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
155242: sql/stats: avoid mutating input buckets in stripOuterBuckets r=Uzair5162 a=Uzair5162
This commit changes `stripOuterBuckets` to modify and return a copy of the given histogram buckets if it finds outer buckets to remove. Previously, we would mutate the caller's histograms with leading outer buckets in-place by zeroing the range counts on the first non-outer bucket. This effectively corrupts the first histogram bucket in stats passed in from the stats cache.
Although this bug has existed since 90e311d (which zeroes the first buckets range counts), it would only impact full statistics that we tried merging with partial stats, as that was the only case in which `stripOuterBuckets` was called. The surface area of this bug increased after db9a344, which calls `stripOuterBuckets` on every full statistic, regardless of whether we end up merging it with partial stats or not.
Fixes: #155184
Release note (bug fix): Previously, we could corrupt the first bucket of table statistic histograms in certain cases, causing underestimates for range counts near the lower end of the domain, which is now fixed.
155390: dev: refine behavior of `--cpus` further r=annaw2193 a=annaw2193
- `--local_cpu_resources` changed to `--local_resources=cpu=`
- `--cpus` now additionaly implies `--local_test_jobs`, which should be set to an equivalent value.
Also, using `--cpus` gives:
Epic: None
Fixes: #151139
Release Note: None
155402: sql: harden recent change about DO block recursion r=yuzefovich a=yuzefovich
This commit reduces the max depth of recursion when evaluating routines with `tail-call-optimization-enabled=false` (recently added in 7b879ef) from 10k to 100. We've just seen a few cases where TestRandomSyntaxSQLSmith failed because DO block didn't respect context cancellation within 5s. I've manually tried it out a few times, and things worked, so my hypothesis is that extremely deep stacks (that are produced with the TCO disabled) is the root cause for slow cancellation, so let's just error out sooner.
This commit also brings back the skip of DO blocks in TestComposeCompare (thinking there is that we should stabilize them in other tests first).
Fixes: #155208.
Fixes: #155210.
Release note: None
155406: sql/hints: remove noisy logging r=DrewKimball a=DrewKimball
This commit removes a noisy source of log messages by only logging for incremental updates with a non-empty list of events. This prevents spamming log messages every time the resolved timestamp for the rangefeed is incremented.
Epic: None
Release note: None
Co-authored-by: Uzair Ahmad <uzair.ahmad@cockroachlabs.com>
Co-authored-by: Anna Wang <anna.wang@cockroachlabs.com>
Co-authored-by: Yahor Yuzefovich <yahor@cockroachlabs.com>
Co-authored-by: Drew Kimball <drewk@cockroachlabs.com>
0 commit comments