[CASSANDRA-10548] OOM in Ref#GlobalState

• Type: Bug
• Status: Resolved
• Priority: Normal
• Resolution: Duplicate
• Severity: Normal
• Since Version: 2.1.5
• Assignee: Unassigned
• Reporter: Maxim Podkolzine
• Created: 19/Oct/15 10:45
• Updated: 16/Apr/19 09:30
• Resolved: 19/Oct/15 13:12

Description

Cassandra eats up all memory (at one time 24Gb) and crashes with OOM.
I was able to analyze one instance of hprof and concluded that ~3Gb were referenced by GlobalState instances, mostly ConcurrentLinkedQueues (see the details in screenshot attached). I can provide the hprof if necessary.

I noticed that in 2.2 the source code changed in this place. Can you confirm it's better in 2.2 in terms of memory management?

Issue Links

• duplicates: CASSANDRA-9549 Memory leak in Ref.GlobalState due to pathological ConcurrentLinkedQueue.remove behaviour