Re: walsummarizer can get stuck when switching timelines

Hello Robert and Andrey,

10.08.2026 21:20, Robert Haas wrote:
> Thanks to both of you for the analysis, and apologies for the slow
> response. Here's a patch. I think the tricky part is justifying the
> change, which I at least find counterintuitive (else I would have done
> it right to begin with!). The attached patch attempts to explain via a
> lengthy comment.

Thank you for paying attention to that test defect!

Meanwhile, other buildfarm animal, morepork, was lucky enough to reveal
one more issue, thanks to its timings: [1], [2].
[01:07:53.816](18.983s) ok 1 - WAL summarization on node3 advanced past timeline switch
[01:07:53.830](0.014s) not ok 2 - no summaries from before LSN 0/0301DC20
[01:07:53.831](0.000s)
[01:07:53.831](0.000s) #   Failed test 'no summaries from before LSN 0/0301DC20'
#   at t/003_tli_switch.pl line 121.
[01:07:53.831](0.000s) #          got: '0/0301DBE8|0/0301DC20'
#     expected: ''
[01:07:53.843](0.012s) not ok 3 - at least one summary from LSN 0/0301DC20 or later
[01:07:53.843](0.000s)
[01:07:53.843](0.000s) #   Failed test 'at least one summary from LSN 0/0301DC20 or later'
#   at t/003_tli_switch.pl line 130.

003_tli_switch_node1.log
2026-08-15 01:07:37.126 CEST [68231:1] LOG:  starting PostgreSQL 19beta3 on x86_64-unknown-openbsd6.9, compiled by
OpenBSD clang version 10.0.1 , 64-bit
...
2026-08-15 01:07:51.657 CEST [9984:4] 003_tli_switch.pl LOG: statement: SELECT pg_switch_wal();
2026-08-15 01:07:51.658 CEST [9984:5] 003_tli_switch.pl LOG: statement: CREATE TABLE dummy ();
..
2026-08-15 01:07:52.200 CEST [4172:4] 003_tli_switch.pl LOG: statement: SELECT pg_current_wal_flush_lsn()
...
2026-08-15 01:07:52.226 CEST [72425:4] 003_tli_switch.pl LOG: statement: SELECT pg_current_wal_insert_lsn()

That is, ~15 seconds passed from the server start to the pg_current_wal_insert_lsn()
call. So I guess, the test failed due to bgwriter's LogStandbySnapshot. I
could reproduce such a failure locally with:
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -70,3 +70,3 @@ int            BgWriterDelay = 200;
  */
-#define LOG_SNAPSHOT_INTERVAL_MS 15000
+#define LOG_SNAPSHOT_INTERVAL_MS 1

@@ -308,3 +308,3 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
                        WL_LATCH_SET | WL_TIMEOUT | WL_EXIT_ON_PM_DEATH,
-                       BgWriterDelay /* ms */ , WAIT_EVENT_BGWRITER_MAIN);
+                       1 /* ms */ , WAIT_EVENT_BGWRITER_MAIN);

make -s check -C src/bin/pg_walsummary

# +++ tap check in src/bin/pg_walsummary +++
t/001_basic.pl ....... ok
t/002_blocks.pl ...... ok
t/003_tli_switch.pl .. 1/?
#   Failed test 'no summaries from before LSN 0/0401DD28'
#   at t/003_tli_switch.pl line 121.
#          got: '0/0401DCF0|0/0401DD28'
#     expected: ''

#   Failed test 'at least one summary from LSN 0/0401DD28 or later'
#   at t/003_tli_switch.pl line 130.
# Looks like you failed 2 tests of 3.
t/003_tli_switch.pl .. Dubious, test returned 2 (wstat 512, 0x200)
Failed 2/3 subtests

Test Summary Report
-------------------
t/003_tli_switch.pl (Wstat: 512 (exited 2) Tests: 3 Failed: 2)
  Failed tests:  2-3
  Non-zero exit status: 2
Files=3, Tests=22,  3 wallclock secs ( 0.01 usr  0.00 sys +  0.21 cusr  0.38 csys =  0.60 CPU)
Result: FAIL

(Basic check-world passes with this change for me.)

Could you please look at this anomaly too?

[1] https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=morepork&dt=2026-08-12%2018%3A51%3A34
[2] https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=morepork&dt=2026-08-14%2022%3A32%3A27

Best regards,
Alexander