null
+++ b/src/backend/access/heap/README.XID64
@@ -0,0 +1,128 @@
+src/backend/access/heap/README.XID64
+
+64-bit Transaction ID's (XID)
+=============================
+
+A limited number (N = 2^32) of XID's required to do vacuum freeze to prevent
+wraparound every N/2 transactions. This causes performance degradation due
+to the need to exclusively lock tables while being vacuumed. In each
+wraparound cycle, SLRU buffers are also being cut.
+
+With 64-bit XID's wraparound is effectively postponed to a very distant
+future. Even in highly loaded systems that had 2^32 transactions per day
+it will take huge 2^31 days before the first enforced "vacuum to prevent
+wraparound"). Buffers cutting and routine vacuum are not enforced, and DBA
+can plan them independently at the time with the least system load and least
+critical for database performance. Also, it can be done less frequently
+(several times a year vs every several days) on systems with transaction rates
+similar to those mentioned above.
+
+On-disk tuple and page format
+-----------------------------
+
+On-disk tuple format remains unchanged. 32-bit t_xmin and t_xmax store the
+lower parts of 64-bit XMIN and XMAX values. Each heap page has additional
+64-bit pd_xid_base and pd_multi_base which are common for all tuples on a page.
+They are placed into a pd_special area - 16 bytes in the end of a heap page.
+Actual XMIN/XMAX for a tuple are calculated upon reading a tuple from a page
+as follows:
+
+XMIN = t_xmin + pd_xid_base. 					(1)
+XMAX = t_xmax + pd_xid_base/pd_multi_base.		(2)
+
+"Double XMAX" page format
+---------------------------------
+
+At first read of a heap page after pg_upgrade from 32-bit XID PostgreSQL
+version pd_special area with a size of 16 bytes should be added to a page.
+Though a page may not have space for this. Then it can be converted to a
+temporary format called "double XMAX".
+
+All tuples after pg-upgrade would necessarily have xmin = FrozenTransactionId.
+So we don't need tuple header t_xmin field and we reuse t_xmin to store higher
+32 bits of its XMAX.
+
+Double XMAX format is only for full pages that don't have 16 bytes for
+pd_special. So it neither has a�place for a single tuple. Insert and HOT update
+for double XMAX pages is impossible and not supported. We can only read or
+delete tuples from it.
+
+When we are able to prune page double XMAX it will be converted from it to
+general 64-bit XID page format with all operations on its tuples supported.
+
+In-memory tuple format
+----------------------
+
+In-memory tuple representation consists of two parts:
+- HeapTupleHeader from disk page (contains all heap tuple contents, not only
+header)
+- HeapTuple with additional in-memory fields
+
+HeapTuple for each tuple in memory stores t_xid_base/t_multi_base - a copies of
+page's pd_xid_base/pd_multi_base. With tuple's 32-bit t_xmin and t_xmax from
+HeapTupleHeader they are used to calculate actual 64-bit XMIN and XMAX:
+
+XMIN = t_xmin + t_xid_base. 					(3)
+XMAX = t_xmax + t_xid_base/t_multi_base.		(4)
+
+The downside of this is that we can not use tuple's XMIN and XMAX right away.
+We often need to re-read t_xmin and t_xmax - which could actually be pointers
+into a page in shared buffers and therefore they could be updated by any other
+backend.
+
+Update/delete with 64-bit XIDs and 32-bit t_xmin/t_xmax
+--------------------------------------------------------------
+
+When we try to delete/update a tuple, we check that XMAX for a page fits (2).
+I.e. that t_xmax will not be over MaxShortTransactionId relative to
+pd_xid_base/pd_multi_base of a its page.
+
+If the current XID doesn't fit a range
+(pd_xid_base, pd_xid_base + MaxShortTransactionId) (5):
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base/pd_multi_base on
+a page and update all t_xmin/t_xmax of the other tuples on the page to
+correspond new pd_xid_base/pd_multi_base.
+
+- If it was impossible, it will try to prune and freeze tuples on a page.
+
+- If this is unsuccessful it will throw an error. Normally this is very
+unlikely but if there is a very old living transaction with an age of around
+2^32 this can arise. Basically, this is a behavior similar to one during the
+vacuum to prevent wraparound when XID was 32-bit. Dba should take care and
+avoid very-long-living transactions with an age close to 2^32. So long-living
+transactions often they are most likely defunct.
+
+Insert with 64-bit XIDs and 32-bit t_xmin/t_xmax
+------------------------------------------------
+
+On insert we check if current XID fits a range (5). Otherwise:
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base for t_xmin will
+not be over MaxShortTransactionId.
+
+- If it is impossible, then it will try to prune and freeze tuples on a page.
+
+Known issue: if pd_xid_base could not be shifted to accommodate a tuple being
+inserted due to a very long-running transaction, we just throw an error. We
+neither try to insert a�tuple into another page nor mark the current page as
+full. So, in this (unlikely) case we will get regular insert errors on the next
+tries to insert to the page 'locked' by this very long-running transaction.
+
+Upgrade from 32-bit XID versions
+--------------------------------
+
+pg_upgrade doesn't change pages format itself. It is done lazily after.
+
+1. At first heap page read, tuples on a page are repacked to free 16 bytes
+at the end of a page, possibly freeing space from dead tuples.
+
+2A. 16 bytes of pd_special is added if there is a place for it
+
+2B. Page is converted to "Double XMAX" format if there is no place for
+pd_special
+
+3. If a page is in double XMAX format after its first read, and vacuum (or
+micro-vacuum at select query) could prune some tuples and free space for
+pd_special, prune_page will add pd_special and convert page from double XMAX
+to general 64-bit XID page format.
-- 
2.24.3 (Apple Git-128)


