problematic upgrade 2.3.16 -> 2.4.3

[Paul Dekkers]

Hi Bron,

On 11-11-10 03:45, Bron Gondwana wrote:
> On Wed, Nov 10, 2010 at 09:38:34PM +0100, Paul Dekkers wrote:
>> On 10-11-2010 21:27, Bron Gondwana wrote:
>>
>>>> (Can I downgrade? I just need to reconstruct everything, right, because
>>>> the index format changed? :-S)
>>>
>>> Yes, you will need to reconstruct anything which successfully upgraded.
>>
>> Ok, there's no way of telling what folders got succesfully upgraded,
>> right? I only see messages "Index upgrade failed: user.paul",
>> indicating the ones that failed :-S
>>
>> (A selective reconstruct would be faster.)
> 
> You can run 'reconstruct -s', which is pretty fast, but still
> detects the worst cases.
>  
>>> This issue got reported last night, and I wrote a patch but haven't
>>
>> So, just the reconstruct problem, or more of my symptoms? (I've
>> looked at the devel archive and the info-cyrus, but did not see
>> anything similar, or from last night anyway ;-) so I have no idea
>> what issue exactly.)
> 
> This will fix all your broken upgrades.

Tusen takk! Thanks for the patches; I applied all, and did a reconstruct -s.

There are no strange errors anymore in the logs anymore, BUT it seems
that people are missing messages that where previously still in their inbox.

During the reconstruct I saw for instance:

user.bla removed 11402 records from stale cyrus.expunge
user.bla uid 10586 not found
user.bla uid 22003 not found
user.bla uid 22010 not found
user.bla uid 22880 not found
user.bla uid 22972 not found
...

So, the stale removal from expunge is good :-)

Yes, of course I made backups ;-)

zfs clone data/imap at daily.1 data/imap-oud
zfs clone data/imap-meta at daily.1 data/imap-meta-oud

... but The other files were there (from the now imap-oud (oud=old)):

teletubbie# ls -l 10586.
-rw-------  1 cyrus  cyrus  2727792 Nov 19  2006 10586.

And seem to contain the messages this user is missing now.

Any clue what went wrong?

Regards,
Paul

P.S. Other errors I see are like: "sync_client[24062]: RECORD MISMATCH
WITH REPLICA: more recent on replica" because - ahem - time was not in
sync (which was not an issue, but is now) but this is just a first-time
warning and it does a full-sync I assume.

Not sure what this means but I see a bunch too, not so worried:
sync_client[24062]: inefficient replication (25 > 24) user.linda