Failed to open file "xxxx/00000001000000870000001D" for wal backup

[Lzjing-1997]

Steps to reproduce

(1) Increase the WAL_SEGMENT_SIZE to 64MB; // Larger WAL segment is more likely to trigger this problem
(2) Enable safekeeper WAL backup;
(3) psql
(4) => select pg_switch_wal(); // It may need to be executed multiple times

Expected result

safekeeper wal backup successfully

Actual result

The following error was found in the safekeeper log:

2025-03-05T03:30:27.232566Z ERROR wal_backup{ttid=fee6933a27ec11f88ff516873b696d68/bea52794b362ed3f62d0c794a011765f}: failed while offloading range 0/12000000-0/13000000, backup_lsn 0/12000000: Failed to open file "/Users/jinfeng/Desktop/workplace/neondatabase/neon_2/neon/.neon/safekeepers/sk1/fee6933a27ec11f88ff516873b696d68/bea52794b362ed3f62d0c794a011765f/000000010000000000000012" for wal backup

Caused by:
No such file or directory (os error 2)

(Eventually, the backup will be successful)

Environment

macbook;
Latest Neon code.

Logs, links

Analysis

(1) Execute "select pg_switch_wal();"，will produce "SWITCH" WAL Record;
(2) When safekeeper received the "SWITCH" WAL Record，the 'write_record_lsn' will be set to the start lsn of the next WAL segment; but safekeeper will continue to receive '\0' data until the current segment is filled;
(3) 'flush_ticker’ periodically triggers the flush operation，which updates the 'flush_record_lsn'/'commit_lsn'，and notify the 'WAL backup' process;
(4) If the current segment is not filled, 'WAL backup' process will not find the WAL segment.

[arssher]

Yeah, I think not making flush_record_lsn update when self.write_lsn < self.write_record_lsn in flush_wal would be a fix.

[Lzjing-1997]

Yeah, I think not making flush_record_lsn update when self.write_lsn < self.write_record_lsn in flush_wal would be a fix.

Yes, I think it's a solution