[OmniOS-discuss] Host is not being rebooted if it uses ZFS over iSCSI

Jim Klimov jimklimov at cos.ru
Wed Nov 30 19:47:34 UTC 2016


30 ноября 2016 г. 17:45:53 CET, "Дмитрий Глушенок" <glush at jet.msk.su> пишет:
>The 'reboot' works, but it kills all processes at once (including
>iscsid), leaving no chances for ZFS to flush write cache. And I suspect
>that with current iscsi/initiator SMF service implementation the same
>happens (but for some reason sync(), which is being called by 'reboot'
>after killing all processes, is not hangs).
>
>> 30 нояб. 2016 г., в 18:05, Chris Siebenmann <cks at cs.toronto.edu>
>написал(а):
>> 
>>> Does anybody uses ZFS over iSCSI?
>>> 
>>> There is problem with reboots as iscsi-initiator service does not
>>> taking care of ZFS while shutting down. It leads to zpool goes into
>>> UNAVAIL state and then first sync() issued gets blocked with
>following
>>> stack:
>> 
>> We have a significant ZFS-over-iSCSI fileserver environment and
>> haven't seen this, but we may be shutting our systems down in a
>> different way than you are.
>> 
>> We're on OmniOS r151014 and we haven't seen this behavior when we
>> reboot or shut down our systems via 'reboot' or 'poweroff'. For us it
>> seems to go the other way around; running 'reboot' won't completely
>shut
>> down the iSCSI initiator before the machine reboots, apparently
>because
>> it's being held busy by our active ZFS pools.
>> 
>> It may be relevant that we're using OmniOS multipathing across two
>> IPs for each target (configured statically); this might change the
>> device stack inside the kernel.
>> 
>> I don't think we've ever tried 'shutdown' itself on a live production
>> system, so I don't know how that would work.
>> 
>> 	- cks
>
>--
>Dmitry Glushenok
>Jet Infosystems
>
>
>
>------------------------------------------------------------------------
>
>_______________________________________________
>OmniOS-discuss mailing list
>OmniOS-discuss at lists.omniti.com
>http://lists.omniti.com/mailman/listinfo/omnios-discuss

Shameless plug: did you try an approach like https://github.com/jimklimov/illumos-smf-zfspools to define an SMF dependency between a pool whose vdevs reside on an iscsi initiator, or an iscsi target residing on a pool, and perhaps other services like zones to depend on the non-rpool zfs pool (SMFized)?

At least for the case of well ordered and synced graceful shutdowns and startups, this should help you.

HTH, Jim Klimov
--
Typos courtesy of K-9 Mail on my Samsung Android


More information about the OmniOS-discuss mailing list