a strange and terrible saga of the cursed iSCSI ZFS SAN

[Eugene M. Zheganin]

On 05.08.2017 22:08, Eugene M. Zheganin wrote:
> Hi,
>
>
> I got a problem that I cannot solve just by myself. I have a iSCSI zfs 
> SAN system that crashes, corrupting it's data. I'll be short, and try 
> to describe it's genesis shortly:
>
> 1) autumn 2016, SAN is set up, supermicro server, external JBOD, 
> sandisk ssds, several redundant pools, FreeBSD 11.x (probably release, 
> don't really remember - see below).
>
> 2) this is working just fine until early spring 2017
>
> 3) system starts to crash (various panics):
>
> panic: general protection fault
> panic: page fault
> panic: Solaris(panic): zfs: allocating allocated 
> segment(offset=6599069589504 size=81920)
> panic: page fault
> panic: page fault
> panic: Solaris(panic): zfs: allocating allocated 
> segment(offset=8245779054592 size=8192)
> panic: page fault
> panic: page fault
> panic: page fault
> panic: Solaris(panic): zfs: allocating allocated 
> segment(offset=1792100934656 size=46080)
>
> 4) we memtested it immidiately, no problems found.
>
> 5) we switch sandisks to toshibas, we switch also the server to an 
> identical one, JBOD to an identical one, leaving same cables.
>
> 6) crashes don't stop.
>
> 7) we found that field engineers physically damaged (sic!) the SATA 
> cables (main one and spare ones), and that 90% of the disks show ICRC 
> SMART errors.
>
> 8) we replaced the cable (brand new HP one).
>
> 9) ATA SMART errors stopped increasing.
>
> 10) crashes continue.
>
> 11) we decided that probably when ZFS was moved over damaged cables 
> between JBODs it was somehow damaged too, so now it's panicking 
> because of that. so we wiped the data completely, reinitialized the 
> SAN system and put it back into the production. we even dd'ed each 
> disk with zeroes (!) - just in case. Important note: the data was 
> imported using zfs send from another, stable system that is runing in 
> production in another DC.
>
> 12) today we got another panic.
>
> btw the pools look now like this:
>
>
> # zpool status -v
>   pool: data
>  state: ONLINE
> status: One or more devices has experienced an error resulting in data
>         corruption.  Applications may be affected.
> action: Restore the file in question if possible.  Otherwise restore the
>         entire pool from backup.
>    see: http://illumos.org/msg/ZFS-8000-8A
>   scan: none requested
> config:
>
>         NAME        STATE     READ WRITE CKSUM
>         data        ONLINE       0     0    62
>           raidz1-0  ONLINE       0     0     0
>             da2     ONLINE       0     0     0
>             da3     ONLINE       0     0     0
>             da4     ONLINE       0     0     0
>             da5     ONLINE       0     0     0
>             da6     ONLINE       0     0     0
>           raidz1-1  ONLINE       0     0     0
>             da7     ONLINE       0     0     0
>             da8     ONLINE       0     0     0
>             da9     ONLINE       0     0     0
>             da10    ONLINE       0     0     0
>             da11    ONLINE       0     0     0
>           raidz1-2  ONLINE       0     0    62
>             da12    ONLINE       0     0     0
>             da13    ONLINE       0     0     0
>             da14    ONLINE       0     0     0
>             da15    ONLINE       0     0     0
>             da16    ONLINE       0     0     0
>
> errors: Permanent errors have been detected in the following files:
>
>         data/userdata/worker208:<0x1>
>
>   pool: userdata
>  state: ONLINE
> status: One or more devices has experienced an error resulting in data
>         corruption.  Applications may be affected.
> action: Restore the file in question if possible.  Otherwise restore the
>         entire pool from backup.
>    see: http://illumos.org/msg/ZFS-8000-8A
>   scan: none requested
> config:
>
>         NAME               STATE     READ WRITE CKSUM
>         userdata           ONLINE       0     0  216K
>           mirror-0         ONLINE       0     0  432K
>             gpt/userdata0  ONLINE       0     0  432K
>             gpt/userdata1  ONLINE       0     0  432K
>
> errors: Permanent errors have been detected in the following files:
>
>         userdata/worker36:<0x1>
>         userdata/worker30:<0x1>
>         userdata/worker31:<0x1>
>         userdata/worker35:<0x1>
>
> 12) somewhere between p.5 and p.10 the pool became deduplicated (not 
> directly connected to the problem, just for production reasons).
>
>
> So, concluding: we had bad hardware, we replaced EACH piece (server, 
> adapter, JBOD, cable, disks), and crashes just don't stop. We have 5 
> another iSCSI SAN systems, almost fully identical that don't crash. 
> Crashes on this particular system began when it was running same set 
> of versions that stable systems.
>
>
So far my priority version is that something was broken in the iSCSI+zfs 
stack somewhere between r310734 (most recent version on my SAN systems 
that works) and r320056 (probably earlier, but r320056 is the first 
revision with documented crash).

So I downgraded back to r310734 (from a 11.1-RELEASE, which is affected, 
if I'm right).

Some things speak pro this version:

- the system was stable pre-spring 2017, before the upgrade happened

- zfs corruption happens _only_ on the pools that the iSCSI is serving 
from, no corruption happens on the zfs pools that have nothing to do 
with providing zvils as iSCSI targets (and this seems to be the most 
convincing point).

- the faulty hardware was changed. though it was changed to a identical 
hardware, BUT I have the very same set of identical hardware working in 
almost identical environment under r310734 in another DC.


so far I'm not sure, because only 20 hours passed since the downgrade. 
However, if the system will be stable for more than a week (was never 
stable that long on recent revisions), it will prove I'm right and I'll 
file the PR.


Thanks.

Eugene.