Fix #917 (Lite) — base FinOps memory recommendation on 7-day P95#920
Merged
erikdarlingdata merged 1 commit intodevfrom May 2, 2026
Merged
Conversation
… a snapshot Mirrors the Dashboard fix in 7cc2265 for the Lite collector. Both the Memory right-sizing check (#3) and the VM right-sizing memory prescription (#12) read util.BufferPoolMb — a single-sample reading of perfmon "Database Cache Memory", which is only the data-cache slice of the buffer pool and could trigger right after a service restart or on servers where plan cache / workspace memory dominates. - Both checks now query DuckDB for the 7-day P95 of total_server_memory_mb (perfmon "Total Server Memory" — the full set of memory SQL has committed) from v_memory_stats. - Both require >= 500 samples (~1 day at 1/min) before firing. - Recommendation text now says "P95 SQL memory" rather than "buffer pool" to reflect what is actually being measured. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This was referenced May 2, 2026
pull Bot
pushed a commit
to ehtick/PerformanceMonitor
that referenced
this pull request
May 5, 2026
…ests PRs erikdarlingdata#918 and erikdarlingdata#920 added a sample-count guard (>= 500) to the new 7-day P95 memory recommendation logic. 500 was overly conservative and broke the Lite FinOps test OverProvisionedEnterprise_MemoryRightSizingFires, which seeds a single memory_stats row. The real protection added by those PRs was switching from TOP 1 to P95; the sample minimum is just a sanity check against degenerate single-point inputs. ~16 samples is enough to compute a meaningful P95 and matches the shape of SeedCpuUtilizationAsync's 16-row fixture, so tests can fire the recommendation without artificial inflation. - Lower threshold from 500 to 16 in both Dashboard and Lite (checks #3 and erikdarlingdata#12) so the value reflects the actual ask: "more than one reading" rather than "8+ hours of data." - Update Lite's SeedMemoryStatsAsync to insert 16 rows across the test period (matching SeedCpuUtilizationAsync's pattern). This makes OverProvisionedEnterprise_MemoryRightSizingFires pass again and keeps CleanServer_NoDuckDbRecommendations green (still no rows seeded for the clean scenario → P95 returns NULL → no recommendation). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
util.BufferPoolMb— a single-sample of perfmon "Database Cache Memory", which is only the data-cache slice and triggers misleadingly after restarts or on servers where plan cache / workspace memory dominate.total_server_memory_mbfrom DuckDB'sv_memory_stats(perfmon "Total Server Memory" — the full set of memory SQL has committed), and requires ~1 day of samples before firing.Related to #917
Test plan
v_memory_statshas < 500 samples (cold install)🤖 Generated with Claude Code