Well after spending the whole day re-arranging PATH variables and hacking to get this relogin safe I gave up as some commands required /home related paths to be relatively high in the lookup list and those caused a huge amount of ENOENT failures for syscalls and a huge overhead.
More discussions on #gluster IRC led to the fact that as gluster native mount doesn't support negative lookup caching, then this means that this is going to be forever slow until that's changed. The recommendation however was to move to NFS mount. It doesn't give the same flexibility as there's some proxy work going on through the NFS server, but for /home the difference should be marginal (we're not hitting the network limits anyway). So the result was that I moved from native glusterfs mount to NFS mount, which btw required on CentOS 6.3 some additional options:
# grep home /etc/fstab
192.168.1.241:/home0 /home nfs tcp,vers=3 0 0
The reason is that glusterfs NFS servers don't support udp protocol. However since moving to NFS based /home the lookup issues have disappeared and all work is nice and fast as it was before. Thanks to jdarcy on #gluster
No comments:
Post a Comment