--cpok4wp6gsarlzvp--





^ permalink  raw  reply  [nested|flat] 265+ messages in thread

* [PATCH v6] README for 64bit xid
@ 2022-01-10 19:20  Pavel Borisov <[email protected]>
  0 siblings, 0 replies; 265+ messages in thread

From: Pavel Borisov @ 2022-01-10 19:20 UTC (permalink / raw)

Authors:
- Pavel Borisov <[email protected]>
- Maxim Orlov <[email protected]>
- Yura Sokolov <[email protected]> <[email protected]>
---
 src/backend/access/heap/README.XID64 | 128 +++++++++++++++++++++++++++
 1 file changed, 128 insertions(+)
 create mode 100644 src/backend/access/heap/README.XID64

diff --git a/src/backend/access/heap/README.XID64 b/src/backend/access/heap/README.XID64
new file mode 100644
index 00000000000..457ba9b9ef5
--- /dev/null
+++ b/src/backend/access/heap/README.XID64
@@ -0,0 +1,128 @@
+src/backend/access/heap/README.XID64
+
+64-bit Transaction ID's (XID)
+=============================
+
+A limited number (N = 2^32) of XID's required to do vacuum freeze to prevent
+wraparound every N/2 transactions. This causes performance degradation due
+to the need to exclusively lock tables while being vacuumed. In each
+wraparound cycle, SLRU buffers are also being cut.
+
+With 64-bit XID's wraparound is effectively postponed to a very distant
+future. Even in highly loaded systems that had 2^32 transactions per day
+it will take huge 2^31 days before the first enforced "vacuum to prevent
+wraparound"). Buffers cutting and routine vacuum are not enforced, and DBA
+can plan them independently at the time with the least system load and least
+critical for database performance. Also, it can be done less frequently
+(several times a year vs every several days) on systems with transaction rates
+similar to those mentioned above.
+
+On-disk tuple and page format
+-----------------------------
+
+On-disk tuple format remains unchanged. 32-bit t_xmin and t_xmax store the
+lower parts of 64-bit XMIN and XMAX values. Each heap page has additional
+64-bit pd_xid_base and pd_multi_base which are common for all tuples on a page.
+They are placed into a pd_special area - 16 bytes in the end of a heap page.
+Actual XMIN/XMAX for a tuple are calculated upon reading a tuple from a page
+as follows:
+
+XMIN = t_xmin + pd_xid_base. 					(1)
+XMAX = t_xmax + pd_xid_base/pd_multi_base.		(2)
+
+"Double XMAX" page format
+---------------------------------
+
+At first read of a heap page after pg_upgrade from 32-bit XID PostgreSQL
+version pd_special area with a size of 16 bytes should be added to a page.
+Though a page may not have space for this. Then it can be converted to a
+temporary format called "double XMAX".
+
+All tuples after pg-upgrade would necessarily have xmin = FrozenTransactionId.
+So we don't need tuple header t_xmin field and we reuse t_xmin to store higher
+32 bits of its XMAX.
+
+Double XMAX format is only for full pages that don't have 16 bytes for
+pd_special. So it neither has a�place for a single tuple. Insert and HOT update
+for double XMAX pages is impossible and not supported. We can only read or
+delete tuples from it.
+
+When we are able to prune page double XMAX it will be converted from it to
+general 64-bit XID page format with all operations on its tuples supported.
+
+In-memory tuple format
+----------------------
+
+In-memory tuple representation consists of two parts:
+- HeapTupleHeader from disk page (contains all heap tuple contents, not only
+header)
+- HeapTuple with additional in-memory fields
+
+HeapTuple for each tuple in memory stores t_xid_base/t_multi_base - a copies of
+page's pd_xid_base/pd_multi_base. With tuple's 32-bit t_xmin and t_xmax from
+HeapTupleHeader they are used to calculate actual 64-bit XMIN and XMAX:
+
+XMIN = t_xmin + t_xid_base. 					(3)
+XMAX = t_xmax + t_xid_base/t_multi_base.		(4)
+
+The downside of this is that we can not use tuple's XMIN and XMAX right away.
+We often need to re-read t_xmin and t_xmax - which could actually be pointers
+into a page in shared buffers and therefore they could be updated by any other
+backend.
+
+Update/delete with 64-bit XIDs and 32-bit t_xmin/t_xmax
+--------------------------------------------------------------
+
+When we try to delete/update a tuple, we check that XMAX for a page fits (2).
+I.e. that t_xmax will not be over MaxShortTransactionId relative to
+pd_xid_base/pd_multi_base of a its page.
+
+If the current XID doesn't fit a range
+(pd_xid_base, pd_xid_base + MaxShortTransactionId) (5):
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base/pd_multi_base on
+a page and update all t_xmin/t_xmax of the other tuples on the page to
+correspond new pd_xid_base/pd_multi_base.
+
+- If it was impossible, it will try to prune and freeze tuples on a page.
+
+- If this is unsuccessful it will throw an error. Normally this is very
+unlikely but if there is a very old living transaction with an age of around
+2^32 this can arise. Basically, this is a behavior similar to one during the
+vacuum to prevent wraparound when XID was 32-bit. Dba should take care and
+avoid very-long-living transactions with an age close to 2^32. So long-living
+transactions often they are most likely defunct.
+
+Insert with 64-bit XIDs and 32-bit t_xmin/t_xmax
+------------------------------------------------
+
+On insert we check if current XID fits a range (5). Otherwise:
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base for t_xmin will
+not be over MaxShortTransactionId.
+
+- If it is impossible, then it will try to prune and freeze tuples on a page.
+
+Known issue: if pd_xid_base could not be shifted to accommodate a tuple being
+inserted due to a very long-running transaction, we just throw an error. We
+neither try to insert a�tuple into another page nor mark the current page as
+full. So, in this (unlikely) case we will get regular insert errors on the next
+tries to insert to the page 'locked' by this very long-running transaction.
+
+Upgrade from 32-bit XID versions
+--------------------------------
+
+pg_upgrade doesn't change pages format itself. It is done lazily after.
+
+1. At first heap page read, tuples on a page are repacked to free 16 bytes
+at the end of a page, possibly freeing space from dead tuples.
+
+2A. 16 bytes of pd_special is added if there is a place for it
+
+2B. Page is converted to "Double XMAX" format if there is no place for
+pd_special
+
+3. If a page is in double XMAX format after its first read, and vacuum (or
+micro-vacuum at select query) could prune some tuples and free space for
+pd_special, prune_page will add pd_special and convert page from double XMAX
+to general 64-bit XID page format.
-- 
2.24.3 (Apple Git-128)


