From: Rick M. <rf...@os...> - 2003-08-09 22:21:18
|
I noticed on our cluster that all nodes had the same "Most Full Disk Partition" percentage listed. Turns out that ganglia isn't discerning remote NFS filesystems quite right. The remote_mount function in gmond/machines/linux.c looks for a ":" to determine if a partition is NFS mounted. This of course works great most of the time. However, our cluster uses NFS mounted root partitions that are shared by all nodes. In /proc/mount, the lines look like rootfs / rootfs rw 0 0 /dev/root / nfs ro,v2,... There is no ":", so it treats this filesystem as local, and since it is the most full, it is reported by every node. It would be better to leave this out. Obviously the second line is easily weeded out because of the "nfs" filesystem type. However, the first line will be the same even is the / filesystem is local. Has this been brought to anyone's attention before? Has anyone attempted to add some logic to remote_mount() to fix this? I just wanted to throw this observation out there. --Rick |