On Wed, Feb 4, 2026 at 6:39 PM Amul Sul <[email protected]> wrote: > > On Wed, Jan 28, 2026 at 2:41 AM Robert Haas <[email protected]> wrote: > > > > On Tue, Jan 27, 2026 at 7:07 AM Amul Sul <[email protected]> wrote: > > > In the attached version, I am using the WAL segment name as the hash > > > key, which is much more straightforward. I have rewritten > > > read_archive_wal_page(), and it looks much cleaner than before. The > > > logic to discard irrelevant WAL files is still within > > > get_archive_wal_entry. I added an explanation for setting cur_wal to > > > NULL, which is now handled in the separate function I mentioned > > > previously. > > > > > > Kindly have a look at the attached version; let me know if you are > > > still not happy with the current approach for filtering/discarding > > > irrelevant WAL segments. It isn't much different from the previous > > > version, but I have tried to keep it in a separate routine for better > > > code readability, with comments to make it easier to understand. I > > > also added a comment for ArchivedWALFile. > > > > I feel like the division of labor between get_archive_wal_entry() and > > read_archive_wal_page() is odd. I noticed this in the last version, > > too, and it still seems to be the case. get_archive_wal_entry() first > > calls ArchivedWAL_lookup(). If that finds an entry, it just returns. > > If it doesn't, it loops until an entry for the requested file shows up > > and then returns it. Then control returns to read_archive_wal_page() > > which loops some more until we have all the data we need for the > > requested file. But it seems odd to me to have two separate loops > > here. I think that the first loop is going to call read_archive_file() > > until we find the beginning of the file that we care about and then > > the second one is going to call read_archive_file() some more until we > > have read enough of it to satisfy the request. It feels odd to me to > > do it that way, as if we told somebody to first wait until 9 o'clock > > and then wait another 30 minutes, instead of just telling them to wait > > until 9:30. I realize it's not quite the same thing, because apart > > from calling read_archive_file(), the two loops do different things, > > but I still think it looks odd. > > > > + /* > > + * Ignore if the timeline is different or the current segment is not > > + * the desired one. > > + */ > > + XLogFromFileName(entry->fname, &curSegTimeline, &curSegNo, WalSegSz); > > + if (privateInfo->timeline != curSegTimeline || > > + privateInfo->startSegNo > curSegNo || > > + privateInfo->endSegNo < curSegNo || > > + segno > curSegNo) > > + { > > + free_archive_wal_entry(entry->fname, privateInfo); > > + continue; > > + } > > > > The comment doesn't match the code. If it did, the test would be > > (privateInfo->timeline != curSegTimeline || segno != curSegno). But > > instead the segno test is > rather than !=, and the checks against > > startSegNo and endSegNo aren't explained at all. I think I understand > > why the segno test uses > rather than !=, but it's the point of the > > comment to explain things like that, rather than leaving the reader to > > guess. And I don't know why we also need to test startSegNo and > > endSegNo. > > > > I also wonder what the point is of doing XLogFromFileName() on the > > fname provided by the caller and then again on entry->fname. Couldn't > > you just compare the strings? > > > > Again, the division of labor is really odd here. It's the job of > > astreamer_waldump_content() to skip things that aren't WAL files at > > all, but it's the job of get_archive_wal_entry() to skip things that > > are WAL files but not the one we want. I disagree with putting those > > checks in completely separate parts of the code. > > > > Keeping the timeline and segment start-end range checks inside the > archive streamer creates a circular dependency that cannot be resolved > without a 'dirty hack'. We must read the first available WAL file page > to determine the wal_segment_size before it can calculate the target > segment range. Moving the checks inside the streamer would make it > impossible to process that initial file, as the necessary filtering > parameters -- would still be unknown which would need to be skipped > for the first read somehow. What if later we realized that the first > WAL file which was allowed to be streamed by skipping that check is > irrelevant and doesn't fall under the start-end segment range? >
Please have a look at the attached version, specifically patch 0005. In astreamer_waldump_content(), I have moved the WAL file filtration check from get_archive_wal_entry(). This check will be skipped during the initial read in init_archive_reader(), which instead performs it explicitly once it determines the WAL segment size and the start/end segments. To access the WAL segment size inside astreamer_waldump_content(), I have moved the WAL segment size variable into the XLogDumpPrivate structure in the separate 0004 patch. Regards, Amul
From 0731db48bb8d154aa72d2c956dec95a8127ae07d Mon Sep 17 00:00:00 2001 From: Amul Sul <[email protected]> Date: Thu, 22 Jan 2026 10:28:32 +0530 Subject: [PATCH v12 1/9] Refactor: pg_waldump: Move some declarations to new pg_waldump.h This change prepares for a second source file in this directory to support reading WAL from tar files. Common structures, declarations, and functions are being exported through this include file so they can be used in both files. --- src/bin/pg_waldump/pg_waldump.c | 9 +-------- src/bin/pg_waldump/pg_waldump.h | 25 +++++++++++++++++++++++++ 2 files changed, 26 insertions(+), 8 deletions(-) create mode 100644 src/bin/pg_waldump/pg_waldump.h diff --git a/src/bin/pg_waldump/pg_waldump.c b/src/bin/pg_waldump/pg_waldump.c index f3446385d6a..4b7411a6498 100644 --- a/src/bin/pg_waldump/pg_waldump.c +++ b/src/bin/pg_waldump/pg_waldump.c @@ -29,6 +29,7 @@ #include "common/logging.h" #include "common/relpath.h" #include "getopt_long.h" +#include "pg_waldump.h" #include "rmgrdesc.h" #include "storage/bufpage.h" @@ -43,14 +44,6 @@ static volatile sig_atomic_t time_to_stop = false; static const RelFileLocator emptyRelFileLocator = {0, 0, 0}; -typedef struct XLogDumpPrivate -{ - TimeLineID timeline; - XLogRecPtr startptr; - XLogRecPtr endptr; - bool endptr_reached; -} XLogDumpPrivate; - typedef struct XLogDumpConfig { /* display options */ diff --git a/src/bin/pg_waldump/pg_waldump.h b/src/bin/pg_waldump/pg_waldump.h new file mode 100644 index 00000000000..b88543856e5 --- /dev/null +++ b/src/bin/pg_waldump/pg_waldump.h @@ -0,0 +1,25 @@ +/*------------------------------------------------------------------------- + * + * pg_waldump.h - decode and display WAL + * + * Copyright (c) 2026, PostgreSQL Global Development Group + * + * IDENTIFICATION + * src/bin/pg_waldump/pg_waldump.h + *------------------------------------------------------------------------- + */ +#ifndef PG_WALDUMP_H +#define PG_WALDUMP_H + +#include "access/xlogdefs.h" + +/* Contains the necessary information to drive WAL decoding */ +typedef struct XLogDumpPrivate +{ + TimeLineID timeline; + XLogRecPtr startptr; + XLogRecPtr endptr; + bool endptr_reached; +} XLogDumpPrivate; + +#endif /* end of PG_WALDUMP_H */ -- 2.47.1
From 09f887f7f6b4c142b527dc6410bc781884646681 Mon Sep 17 00:00:00 2001 From: Amul Sul <[email protected]> Date: Thu, 22 Jan 2026 10:38:16 +0530 Subject: [PATCH v12 2/9] Refactor: pg_waldump: Separate logic used to calculate the required read size. This refactoring prepares the codebase for an upcoming patch that will support reading WAL from tar files. The logic for calculating the required read size has been updated to handle both normal WAL files and WAL files located inside a tar archive. --- src/bin/pg_waldump/pg_waldump.c | 43 +++++++++++++++++++++++---------- 1 file changed, 30 insertions(+), 13 deletions(-) diff --git a/src/bin/pg_waldump/pg_waldump.c b/src/bin/pg_waldump/pg_waldump.c index 4b7411a6498..958a71a01cf 100644 --- a/src/bin/pg_waldump/pg_waldump.c +++ b/src/bin/pg_waldump/pg_waldump.c @@ -326,6 +326,32 @@ identify_target_directory(char *directory, char *fname, int *WalSegSz) return NULL; /* not reached */ } +/* + * Returns the size in bytes of the data to be read. Returns -1 if the end + * point has already been reached. + */ +static inline int +required_read_len(XLogDumpPrivate *private, XLogRecPtr targetPagePtr, + int reqLen) +{ + int count = XLOG_BLCKSZ; + + if (XLogRecPtrIsValid(private->endptr)) + { + if (targetPagePtr + XLOG_BLCKSZ <= private->endptr) + count = XLOG_BLCKSZ; + else if (targetPagePtr + reqLen <= private->endptr) + count = private->endptr - targetPagePtr; + else + { + private->endptr_reached = true; + return -1; + } + } + + return count; +} + /* pg_waldump's XLogReaderRoutine->segment_open callback */ static void WALDumpOpenSegment(XLogReaderState *state, XLogSegNo nextSegNo, @@ -383,21 +409,12 @@ WALDumpReadPage(XLogReaderState *state, XLogRecPtr targetPagePtr, int reqLen, XLogRecPtr targetPtr, char *readBuff) { XLogDumpPrivate *private = state->private_data; - int count = XLOG_BLCKSZ; + int count = required_read_len(private, targetPagePtr, reqLen); WALReadError errinfo; - if (XLogRecPtrIsValid(private->endptr)) - { - if (targetPagePtr + XLOG_BLCKSZ <= private->endptr) - count = XLOG_BLCKSZ; - else if (targetPagePtr + reqLen <= private->endptr) - count = private->endptr - targetPagePtr; - else - { - private->endptr_reached = true; - return -1; - } - } + /* Bail out if the count to be read is not valid */ + if (count < 0) + return -1; if (!WALRead(state, readBuff, targetPagePtr, count, private->timeline, &errinfo)) -- 2.47.1
