[ https://issues.apache.org/jira/browse/CASSANDRA-14475?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16493474#comment-16493474 ]
Thomas Steinmaurer commented on CASSANDRA-14475: ------------------------------------------------ [~jasobrown], thanks! Sorry for being a pain, but any chance to get this back-ported for 2.1+? > nodetool - Occasional high CPU on large, CPU capable machines > ------------------------------------------------------------- > > Key: CASSANDRA-14475 > URL: https://issues.apache.org/jira/browse/CASSANDRA-14475 > Project: Cassandra > Issue Type: Bug > Components: Tools > Reporter: Thomas Steinmaurer > Assignee: Thomas Steinmaurer > Priority: Major > Fix For: 4.0, 3.11.3 > > Attachments: nodetool_gc_threads.patch, > nodetool_highcpu_gcthreads1_cassandra_JIRA.png > > > Periodically calling nodetool every 5 min results in increased CPU usage by > nodetool only on a machine with 32 physical cores (64 vCPUs) according to our > monitoring: > !nodetool_highcpu_gcthreads1_cassandra_JIRA.png|width=600! > Investigation and testing has shown that it is related to running with > default number of parallel GC threads which is 43 on this particular machine. > We see a System.gc() according to flight recorder but no real evidence from > where it comes from. The nodetool call in question is simply gathering e.g. > the version with "nodetool version". > After explicitly setting the number of parallel GC threads to 1, the high CPU > is entirely gone (see chart above), without impacting nodetool being executed > successfully. 1 parallel GC thread should be sufficient for nodetool anyway I > think. -- This message was sent by Atlassian JIRA (v7.6.3#76005) --------------------------------------------------------------------- To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org For additional commands, e-mail: commits-h...@cassandra.apache.org