Don't reset 'latest_page_number' when replaying multixid truncation

'latest_page_number' is set to the correct value, according to
nextOffset, early at system startup. Contrary to the comment, it hence
should be set up correctly by the time we get to WAL replay.

This fixes a failure to replay WAL generated on older minor versions,
before commit 789d65364c (18.2, 17.8, 16.12, 15.16, 14.21). The
failure occurs after a truncation record has been replayed and looks
like this:

    FATAL:  could not access status of transaction 858112
    DETAIL:  Could not read from file "pg_multixact/offsets/000D" at offset 24576: read too few bytes.
    CONTEXT:  WAL redo at 3/2A3AB408 for MultiXact/CREATE_ID: 858111 offset 6695072 nmembers 5: 1048228 (sh) 1048271 (keysh) 1048316 (sh) 1048344 (keysh) 1048370 (sh)

Reported-by: Sebastian Webber <sebastian@swebber.me>
Reviewed-by: Andrey Borodin <x4mmm@yandex-team.ru>
Reviewed-by: Kirill Reshke <reshkekirill@gmail.com>
Discussion: https://www.postgresql.org/message-id/20260214090150.GC2297@p46.dedyn.io;lightning.p46.dedyn.io
Backpatch-through: 14-18

diff --git a/src/backend/access/transam/multixact.c b/src/backend/access/transam/multixact.c
index c863e4e0556743e1172a67e7b32e00692f54148e..e45ec0d724782c3241449ecd208168d6af822da2 100644 (file)
--- a/src/backend/access/transam/multixact.c
+++ b/src/backend/access/transam/multixact.c
@@ -3571,7 +3571,6 @@ multixact_redo(XLogReaderState *record)
        else if (info == XLOG_MULTIXACT_TRUNCATE_ID)
        {
                xl_multixact_truncate xlrec;
-               int64           pageno;
 
                memcpy(&xlrec, XLogRecGetData(record),
                           SizeOfMultiXactTruncate);
@@ -3596,15 +3595,6 @@ multixact_redo(XLogReaderState *record)
                SetMultiXactIdLimit(xlrec.endTruncOff, xlrec.oldestMultiDB, false);
 
                PerformMembersTruncation(xlrec.startTruncMemb, xlrec.endTruncMemb);
-
-               /*
-                * During XLOG replay, latest_page_number isn't necessarily set up
-                * yet; insert a suitable value to bypass the sanity test in
-                * SimpleLruTruncate.
-                */
-               pageno = MultiXactIdToOffsetPage(xlrec.endTruncOff);
-               pg_atomic_write_u64(&MultiXactOffsetCtl->shared->latest_page_number,
-                                                       pageno);
                PerformOffsetsTruncation(xlrec.startTruncOff, xlrec.endTruncOff);
 
                LWLockRelease(MultiXactTruncationLock);

diff --git a/src/include/access/slru.h b/src/include/access/slru.h
index 9409f68035a052942c6b19cb53eb677cbe7f3432..8edc95b4962d75de1e82e576140f7a31525106ca 100644 (file)
--- a/src/include/access/slru.h
+++ b/src/include/access/slru.h
@@ -110,7 +110,9 @@ typedef struct SlruSharedData
        /*
         * latest_page_number is the page number of the current end of the log;
         * this is not critical data, since we use it only to avoid swapping out
-        * the latest page.
+        * the latest page.  (An exception: an accurate latest_page_number is
+        * needed on pg_multixact/offsets to replay WAL generated with older minor
+        * versions correctly.  See RecordNewMultiXact().)
         */
        pg_atomic_uint64 latest_page_number;