Metrics still active after "DROP TABLE"

[yakirgb]

Metrics still active after "DROP TABLE"

[erebe]

Hello,

I am not part of Criteo (the org owning this repo) anymore, so I don't have more rights than you on this repository at this time : )
So I guess PR is welcome ^^

Regarding the issue, you are right. Once a metric is stored inside the Prometheus registry (the object that contains the metrics internally) is never cleand/diffed with what exist inside cassandra.
https://github.com/criteo/cassandra_exporter/blob/master/src/main/java/com/criteo/nosql/cassandra/exporter/JmxScraper.java#L49

To solve the issue, either we should erase the content of the registry from time ot time somewhere after
https://github.com/criteo/cassandra_exporter/blob/master/src/main/java/com/criteo/nosql/cassandra/exporter/JmxScraper.java#L183

or try to accumulate new metrics in a temporary Prometheus register that we swap with the default one in the end of the run (should be a bit more rewrite)

[egoroof]

I also have issue with active metrics when they should gone, but with repair treadpool metrics - they stay alive even after repair is done.

cassandra_stats{cluster="Testing Tiles",datacenter="n3-staging",keyspace="",table="",name="org:apache:cassandra:metrics:threadpools:internal:repair#22:activetasks:value",} 1.0

After I restart the exporter its gone

exporter:2.3.5
cassandra:3.11.8