sqlstats: fix reset stats container on flush

[xinhaoz]

cockroach/pkg/sql/sqlstats/ssmemstorage/ss_mem_storage.go

Lines 607 to 613 in a44e43c

	s.mu.stmts = make(map[stmtKey]*stmtStats, len(s.mu.stmts)/2)
	s.mu.txns = make(map[appstatspb.TransactionFingerprintID]*txnStats, len(s.mu.txns)/2)
	s.mu.sampledPlanMetadataCache = make(map[sampledPlanKey]time.Time, len(s.mu.sampledPlanMetadataCache)/2)
	s.freeLocked(ctx)
	if s.knobs != nil && s.knobs.OnAfterClear != nil {
	s.knobs.OnAfterClear()
	}

In the above, freeLocked is called after clearing the containers. Within freeLocked, the size of each container cleared is used to decrement uniqueServerCount which is used to track the total number of in-memory fingerprints tracked by sql stats. Due to these steps we never actually decrement the counter after each flush, causing the uniqueServerCount to grow erroneously. This can cause us to incorrectly spam sending the memory pressure signal to the sql stats flush worker, causing rapid flushing of empty containers.

Jira issue: CRDB-37092

[blathers-crl]

Hi @xinhaoz, please add a C-ategory label to your issue. Check out the label system docs.

While you're here, please consider adding an A- label to help keep our repository tidy.

_{🦉 Hoot! I am a Blathers, a bot for CockroachDB. My owner is dev-inf.}