https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=288777

--- Comment #7 from Bren <[email protected]> ---
I moved our remote backup process to rsync'ing the live filesystem vs. the
snapdirs. Now I'm seeing similar but slightly different behavior:

- After running the rsync process for some time, disk IO freezes completely on
both the rpool and storage pools.

- I can get logged in, but if I run even "dmesg" the command hangs forever (the
OS drives are separate SSDs from the storage pool drives to that's odd). I
can't ctrl-c out of it.

- Running "top -mio" I see a zfs process at 100%.

- This gets logged:

swap_pager: indefinite wait buffer: bufobj: 0, blkno: 260988, size: 4096

- watchdogd doesn't always trigger a reboot. I'm running that with the these
flags:

-t 120 --pretimeout 60 --pretimeout-action log,printf,panic -e '/bin/ps' -w -T
15

- I capped arc_max to 50% physmem which didn't help:

# sysctl -a | rg 'arc_(?:max|min)|physmem'
vfs.zfs.arc_min: 2147483648
vfs.zfs.arc_max: 51339679744
hw.physmem: 102679359488

Last time this happened, I showed 50G free, 34G ARC total, 1954MB swap free.

Before changing these settings, I did see memory use going above 90% but still
several gig free. Most of this was ARC I imagine. As I understand it, ARC is
evictable, so even if the system started to run out of available memory, it
would free up memory from ARC before swapping or anything like that. I will
revert these settings.

- If I stop the rsync process the system recovers after a few minutes.

Oddly enough (and thankfully) our production server is on 14.3-RELEASE and
doesn't have this problem. It's mainly serving NFS and is fairly loaded, but
probably not as much as the backup server running 2-4 x constant rsync
processes, backups over NFS, plus regular ZFS snapshot sends from primary to
backup.

I think I'll reboot into 14.2 later today. Let me know if I can help with this
in any other way.

-- 
You are receiving this mail because:
You are the assignee for the bug.

Reply via email to