[Bug 241980] panic: I/O to pool appears to be hung on vdev

[bugzilla-noreply at freebsd.org]

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=241980

Eugene Grosbein <eugen at freebsd.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|New                         |Open

--- Comment #17 from Eugene Grosbein <eugen at freebsd.org> ---
Now I'm pretty sure the hardware is fune and there is some bug in ZFS logic. I
suspect this system triggers it by combination of multiple ZVOL snapshots
created/removed every hour (tens or hundreds) combined with SSD-only RAIDZ1
pool (no HDDs), so huge amount of ATA TRIM operations are added every hours to
the queues (avl trees, in fact).

Because of kernel hungs and inability to produce crashdump, I've patched ZFS
code changing vfs.zfs.deadman_enabled to r/w sysctl. Also, the patch disables
panic if vfs.zfs.deadman_enabled=2 and adds additional logging to syslog (not
via zfs_dbgmsg) early when number of queued zio's hits 1/8th of standard limit
of 1000 seconds.

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