Notification for excessive Full Garbage Collections in a JVM

[SteveHall]

One of our applications that Dynatrace monitors recently had 172 Full GCs in a one hour period.

256 garbage collections were performed in a one hour period and 172 of those (67.1875%) were Full GCs.

While our organization does not subscribe to Dynatrace's log monitoring, you can see from the verbose garbage collection log that indeed, 172 of 256 GCs in a one hour window were Full GCs.

172 of 256 garbage collections in a one hour period were FULL GCs.

It's disappointing that Dynatrace doesn't have a ready-made alert for excessive Full GCs. I am less concerned that there are x GCs in a certain period than I am that there are y GCs in that same period.

This seems to be a perfect use case for Dynatrace monitoring a JVM and to require the log monitoring solution on top doesn't fit our current needs.

Please find a way to help monitor Oracle's Java 8 Hotspot JVMs for excessive Full GCs in a future OneAgent.

Thank you!

[matthias_dillie · Accepted solution]

Hi Steve

Dynatrace has an anomaly detection setting (Extension events) for 'High GC time'. I think this is what you should monitor.
To see full GCs, maybe you can use a java method sensor? You also can try jmx measures. Maybe you find some hints here:
https://gist.github.com/huynhbaoan/1bb4cd87313f56bc63cc4e19ce5c3ec7
https://stackoverflow.com/questions/4633582/is-it-possible-to-monitor-full-gc-frequency-in-jmx-on-ho...

Regards

Matthias