[OpenIndiana-discuss] Zpool crashes system on reboot and import

[Stephan Budach]

Hi all,

Am 11.12.13 21:28, schrieb Jim Klimov:
> Also, that thread mentions that you may use this work around with
> the read-only dataset with the pool to enable writes to the dataset
> and keeping it read-only before exporting the pool. Still, yes,
> "budy" mentions setting the dataset attribute while the pool is
> imported read-only (Stephan? would you chime in with more details,
> if that's really you?)
>
Yes, that's me.  And I have to confess, that I am still running with 
this zpool since. I had a long running SR with Oracle about this, but in 
the end I was told by Oracle engineering to re-create the zpool, which I 
refused to do.
So basically, I am still running with this work around of setting the 
affected fs to read-only, before I export the zpool.
This server is under constant load and I just don't have the time and 
resources to move all 370+ ZFS fs onto another storage.

To make things even worse, this error is inside the data structure of 
the ZFS fs, so zfs send/revc, doesn't help here and the data would have 
to be copied "manually" - nasty indeed.

However, I seem to recall, that Oracle Support told me, that this bug 
had been fixed in S11SRU13. I am not sure, but I could look that up in 
the SR. Of course, this only prevents this from happening to fs that are 
not yet affected by this issue - there's currently no cure, afaik.

> If you manage to reproduce this trick in command-line and if it does
> indeed help (and if you want to keep using this pool i.e. to help
> the developers reproduce and fix the core problem) instead of just
> remaking the pool, you might build on the (unbaked as of yet) scripts
> and SMF manifests here:
>
> http://wiki.openindiana.org/oi/Advanced+-+ZFS+Pools+as+SMF+services+and+iSCSI+loopback+mounts 
>
I would go for that, but I don't know, if that'd be to any avail, since 
I am still running S11.
>
> This would enable you to code all this logic about setting and removing
> the readonly bits from your pool around the import-as-a-service, so that
> your system would "gracefully" import the pool, enable the dataset, etc.
> and undo this upon proper shutdown.
>
> HTH,
> //Jim Klimov

Cheers,
budy