Tune up thresholds of data quality hourly alarms. Since deploying the hourly alarms we have seen a lot of hours in which there were (supposedly) "issues". This warrants some research and also, likely, tuning up thresholds
Description
Details
Status | Subtype | Assigned | Task | ||
---|---|---|---|---|---|
Resolved | mforns | T249759 Add hourly resolution to data quality outage/censhorship alarms | |||
Resolved | mforns | T251814 Tune up thresholds of data quality hourly alarms |
Event Timeline
Change 620767 had a related patch set uploaded (by Mforns; owner: Mforns):
[analytics/refinery@master] Tune threshold for hourly traffic anomaly detection
As discussed in the last sync up meeting, I bumped up the threshold for hourly traffic anomaly detection to 15 times the standard deviation |p75 - median|.
Once deployed, we should see only alerts with such deviations or higher.
Change 620767 merged by Nuria:
[analytics/refinery@master] Tune threshold for hourly traffic anomaly detection
Change 627640 had a related patch set uploaded (by Nuria; owner: Nuria):
[analytics/refinery@master] Doubling threshold to reduce false positive alarms
Change 627640 merged by Mforns:
[analytics/refinery@master] Doubling threshold to reduce false positive alarms
Change 628169 had a related patch set uploaded (by Nuria; owner: Nuria):
[analytics/refinery@master] Removing oversampled data that can trigger false positives
Change 628169 merged by Mforns:
[analytics/refinery@master] Removing oversampled data that can trigger false positives