Table swap fails due to exclusive lock

[Alytv13]

Hi!
When running pg_repack on a table ~1TB with --order-by, the process reaches the APPLY stage. When the row count drops below --switch-threshold, pg_repack should acquire an exclusive lock for the final table swap, but fails to do so. The row count then increases again, causing an infinite loop.
The problem occurred with default --switch-threshold settings. We tried increasing --switch-threshold to 4900 and --apply-count to 5000, but the issue persists.
Manually acquiring an exclusive lock on the table for 0.5 seconds allows pg_repack to complete successfully.
pg_repack version:
pg_repack | 1.5.2 | public | Reorganize tables in PostgreSQL databases with minimal locks

[za-arthur]

Hey @Alytv13 ,
This is known edge case, there is a risk that repacking a big table which has high write rate never finishes. You might try to different values for --switch-threshold and --apply-count. Or you can try to reduce write rate or number of workers writing to your table.
Otherwise do you have any other suggestions to handle this?

[Alytv13]

@za-arthur thanks!
We tried various values for --switch-threshold and --apply-count and run the repack under minimal possible load on the table. Currently, the only working method is to manually acquire an exclusive lock on the table.