--cpok4wp6gsarlzvp--





^ permalink  raw  reply  [nested|flat] 265+ messages in thread

* [PATCH v6] README for 64bit xid
@ 2022-01-10 19:20  Pavel Borisov <[email protected]>
  0 siblings, 0 replies; 265+ messages in thread

From: Pavel Borisov @ 2022-01-10 19:20 UTC (permalink / raw)

Authors:
- Pavel Borisov <[email protected]>
- Maxim Orlov <[email protected]>
- Yura Sokolov <[email protected]> <[email protected]>
---
 src/backend/access/heap/README.XID64 | 128 +++++++++++++++++++++++++++
 1 file changed, 128 insertions(+)
 create mode 100644 src/backend/access/heap/README.XID64

diff --git a/src/backend/access/heap/README.XID64 b/src/backend/access/heap/README.XID64
new file mode 100644
index 00000000000..457ba9b9ef5
--- /dev/null
+++ b/src/backend/access/heap/README.XID64
@@ -0,0 +1,128 @@
+src/backend/access/heap/README.XID64
+
+64-bit Transaction ID's (XID)
+=============================
+
+A limited number (N = 2^32) of XID's required to do vacuum freeze to prevent
+wraparound every N/2 transactions. This causes performance degradation due
+to the need to exclusively lock tables while being vacuumed. In each
+wraparound cycle, SLRU buffers are also being cut.
+
+With 64-bit XID's wraparound is effectively postponed to a very distant
+future. Even in highly loaded systems that had 2^32 transactions per day
+it will take huge 2^31 days before the first enforced "vacuum to prevent
+wraparound"). Buffers cutting and routine vacuum are not enforced, and DBA
+can plan them independently at the time with the least system load and least
+critical for database performance. Also, it can be done less frequently
+(several times a year vs every several days) on systems with transaction rates
+similar to those mentioned above.
+
+On-disk tuple and page format
+-----------------------------
+
+On-disk tuple format remains unchanged. 32-bit t_xmin and t_xmax store the
+lower parts of 64-bit XMIN and XMAX values. Each heap page has additional
+64-bit pd_xid_base and pd_multi_base which are common for all tuples on a page.
+They are placed into a pd_special area - 16 bytes in the end of a heap page.
+Actual XMIN/XMAX for a tuple are calculated upon reading a tuple from a page
+as follows:
+
+XMIN = t_xmin + pd_xid_base. 					(1)
+XMAX = t_xmax + pd_xid_base/pd_multi_base.		(2)
+
+"Double XMAX" page format
+---------------------------------
+
+At first read of a heap page after pg_upgrade from 32-bit XID PostgreSQL
+version pd_special area with a size of 16 bytes should be added to a page.
+Though a page may not have space for this. Then it can be converted to a
+temporary format called "double XMAX".
+
+All tuples after pg-upgrade would necessarily have xmin = FrozenTransactionId.
+So we don't need tuple header t_xmin field and we reuse t_xmin to store higher
+32 bits of its XMAX.
+
+Double XMAX format is only for full pages that don't have 16 bytes for
+pd_special. So it neither has a�place for a single tuple. Insert and HOT update
+for double XMAX pages is impossible and not supported. We can only read or
+delete tuples from it.
+
+When we are able to prune page double XMAX it will be converted from it to
+general 64-bit XID page format with all operations on its tuples supported.
+
+In-memory tuple format
+----------------------
+
+In-memory tuple representation consists of two parts:
