v1.4.2 agent hangs on timed out UserParameter (caused by unreachable NFS mount)

[just2blue4u]

(As posted in "call to discussion for 1.6")

Today our NFS Server was shut down without unmounting it on all NFS Clients. We have a Zabbix UserParameter "files.open", that contains

#UserParameter=files.open,/usr/sbin/lsof|grep -v lsof|grep -v grep|wc -l|sed s/" "//g

to count all open files on the Client.
This Command hung on the mounted (and unreachable) NFS dir. That caused our agents to hang, too. After killing the lsof procs and restarting the agents, everything was fine again (deactivated the UserParameter line in config).

I'd really like to see Zabbix killing alle UserParameters that time out or at least being more immune/resistent/independent to hanging UserParameters.

(Issue seen on Server 1.4.4, agent 1.4.2)

[just2blue4u]

first, no, you are not the only one
second, that's a hard nfs mount. that's just how they behave.
you could use softmounts, but things can break badly if you do.
i'm not sure how much would some timeouts help in case of zabbix agent - there are very few, if any, applications that behave really well with stalled nfs mounts. on the positive note, all apps resume nicely when nfs connection is resumed

Thanks for your Answer!
I know NFS mounts are a really tricky thing. But i don't have the problem with my nfs, it's the fact that the complete agent hangs if one UserParameter does so...

[richlv]

ok, that's a reasonable request, i think. though it would still require timeouting of such checks, otherwise pooling them in the background could result in a lot of 'hung' checks.