Hello :)
I was developing some tooling to restore the wal-e backup without having to install wal-e, and an issue after finishing the restore.
The restore itself would work fine, the database would start fine, but some tables would be unreadable, for example running a vacuum analyze on the whole database would cause the following error :
2022-05-13 13:33:20.038 UTC [19465] postgres@production STATEMENT: SELECT anonymize_database(false);
2022-05-13 13:33:23.797 UTC [19465] postgres@production LOG: request to flush past end of generated WAL; request 6A4/3A579D98, currpos 6A4/3A0380B0
2022-05-13 13:33:23.797 UTC [19465] postgres@production CONTEXT: writing block 5339 of relation base/16387/546689
2022-05-13 13:33:23.797 UTC [19465] postgres@production STATEMENT: ANALYZE VERBOSE;
2022-05-13 13:33:23.798 UTC [19465] postgres@production ERROR: xlog flush request 6A4/3A579D98 is not satisfied --- flushed only to 6A4/3A0380B0
2022-05-13 13:33:23.798 UTC [19465] postgres@production CONTEXT: writing block 5339 of relation base/16387/546689
2022-05-13 13:33:23.798 UTC [19465] postgres@production STATEMENT: ANALYZE VERBOSE;
I first though my tool had an issue, so I reproducing the same issue with wal-e calls directly, it happens when I give a backup it instead of LATEST or providing a timestamp
This is the call I have used to reproduce the same issue with wal-e :
WABS_ACCESS_KEY="xxxx" WABS_ACCOUNT_NAME=xxxx WALE_WABS_PREFIX=wabs://proddb/dbk8s wal-e -a xxxx backup-fetch -p 8 /var/lib/replika/data/10 base_00000066000006A1000000D6_00000080
However it's working fine when specifying LATEST or a timestamp.
Another detail: This worked perfectly fine with "small" 200GB database, but not on a 900GB database
It's not clear why it's not working when only fetching the full backup ? What should be done more to make it useable without having to fetch the WALs ?
Hello :)
I was developing some tooling to restore the wal-e backup without having to install wal-e, and an issue after finishing the restore.
The restore itself would work fine, the database would start fine, but some tables would be unreadable, for example running a
vacuum analyzeon the whole database would cause the following error :I first though my tool had an issue, so I reproducing the same issue with wal-e calls directly, it happens when I give a backup it instead of
LATESTor providing a timestampThis is the call I have used to reproduce the same issue with wal-e :
However it's working fine when specifying
LATESTor a timestamp.Another detail: This worked perfectly fine with "small" 200GB database, but not on a 900GB database
It's not clear why it's not working when only fetching the full backup ? What should be done more to make it useable without having to fetch the WALs ?