Re: ZFS pool hangs (live-locks?) after adding L2ARC

On 28 Dec 2023, at 16:03, Lev Serebryakov <lev@freebsd.org> wrote:
> 
> On 28.12.2023 15:16, Dimitry Andric wrote:
> 
>>>>   I can not find any discussion about such problem on Internet. Also, "live" system doesn't have these "z_int_X_Y" threads at all.
>>>>   I want my L2ARC, I've payed for this NVMe!
>>> 
>>> Any ideas?
>> Do you know 100% sure that this new NVMe device is working correctly? Are there any errors to be found in SMART information, dmesg or syslog?
> 
> I'm running "ssd-steadystate.fio" on this SSD now (not whole SSD but only l2arc partition, which is 85% of whole SSD to have 15% over-provisioning), after running `trim` on it.
> 
> Let see...
> 
> If you could advice better scenario for `fio` (with some TRIM added for good measure?) I can run it too.

If these tests all work, I would first try making a plain new zpool, with just this one device under it, and with the same settings you are using for the other zpools.

Then mount the pool somewhere, and exercise it with whatever workload you can find, to see if that too "locks up". If not, you may have found some sort of problem specific to l2arc, and I would guess it is best to take that upstream?

If it does lock up, there is probably some sort of incompatibility between your NVMe device and the FreeBSD driver. You could then try using another driver, IIRC there are two different ones now?

-Dimitry