insights: enable the anomaly detector#86673
insights: enable the anomaly detector#86673craig[bot] merged 1 commit intocockroachdb:masterfrom matthewtodd:insights-enable-anomaly-detector
Conversation
|
This change is |
This change turns on the insights "anomaly detector" by default, catching statement executions in the >p99 latency for their fingerprint that are also far enough away from median latency and above a (default 50ms) threshold. Release justification: Category 2: Bug fixes and low-risk updates to new functionality. Release note (ops change): The `sql.insights.anomaly_detection.enabled` cluster setting now defaults to true, and the `sql.insights.anomaly_detection.latency_threshold` cluster setting now defaults to 50ms, down from 100ms to complement the fixed-threshold detector's default of 100ms.
maryliag
left a comment
There was a problem hiding this comment.
if the user fits both the anomaly detector and the "regular" detector, you will see one row for each, or just one row?
Reviewed 4 of 4 files at r1, all commit messages.
Reviewable status:complete! 0 of 0 LGTMs obtained (waiting on @matthewtodd)
maryliag
left a comment
There was a problem hiding this comment.
*if the query fits
Reviewable status:
matthewtodd
left a comment
There was a problem hiding this comment.
Good q! Just the one row. (That's the point of the compositeDetector.)
Reviewable status:
maryliag
left a comment
There was a problem hiding this comment.
Awesome!
Reviewable status:
|
bors r+ |
|
I watched the anomaly detector's memory usage while running the ycsb-b workload. With 22 total statement fingerprints, memory reached 37KiB then settled down at just under 15 KiB. Our default memory limit of 1MiB means we would track somewhere in the range of 600-1500 distinct fingerprints.
|
|
Build failed (retrying...): |
|
Build succeeded: |
This change turns on the insights "anomaly detector" by default,
catching statement executions in the >p99 latency for their fingerprint
that are also far enough away from median latency and above a (default
50ms) threshold.
Release justification: Category 2: Bug fixes and low-risk updates to new
functionality.
Release note (ops change): The
sql.insights.anomaly_detection.enabledcluster setting now defaults to true, and the
sql.insights.anomaly_detection.latency_thresholdcluster setting nowdefaults to 50ms, down from 100ms to complement the fixed-threshold
detector's default of 100ms.