+- HeapTupleHeader from disk page (contains all heap tuple contents, not only
+header)
+- HeapTuple with additional in-memory fields
+
+HeapTuple for each tuple in memory stores t_xid_base/t_multi_base - a copies of
+page's pd_xid_base/pd_multi_base. With tuple's 32-bit t_xmin and t_xmax from
+HeapTupleHeader they are used to calculate actual 64-bit XMIN and XMAX:
+
+XMIN = t_xmin + t_xid_base. 					(3)
+XMAX = t_xmax + t_xid_base/t_multi_base.		(4)
+
+The downside of this is that we can not use tuple's XMIN and XMAX right away.
+We often need to re-read t_xmin and t_xmax - which could actually be pointers
+into a page in shared buffers and therefore they could be updated by any other
+backend.
+
+Update/delete with 64-bit XIDs and 32-bit t_xmin/t_xmax
+--------------------------------------------------------------
+
+When we try to delete/update a tuple, we check that XMAX for a page fits (2).
+I.e. that t_xmax will not be over MaxShortTransactionId relative to
+pd_xid_base/pd_multi_base of a its page.
+
+If the current XID doesn't fit a range
+(pd_xid_base, pd_xid_base + MaxShortTransactionId) (5):
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base/pd_multi_base on
+a page and update all t_xmin/t_xmax of the other tuples on the page to
+correspond new pd_xid_base/pd_multi_base.
+
+- If it was impossible, it will try to prune and freeze tuples on a page.
+
+- If this is unsuccessful it will throw an error. Normally this is very
+unlikely but if there is a very old living transaction with an age of around
+2^32 this can arise. Basically, this is a behavior similar to one during the
+vacuum to prevent wraparound when XID was 32-bit. Dba should take care and
+avoid very-long-living transactions with an age close to 2^32. So long-living
+transactions often they are most likely defunct.
+
+Insert with 64-bit XIDs and 32-bit t_xmin/t_xmax
+------------------------------------------------
+
+On insert we check if current XID fits a range (5). Otherwise:
+
+- heap_page_prepare_for_xid() will try to increase pd_xid_base for t_xmin will
+not be over MaxShortTransactionId.
+
+- If it is impossible, then it will try to prune and freeze tuples on a page.
+
+Known issue: if pd_xid_base could not be shifted to accommodate a tuple being
+inserted due to a very long-running transaction, we just throw an error. We
+neither try to insert a�tuple into another page nor mark the current page as
+full. So, in this (unlikely) case we will get regular insert errors on the next
+tries to insert to the page 'locked' by this very long-running transaction.
+
+Upgrade from 32-bit XID versions
+--------------------------------
+
+pg_upgrade doesn't change pages format itself. It is done lazily after.
+
+1. At first heap page read, tuples on a page are repacked to free 16 bytes
+at the end of a page, possibly freeing space from dead tuples.
+
+2A. 16 bytes of pd_special is added if there is a place for it
+
+2B. Page is converted to "Double XMAX" format if there is no place for
+pd_special
+
+3. If a page is in double XMAX format after its first read, and vacuum (or
+micro-vacuum at select query) could prune some tuples and free space for
+pd_special, prune_page will add pd_special and convert page from double XMAX
+to general 64-bit XID page format.
-- 
2.24.3 (Apple Git-128)


--cpok4wp6gsarlzvp--





^ permalink  raw  reply  [nested|flat] 265+ messages in thread