Bug #4408
closed
Sometimes there are too many cf-agent processes running
Added by Dennis Cabooter almost 11 years ago.
Updated almost 11 years ago.
Description
[root@node ~]# free -m
total used free shared buffers cached
Mem: 3824 3798 25 0 0 13
-/+ buffers/cache: 3784 39
Swap: 2047 2047 0
[root@node ~]# ps wwwuax | grep cf-|wc -l
5988
[root@node ~]# kill -9 `ps wwwuax | grep cf- | awk '{ print $2 }'`
-bash: kill: (8484) - No such process
[root@node ~]# ps wwwuax | grep cf-|wc -l
5
[root@node ~]# free -m
total used free shared buffers cached
Mem: 3824 344 3479 0 2 188
-/+ buffers/cache: 153 3670
Swap: 2047 61 1986
- Assignee set to Nicolas CHARLES
- Target version set to 2.8.3
That is very very much ...
Is it happening only on root server ?
A fix was merged this weekend #3928 that could fix that problem ... (maybe linked to the tokyo cabinet locks ...)
Nicolas, what do you think about that??
As i Suspect tcdb, i target branch 2.8
Sometimes it's on the server as well, sometimes on the nodes. This times it was on a node. And as you can understand ~ 6000 cf-agent processes absorb almost al the memory and makes the node unworkable. The nodes (included the server) report NoAnswer state when this happens.
- Status changed from New to Rejected
This looks very much like a duplicate of #3928, and even if it is not, the fix for #3928 will fix this: the cron script that is run every 5 minutes checks if there are more than 8 CFEngine processes running, and kills them if so. It also cleans up the TokyoCabinet cf_lock.tcdb database, which is the cause of this.
Thanks for the report Dennis. I'm closing this ticket as a duplicate of #3928, which will be released in 2.8.3 and 2.9.3 very shortly.
Also available in: Atom
PDF