You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
147933: importer: remove tests for PGDUMP and MYSQLDUMP r=yuzefovich a=yuzefovich
This commit only touches the tests for deprecated IMPORT formats. It removes the tests that are specific to these formats (which includes things like creating stub table statistics for newly-created tables, which other formats can't do).
It also skips `TestImportMultiRegion` under race (similar to all other multi-node IMPORT tests) since I've seen some timeouts on it in CI.
Epic: None
Release note: None
148234: row: remove InitChecksum for each KV inserted by IMPORT r=yuzefovich a=yuzefovich
I noticed `InitChecksum` call show up in the CPU profile of a node from 300 node cluster when IMPORT is running, and I was confused by it. I did some archaeology, and I think it hasn't been needed for a long time.
For context, prior to 2.0, we effectively required setting the checksum on the Value since evaluation of CPuts compared RawBytes directly, which included the checksum comparison. This was an oversight because the checksum should be optional, and it became such in 33957ae.
However, in order to support IMPORT in mixed-version 1.1-2.0 cluster, in 5ae1bb2 we added the population of the checksum for all KVs produced for insertion. There is a discussion on the PR that this is no longer needed as of the optional checksum change mentioned above.
There is also a very recent discussion about removing this checksum in non-test code altogether since it probably doesn't give us much since we do checksum verification at the storage / network level. See #145541.
It should be safe to not populate the checksum for IMPORT as it seems to be the only place we do so in the SQL land.
Epic: None
Release note: None
Co-authored-by: Yahor Yuzefovich <[email protected]>
0 commit comments