Anomaly Detection's band spikes, triggering false alarms
I have an Aurora Serverless database, which I like to be notified whenever it scales. I've been using a band to do this, and it's fine 99% of the time. But sometimes, it triggers, and the database capacity hasn't changed. Digging into it in cloudwatch, I see this:
It looks like the band itself bounces around instead of remaining stable. Can anyone explain this? For now I've switched to a static alarm, so i stop getting 5 messages every night, but this isn't sufficient for longer term and our high load periods.
- 최신
- 최다 투표
- 가장 많은 댓글
Hi, Is it possible for you to share the metric last 2 week and last month behavior. If the patter is stable in the last 2 weeks, can you try to delete and recreate the anomaly detection bands to retrain the model. Here is the information for creation and deletion: https://docs.aws.amazon.com/AmazonCloudWatch/latest/monitoring/Create_Anomaly_Detection_Alarm.html
If the issue still persist, we can dive deep into it.
As you can see, this has been going on for a while. Before this 2 week period, we did have some actual scaling fluctuation, but in this two weeks, not at all. Does the band take historical data into account for its prediction? I thought it just did some beizer-esque curve fitting.
I will try delete it and recreate it and see if that helps.
No that didn't help, still happening last night. Similar sort of pattern, spiking band, stable value. Note that since i recreated this alarm, the value has not changed at all from 2, so the band spiking cannot be because of the value changing historically.
this was triggering all day yesterday, so i've switched it back to a threshold alarm. The band is totally unusable for a stable value for me.
