[Bug 229614] ZFS lockup in zil_commit_impl

[bugzilla-noreply at freebsd.org]

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=229614

--- Comment #16 from Andreas Sommer <andreas.sommer87 at googlemail.com> ---
Hi all,

within the last weeks I've managed to strip my host's workload down to a
reproducible example of the interlock, and here it is:
https://github.com/AndiDog/zfs-lockup

Please have a look at the `README` and `scripts/all.sh` files at least to
understand the very simple setup. Sure it can be reproduced with even simpler
tricks. I talked to Allan Jude (CC) at EuroBSDcon and he thought that any
synchronous write might trigger the issue (and had some ideas on debugging it)
– for me, the buildbot master did a good job of triggering the interlock by
writing lots of stuff to its SQLite database.

It would be great if some ZFS experts could take a look into this, now that we
have a way to reproduce it. Since several people already claimed to have this
problem, it remains a pretty risky issue for production.

(P.S. if you find non-best-practice things in the setup, I'm interested to hear
about it by short e-mail or @AndiDog87, just for the sake of learning)

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