Index upgrade problems after cyrus upgrade

[Zachariah Mully]

On 10/13/2011 01:46 AM, Bron Gondwana wrote:
> On Thu, Oct 13, 2011 at 07:00:42AM +0530, Ramprasad wrote:
>> On 10/13/2011 6:39 AM, Adam Tauno Williams wrote:
>>> On Thu, 2011-10-13 at 05:33 +0530, Ramprasad wrote:
>>>> I just upgraded cyrus on a busy Imap server from 2.3.x to 2.4.6 ,
>>>> problem is with lmtp delivery.
>>>> All lmtp deliveries from postfix are hanging forever.
>>>> I think this is due to the "Index upgrade" that is going on .. Is there
>>>> a way I can speed up the index upgrade
>>>> I have 2000 users and only 200 users index upgrade is complete till now
>>>> in last 4 hours
>>> Are you letting index reconstructs happen on-the-fly or are you doing a
>>> reconstruct of the mailbox(es).  A reconstruct will rebuild the indexes.
>>
>> I just sent a mail to all mailboxes , so lmtp delivery will do a Index
>> upgrade.
>> Reconstructing will be much heavier , I think.
>>
>> Is there a way I can do a Index-upgrade only ?
>>
>> Also I have a few more servers to upgrade .. but if this is going to
>> take so long , upgrade may not be possible. Only if I could upgrade the
>> Index before really upgrading cyrus then this would work
>
> Actually, reconstruct isn't much heavier.
>
> Bloody 2.4.6, it's half a year old now.  There are quite a few known bugs
> solved since, but your upgrade cost isn't one of them.  The only real
> way to do save the upgrade cost would be a delayed rebuild, which is...
> tricky.

We're getting bitten by this index upgrade as well. Unfortunately, we're 
going 2.2.13->2.4.12, so don't have replication to assist. We're also 
moving into a new data center, so the bandwidth is complicating everything.

If we're rsync'ing our spools from our 2.2 to our 2.4 instance, should 
we exclude the 2.2 cyrus.index files? I'm worried now that we're undoing 
all the reindexing with every rsync run.

Z