[CASSANDRA-8932] Update default Garbage Collector GC on cassandra ps: Use G1GC instead of CMS

• Type: Improvement
• Status: Resolved
• Priority: Normal
• Resolution: Duplicate
• Assignee: Rekha Joshi
• Reporter: Rekha Joshi

Authors

Rekha Joshi

• Created: 08/Mar/15 19:18
• Updated: 16/Apr/19 09:31
• Resolved: 08/Mar/15 19:44

Description

On Cassandra 2.0.8.39/DSE 4.5.1/Java HotSpot(TM) 64-Bit Server VM/1.7.0_67 encountered long running GC before dse was shutdown.
This was a ring of 6 nodes, and no unexpected load/processes were run.The error log attached.

Suspect was that it has hit hotspot/java GC issue.But in this case the class unloading is enabled.
JVM_OPTS="$JVM_OPTS -XX:+CMSClassUnloadingEnabled”
XX:CMSInitiatingOccupancyFraction=75

On a general note, what would make things better is to make G1GC as default garbage collector (for cassandra running > java 7 update 9) instead of CMS., as G1GC is far better.For eg: -XX:+UseG1GC -XX:MaxGCPauseMillis=50
http://www.oracle.com/technetwork/tutorials/tutorials-1876574.html

Perhaps also can be looked if 1.when CPU utilization kicks or heap size leak/long running GC is sensed, the nodetool -h localhost flush can auto kick in?
2. force log the query when long running GC kicks in to make debuging easier.

Issue Links

• duplicates: CASSANDRA-7486 Migrate to G1GC by default