From 05d2c4218d9c3496878f342c9c32ff5148d3d6a4 Mon Sep 17 00:00:00 2001 From: Amul Sul <[email protected]> Date: Thu, 22 Jan 2026 11:06:05 +0530 Subject: [PATCH v12 3/9] Refactor: pg_waldump: Restructure TAP tests. Restructured tests that do not have a WAL file argument to run within a loop, facilitating their re-execution for decoding WAL from tar archives. == NOTE == This is not intended to be committed separately. It can be merged with the next patch, which is the main patch implementing this feature. --- src/bin/pg_waldump/t/001_basic.pl | 123 ++++++++++++++++-------------- 1 file changed, 67 insertions(+), 56 deletions(-) diff --git a/src/bin/pg_waldump/t/001_basic.pl b/src/bin/pg_waldump/t/001_basic.pl index 5db5d20136f..3288fadcf48 100644 --- a/src/bin/pg_waldump/t/001_basic.pl +++ b/src/bin/pg_waldump/t/001_basic.pl @@ -198,28 +198,6 @@ command_like( ], qr/./, 'runs with start and end segment specified'); -command_fails_like( - [ 'pg_waldump', '--path' => $node->data_dir ], - qr/error: no start WAL location given/, - 'path option requires start location'); -command_like( - [ - 'pg_waldump', - '--path' => $node->data_dir, - '--start' => $start_lsn, - '--end' => $end_lsn, - ], - qr/./, - 'runs with path option and start and end locations'); -command_fails_like( - [ - 'pg_waldump', - '--path' => $node->data_dir, - '--start' => $start_lsn, - ], - qr/error: error in WAL record at/, - 'falling off the end of the WAL results in an error'); - command_like( [ 'pg_waldump', '--quiet', @@ -227,15 +205,6 @@ command_like( ], qr/^$/, 'no output with --quiet option'); -command_fails_like( - [ - 'pg_waldump', '--quiet', - '--path' => $node->data_dir, - '--start' => $start_lsn - ], - qr/error: error in WAL record at/, - 'errors are shown with --quiet'); - # Test for: Display a message that we're skipping data if `from` # wasn't a pointer to the start of a record. @@ -272,7 +241,6 @@ sub test_pg_waldump my $result = IPC::Run::run [ 'pg_waldump', - '--path' => $node->data_dir, '--start' => $start_lsn, '--end' => $end_lsn, @opts @@ -288,38 +256,81 @@ sub test_pg_waldump my @lines; -@lines = test_pg_waldump; -is(grep(!/^rmgr: \w/, @lines), 0, 'all output lines are rmgr lines'); +my @scenarios = ( + { + 'path' => $node->data_dir + }); -@lines = test_pg_waldump('--limit' => 6); -is(@lines, 6, 'limit option observed'); +for my $scenario (@scenarios) +{ + my $path = $scenario->{'path'}; -@lines = test_pg_waldump('--fullpage'); -is(grep(!/^rmgr:.*\bFPW\b/, @lines), 0, 'all output lines are FPW'); + SKIP: + { + command_fails_like( + [ 'pg_waldump', '--path' => $path ], + qr/error: no start WAL location given/, + 'path option requires start location'); + command_like( + [ + 'pg_waldump', + '--path' => $path, + '--start' => $start_lsn, + '--end' => $end_lsn, + ], + qr/./, + 'runs with path option and start and end locations'); + command_fails_like( + [ + 'pg_waldump', + '--path' => $path, + '--start' => $start_lsn, + ], + qr/error: error in WAL record at/, + 'falling off the end of the WAL results in an error'); -@lines = test_pg_waldump('--stats'); -like($lines[0], qr/WAL statistics/, "statistics on stdout"); -is(grep(/^rmgr:/, @lines), 0, 'no rmgr lines output'); + command_fails_like( + [ + 'pg_waldump', '--quiet', + '--path' => $path, + '--start' => $start_lsn + ], + qr/error: error in WAL record at/, + 'errors are shown with --quiet'); -@lines = test_pg_waldump('--stats=record'); -like($lines[0], qr/WAL statistics/, "statistics on stdout"); -is(grep(/^rmgr:/, @lines), 0, 'no rmgr lines output'); + @lines = test_pg_waldump('--path' => $path); + is(grep(!/^rmgr: \w/, @lines), 0, 'all output lines are rmgr lines'); -@lines = test_pg_waldump('--rmgr' => 'Btree'); -is(grep(!/^rmgr: Btree/, @lines), 0, 'only Btree lines'); + @lines = test_pg_waldump('--path' => $path, '--limit' => 6); + is(@lines, 6, 'limit option observed'); -@lines = test_pg_waldump('--fork' => 'init'); -is(grep(!/fork init/, @lines), 0, 'only init fork lines'); + @lines = test_pg_waldump('--path' => $path, '--fullpage'); + is(grep(!/^rmgr:.*\bFPW\b/, @lines), 0, 'all output lines are FPW'); -@lines = test_pg_waldump( - '--relation' => "$default_ts_oid/$postgres_db_oid/$rel_t1_oid"); -is(grep(!/rel $default_ts_oid\/$postgres_db_oid\/$rel_t1_oid/, @lines), - 0, 'only lines for selected relation'); + @lines = test_pg_waldump('--path' => $path, '--stats'); + like($lines[0], qr/WAL statistics/, "statistics on stdout"); + is(grep(/^rmgr:/, @lines), 0, 'no rmgr lines output'); -@lines = test_pg_waldump( - '--relation' => "$default_ts_oid/$postgres_db_oid/$rel_i1a_oid", - '--block' => 1); -is(grep(!/\bblk 1\b/, @lines), 0, 'only lines for selected block'); + @lines = test_pg_waldump('--path' => $path, '--stats=record'); + like($lines[0], qr/WAL statistics/, "statistics on stdout"); + is(grep(/^rmgr:/, @lines), 0, 'no rmgr lines output'); + @lines = test_pg_waldump('--path' => $path, '--rmgr' => 'Btree'); + is(grep(!/^rmgr: Btree/, @lines), 0, 'only Btree lines'); + + @lines = test_pg_waldump('--path' => $path, '--fork' => 'init'); + is(grep(!/fork init/, @lines), 0, 'only init fork lines'); + + @lines = test_pg_waldump('--path' => $path, + '--relation' => "$default_ts_oid/$postgres_db_oid/$rel_t1_oid"); + is(grep(!/rel $default_ts_oid\/$postgres_db_oid\/$rel_t1_oid/, @lines), + 0, 'only lines for selected relation'); + + @lines = test_pg_waldump('--path' => $path, + '--relation' => "$default_ts_oid/$postgres_db_oid/$rel_i1a_oid", + '--block' => 1); + is(grep(!/\bblk 1\b/, @lines), 0, 'only lines for selected block'); + } +} done_testing(); -- 2.47.1
From c00a6728cde927f4bb092a95694d1fcd2c17205f Mon Sep 17 00:00:00 2001 From: Amul Sul <[email protected]> Date: Wed, 4 Feb 2026 15:31:51 +0530 Subject: [PATCH v12 4/9] Refactor: pg_waldump: Move WAL segment size to XLogDumpPrivate. Relocate the WAL segment size variable to the XLogDumpPrivate structure and rename it to segsize for consistency. This change is required to make the segment size accessible to the archive streamer code, where passing it as a function argument is not feasible. --- src/bin/pg_waldump/pg_waldump.c | 26 +++++++++++++------------- src/bin/pg_waldump/pg_waldump.h | 1 + 2 files changed, 14 insertions(+), 13 deletions(-) diff --git a/src/bin/pg_waldump/pg_waldump.c b/src/bin/pg_waldump/pg_waldump.c index 958a71a01cf..5d31b15dbd8 100644 --- a/src/bin/pg_waldump/pg_waldump.c +++ b/src/bin/pg_waldump/pg_waldump.c @@ -811,7 +811,6 @@ main(int argc, char **argv) XLogRecPtr first_record; char *waldir = NULL; char *errormsg; - int WalSegSz; static struct option long_options[] = { {"bkp-details", no_argument, NULL, 'b'}, @@ -865,6 +864,7 @@ main(int argc, char **argv) memset(&stats, 0, sizeof(XLogStats)); private.timeline = 1; + private.segsize = 0; private.startptr = InvalidXLogRecPtr; private.endptr = InvalidXLogRecPtr; private.endptr_reached = false; @@ -1138,18 +1138,18 @@ main(int argc, char **argv) pg_fatal("could not open directory \"%s\": %m", waldir); } - waldir = identify_target_directory(waldir, fname, &WalSegSz); + waldir = identify_target_directory(waldir, fname, &private.segsize); fd = open_file_in_directory(waldir, fname); if (fd < 0) pg_fatal("could not open file \"%s\"", fname); close(fd); /* parse position from file */ - XLogFromFileName(fname, &private.timeline, &segno, WalSegSz); + XLogFromFileName(fname, &private.timeline, &segno, private.segsize); if (!XLogRecPtrIsValid(private.startptr)) - XLogSegNoOffsetToRecPtr(segno, 0, WalSegSz, private.startptr); - else if (!XLByteInSeg(private.startptr, segno, WalSegSz)) + XLogSegNoOffsetToRecPtr(segno, 0, private.segsize, private.startptr); + else if (!XLByteInSeg(private.startptr, segno, private.segsize)) { pg_log_error("start WAL location %X/%08X is not inside file \"%s\"", LSN_FORMAT_ARGS(private.startptr), @@ -1159,7 +1159,7 @@ main(int argc, char **argv) /* no second file specified, set end position */ if (!(optind + 1 < argc) && !XLogRecPtrIsValid(private.endptr)) - XLogSegNoOffsetToRecPtr(segno + 1, 0, WalSegSz, private.endptr); + XLogSegNoOffsetToRecPtr(segno + 1, 0, private.segsize, private.endptr); /* parse ENDSEG if passed */ if (optind + 1 < argc) @@ -1175,14 +1175,14 @@ main(int argc, char **argv) close(fd); /* parse position from file */ - XLogFromFileName(fname, &private.timeline, &endsegno, WalSegSz); + XLogFromFileName(fname, &private.timeline, &endsegno, private.segsize); if (endsegno < segno) pg_fatal("ENDSEG %s is before STARTSEG %s", argv[optind + 1], argv[optind]); if (!XLogRecPtrIsValid(private.endptr)) - XLogSegNoOffsetToRecPtr(endsegno + 1, 0, WalSegSz, + XLogSegNoOffsetToRecPtr(endsegno + 1, 0, private.segsize, private.endptr); /* set segno to endsegno for check of --end */ @@ -1190,8 +1190,8 @@ main(int argc, char **argv) } - if (!XLByteInSeg(private.endptr, segno, WalSegSz) && - private.endptr != (segno + 1) * WalSegSz) + if (!XLByteInSeg(private.endptr, segno, private.segsize) && + private.endptr != (segno + 1) * private.segsize) { pg_log_error("end WAL location %X/%08X is not inside file \"%s\"", LSN_FORMAT_ARGS(private.endptr), @@ -1200,7 +1200,7 @@ main(int argc, char **argv) } } else - waldir = identify_target_directory(waldir, NULL, &WalSegSz); + waldir = identify_target_directory(waldir, NULL, &private.segsize); /* we don't know what to print */ if (!XLogRecPtrIsValid(private.startptr)) @@ -1213,7 +1213,7 @@ main(int argc, char **argv) /* we have everything we need, start reading */ xlogreader_state = - XLogReaderAllocate(WalSegSz, waldir, + XLogReaderAllocate(private.segsize, waldir, XL_ROUTINE(.page_read = WALDumpReadPage, .segment_open = WALDumpOpenSegment, .segment_close = WALDumpCloseSegment), @@ -1234,7 +1234,7 @@ main(int argc, char **argv) * a segment (e.g. we were used in file mode). */ if (first_record != private.startptr && - XLogSegmentOffset(private.startptr, WalSegSz) != 0) + XLogSegmentOffset(private.startptr, private.segsize) != 0) pg_log_info(ngettext("first record is after %X/%08X, at %X/%08X, skipping over %u byte", "first record is after %X/%08X, at %X/%08X, skipping over %u bytes", (first_record - private.startptr)), diff --git a/src/bin/pg_waldump/pg_waldump.h b/src/bin/pg_waldump/pg_waldump.h index b88543856e5..4f1b2ab668b 100644 --- a/src/bin/pg_waldump/pg_waldump.h +++ b/src/bin/pg_waldump/pg_waldump.h @@ -17,6 +17,7 @@ typedef struct XLogDumpPrivate { TimeLineID timeline; + int segsize; XLogRecPtr startptr; XLogRecPtr endptr; bool endptr_reached; -- 2.47.1
From d1bb58eb796b8d1d25eb836c21dd874cb46f6359 Mon Sep 17 00:00:00 2001 From: Amul Sul <[email protected]> Date: Tue, 10 Feb 2026 11:42:36 +0530 Subject: [PATCH v12 5/9] pg_waldump: Add support for archived WAL decoding. pg_waldump can now accept the path to a tar archive containing WAL files and decode them. This feature was added primarily for pg_verifybackup, which previously disabled WAL parsing for tar-formatted backups. Note that this patch requires that the WAL files within the archive be in sequential order; an error will be reported otherwise. The next patch is planned to remove this restriction. --- doc/src/sgml/ref/pg_waldump.sgml | 8 +- src/bin/pg_waldump/Makefile | 7 +- src/bin/pg_waldump/archive_waldump.c | 669 +++++++++++++++++++++++++++ src/bin/pg_waldump/meson.build | 4 +- src/bin/pg_waldump/pg_waldump.c | 242 +++++++--- src/bin/pg_waldump/pg_waldump.h | 45 ++ src/bin/pg_waldump/t/001_basic.pl | 83 +++- src/tools/pgindent/typedefs.list | 3 + 8 files changed, 986 insertions(+), 75 deletions(-) create mode 100644 src/bin/pg_waldump/archive_waldump.c diff --git a/doc/src/sgml/ref/pg_waldump.sgml b/doc/src/sgml/ref/pg_waldump.sgml index ce23add5577..d004bb0f67e 100644 --- a/doc/src/sgml/ref/pg_waldump.sgml +++ b/doc/src/sgml/ref/pg_waldump.sgml @@ -141,13 +141,17 @@ PostgreSQL documentation <term><option>--path=<replaceable>path</replaceable></option></term> <listitem> <para> - Specifies a directory to search for WAL segment files or a - directory with a <literal>pg_wal</literal> subdirectory that + Specifies a tar archive or a directory to search for WAL segment files + or a directory with a <literal>pg_wal</literal> subdirectory that contains such files. The default is to search in the current directory, the <literal>pg_wal</literal> subdirectory of the current directory, and the <literal>pg_wal</literal> subdirectory of <envar>PGDATA</envar>. </para> + <para> + If a tar archive is provided, its WAL segment files must be in + sequential order; otherwise, an error will be reported. + </para> </listitem> </varlistentry> diff --git a/src/bin/pg_waldump/Makefile b/src/bin/pg_waldump/Makefile index 4c1ee649501..aabb87566a2 100644 --- a/src/bin/pg_waldump/Makefile +++ b/src/bin/pg_waldump/Makefile @@ -3,6 +3,9 @@ PGFILEDESC = "pg_waldump - decode and display WAL" PGAPPICON=win32 +# make these available to TAP test scripts +export TAR + subdir = src/bin/pg_waldump top_builddir = ../../.. include $(top_builddir)/src/Makefile.global @@ -10,13 +13,15 @@ include $(top_builddir)/src/Makefile.global OBJS = \ $(RMGRDESCOBJS) \ $(WIN32RES) \ + archive_waldump.o \ compat.o \ pg_waldump.o \ rmgrdesc.o \ xlogreader.o \ xlogstats.o -override CPPFLAGS := -DFRONTEND $(CPPFLAGS) +override CPPFLAGS := -DFRONTEND -I$(libpq_srcdir) $(CPPFLAGS) +LDFLAGS_INTERNAL += -L$(top_builddir)/src/fe_utils -lpgfeutils RMGRDESCSOURCES = $(sort $(notdir $(wildcard $(top_srcdir)/src/backend/access/rmgrdesc/*desc*.c))) RMGRDESCOBJS = $(patsubst %.c,%.o,$(RMGRDESCSOURCES)) diff --git a/src/bin/pg_waldump/archive_waldump.c b/src/bin/pg_waldump/archive_waldump.c new file mode 100644 index 00000000000..27a5a5c6d5d --- /dev/null +++ b/src/bin/pg_waldump/archive_waldump.c @@ -0,0 +1,669 @@ +/*------------------------------------------------------------------------- + * + * archive_waldump.c + * A generic facility for reading WAL data from tar archives via archive + * streamer. + * + * Portions Copyright (c) 2026, PostgreSQL Global Development Group + * + * IDENTIFICATION + * src/bin/pg_waldump/archive_waldump.c + * + *------------------------------------------------------------------------- + */ + +#include "postgres_fe.h" + +#include <unistd.h> + +#include "access/xlog_internal.h" +#include "common/hashfn.h" +#include "common/logging.h" +#include "fe_utils/simple_list.h" +#include "pg_waldump.h" + +/* + * How many bytes should we try to read from a file at once? + */ +#define READ_CHUNK_SIZE (128 * 1024) + +/* + * Check if the start segment number is zero; this indicates a request to read + * any WAL file. + */ +#define READ_ANY_WAL(privateInfo) ((privateInfo)->start_segno == 0) + +/* + * Hash entry representing a WAL segment retrieved from the archive. + * + * While WAL segments are typically read sequentially, individual entries + * maintain their own buffers for the following reasons: + * + * 1. Boundary Handling: The archive streamer provides a continuous byte + * stream. A single streaming chunk may contain the end of one WAL segment + * and the start of the next. Separate buffers allow us to easily + * partition and track these bytes by their respective segments. + * + * 2. Out-of-Order Support: Dedicated buffers simplify logic if segments + * are ever archived or retrieved out of sequence. + * + * To minimize the memory footprint, entries and their associated buffers are + * freed immediately once consumed. Since pg_waldump does not request the same + * bytes twice, a segment is discarded as soon as it moves past it. + */ +typedef struct ArchivedWALFile +{ + uint32 status; /* hash status */ + const char *fname; /* hash key: WAL segment name */ + + StringInfo buf; /* holds WAL bytes read from archive */ + + int read_len; /* total bytes of a WAL read from archive */ +} ArchivedWALFile; + +static uint32 hash_string_pointer(const char *s); +#define SH_PREFIX ArchivedWAL +#define SH_ELEMENT_TYPE ArchivedWALFile +#define SH_KEY_TYPE const char * +#define SH_KEY fname +#define SH_HASH_KEY(tb, key) hash_string_pointer(key) +#define SH_EQUAL(tb, a, b) (strcmp(a, b) == 0) +#define SH_SCOPE static inline +#define SH_RAW_ALLOCATOR pg_malloc0 +#define SH_DECLARE +#define SH_DEFINE +#include "lib/simplehash.h" + +typedef struct astreamer_waldump +{ + astreamer base; + XLogDumpPrivate *privateInfo; +} astreamer_waldump; + +static ArchivedWALFile *get_archive_wal_entry(const char *fname, + XLogDumpPrivate *privateInfo, + int WalSegSz); +static int read_archive_file(XLogDumpPrivate *privateInfo, Size count); + +static astreamer *astreamer_waldump_new(XLogDumpPrivate *privateInfo); +static void astreamer_waldump_content(astreamer *streamer, + astreamer_member *member, + const char *data, int len, + astreamer_archive_context context); +static void astreamer_waldump_finalize(astreamer *streamer); +static void astreamer_waldump_free(astreamer *streamer); + +static bool member_is_wal_file(astreamer_waldump *mystreamer, + astreamer_member *member, + char **fname); + +static const astreamer_ops astreamer_waldump_ops = { + .content = astreamer_waldump_content, + .finalize = astreamer_waldump_finalize, + .free = astreamer_waldump_free +}; + +/* + * Returns true if the given file is a tar archive and outputs its compression + * algorithm. + */ +bool +is_archive_file(const char *fname, pg_compress_algorithm *compression) +{ + int fname_len = strlen(fname); + + /* Now, check the compression type of the tar */ + if (fname_len > 4 && + strcmp(fname + fname_len - 4, ".tar") == 0) + *compression = PG_COMPRESSION_NONE; + else if (fname_len > 4 && + strcmp(fname + fname_len - 4, ".tgz") == 0) + *compression = PG_COMPRESSION_GZIP; + else if (fname_len > 7 && + strcmp(fname + fname_len - 7, ".tar.gz") == 0) + *compression = PG_COMPRESSION_GZIP; + else if (fname_len > 8 && + strcmp(fname + fname_len - 8, ".tar.lz4") == 0) + *compression = PG_COMPRESSION_LZ4; + else if (fname_len > 8 && + strcmp(fname + fname_len - 8, ".tar.zst") == 0) + *compression = PG_COMPRESSION_ZSTD; + else + return false; + + return true; +} + +/* + * Initializes the tar archive reader, creates a hash table for WAL entries, + * checks for existing valid WAL segments in the archive file and retrieves the + * segment size, and sets up filters for relevant entries. + */ +void +init_archive_reader(XLogDumpPrivate *privateInfo, const char *waldir, + int *WalSegSz, pg_compress_algorithm compression) +{ + int fd; + astreamer *streamer; + ArchivedWALFile *entry = NULL; + XLogLongPageHeader longhdr; + XLogSegNo segno; + TimeLineID timeline; + + /* Open tar archive and store its file descriptor */ + fd = open_file_in_directory(waldir, privateInfo->archive_name); + + if (fd < 0) + pg_fatal("could not open file \"%s\"", privateInfo->archive_name); + + privateInfo->archive_fd = fd; + + streamer = astreamer_waldump_new(privateInfo); + + /* Before that we must parse the tar archive. */ + streamer = astreamer_tar_parser_new(streamer); + + /* Before that we must decompress, if archive is compressed. */ + if (compression == PG_COMPRESSION_GZIP) + streamer = astreamer_gzip_decompressor_new(streamer); + else if (compression == PG_COMPRESSION_LZ4) + streamer = astreamer_lz4_decompressor_new(streamer); + else if (compression == PG_COMPRESSION_ZSTD) + streamer = astreamer_zstd_decompressor_new(streamer); + + privateInfo->archive_streamer = streamer; + + /* + * Hash table storing WAL entries read from the archive with an arbitrary + * initial size + */ + privateInfo->archive_wal_htab = ArchivedWAL_create(8, NULL); + + /* + * Verify that the archive contains valid WAL files and fetch WAL segment + * size + */ + while (entry == NULL || entry->buf->len < XLOG_BLCKSZ) + { + if (read_archive_file(privateInfo, XLOG_BLCKSZ) == 0) + pg_fatal("could not find WAL in archive \"%s\"", + privateInfo->archive_name); + + entry = privateInfo->cur_file; + } + + /* Set WalSegSz if WAL data is successfully read */ + longhdr = (XLogLongPageHeader) entry->buf->data; + + if (!IsValidWalSegSize(longhdr->xlp_seg_size)) + { + pg_log_error(ngettext("invalid WAL segment size in WAL file from archive \"%s\" (%d byte)", + "invalid WAL segment size in WAL file from archive \"%s\" (%d bytes)", + longhdr->xlp_seg_size), + privateInfo->archive_name, longhdr->xlp_seg_size); + pg_log_error_detail("The WAL segment size must be a power of two between 1 MB and 1 GB."); + exit(1); + } + + *WalSegSz = longhdr->xlp_seg_size; + + /* + * With the WAL segment size available, we can now initialize the + * dependent start and end segment numbers. + */ + Assert(!XLogRecPtrIsInvalid(privateInfo->startptr)); + XLByteToSeg(privateInfo->startptr, privateInfo->start_segno, *WalSegSz); + + if (!XLogRecPtrIsInvalid(privateInfo->endptr)) + XLByteToSeg(privateInfo->endptr, privateInfo->end_segno, *WalSegSz); + + /* + * This WAL record was fetched before the filtering parameters + * (start_segno and end_segno) were fully initialized. Perform the + * relevance check against the user-provided range now; if the WAL falls + * outside this range, remove it from the hash table. Subsequent WAL will + * be filtered automatically by the archived streamer using the updated + * start_segno and end_segno values. + */ + XLogFromFileName(entry->fname, &timeline, &segno, privateInfo->segsize); + if (privateInfo->timeline != timeline || + privateInfo->start_segno > segno || + privateInfo->end_segno < segno) + free_archive_wal_entry(entry->fname, privateInfo); +} + +/* + * Release the archive streamer chain and close the archive file. + */ +void +free_archive_reader(XLogDumpPrivate *privateInfo) +{ + /* + * NB: Normally, astreamer_finalize() is called before astreamer_free() to + * flush any remaining buffered data or to ensure the end of the tar + * archive is reached. However, when decoding a WAL file, once we hit the + * end LSN, any remaining WAL data in the buffer or the tar archive's + * unreached end can be safely ignored. + */ + astreamer_free(privateInfo->archive_streamer); + + /* Close the file. */ + if (close(privateInfo->archive_fd) != 0) + pg_log_error("could not close file \"%s\": %m", + privateInfo->archive_name); +} + +/* + * Copies WAL data from astreamer to readBuff; if unavailable, fetches more + * from the tar archive via astreamer. + */ +int +read_archive_wal_page(XLogDumpPrivate *privateInfo, XLogRecPtr targetPagePtr, + Size count, char *readBuff, int WalSegSz) +{ + char *p = readBuff; + Size nbytes = count; + XLogRecPtr recptr = targetPagePtr; + XLogSegNo segno; + char fname[MAXFNAMELEN]; + ArchivedWALFile *entry; + + /* Identify the segment and locate its entry in the archive hash */ + XLByteToSeg(targetPagePtr, segno, WalSegSz); + XLogFileName(fname, privateInfo->timeline, segno, WalSegSz); + entry = get_archive_wal_entry(fname, privateInfo, WalSegSz); + + while (nbytes > 0) + { + char *buf = entry->buf->data; + int bufLen = entry->buf->len; + XLogRecPtr endPtr; + XLogRecPtr startPtr; + + /* Calculate the LSN range currently residing in the buffer */ + XLogSegNoOffsetToRecPtr(segno, entry->read_len, WalSegSz, endPtr); + startPtr = endPtr - bufLen; + + /* + * Copy the requested WAL record if it exists in the buffer. + */ + if (bufLen > 0 && startPtr <= recptr && recptr < endPtr) + { + int copyBytes; + int offset = recptr - startPtr; + + /* + * Given startPtr <= recptr < endPtr and a total buffer size + * 'bufLen', the offset (recptr - startPtr) will always be less + * than 'bufLen'. + */ + Assert(offset < bufLen); + + copyBytes = Min(nbytes, bufLen - offset); + memcpy(p, buf + offset, copyBytes); + + /* Update state for read */ + recptr += copyBytes; + nbytes -= copyBytes; + p += copyBytes; + } + else + { + /* + * Before starting the actual decoding loop, pg_waldump tries to + * locate the first valid record from the user-specified start + * position, which might not be the start of a WAL record and + * could fall in the middle of a record that spans multiple pages. + * Consequently, the valid start position the decoder is looking + * for could be far away from that initial position. + * + * This may involve reading across multiple pages, and this + * pre-reading fetches data in multiple rounds from the archive + * streamer; normally, we would throw away existing buffer + * contents to fetch the next set of data, but that existing data + * might be needed once the main loop starts. Because