storage: delake TestPebbleMapClose#51810
Conversation
|
This change is |
jbowens
left a comment
There was a problem hiding this comment.
I added a second commit that deflakes the test without changes to Pebble by no longer requiring the LSM to be completely empty. Instead it just requires all the tables that existed before the disk map was closed no longer exist. It's a little less sensitive to Pebble compaction heuristic idiosyncrasies, but not as strong an assurance we deleted all previous data. Thoughts?
Reviewable status:
complete! 0 of 0 LGTMs obtained (waiting on @itsbilal and @petermattis)
petermattis
left a comment
There was a problem hiding this comment.
I'm ok with this adjustment to the test passing heuristics, but curious about why the sstable containing only a range tombstone was not compacted out of existence? Was this sstable in L6? If yes, why was the range tombstone not dropped?
Reviewable status:
The TestPebbleMapClose test failed in CI with a sstable remaining after the clearing of the disk map (cockroachdb#51786). A delete-only only compaction followed by a move compaction resulted in a single remaining sstable containing the range tombstone. Adding compactions to elide deletions in L6 in Pebble would allow us to revert to the old test condition. See cockroachdb/pebble#838. Also, update this test to print a hex representation of the raw, prefixed sstable bounds as well for easier debugging. Release note: none
c3c0aab to
0c905e1
Compare
jbowens
left a comment
There was a problem hiding this comment.
Squashed the commits and added the explanation for the flakiness to the commit message.
Reviewable status:
petermattis
left a comment
There was a problem hiding this comment.
Reviewable status:
|
TFTR! bors r+ |
|
Build succeeded: |
The TestPebbleMapClose test failed in CI with a sstable remaining after
the clearing of the disk map (#51786). A delete-only only compaction
followed by a move compaction resulted in a single remaining sstable
containing the range tombstone.
Adding compactions to elide deletions in L6 in Pebble would allow us to
revert to the old test condition. See cockroachdb/pebble#838.
Also, update this test to print a hex representation of
the raw, prefixed sstable bounds as well for easier debugging.
Fixes #51786.
Release note: none