On 04/01/2022 05:47, David Brodbeck wrote:
[SNIP]
I found bacula-sd and NFS were not a very happy combination. It was very
quick to react to network glitches by marking the entire volume as
failed, then trying and failing several more volumes before things
finally got moving again. I was constantly having to manually recycle
failed volumes.
That can sometimes be fixed by setting the mount to be soft rather than
hard, or vice-versa, but the Linux implementation of NFS leaves a lot to
be desired, at both ends. (Speaking as a person who was a tech lead at
SGI's "File Serving Systems" and wrote many, many tests for NFS and
really developed a distaste for the subtle gotchas in NFS. If I ever
hear someone defending the, "Oh, it can't fall back from v4 to v3
because they're different protocols", again, I will strangle the
<profanity elided>.)
Or carefully sequentially re-booting all the systems, in the right
order, so that the NFS state is nice and fresh before you start
hammering it. (And what about a circular mount state? This server
mounts that server mounts another server mounts this server. Fun. Not.)
I could see iSCSI maybe working better, but I haven't tried it.
iSCSI is not the same as NFS (or SMB), it shares a "raw device" which
the mounting system then formats and uses, the device can't be shared
on/from the iSCSI server. (Well, it can be shared by simulating
sneaker-net, mount-fiddle about-umount-mount elsewhere-fiddle
about-umount-etc.)
Cheers,
Gary B-)
_______________________________________________
Bacula-users mailing list
Bacula-users@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/bacula-users
