release-24.3: ccl/multiregionccl: deflake TestMultiRegionDataDriven_global_tables#166248
release-24.3: ccl/multiregionccl: deflake TestMultiRegionDataDriven_global_tables#166248blathers-crl[bot] wants to merge 1 commit intorelease-24.3from
Conversation
The test was flaky because the cluster setting `kv.closed_timestamp.lead_for_global_reads_auto_tune.enabled` uses a metamorphic constant that can randomly enable auto-tuning during tests. When enabled on a localhost test cluster (near-zero inter-node latency), auto-tuning selects a lower-latency policy bucket, reducing the closed timestamp lead time. This reduced lead time provides insufficient buffer for follower reads to be served consistently, causing the closed timestamp to be too low (~27ms) when the test asserts that a GLOBAL table read is served via follower read. Explicitly disable auto-tuning in the test's cluster setup to ensure a stable closed timestamp lead time. Resolves: #165405 Resolves: #165266 Release note: None Generated by Claude Code Auto-Solver Co-Authored-By: Claude <noreply@anthropic.com>
|
Merging to
|
|
Thanks for opening a backport. Before merging, please confirm that it falls into one of the following categories (select one):
Add a brief release justification to the PR description explaining your selection. Also, confirm that the change does not break backward compatibility and complies with all aspects of the backport policy. All backports must be reviewed by the TL and EM for the owning area. |
|
✅ PR #166248 is compliant with backport policy Confidence: high ❌ ENGREQ Check Failed: Please check one of the backport category checkboxes. 🦉 Hoot! I am a Blathers, a bot for CockroachDB. My owner is dev-inf. |
Backport 1/1 commits from #166247 on behalf of @cockroach-teamcity.
The test was flaky because the cluster setting
kv.closed_timestamp.lead_for_global_reads_auto_tune.enableduses ametamorphic constant that can randomly enable auto-tuning during tests.
When enabled on a localhost test cluster (near-zero inter-node latency),
auto-tuning selects a lower-latency policy bucket, reducing the closed
timestamp lead time. This reduced lead time provides insufficient buffer
for follower reads to be served consistently, causing the closed timestamp
to be too low (~27ms) when the test asserts that a GLOBAL table read is
served via follower read.
Explicitly disable auto-tuning in the test's cluster setup to ensure a
stable closed timestamp lead time.
Resolves: #165405
Resolves: #165266
Release note: None
Generated by Claude Code Auto-Solver
Co-Authored-By: Claude noreply@anthropic.com
This PR was auto-generated by issue-autosolve using Claude Code.
Please review carefully before approving.
Release justification: