IOERROR: writing cache file

[Rudy Gevaert]

Bron Gondwana wrote:
> On Thu, 27 Sep 2007 16:08:19 +0200, "Rudy Gevaert" <Rudy.Gevaert at UGent.be> said:
>> Bron Gondwana wrote:
>>> On Thu, Sep 27, 2007 at 02:51:09PM +0200, Rudy Gevaert wrote:
>>>> Hello again,
>>>>
>>>> Does somebody know what could be the source of the following error messages?
>>>>
>>>> Thu Sep 27 13:24:09 2007 manaslu mail6/imaps/mail6/imaps[27763]: 
>>>> IOERROR: writing cache file for ugent.be!user.heleen^nailis.Labo.Rosina: 
>>>> Bad address
>>>> Thu Sep 27 13:24:26 2007 manaslu mail6/imaps/mail6/imaps[4600]: IOERROR: 
>>>> writing cache file for ugent.be!user.heleen^nailis.Labo.Rosina: Bad address
>>>> Thu Sep 27 13:26:37 2007 manaslu mail6/imaps/mail6/imaps[15730]: 
>>>> IOERROR: writing cache file for 
>>>> ugent.be!user.heleen^nailis.maatjes.Gwenn: Bad address
>>>> Thu Sep 27 13:26:51 2007 manaslu mail6/imaps/mail6/imaps[2402]: IOERROR: 
>>>> writing cache file for ugent.be!user.heleen^nailis.maatjes.Gwenn: Bad 
>>>> address
>>> heleen^nailis ?   Is that a carat or a control character being printed?
>> That's a carat.
>>
>>> Are you able to reproduce reliably?  May be worth trying to grab
>>> an strace/truss of the process and see what syscalls it's making.
>> I have no idea how it happens.  I just saw it today, a couple of times 
>> from the same user.  She has it on two folders. The user has already 
>> logged out so I can't attach to that process. :(
> 
> Can you log in as the same user and test (or maybe copy the exact
> contents of the mailboxes somewhere else and try to recreate - that
> is if you care that much to figure out the underlying cause...)

Ok, I logged in as the user and went to the mailbox.  Did some selects 
but that didn't show any errors.

I also copied here folders to an other location and all went well to.
I'm keeping an eye on the log files, so if it happens again I'll 
reinvestigate.

> 
>>> Alternatively, you could have filesystem corruption in the meta
>>> directories - have you tried moving them aside and creating new
>>> ones?
>> Hmm.  Do you mean the cyrus.cache file in those folders?  Is a 
>> reconstruct enough to recreate them or doesn't happen automatically.
> 
> I had a filesystem which had one directory that was totally corrupted
> inside due to random gremlins (not sure when it happened, but with
> terabytes of storage, these things do happen occasionally, even with
> RAID you still get occasional errors on the wire - the research on
> this really quite disturbing!)
> 
> I moved it out somewhere else and recreated the meta directory then
> reconstructed the mailbox.  Worked fine.  You do need to reconstruct
> and if you can copy the cyrus.index back in then you'll find that
> no information (flags, etc) should be lost.
> 
> I failed over to the replica and took the filesystem offline and ran
> a full fsck (reiserfsck in this case) which only found the one error
> on that folder.  Ran with rebuild-tree to remove it from the tree
> (it wasn't deleteable otherwise) and all was good.
> 
> And yeah - for pretty much anything involving the cache file,
> reconstruct is your friend.  Everything in the cache file is
> copies of data in the original messages, so reconstruct can
> rebuild it precisely.

Unfortunately taking the mailstore offline isn't that easy here :(.  We 
are running replication but for now particular reason because it isn't 
stable yet.  I haven't got round to implement a checking system. 
Hopefully if we upgrade to 2.3.9 with your md5 hash I can get round to that.


-- 
-- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --
Rudy Gevaert          Rudy.Gevaert at UGent.be          tel:+32 9 264 4734
Directie ICT, afd. Infrastructuur ICT Department, Infrastructure office
Groep Systemen                    Systems group
Universiteit Gent                 Ghent University
Krijgslaan 281, gebouw S9, 9000 Gent, Belgie               www.UGent.be
-- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --