You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Hello.
I installed readonlyrest-1.16.14_es5.6.4.zip. When requests that have no permission to the cluster were forbidden by the plugins, I noticed that they remained in the cluster by using task management API, and the number of them increases when more HTTP requests are forbidden. After a few days, i found that there were millions of tasks remained in memory and causing continuous full GC.
Hi @viewsite thanks for reporting this. It's a regression, as we found this issue before, months ago before a couple of major refactoring sessions which must have reintroduced this behaviour.
Just to let you know I'm right on this at the moment. Will update soon with progress. 👍
Hello.
I installed readonlyrest-1.16.14_es5.6.4.zip. When requests that have no permission to the cluster were forbidden by the plugins, I noticed that they remained in the cluster by using task management API, and the number of them increases when more HTTP requests are forbidden. After a few days, i found that there were millions of tasks remained in memory and causing continuous full GC.
curl "localhost:9200/_tasks?pretty&filter_path=nodes.*.tasks.*.action" | grep action | sort | uniq -c
236 "action" : "cluster:monitor/health"
8191197 "action" : "cluster:monitor/main"
log:
[2017-12-28T12:14:44,526][WARN ][o.e.m.j.JvmGcMonitorService] [1498146454000002009] [gc][old][240145][905] duration [16.8s], collections [1]/[17s], total [16.8s]/[2.1h], memory [29.1gb]->[12.1gb]/[29.9gb], all_pools {[young] [416.5mb]->[3.8mb]/[665.6mb]}{[survivor] [83.1mb]->[0b]/[83.1mb]}{[old] [28.6gb]->[12.1gb]/[29.1gb]}
It is easy to make this reproduce. All you have to do is sending forbidden requests to cluster continuously and check tasks of the cluster.
The text was updated successfully, but these errors were encountered: