trouble by file-nr / file-max

[jms1000]

hello,

A few weeks ago we added a raid system (synologic) via nfs on our proxmox cluster. Before all system ran on glusterfs (which is integrated on the os level).

Now we have the problem that the file-nr value goes up. Estimated approx. 5000 in 30 min. At some point the buffer at the file-max value and the server smears. We have already set 956,000 with file-max - but this can not be a solution ...

Who can help me to find the error or debug it?

 

[dcsapak]

do you make container backups on the nfs server?
if yes, you could add a tmpdir in /etc/vzdump.conf
if you do not do this,
the container files get copied to the nfs server, then back, then again to the nfs server

 

[jms1000]

The nfs of the raid I probably exclude. In addition, the backups run only once in the night and the counter runs even if no container actively ever higher.

I interzetlich something further searched and the process found the (so my guess) the file handles does not close again. It is the clusterfsd process that replicates between two proxmox nodes in the cluster (on os level). In the cluster there are still two other nodes, these access via nfs on each one of these servers.

Why the error with the file-nr only occurs after we have integrated the raid I unfortunately do not understand. Also why the file-nr counter only on one of the two glusterfs nodes autriff, I do not understand.

ps. sorry for my bad (Google) english ...

Mfg.Jörg