execbuilder: de-flake a logic test#113159
Conversation
|
This change is |
We recently made a change to use the InOrder mode of the streamer in more cases. In particular, when we have at least 3 column families and we might end up with "family-specific" spans in the lookup and index joins, we'll now use the InOrder mode. That mode needs disk-backed results buffer which can show up as `estimated max sql temp disk usage` line in EXPLAIN ANALYZE output. This made one opt logic test a bit flaky due to randomization of column families, so this commit adjusts a single table with 3 columns to have a single column family to guarantee deterministic output. Release note: None
55aabd5 to
ed9b04f
Compare
DrewKimball
left a comment
There was a problem hiding this comment.
Reviewed 1 of 1 files at r1, all commit messages.
Reviewable status:complete! 1 of 0 LGTMs obtained (waiting on @michae2 and @yuzefovich)
-- commits line 8 at r1:
Should we just hard-code estimated max sql temp disk usage as well? I guess this doesn't lose too much coverage, so it's probably alright.
yuzefovich
left a comment
There was a problem hiding this comment.
TFTR!
bors r+
Reviewable status:
Previously, DrewKimball (Drew Kimball) wrote…
Should we just hard-code
estimated max sql temp disk usageas well? I guess this doesn't lose too much coverage, so it's probably alright.
The question is whether this line shows up or not at all. With OutOfOrder mode it doesn't, with InOrder mode it might (when non-zero disk space is used). We already deflake the value to be 0B in case it does show, but that's insufficient. So I do think we just want to make sure that fixed mode is used.
|
Build succeeded: |
We recently made a change to use the InOrder mode of the streamer in more cases. In particular, when we have at least 3 column families and we might end up with "family-specific" spans in the lookup and index joins, we'll now use the InOrder mode. That mode needs disk-backed results buffer which can show up as
estimated max sql temp disk usageline in EXPLAIN ANALYZE output. This made one opt logic test a bit flaky due to randomization of column families, so this commit adjusts a single table with 3 columns to have a single column family to guarantee deterministic output.Fixes: #113112.
Release note: None