execbuilder: re-enable skipped tests#75972
Conversation
This commit de-flakes tests that were previously disabled in cockroachdb#75962 by setting column families explicitly. I originally thought these execbuilder tests were flaky because there were two query plans with the same cost, but the flakiness was actually the result of randomized families in logic tests. Fixes cockroachdb#75960 Release note: None
cucaroach
left a comment
There was a problem hiding this comment.
Reviewed 1 of 1 files at r1, all commit messages.
Reviewable status:complete! 0 of 0 LGTMs obtained (waiting on @RaduBerinde)
cucaroach
left a comment
There was a problem hiding this comment.
Reviewable status:
complete! 1 of 0 LGTMs obtained (waiting on @RaduBerinde)
rharding6373
left a comment
There was a problem hiding this comment.
Reviewable status:
complete! 1 of 0 LGTMs obtained (waiting on @mgartner and @RaduBerinde)
pkg/sql/opt/exec/execbuilder/testdata/hash_sharded_index, line 203 at r1 (raw file):
a INT8 PRIMARY KEY USING HASH WITH BUCKET_COUNT = 8, b INT8 NOT NULL, FAMILY (a, b)
Drive-by question: So without specifying the column family, we will store all columns as a single family but randomly decide whether to store them in (a,b) order or (b,a) order?
|
pkg/sql/opt/exec/execbuilder/testdata/hash_sharded_index, line 203 at r1 (raw file): Previously, rharding6373 (Rachael Harding) wrote…
The logic tests use a column family mutator here: cockroach/pkg/sql/logictest/logic.go Line 2791 in 73b6819 This mutates cockroach/pkg/sql/randgen/mutator.go Lines 821 to 870 in 8b5f53c This is only happens for logic tests. In production, column families are always deterministic. I believe the default is to include all columns in a single family. |
|
bors r+ |
|
Build failed (retrying...): |
|
Build failed: |
|
Looks like a flake. bors r+ |
|
Build succeeded: |
This commit de-flakes tests that were previously disabled in #75962 by
setting column families explicitly. I originally thought these
execbuilder tests were flaky because there were two query plans with the
same cost, but the flakiness was actually the result of randomized
families in logic tests.
Fixes #75960
Release note: None