Resilver breaks nfs

I have a zpool which is a mirror made out of 3 disks. Those are again located on three different iscsi-servers. Other hosts are accessing this pool via nfs and use it as a mail-storage. Thus dovecot (with director) and exim are accessing it.
After one disk failed, I replaced it, using the drive locally makes virtually no difference. But when I try to access it via nfs and there are loads of users on the system, nfs is unbareably slow. I do not have this problems with other pools (I have multiple pools for that purpose, spread over multiple nfs-servers and disks are spread over several more iscsi-servers). Everything is configured the same, the same applications are accessing it, roughly the same number of users are on the pools.
Currently I am also resilvering on another pool which is on another server and there I do not see those problems.
I just stopped the resilver and operations via nfs seem to be immediately at normal speed again.
I don't see any packet loss on the ethernet-devices for nfs-traffic and no packet loss with the iscsi-traffic.
Any ideas what could cause this?
 
Can it be that the iscsi traffic takes up all the network resources somehow? Have you configured QoS/IP prioritization?
 
No, I don't. But the iscsi-connections are direct connections between the servers, no switch in between.
But it seems already that it is something different than the resilver because stopping the resilver didn't help much alleviating the problem :/
 
Tja, try using performance monitoring tools like:
Code:
# Memory, resource info:
vmstat 3
iostat -w3
systat -iostat
gstat
These might show some hints if the storage has unexpectedly high load for some reason.

And for the network load maybe try net-mgmt/iftop.
 
Back
Top