previously + * read data cannot be re-read by the archive streamer, we delay + * resetting the buffer until the main decoding loop is entered. + * + * Once pg_waldump has entered the main loop, it may re-read the + * currently active page, but never an older one; therefore, any + * fully consumed WAL data preceding the current page can then be + * safely discarded. + */ + if (privateInfo->decoding_started) + { + resetStringInfo(entry->buf); + + /* + * Push back the partial page data for the current page to the + * buffer, ensuring it remains full page available for + * re-reading if requested. + */ + if (p > readBuff) + { + Assert((count - nbytes) > 0); + appendBinaryStringInfo(entry->buf, readBuff, count - nbytes); + } + } + + /* + * Now, fetch more data; raise an error if it's not the current + * segment being read by the archive streamer or if reading of the + * archived file has finished. + */ + if (privateInfo->cur_file != entry || + read_archive_file(privateInfo, READ_CHUNK_SIZE) == 0) + pg_fatal("could not read file \"%s\" from archive \"%s\": read %lld of %lld", + fname, privateInfo->archive_name, + (long long int) count - nbytes, + (long long int) nbytes); + } + } + + /* + * Should have either have successfully read all the requested bytes or + * reported a failure before this point. + */ + Assert(nbytes == 0); + + /* + * NB: We return the fixed value provided as input. Although we could + * return a boolean since we either successfully read the WAL page or + * raise an error, but the caller expects this value to be returned. The + * routine that reads WAL pages from the physical WAL file follows the + * same convention. + */ + return count; +} + +/* + * Clears the buffer of a WAL entry that is being ignored. This frees up memory + * and prevents the accumulation of irrelevant WAL data. Additionally, + * conditionally setting cur_file within privateinfo to NULL ensures the + * archive streamer skips unnecessary copy operations + */ +void +free_archive_wal_entry(const char *fname, XLogDumpPrivate *privateInfo) +{ + ArchivedWALFile *entry; + + entry = ArchivedWAL_lookup(privateInfo->archive_wal_htab, fname); + + if (entry == NULL) + return; + + /* Destroy the buffer */ + destroyStringInfo(entry->buf); + entry->buf = NULL; + + /* Set cur_file to NULL if it matches the entry being ignored */ + if (privateInfo->cur_file == entry) + privateInfo->cur_file = NULL; + + ArchivedWAL_delete_item(privateInfo->archive_wal_htab, entry); +} + +/* + * Returns the archived WAL entry from the hash table if it exists. Otherwise, + * it invokes the routine to read the archived file, which then populates the + * entry in the hash table if that WAL exists in the archive. + */ +static ArchivedWALFile * +get_archive_wal_entry(const char *fname, XLogDumpPrivate *privateInfo, + int WalSegSz) +{ + ArchivedWALFile *entry = NULL; + + /* Search hash table */ + entry = ArchivedWAL_lookup(privateInfo->archive_wal_htab, fname); + + if (entry != NULL) + return entry; + + /* + * The requested WAL entry has not been read from the archive yet; invoke + * the archive streamer to read it. + */ + while (1) + { + /* Fetch more data */ + if (read_archive_file(privateInfo, READ_CHUNK_SIZE) == 0) + break; /* archive file ended */ + + /* + * Archived streamer is reading a non-WAL file or an irrelevant WAL + * file. + */ + if (privateInfo->cur_file == NULL) + continue; + + entry = privateInfo->cur_file; + + /* Found the required entry */ + if (strcmp(fname, entry->fname) == 0) + return entry; + + /* WAL segments must be archived in order */ + pg_log_error("WAL files are not archived in sequential order"); + pg_log_error_detail("Expecting segment \"%s\" but found \"%s\".", + fname, entry->fname); + exit(1); + } + + /* Requested WAL segment not found */ + pg_fatal("could not find WAL \"%s\" in archive \"%s\"", + fname, privateInfo->archive_name); +} + +/* + * Reads the archive file and passes it to the archive streamer for + * decompression. + */ +static int +read_archive_file(XLogDumpPrivate *privateInfo, Size count) +{ + int rc; + char *buffer; + + buffer = pg_malloc(count * sizeof(uint8)); + + rc = read(privateInfo->archive_fd, buffer, count); + if (rc < 0) + pg_fatal("could not read file \"%s\": %m", + privateInfo->archive_name); + + /* + * Decompress (if required), and then parse the previously read contents + * of the tar file. + */ + if (rc > 0) + astreamer_content(privateInfo->archive_streamer, NULL, + buffer, rc, ASTREAMER_UNKNOWN); + pg_free(buffer); + + return rc; +} + +/* + * Create an astreamer that can read WAL from a tar file. + */ +static astreamer * +astreamer_waldump_new(XLogDumpPrivate *privateInfo) +{ + astreamer_waldump *streamer; + + streamer = palloc0(sizeof(astreamer_waldump)); + *((const astreamer_ops **) &streamer->base.bbs_ops) = + &astreamer_waldump_ops; + + streamer->privateInfo = privateInfo; + + return &streamer->base; +} + +/* + * Main entry point of the archive streamer for reading WAL data from a tar + * file. If a member is identified as a valid WAL file, a hash entry is created + * for it, and its contents are copied into that entry's buffer, making them + * accessible to the decoding routine. + */ +static void +astreamer_waldump_content(astreamer *streamer, astreamer_member *member, + const char *data, int len, + astreamer_archive_context context) +{ + astreamer_waldump *mystreamer = (astreamer_waldump *) streamer; + XLogDumpPrivate *privateInfo = mystreamer->privateInfo; + + Assert(context != ASTREAMER_UNKNOWN); + + switch (context) + { + case ASTREAMER_MEMBER_HEADER: + { + char *fname = NULL; + ArchivedWALFile *entry; + bool found; + + pg_log_debug("reading \"%s\"", member->pathname); + + if (!member_is_wal_file(mystreamer, member, &fname)) + break; + + /* + * Further checks are skipped if any WAL file can be read. + * This typically occurs during initial verification. + */ + if (!READ_ANY_WAL(privateInfo)) + { + XLogSegNo segno; + TimeLineID timeline; + + /* + * Skip the segment if the timeline does not match, if it + * falls outside the caller-specified range. + */ + XLogFromFileName(fname, &timeline, &segno, privateInfo->segsize); + if (privateInfo->timeline != timeline || + privateInfo->start_segno > segno || + privateInfo->end_segno < segno) + { + free(fname); + break; + } + } + + entry = ArchivedWAL_insert(privateInfo->archive_wal_htab, + fname, &found); + + /* + * Shouldn't happen, but if it does, simply ignore the + * duplicate WAL file. + */ + if (found) + { + pg_log_warning("ignoring duplicate WAL \"%s\" found in archive \"%s\"", + member->pathname, privateInfo->archive_name); + break; + } + + entry->buf = makeStringInfo(); + entry->read_len = 0; + privateInfo->cur_file = entry; + } + break; + + case ASTREAMER_MEMBER_CONTENTS: + if (privateInfo->cur_file) + { + appendBinaryStringInfo(privateInfo->cur_file->buf, data, len); + privateInfo->cur_file->read_len += len; + } + break; + + case ASTREAMER_MEMBER_TRAILER: + privateInfo->cur_file = NULL; + break; + + case ASTREAMER_ARCHIVE_TRAILER: + break; + + default: + /* Shouldn't happen. */ + pg_fatal("unexpected state while parsing tar file"); + } +} + +/* + * End-of-stream processing for an astreamer_waldump stream. + */ +static void +astreamer_waldump_finalize(astreamer *streamer) +{ + Assert(streamer->bbs_next == NULL); +} + +/* + * Free memory associated with a astreamer_waldump stream. + */ +static void +astreamer_waldump_free(astreamer *streamer) +{ + Assert(streamer->bbs_next == NULL); + pfree(streamer); +} + +/* + * Returns true if the archive member name matches the WAL naming format. If + * successful, it also outputs the WAL segment name. + */ +static bool +member_is_wal_file(astreamer_waldump *mystreamer, astreamer_member *member, + char **fname) +{ + int pathlen; + char pathname[MAXPGPATH]; + char *filename; + + /* We are only interested in normal files. */ + if (member->is_directory || member->is_link) + return false; + + if (strlen(member->pathname) < XLOG_FNAME_LEN) + return false; + + /* + * For a correct comparison, we must remove any '.' or '..' components + * from the member pathname. Similar to member_verify_header(), we prepend + * './' to the path so that canonicalize_path() can properly resolve and + * strip these references from the tar member name + */ + snprintf(pathname, MAXPGPATH, "./%s", member->pathname); + canonicalize_path(pathname); + pathlen = strlen(pathname); + + /* WAL files from the top-level or pg_wal directory will be decoded */ + if (pathlen > XLOG_FNAME_LEN && + strncmp(pathname, XLOGDIR, strlen(XLOGDIR)) != 0) + return false; + + /* WAL file could be with full path */ + filename = pathname + (pathlen - XLOG_FNAME_LEN); + if (!IsXLogFileName(filename)) + return false; + + *fname = pnstrdup(filename, XLOG_FNAME_LEN); + + return true; +} + +/* + * Helper function for filemap hash table. + */ +static uint32 +hash_string_pointer(const char *s) +{ + unsigned char *ss = (unsigned char *) s; + + return hash_bytes(ss, strlen(s)); +} diff --git a/src/bin/pg_waldump/meson.build b/src/bin/pg_waldump/meson.build index 633a9874bb5..5296f21b82c 100644 --- a/src/bin/pg_waldump/meson.build +++ b/src/bin/pg_waldump/meson.build @@ -1,6 +1,7 @@ # Copyright (c) 2022-2026, PostgreSQL Global Development Group pg_waldump_sources = files( + 'archive_waldump.c', 'compat.c', 'pg_waldump.c', 'rmgrdesc.c', @@ -18,7 +19,7 @@ endif pg_waldump = executable('pg_waldump', pg_waldump_sources, - dependencies: [frontend_code, lz4, zstd], + dependencies: [frontend_code, libpq, lz4, zstd], c_args: ['-DFRONTEND'], # needed for xlogreader et al kwargs: default_bin_args, ) @@ -29,6 +30,7 @@ tests += { 'sd': meson.current_source_dir(), 'bd': meson.current_build_dir(), 'tap': { + 'env': {'TAR': tar.found() ? tar.full_path() : ''}, 'tests': [ 't/001_basic.pl', 't/002_save_fullpage.pl', diff --git a/src/bin/pg_waldump/pg_waldump.c b/src/bin/pg_waldump/pg_waldump.c index 5d31b15dbd8..6d04462d039 100644 --- a/src/bin/pg_waldump/pg_waldump.c +++ b/src/bin/pg_waldump/pg_waldump.c @@ -176,7 +176,7 @@ split_path(const char *path, char **dir, char **fname) * * return a read only fd */ -static int +int open_file_in_directory(const char *directory, const char *fname) { int fd = -1; @@ -440,6 +440,67 @@ WALDumpReadPage(XLogReaderState *state, XLogRecPtr targetPagePtr, int reqLen, return count; } +/* + * pg_waldump's XLogReaderRoutine->segment_open callback to support dumping WAL + * files from tar archives. + */ +static void +TarWALDumpOpenSegment(XLogReaderState *state, XLogSegNo nextSegNo, + TimeLineID *tli_p) +{ + /* No action needed */ +} + +/* + * pg_waldump's XLogReaderRoutine->segment_close callback. + */ +static void +TarWALDumpCloseSegment(XLogReaderState *state) +{ + /* No action needed */ +} + +/* + * pg_waldump's XLogReaderRoutine->page_read callback to support dumping WAL + * files from tar archives. + */ +static int +TarWALDumpReadPage(XLogReaderState *state, XLogRecPtr targetPagePtr, int reqLen, + XLogRecPtr targetPtr, char *readBuff) +{ + XLogDumpPrivate *private = state->private_data; + int count = required_read_len(private, targetPagePtr, reqLen); + int WalSegSz = state->segcxt.ws_segsize; + XLogSegNo nextSegNo; + + /* Bail out if the count to be read is not valid */ + if (count < 0) + return -1; + + /* + * If the target page is in a different segment, free the buffer space + * occupied by the previous segment data. Since pg_waldump never requests + * the same WAL bytes twice, moving to a new segment implies the previous + * buffer's data and that segment will not be needed again. + */ + nextSegNo = state->seg.ws_segno; + if (!XLByteInSeg(targetPagePtr, nextSegNo, WalSegSz)) + { + char fname[MAXFNAMELEN]; + + XLogFileName(fname, state->seg.ws_tli, nextSegNo, WalSegSz); + free_archive_wal_entry(fname, private); + + XLByteToSeg(targetPagePtr, nextSegNo, WalSegSz); + state->seg.ws_tli = private->timeline; + state->seg.ws_segno = nextSegNo; + } + + /* Read the WAL page from the archive streamer */ + return read_archive_wal_page(private, targetPagePtr, count, readBuff, + WalSegSz); +} + /* * Boolean to return whether the given WAL record matches a specific relation * and optionally block. @@ -777,8 +838,8 @@ usage(void) printf(_(" -F, --fork=FORK only show records that modify blocks in fork FORK;\n" " valid names are main, fsm, vm, init\n")); printf(_(" -n, --limit=N number of records to display\n")); - printf(_(" -p, --path=PATH directory in which to find WAL segment files or a\n" - " directory with a ./pg_wal that contains such files\n" + printf(_(" -p, --path=PATH tar archive or a directory in which to find WAL segment files or\n" + " a directory with a ./pg_wal that contains such files\n" " (default: current directory, ./pg_wal, $PGDATA/pg_wal)\n")); printf(_(" -q, --quiet do not print any output, except for errors\n")); printf(_(" -r, --rmgr=RMGR only show records generated by resource manager RMGR;\n" @@ -810,7 +871,9 @@ main(int argc, char **argv) XLogRecord *record; XLogRecPtr first_record; char *waldir = NULL; + char *walpath = NULL; char *errormsg; + pg_compress_algorithm compression; static struct option long_options[] = { {"bkp-details", no_argument, NULL, 'b'}, @@ -868,6 +931,10 @@ main(int argc, char **argv) private.startptr = InvalidXLogRecPtr; private.endptr = InvalidXLogRecPtr; private.endptr_reached = false; + private.decoding_started = false; + private.archive_name = NULL; + private.start_segno = 0; + private.end_segno = UINT64_MAX; config.quiet = false; config.bkp_details = false; @@ -943,7 +1010,7 @@ main(int argc, char **argv) } break; case 'p': - waldir = pg_strdup(optarg); + walpath = pg_strdup(optarg); break; case 'q': config.quiet = true; @@ -1107,10 +1174,19 @@ main(int argc, char **argv) goto bad_argument; } - if (waldir != NULL) + if (walpath != NULL) { + /* validate path points to tar archive */ + if (is_archive_file(walpath, &compression)) + { + char *fname = NULL; + + split_path(walpath, &waldir, &fname); + + private.archive_name = fname; + } /* validate path points to directory */ - if (!verify_directory(waldir)) + else if (!verify_directory(walpath)) { pg_log_error("could not open directory \"%s\": %m", waldir); goto bad_argument; @@ -1128,6 +1204,17 @@ main(int argc, char **argv) int fd; XLogSegNo segno; + /* + * If a tar archive is passed using the --path option, all other + * arguments become unnecessary. + */ + if (private.archive_name) + { + pg_log_error("unnecessary command-line arguments specified with tar archive (first is \"%s\")", + argv[optind]); + goto bad_argument; + } + split_path(argv[optind], &directory, &fname); if (waldir == NULL && directory != NULL) @@ -1138,69 +1225,76 @@ main(int argc, char **argv) pg_fatal("could not open directory \"%s\": %m", waldir); } - waldir = identify_target_directory(waldir, fname, &private.segsize); - fd = open_file_in_directory(waldir, fname); - if (fd < 0) - pg_fatal("could not open file \"%s\"", fname); - close(fd); - - /* parse position from file */ - XLogFromFileName(fname, &private.timeline, &segno, private.segsize); - - if (!XLogRecPtrIsValid(private.startptr)) - XLogSegNoOffsetToRecPtr(segno, 0, private.segsize, private.startptr); - else if (!XLByteInSeg(private.startptr, segno, private.segsize)) + if (fname != NULL && is_archive_file(fname, &compression)) { - pg_log_error("start WAL location %X/%08X is not inside file \"%s\"", - LSN_FORMAT_ARGS(private.startptr), - fname); - goto bad_argument; + private.archive_name = fname; } - - /* no second file specified, set end position */ - if (!(optind + 1 < argc) && !XLogRecPtrIsValid(private.endptr)) - XLogSegNoOffsetToRecPtr(segno + 1, 0, private.segsize, private.endptr); - - /* parse ENDSEG if passed */ - if (optind + 1 < argc) + else { - XLogSegNo endsegno; - - /* ignore directory, already have that */ - split_path(argv[optind + 1], &directory, &fname); - + waldir = identify_target_directory(waldir, fname, &private.segsize); fd = open_file_in_directory(waldir, fname); if (fd < 0) pg_fatal("could not open file \"%s\"", fname); close(fd); /* parse position from file */ - XLogFromFileName(fname, &private.timeline, &endsegno, private.segsize); + XLogFromFileName(fname, &private.timeline, &segno, private.segsize); - if (endsegno < segno) - pg_fatal("ENDSEG %s is before STARTSEG %s", - argv[optind + 1], argv[optind]); + if (!XLogRecPtrIsValid(private.startptr)) + XLogSegNoOffsetToRecPtr(segno, 0, private.segsize, private.startptr); + else if (!XLByteInSeg(private.startptr, segno, private.segsize)) + { + pg_log_error("start WAL location %X/%08X is not inside file \"%s\"", + LSN_FORMAT_ARGS(private.startptr), + fname); + goto bad_argument; + } - if (!XLogRecPtrIsValid(private.endptr)) - XLogSegNoOffsetToRecPtr(endsegno + 1, 0, private.segsize, - private.endptr); + /* no second file specified, set end position */ + if (!(optind + 1 < argc) && !XLogRecPtrIsValid(private.endptr)) + XLogSegNoOffsetToRecPtr(segno + 1, 0, private.segsize, private.endptr); - /* set segno to endsegno for check of --end */ - segno = endsegno; - } + /* parse ENDSEG if passed */ + if (optind + 1 < argc) + { + XLogSegNo endsegno; + /* ignore directory, already have that */ + split_path(argv[optind + 1], &directory, &fname); - if (!XLByteInSeg(private.endptr, segno, private.segsize) && - private.endptr != (segno + 1) * private.segsize) - { - pg_log_error("end WAL location %X/%08X is not inside file \"%s\"", - LSN_FORMAT_ARGS(private.endptr), - argv[argc - 1]); - goto bad_argument; + fd = open_file_in_directory(waldir, fname); + if (fd < 0) + pg_fatal("could not open file \"%s\"", fname); + close(fd); + + /* parse position from file */ + XLogFromFileName(fname, &private.timeline, &endsegno, private.segsize); + + if (endsegno < segno) + pg_fatal("ENDSEG %s is before STARTSEG %s", + argv[optind + 1], argv[optind]); + + if (!XLogRecPtrIsValid(private.endptr)) + XLogSegNoOffsetToRecPtr(endsegno + 1, 0, private.segsize, + private.endptr); + + /* set segno to endsegno for check of --end */ + segno = endsegno; + } + + + if (!XLByteInSeg(private.endptr, segno, private.segsize) && + private.endptr != (segno + 1) * private.segsize) + { + pg_log_error("end WAL location %X/%08X is not inside file \"%s\"", + LSN_FORMAT_ARGS(private.endptr), + argv[argc - 1]); + goto bad_argument; + } } } - else - waldir = identify_target_directory(waldir, NULL, &private.segsize); + else if (!private.archive_name) + waldir = identify_target_directory(walpath, NULL, &private.segsize); /* we don't know what to print */ if (!XLogRecPtrIsValid(private.startptr)) @@ -1212,12 +1306,36 @@ main(int argc, char **argv) /* done with argument parsing, do the actual work */ /* we have everything we need, start reading */ - xlogreader_state = - XLogReaderAllocate(private.segsize, waldir, - XL_ROUTINE(.page_read = WALDumpReadPage, - .segment_open = WALDumpOpenSegment, - .segment_close = WALDumpCloseSegment), - &private); + if (private.archive_name) + { + /* + * A NULL WAL directory indicates that the archive file is located in + * the current working directory of the pg_waldump execution + */ + if (waldir == NULL) + waldir = pg_strdup("."); + + /* Set up for reading tar file */ + init_archive_reader(&private, waldir, &private.segsize, compression); + + /* Routine to decode WAL files in tar archive */ + xlogreader_state = + XLogReaderAllocate(private.segsize, waldir, + XL_ROUTINE(.page_read = TarWALDumpReadPage, + .segment_open = TarWALDumpOpenSegment, + .segment_close = TarWALDumpCloseSegment), + &private); + } + else + { + xlogreader_state = + XLogReaderAllocate(private.segsize, waldir, + XL_ROUTINE(.page_read = WALDumpReadPage, + .segment_open = WALDumpOpenSegment, + .segment_close = WALDumpCloseSegment), + &private); + } + if (!xlogreader_state) pg_fatal("out of memory while allocating a WAL reading processor"); @@ -1245,6 +1363,9 @@ main(int argc, char **argv) if (config.stats == true && !config.quiet) stats.startptr = first_record; + /* Flag indicating that the decoding loop has been entered */ + private.decoding_started = true; + for (;;) { if (time_to_stop) @@ -1326,6 +1447,9 @@ main(int argc, char **argv) XLogReaderFree(xlogreader_state); + if (private.archive_name) + free_archive_reader(&private); + return EXIT_SUCCESS; bad_argument: diff --git a/src/bin/pg_waldump/pg_waldump.h b/src/bin/pg_waldump/pg_waldump.h index 4f1b2ab668b..02da2c43b08 100644 --- a/src/bin/pg_waldump/pg_waldump.h +++ b/src/bin/pg_waldump/pg_waldump.h @@ -12,6 +12,11 @@ #define PG_WALDUMP_H #include "access/xlogdefs.h" +#include "fe_utils/astreamer.h" + +/* Forward declaration */ +struct ArchivedWALFile; +struct ArchivedWAL_hash; /* Contains the necessary information to drive WAL decoding */ typedef struct XLogDumpPrivate @@ -21,6 +26,46 @@ typedef struct XLogDumpPrivate XLogRecPtr startptr; XLogRecPtr endptr; bool endptr_reached; + bool decoding_started; + + /* Fields required to read WAL from archive */ + char *archive_name; /* Tar archive name */ + int archive_fd; /* File descriptor for the open tar file */ + + astreamer *archive_streamer; + + /* What the archive streamer is currently reading */ + struct ArchivedWALFile *cur_file; + + /* + * Hash table of all WAL files that the archive stream has read, including + * the one currently in progress. + */ + struct ArchivedWAL_hash *archive_wal_htab; + + /* + * Although these values can be easily derived from startptr and endptr, + * doing so repeatedly for each archived member would be inefficient, as + * it would involve recalculating and filtering out irrelevant WAL + * segments. + */ + XLogSegNo start_segno; + XLogSegNo end_segno; } XLogDumpPrivate; +extern int open_file_in_directory(const char *directory, const char *fname); + +extern bool is_archive_file(const char *fname, + pg_compress_algorithm *compression); +extern void init_archive_reader(XLogDumpPrivate *privateInfo, + const char *waldir, int *WalSegSz, + pg_compress_algorithm compression); +extern void free_archive_reader(XLogDumpPrivate *privateInfo); +extern int read_archive_wal_page(XLogDumpPrivate *privateInfo, + XLogRecPtr targetPagePtr, + Size count, char *readBuff, + int WalSegSz); +extern void free_archive_wal_entry(const char *fname, + XLogDumpPrivate *privateInfo); + #endif /* end of PG_WALDUMP_H */ diff --git a/src/bin/pg_waldump/t/001_basic.pl b/src/bin/pg_waldump/t/001_basic.pl index 3288fadcf48..cae543c8990 100644 --- a/src/bin/pg_waldump/t/001_basic.pl +++ b/src/bin/pg_waldump/t/001_basic.pl @@ -3,10 +3,13 @@ use strict; use warnings FATAL => 'all'; +use Cwd; use PostgreSQL::Test::Cluster; use PostgreSQL::Test::Utils; use Test::More; +my $tar = $ENV{TAR}; + program_help_ok('pg_waldump'); program_version_ok('pg_waldump'); program_options_handling_ok('pg_waldump'); @@ -235,7 +238,7 @@ command_like( sub test_pg_waldump { local $Test::Builder::Level = $Test::Builder::Level + 1; - my @opts = @_; + my ($path, @opts) = @_; my ($stdout, $stderr); @@ -243,6 +246,7 @@ sub test_pg_waldump 'pg_waldump', '--start' => $start_lsn, '--end' => $end_lsn, + '--path' => $path, @opts ], '>' => \$stdout, @@ -254,11 +258,50 @@ sub test_pg_waldump return @lines; } -my @lines; +# Create a tar archive, sorting the file order +sub generate_archive +{ + my ($archive, $directory, $compression_flags) = @_; + + my @files; + opendir my $dh, $directory or die "opendir: $!"; + while (my $entry = readdir $dh) { + # Skip '.' and '..' + next if $entry eq '.' || $entry eq '..'; + push @files, $entry; + } + closedir $dh; + + @files = sort @files; + + # move into the WAL directory before archiving files + my $cwd = getcwd; + chdir($directory) || die "chdir: $!"; + command_ok([$tar, $compression_flags, $archive, @files]); + chdir($cwd) || die "chdir: $!"; +} + +my $tmp_dir = PostgreSQL::Test::Utils::tempdir_short(); my @scenarios = ( { - 'path' => $node->data_dir + 'path' => $node->data_dir, + 'is_archive' => 0, + 'enabled' => 1 + }, + { + 'path' => "$tmp_dir/pg_wal.tar", + 'compression_method' => 'none', + 'compression_flags' => '-cf', + 'is_archive' => 1, + 'enabled' => 1 + }, + { + 'path' => "$tmp_dir/pg_wal.tar.gz", + 'compression_method' => 'gzip', + 'compression_flags' => '-czf', + 'is_archive' => 1, + 'enabled' => check_pg_config("#define HAVE_LIBZ 1") }); for my $scenario (@scenarios) @@ -267,6 +310,19 @@ for my $scenario (@scenarios) SKIP: { + skip "tar command is not available", 3 + if !defined $tar; + skip "$scenario->{'compression_method'} compression not supported by this build", 3 + if !$scenario->{'enabled'} && $scenario->{'is_archive'}; + + # create pg_wal archive + if ($scenario->{'is_archive'}) + { + generate_archive($path, + $node->data_dir . '/pg_wal', + $scenario->{'compression_flags'}); + } + command_fails_like( [ 'pg_waldump', '--path' => $path ], qr/error: no start WAL location given/, @@ -298,38 +354,41 @@ for my $scenario (@scenarios) qr/error: error in WAL record at/, 'errors are shown with --quiet'); - @lines = test_pg_waldump('--path' => $path); + my @lines = test_pg_waldump($path); is(grep(!/^rmgr: \w/, @lines), 0, 'all output lines are rmgr lines'); - @lines = test_pg_waldump('--path' => $path, '--limit' => 6); + @lines = test_pg_waldump($path, '--limit' => 6); is(@lines, 6, 'limit option observed'); - @lines = test_pg_waldump('--path' => $path, '--fullpage'); + @lines = test_pg_waldump($path, '--fullpage'); is(grep(!/^rmgr:.*\bFPW\b/, @lines), 0, 'all output lines are FPW'); - @lines = test_pg_waldump('--path' => $path, '--stats'); + @lines = test_pg_waldump($path, '--stats'); like($lines[0], qr/WAL statistics/, "statistics on stdout"); is(grep(/^rmgr:/, @lines), 0, 'no rmgr lines output'); - @lines = test_pg_waldump('--path' => $path, '--stats=record'); + @lines = test_pg_waldump($path, '--stats=record'); like($lines[0], qr/WAL statistics/, "statistics on stdout"); is(grep(/^rmgr:/, @lines), 0, 'no rmgr lines output'); - @lines = test_pg_waldump('--path' => $path, '--rmgr' => 'Btree'); + @lines = test_pg_waldump($path, '--rmgr' => 'Btree'); is(grep(!/^rmgr: Btree/, @lines), 0, 'only Btree lines'); - @lines = test_pg_waldump('--path' => $path, '--fork' => 'init'); + @lines = test_pg_waldump($path, '--fork' => 'init'); is(grep(!/fork init/, @lines), 0, 'only init fork lines'); - @lines = test_pg_waldump('--path' => $path, + @lines = test_pg_waldump($path, '--relation' => "$default_ts_oid/$postgres_db_oid/$rel_t1_oid"); is(grep(!/rel $default_ts_oid\/$postgres_db_oid\/$rel_t1_oid/, @lines), 0, 'only lines for selected relation'); - @lines = test_pg_waldump('--path' => $path, + @lines = test_pg_waldump($path, '--relation' => "$default_ts_oid/$postgres_db_oid/$rel_i1a_oid", '--block' => 1); is(grep(!/\bblk 1\b/, @lines), 0, 'only lines for selected block'); + + # Cleanup. + unlink $path if $scenario->{'is_archive'}; } } diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 9f5ee8fd482..2cd87de84ee 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -144,6 +144,8 @@ ArchiveOpts ArchiveShutdownCB ArchiveStartupCB ArchiveStreamState +ArchivedWALFile +ArchivedWAL_hash ArchiverOutput ArchiverStage ArrayAnalyzeExtraData @@ -3506,6 +3508,7 @@ astreamer_recovery_injector astreamer_tar_archiver astreamer_tar_parser astreamer_verify +astreamer_waldump astreamer_zstd_frame auth_password_hook_typ autovac_table -- 2.47.1
From 77e29d3162afec46d9ed9f3d592bdbeee6347b37 Mon Sep 17 00:00:00 2001 From: Amul Sul <[email protected]> Date: Tue, 27 Jan 2026 15:38:34 +0530 Subject: [PATCH v12 6/9] pg_waldump: Remove the restriction on the order of archived WAL files. With previous patch, pg_waldump would stop decoding if WAL files were not in the required sequence. With this patch, decoding will now continue. Any WAL file that is out of order will be written to a temporary location, from which it will be read later. Once a temporary file has been read, it will be removed. --- doc/src/sgml/ref/pg_waldump.sgml | 8 +- src/bin/pg_waldump/archive_waldump.c | 171 +++++++++++++++++++++++++-- src/bin/pg_waldump/pg_waldump.c | 31 ++++- src/bin/pg_waldump/pg_waldump.h | 3 + src/bin/pg_waldump/t/001_basic.pl | 3 +- 5 files changed, 196 insertions(+), 20 deletions(-) diff --git a/doc/src/sgml/ref/pg_waldump.sgml b/doc/src/sgml/ref/pg_waldump.sgml index d004bb0f67e..27adf77755c 100644 --- a/doc/src/sgml/ref/pg_waldump.sgml +++ b/doc/src/sgml/ref/pg_waldump.sgml @@ -149,8 +149,12 @@ PostgreSQL documentation of <envar>PGDATA</envar>. </para> <para> - If a tar archive is provided, its WAL segment files must be in - sequential order; otherwise, an error will be reported. + If a tar archive is provided and its WAL segment files are not in + sequential order, those files will be written to a temporary directory + named starting with <filename>waldump_tmp</filename>. This directory will be + created inside the directory specified by the <envar>TMPDIR</envar> + environment variable if it is set; otherwise, it will be created within + the same directory as the tar archive. </para> </listitem> </varlistentry> diff --git a/src/bin/pg_waldump/archive_waldump.c b/src/bin/pg_waldump/archive_waldump.c index 27a5a5c6d5d..b1353088c4a 100644 --- a/src/bin/pg_waldump/archive_waldump.c +++ b/src/bin/pg_waldump/archive_waldump.c @@ -17,6 +17,7 @@ #include <unistd.h> #include "access/xlog_internal.h" +#include "common/file_perm.h" #include "common/hashfn.h" #include "common/logging.h" #include "fe_utils/simple_list.h" @@ -27,6 +28,9 @@ */ #define READ_CHUNK_SIZE (128 * 1024) +/* Temporary exported WAL file directory */ +char *TmpWalSegDir = NULL; + /* * Check if the start segment number is zero; this indicates a request to read * any WAL file. @@ -57,6 +61,8 @@ typedef struct ArchivedWALFile const char *fname; /* hash key: WAL segment name */ StringInfo buf; /* holds WAL bytes read from archive */ + bool spilled; /* true if the WAL data was spilled to a + * temporary file */ int read_len; /* total bytes of a WAL read from archive */ } ArchivedWALFile; @@ -84,6 +90,11 @@ static ArchivedWALFile *get_archive_wal_entry(const char *fname, XLogDumpPrivate *privateInfo, int WalSegSz); static int read_archive_file(XLogDumpPrivate *privateInfo, Size count); +static void setup_tmpwal_dir(const char *waldir); +static void cleanup_tmpwal_dir_atexit(void); + +static FILE *prepare_tmp_write(const char *fname); +static void perform_tmp_write(const char *fname, StringInfo buf, FILE *file); static astreamer *astreamer_waldump_new(XLogDumpPrivate *privateInfo); static void astreamer_waldump_content(astreamer *streamer, @@ -137,7 +148,9 @@ is_archive_file(const char *fname, pg_compress_algorithm *compression) /* * Initializes the tar archive reader, creates a hash table for WAL entries, * checks for existing valid WAL segments in the archive file and retrieves the - * segment size, and sets up filters for relevant entries. + * segment size, and sets up filters for relevant entries. It also configures a + * temporary directory for out-of-order WAL data and registers an exit callback + * to clean up temporary files. */ void init_archive_reader(XLogDumpPrivate *privateInfo, const char *waldir, @@ -230,6 +243,13 @@ init_archive_reader(XLogDumpPrivate *privateInfo, const char *waldir, privateInfo->start_segno > segno || privateInfo->end_segno < segno) free_archive_wal_entry(entry->fname, privateInfo); + + /* + * Setup temporary directory to store WAL segments and set up an exit + * callback to remove it upon completion. + */ + setup_tmpwal_dir(waldir); + atexit(cleanup_tmpwal_dir_atexit); } /* @@ -396,6 +416,17 @@ free_archive_wal_entry(const char *fname, XLogDumpPrivate *privateInfo) destroyStringInfo(entry->buf); entry->buf = NULL; + /* Remove temporary file if any */ + if (entry->spilled) + { + char fpath[MAXPGPATH]; + + snprintf(fpath, MAXPGPATH, "%s/%s", TmpWalSegDir, fname); + + if (unlink(fpath) == 0) + pg_log_debug("removed file \"%s\"", fpath); + } + /* Set cur_file to NULL if it matches the entry being ignored */ if (privateInfo->cur_file == entry) privateInfo->cur_file = NULL; @@ -407,12 +438,16 @@ free_archive_wal_entry(const char *fname, XLogDumpPrivate *privateInfo) * Returns the archived WAL entry from the hash table if it exists. Otherwise, * it invokes the routine to read the archived file, which then populates the * entry in the hash table if that WAL exists in the archive. + * If the archive streamer happens to be reading a + * WAL from archive file that is not currently needed, that WAL data is written + * to a temporary file. */ static ArchivedWALFile * get_archive_wal_entry(const char *fname, XLogDumpPrivate *privateInfo, int WalSegSz) { ArchivedWALFile *entry = NULL; + FILE *write_fp = NULL; /* Search hash table */ entry = ArchivedWAL_lookup(privateInfo->archive_wal_htab, fname); @@ -426,28 +461,59 @@ get_archive_wal_entry(const char *fname, XLogDumpPrivate *privateInfo, */ while (1) { + /* + * The WAL file entry currently being processed may change during + * archive streamer execution. Therefore, maintain a local variable to + * reference the previous entry, ensuring that any remaining data in + * its buffer is successfully flushed to the temporary file before + * switching to the next WAL entry. + */ + entry = privateInfo->cur_file; + /* Fetch more data */ - if (read_archive_file(privateInfo, READ_CHUNK_SIZE) == 0) - break; /* archive file ended */ + if (entry == NULL || entry->buf->len == 0) + { + if (read_archive_file(privateInfo, READ_CHUNK_SIZE) == 0) + break; /* archive file ended */ + } /* * Archived streamer is reading a non-WAL file or an irrelevant WAL * file. */ - if (privateInfo->cur_file == NULL) + if (entry == NULL) continue; - entry = privateInfo->cur_file; - /* Found the required entry */ if (strcmp(fname, entry->fname) == 0) return entry; - /* WAL segments must be archived in order */ - pg_log_error("WAL files are not archived in sequential order"); - pg_log_error_detail("Expecting segment \"%s\" but found \"%s\".", - fname, entry->fname); - exit(1); + /* + * Archive streamer is currently reading a file that isn't the one + * asked for, but it's required in the future. It should be written to + * a temporary location for retrieval when needed. + */ + + /* Create a temporary file if one does not already exist */ + if (!entry->spilled) + { + write_fp = prepare_tmp_write(entry->fname); + entry->spilled = true; + } + + /* Flush data from the buffer to the file */ + perform_tmp_write(entry->fname, entry->buf, write_fp); + resetStringInfo(entry->buf); + + /* + * The change in the current segment entry indicates that the reading + * of this file has ended. + */ + if (entry != privateInfo->cur_file && write_fp != NULL) + { + fclose(write_fp); + write_fp = NULL; + } } /* Requested WAL segment not found */ @@ -485,7 +551,88 @@ read_archive_file(XLogDumpPrivate *privateInfo, Size count) } /* - * Create an astreamer that can read WAL from a tar file. + * Set up a temporary directory to temporarily store WAL segments. + */ +static void +setup_tmpwal_dir(const char *waldir) +{ + char *template; + + /* + * Use the directory specified by the TMPDIR environment variable. If it's + * not set, use the provided WAL directory to extract WAL file + * temporarily. + */ + template = psprintf("%s/waldump_tmp-XXXXXX", + getenv("TMPDIR") ? getenv("TMPDIR") : waldir); + TmpWalSegDir = mkdtemp(template); + + if (TmpWalSegDir == NULL) + pg_fatal("could not create directory \"%s\": %m", template); + + canonicalize_path(TmpWalSegDir); + + pg_log_debug("created directory \"%s\"", TmpWalSegDir); +} + +/* + * Remove temporary directory at exit, if any. + */ +static void +cleanup_tmpwal_dir_atexit(void) +{ + rmtree(TmpWalSegDir, true); +} + +/* + * Create an empty placeholder file and return its handle. + */ +static FILE * +prepare_tmp_write(const char *fname) +{ + char fpath[MAXPGPATH]; + FILE *file; + + snprintf(fpath, MAXPGPATH, "%s/%s", TmpWalSegDir, fname); + + /* Create an empty placeholder */ + file = fopen(fpath, PG_BINARY_W); + if (file == NULL) + pg_fatal("could not create file \"%s\": %m", fpath); + +#ifndef WIN32 + if (chmod(fpath, pg_file_create_mode)) + pg_fatal("could not set permissions on file \"%s\": %m", + fpath); +#endif + + pg_log_debug("spilling to temporary file \"%s\"", fpath); + + return file; +} + +/* + * Write buffer data to the given file handle. + */ +static void +perform_tmp_write(const char *fname, StringInfo buf, FILE *file) +{ + Assert(file); + + errno = 0; + if (buf->len > 0 && fwrite(buf->data, buf->len, 1, file) != 1) + { + /* + * If write didn't set errno, assume problem is no disk space + */ + if (errno == 0) + errno = ENOSPC; + pg_fatal("could not write to file \"%s\": %m", fname); + } +} + +/* + * Create an astreamer that can read WAL from tar file. */ static astreamer * astreamer_waldump_new(XLogDumpPrivate *privateInfo) diff --git a/src/bin/pg_waldump/pg_waldump.c b/src/bin/pg_waldump/pg_waldump.c index 6d04462d039..faf300af2be 100644 --- a/src/bin/pg_waldump/pg_waldump.c +++ b/src/bin/pg_waldump/pg_waldump.c @@ -478,25 +478,46 @@ TarWALDumpReadPage(XLogReaderState *state, XLogRecPtr targetPagePtr, int reqLen, return -1; /* - * If the target page is in a different segment, free the buffer space - * occupied by the previous segment data. Since pg_waldump never requests - * the same WAL bytes twice, moving to a new segment implies the previous - * buffer's data and that segment will not be needed again. + * If the target page is in a different segment, free the buffer and/or + * temporary file disk space occupied by the previous segment's data. + * Since pg_waldump never requests the same WAL bytes twice, moving to a + * new segment implies the previous buffer's data and that segment will + * not be needed again. + * + * Afterward, check for the next required WAL segment's physical existence + * in the temporary directory first before invoking the archive streamer. */ nextSegNo = state->seg.ws_segno; if (!XLByteInSeg(targetPagePtr, nextSegNo, WalSegSz)) { char fname[MAXFNAMELEN]; + if (state->seg.ws_file >= 0) + { + close(state->seg.ws_file); + state->seg.ws_file = -1; + } + XLogFileName(fname, state->seg.ws_tli, nextSegNo, WalSegSz); free_archive_wal_entry(fname, private); XLByteToSeg(targetPagePtr, nextSegNo, WalSegSz); state->seg.ws_tli = private->timeline; state->seg.ws_segno = nextSegNo; + + /* + * If the next segment exists, open it and continue reading from there + */ + XLogFileName(fname, state->seg.ws_tli, nextSegNo, WalSegSz); + state->seg.ws_file = open_file_in_directory(TmpWalSegDir, fname); } - /* Read the WAL page from the archive streamer */ + /* Continue reading from the open WAL segment, if any */ + if (state->seg.ws_file >= 0) + return WALDumpReadPage(state, targetPagePtr, count, targetPtr, + readBuff); + + /* Otherwise, read the WAL page from the archive streamer */ return read_archive_wal_page(private, targetPagePtr, count, readBuff, WalSegSz); } diff --git a/src/bin/pg_waldump/pg_waldump.h b/src/bin/pg_waldump/pg_waldump.h index 02da2c43b08..476f74e2846 100644 --- a/src/bin/pg_waldump/pg_waldump.h +++ b/src/bin/pg_waldump/pg_waldump.h @@ -18,6 +18,9 @@ struct ArchivedWALFile; struct ArchivedWAL_hash; +/* Temporary directory */ +extern char *TmpWalSegDir; + /* Contains the necessary information to drive WAL decoding */ typedef struct XLogDumpPrivate { diff --git a/src/bin/pg_waldump/t/001_basic.pl b/src/bin/pg_waldump/t/001_basic.pl index cae543c8990..55a21c71208 100644 --- a/src/bin/pg_waldump/t/001_basic.pl +++ b/src/bin/pg_waldump/t/001_basic.pl @@ -7,6 +7,7 @@ use Cwd; use PostgreSQL::Test::Cluster; use PostgreSQL::Test::Utils; use Test::More; +use List::Util qw(shuffle); my $tar = $ENV{TAR}; @@ -272,7 +273,7 @@ sub generate_archive } closedir $dh; - @files = sort @files; + @files = shuffle @files; # move into the WAL directory before archiving files my $cwd = getcwd; -- 2.47.1
From 8cb70851571e268a6be7763c845d79b9a8b50cc0 Mon Sep 17 00:00:00 2001 From: Amul Sul <[email protected]> Date: Wed, 16 Jul 2025 14:47:43 +0530 Subject: [PATCH v12 7/9] pg_verifybackup: Delay default WAL directory preparation. We are not sure whether to parse WAL from a directory or an archive until the backup format is known. Therefore, we delay preparing the default WAL directory until the point of parsing. This delay is harmless, as the WAL directory is not used elsewhere. --- src/bin/pg_verifybackup/pg_verifybackup.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) diff --git a/src/bin/pg_verifybackup/pg_verifybackup.c b/src/bin/pg_verifybackup/pg_verifybackup.c index f9f2d457f2f..ab01c4d003a 100644 --- a/src/bin/pg_verifybackup/pg_verifybackup.c +++ b/src/bin/pg_verifybackup/pg_verifybackup.c @@ -285,10 +285,6 @@ main(int argc, char **argv) manifest_path = psprintf("%s/backup_manifest", context.backup_directory); - /* By default, look for the WAL in the backup directory, too. */ - if (wal_directory == NULL) - wal_directory = psprintf("%s/pg_wal", context.backup_directory); - /* * Try to read the manifest. We treat any errors encountered while parsing * the manifest as fatal; there doesn't seem to be much point in trying to @@ -368,6 +364,10 @@ main(int argc, char **argv) if (context.format == 'p' && !context.skip_checksums) verify_backup_checksums(&context); + /* By default, look for the WAL in the backup directory, too. */ + if (wal_directory == NULL) + wal_directory = psprintf("%s/pg_wal", context.backup_directory); + /* * Try to parse the required ranges of WAL records, unless we were told * not to do so. -- 2.47.1
From c2b0309deeb401d1cfd0ed6140d0a1d37bdd7e27 Mon Sep 17 00:00:00 2001 From: Amul Sul <[email protected]> Date: Tue, 25 Nov 2025 17:32:14 +0530 Subject: [PATCH v12 8/9] pg_verifybackup: Rename the wal-directory switch to wal-path With previous patches to pg_waldump can now decode WAL directly from tar files. This means you'll be able to specify a tar archive path instead of a traditional WAL directory. To keep things consistent and more versatile, we should also generalize the input switch for pg_verifybackup. It should accept either a directory or a tar file path that contains WALs. This change will also aligning it with the existing manifest-path switch naming. == NOTE == The corresponding PO files require updating due to this change. --- doc/src/sgml/ref/pg_verifybackup.sgml | 2 +- src/bin/pg_verifybackup/pg_verifybackup.c | 22 +++++++++++----------- src/bin/pg_verifybackup/t/007_wal.pl | 4 ++-- 3 files changed, 14 insertions(+), 14 deletions(-) diff --git a/doc/src/sgml/ref/pg_verifybackup.sgml b/doc/src/sgml/ref/pg_verifybackup.sgml index 61c12975e4a..e9b8bfd51b1 100644 --- a/doc/src/sgml/ref/pg_verifybackup.sgml +++ b/doc/src/sgml/ref/pg_verifybackup.sgml @@ -261,7 +261,7 @@ PostgreSQL documentation <varlistentry> <term><option>-w <replaceable class="parameter">path</replaceable></option></term> - <term><option>--wal-directory=<replaceable class="parameter">path</replaceable></option></term> + <term><option>--wal-path=<replaceable class="parameter">path</replaceable></option></term> <listitem> <para> Try to parse WAL files stored in the specified directory, rather than diff --git a/src/bin/pg_verifybackup/pg_verifybackup.c b/src/bin/pg_verifybackup/pg_verifybackup.c index ab01c4d003a..3103d36f1b9 100644 --- a/src/bin/pg_verifybackup/pg_verifybackup.c +++ b/src/bin/pg_verifybackup/pg_verifybackup.c @@ -93,7 +93,7 @@ static void verify_file_checksum(verifier_context *context, uint8 *buffer); static void parse_required_wal(verifier_context *context, char *pg_waldump_path, - char *wal_directory); + char *wal_path); static astreamer *create_archive_verifier(verifier_context *context, char *archive_name, Oid tblspc_oid, @@ -126,7 +126,7 @@ main(int argc, char **argv) {"progress", no_argument, NULL, 'P'}, {"quiet", no_argument, NULL, 'q'}, {"skip-checksums", no_argument, NULL, 's'}, - {"wal-directory", required_argument, NULL, 'w'}, + {"wal-path", required_argument, NULL, 'w'}, {NULL, 0, NULL, 0} }; @@ -135,7 +135,7 @@ main(int argc, char **argv) char *manifest_path = NULL; bool no_parse_wal = false; bool quiet = false; - char *wal_directory = NULL; + char *wal_path = NULL; char *pg_waldump_path = NULL; DIR *dir; @@ -221,8 +221,8 @@ main(int argc, char **argv) context.skip_checksums = true; break; case 'w': - wal_directory = pstrdup(optarg); - canonicalize_path(wal_directory); + wal_path = pstrdup(optarg); + canonicalize_path(wal_path); break; default: /* getopt_long already emitted a complaint */ @@ -365,15 +365,15 @@ main(int argc, char **argv) verify_backup_checksums(&context); /* By default, look for the WAL in the backup directory, too. */ - if (wal_directory == NULL) - wal_directory = psprintf("%s/pg_wal", context.backup_directory); + if (wal_path == NULL) + wal_path = psprintf("%s/pg_wal", context.backup_directory); /* * Try to parse the required ranges of WAL records, unless we were told * not to do so. */ if (!no_parse_wal) - parse_required_wal(&context, pg_waldump_path, wal_directory); + parse_required_wal(&context, pg_waldump_path, wal_path); /* * If everything looks OK, tell the user this, unless we were asked to @@ -1198,7 +1198,7 @@ verify_file_checksum(verifier_context *context, manifest_file *m, */ static void parse_required_wal(verifier_context *context, char *pg_waldump_path, - char *wal_directory) + char *wal_path) { manifest_data *manifest = context->manifest; manifest_wal_range *this_wal_range = manifest->first_wal_range; @@ -1208,7 +1208,7 @@ parse_required_wal(verifier_context *context, char *pg_waldump_path, char *pg_waldump_cmd; pg_waldump_cmd = psprintf("\"%s\" --quiet --path=\"%s\" --timeline=%u --start=%X/%08X --end=%X/%08X\n", - pg_waldump_path, wal_directory, this_wal_range->tli, + pg_waldump_path, wal_path, this_wal_range->tli, LSN_FORMAT_ARGS(this_wal_range->start_lsn), LSN_FORMAT_ARGS(this_wal_range->end_lsn)); fflush(NULL); @@ -1376,7 +1376,7 @@ usage(void) printf(_(" -P, --progress show progress information\n")); printf(_(" -q, --quiet do not print any output, except for errors\n")); printf(_(" -s, --skip-checksums skip checksum verification\n")); - printf(_(" -w, --wal-directory=PATH use specified path for WAL files\n")); + printf(_(" -w, --wal-path=PATH use specified path for WAL files\n")); printf(_(" -V, --version output version information, then exit\n")); printf(_(" -?, --help show this help, then exit\n")); printf(_("\nReport bugs to <%s>.\n"), PACKAGE_BUGREPORT); diff --git a/src/bin/pg_verifybackup/t/007_wal.pl b/src/bin/pg_verifybackup/t/007_wal.pl index 79087a1f6be..8ad2234453d 100644 --- a/src/bin/pg_verifybackup/t/007_wal.pl +++ b/src/bin/pg_verifybackup/t/007_wal.pl @@ -42,10 +42,10 @@ command_ok([ 'pg_verifybackup', '--no-parse-wal', $backup_path ], command_ok( [ 'pg_verifybackup', - '--wal-directory' => $relocated_pg_wal, + '--wal-path' => $relocated_pg_wal, $backup_path ], - '--wal-directory can be used to specify WAL directory'); + '--wal-path can be used to specify WAL directory'); # Move directory back to original location. rename($relocated_pg_wal, $original_pg_wal) || die "rename pg_wal back: $!"; -- 2.47.1
From 8abeadee490f1d7d751d71c3a1490986cbe26f36 Mon Sep 17 00:00:00 2001 From: Amul Sul <[email protected]> Date: Tue, 25 Nov 2025 17:34:26 +0530 Subject: [PATCH v12 9/9] pg_verifybackup: enabled WAL parsing for tar-format backup Now that pg_waldump supports decoding from tar archives, we should leverage this functionality to remove the previous restriction on WAL parsing for tar-backed formats. --- doc/src/sgml/ref/pg_verifybackup.sgml | 5 +- src/bin/pg_verifybackup/pg_verifybackup.c | 66 +++++++++++++------ src/bin/pg_verifybackup/t/002_algorithm.pl | 4 -- src/bin/pg_verifybackup/t/003_corruption.pl | 4 +- src/bin/pg_verifybackup/t/008_untar.pl | 5 +- src/bin/pg_verifybackup/t/010_client_untar.pl | 5 +- 6 files changed, 50 insertions(+), 39 deletions(-) diff --git a/doc/src/sgml/ref/pg_verifybackup.sgml b/doc/src/sgml/ref/pg_verifybackup.sgml index e9b8bfd51b1..16b50b5a4df 100644 --- a/doc/src/sgml/ref/pg_verifybackup.sgml +++ b/doc/src/sgml/ref/pg_verifybackup.sgml @@ -36,10 +36,7 @@ PostgreSQL documentation <literal>backup_manifest</literal> generated by the server at the time of the backup. The backup may be stored either in the "plain" or the "tar" format; this includes tar-format backups compressed with any algorithm - supported by <application>pg_basebackup</application>. However, at present, - <literal>WAL</literal> verification is supported only for plain-format - backups. Therefore, if the backup is stored in tar-format, the - <literal>-n, --no-parse-wal</literal> option should be used. + supported by <application>pg_basebackup</application>. </para> <para> diff --git a/src/bin/pg_verifybackup/pg_verifybackup.c b/src/bin/pg_verifybackup/pg_verifybackup.c index 3103d36f1b9..cc492728ae8 100644 --- a/src/bin/pg_verifybackup/pg_verifybackup.c +++ b/src/bin/pg_verifybackup/pg_verifybackup.c @@ -74,7 +74,9 @@ pg_noreturn static void report_manifest_error(JsonManifestParseContext *context, const char *fmt,...) pg_attribute_printf(2, 3); -static void verify_tar_backup(verifier_context *context, DIR *dir); +static void verify_tar_backup(verifier_context *context, DIR *dir, + char **base_archive_path, + char **wal_archive_path); static void verify_plain_backup_directory(verifier_context *context, char *relpath, char *fullpath, DIR *dir); @@ -83,7 +85,9 @@ static void verify_plain_backup_file(verifier_context *context, char *relpath, static void verify_control_file(const char *controlpath, uint64 manifest_system_identifier); static void precheck_tar_backup_file(verifier_context *context, char *relpath, - char *fullpath, SimplePtrList *tarfiles); + char *fullpath, SimplePtrList *tarfiles, + char **base_archive_path, + char **wal_archive_path); static void verify_tar_file(verifier_context *context, char *relpath, char *fullpath, astreamer *streamer); static void report_extra_backup_files(verifier_context *context); @@ -136,6 +140,8 @@ main(int argc, char **argv) bool no_parse_wal = false; bool quiet = false; char *wal_path = NULL; + char *base_archive_path = NULL; + char *wal_archive_path = NULL; char *pg_waldump_path = NULL; DIR *dir; @@ -327,17 +333,6 @@ main(int argc, char **argv) pfree(path); } - /* - * XXX: In the future, we should consider enhancing pg_waldump to read WAL - * files from an archive. - */ - if (!no_parse_wal && context.format == 't') - { - pg_log_error("pg_waldump cannot read tar files"); - pg_log_error_hint("You must use -n/--no-parse-wal when verifying a tar-format backup."); - exit(1); - } - /* * Perform the appropriate type of verification appropriate based on the * backup format. This will close 'dir'. @@ -346,7 +341,7 @@ main(int argc, char **argv) verify_plain_backup_directory(&context, NULL, context.backup_directory, dir); else - verify_tar_backup(&context, dir); + verify_tar_backup(&context, dir, &base_archive_path, &wal_archive_path); /* * The "matched" flag should now be set on every entry in the hash table. @@ -364,9 +359,28 @@ main(int argc, char **argv) if (context.format == 'p' && !context.skip_checksums) verify_backup_checksums(&context); - /* By default, look for the WAL in the backup directory, too. */ + /* + * By default, WAL files are expected to be found in the backup directory + * for plain-format backups. In the case of tar-format backups, if a + * separate WAL archive is not found, the WAL files are most likely + * included within the main data directory archive. + */ if (wal_path == NULL) - wal_path = psprintf("%s/pg_wal", context.backup_directory); + { + if (context.format == 'p') + wal_path = psprintf("%s/pg_wal", context.backup_directory); + else if (wal_archive_path) + wal_path = wal_archive_path; + else if (base_archive_path) + wal_path = base_archive_path; + else + { + pg_log_error("WAL archive not found"); + pg_log_error_hint("Specify the correct path using the option -w/--wal-path. " + "Or you must use -n/--no-parse-wal when verifying a tar-format backup."); + exit(1); + } + } /* * Try to parse the required ranges of WAL records, unless we were told @@ -787,7 +801,8 @@ verify_control_file(const char *controlpath, uint64 manifest_system_identifier) * close when we're done with it. */ static void -verify_tar_backup(verifier_context *context, DIR *dir) +verify_tar_backup(verifier_context *context, DIR *dir, char **base_archive_path, + char **wal_archive_path) { struct dirent *dirent; SimplePtrList tarfiles = {NULL, NULL}; @@ -816,7 +831,8 @@ verify_tar_backup(verifier_context *context, DIR *dir) char *fullpath; fullpath = psprintf("%s/%s", context->backup_directory, filename); - precheck_tar_backup_file(context, filename, fullpath, &tarfiles); + precheck_tar_backup_file(context, filename, fullpath, &tarfiles, + base_archive_path, wal_archive_path); pfree(fullpath); } } @@ -875,11 +891,13 @@ verify_tar_backup(verifier_context *context, DIR *dir) * * The arguments to this function are mostly the same as the * verify_plain_backup_file. The additional argument outputs a list of valid - * tar files. + * tar files, along with the full paths to the main archive and the WAL + * directory archive. */ static void precheck_tar_backup_file(verifier_context *context, char *relpath, - char *fullpath, SimplePtrList *tarfiles) + char *fullpath, SimplePtrList *tarfiles, + char **base_archive_path, char **wal_archive_path) { struct stat sb; Oid tblspc_oid = InvalidOid; @@ -918,9 +936,17 @@ precheck_tar_backup_file(verifier_context *context, char *relpath, * extension such as .gz, .lz4, or .zst. */ if (strncmp("base", relpath, 4) == 0) + { suffix = relpath + 4; + + *base_archive_path = pstrdup(fullpath); + } else if (strncmp("pg_wal", relpath, 6) == 0) + { suffix = relpath + 6; + + *wal_archive_path = pstrdup(fullpath); + } else { /* Expected a <tablespaceoid>.tar file here. */ diff --git a/src/bin/pg_verifybackup/t/002_algorithm.pl b/src/bin/pg_verifybackup/t/002_algorithm.pl index 0556191ec9d..edc515d5904 100644 --- a/src/bin/pg_verifybackup/t/002_algorithm.pl +++ b/src/bin/pg_verifybackup/t/002_algorithm.pl @@ -30,10 +30,6 @@ sub test_checksums { # Add switch to get a tar-format backup push @backup, ('--format' => 'tar'); - - # Add switch to skip WAL verification, which is not yet supported for - # tar-format backups - push @verify, ('--no-parse-wal'); } # A backup with a bogus algorithm should fail. diff --git a/src/bin/pg_verifybackup/t/003_corruption.pl b/src/bin/pg_verifybackup/t/003_corruption.pl index b1d65b8aa0f..882d75d9dc2 100644 --- a/src/bin/pg_verifybackup/t/003_corruption.pl +++ b/src/bin/pg_verifybackup/t/003_corruption.pl @@ -193,10 +193,8 @@ for my $scenario (@scenario) command_ok([ $tar, '-cf' => "$tar_backup_path/base.tar", '.' ]); chdir($cwd) || die "chdir: $!"; - # Now check that the backup no longer verifies. We must use -n - # here, because pg_waldump can't yet read WAL from a tarfile. command_fails_like( - [ 'pg_verifybackup', '--no-parse-wal', $tar_backup_path ], + [ 'pg_verifybackup', $tar_backup_path ], $scenario->{'fails_like'}, "corrupt backup fails verification: $name"); diff --git a/src/bin/pg_verifybackup/t/008_untar.pl b/src/bin/pg_verifybackup/t/008_untar.pl index ae67ae85a31..161c08c190d 100644 --- a/src/bin/pg_verifybackup/t/008_untar.pl +++ b/src/bin/pg_verifybackup/t/008_untar.pl @@ -47,7 +47,6 @@ my $tsoid = $primary->safe_psql( SELECT oid FROM pg_tablespace WHERE spcname = 'regress_ts1')); my $backup_path = $primary->backup_dir . '/server-backup'; -my $extract_path = $primary->backup_dir . '/extracted-backup'; my @test_configuration = ( { @@ -123,14 +122,12 @@ for my $tc (@test_configuration) # Verify tar backup. $primary->command_ok( [ - 'pg_verifybackup', '--no-parse-wal', - '--exit-on-error', $backup_path, + 'pg_verifybackup', '--exit-on-error', $backup_path, ], "verify backup, compression $method"); # Cleanup. rmtree($backup_path); - rmtree($extract_path); } } diff --git a/src/bin/pg_verifybackup/t/010_client_untar.pl b/src/bin/pg_verifybackup/t/010_client_untar.pl index 1ac7b5db75a..9670fbe4fda 100644 --- a/src/bin/pg_verifybackup/t/010_client_untar.pl +++ b/src/bin/pg_verifybackup/t/010_client_untar.pl @@ -32,7 +32,6 @@ print $jf $junk_data; close $jf; my $backup_path = $primary->backup_dir . '/client-backup'; -my $extract_path = $primary->backup_dir . '/extracted-backup'; my @test_configuration = ( { @@ -137,13 +136,11 @@ for my $tc (@test_configuration) # Verify tar backup. $primary->command_ok( [ - 'pg_verifybackup', '--no-parse-wal', - '--exit-on-error', $backup_path, + 'pg_verifybackup', '--exit-on-error', $backup_path, ], "verify backup, compression $method"); # Cleanup. - rmtree($extract_path); rmtree($backup_path); } } -- 2.47.1
