Bruce Momjian <bruce@momjian.us> writes:
> On Fri, Jul 22, 2016 at 06:29:05PM -0500, Jerry Sievers wrote:
>
>> Has anyone found a clever way to $subject that doesn't involved
>> calculating which are all of the files unnecessary to send and then
>> running rsync with an exclude list?
>>
>> I'm pondering this right now.
>>
>> ISTM for each unlogged table/index/toast in pg_class where
>> relpersistence='u'...
>>
>> rsync --exclude-from... (and exclude-from-file contains)
>>
>> */base/$dbid/$relfilenode
>> */base/$dbid/$relfilenode.*
>> */base/$dbid/$relfilenode_*
>> ...
>>
>> The leading * will match the old/new data directoryes being sent to the
>> standby.
>>
>> MAster upgrade is done using hard-links and the rsync method generally
>> as per the pg docs.
>>
>> I have among others, a 5TB system w/about 40G of unlogged data.
>>
>> The rsync method of upgrade takes only several seconds during a snapshot
>> based test however as the snap was off a live system, the crash recovery
>> startup trunc'd all the unlogged data.
>>
>> This will not be the case during the real upgrade and I just as soon
>> avoid sending the unlogged data only to have it trunc'd away as soon as
>> the standby is started after the upgrade.
>
> Good question --- I don't think there is an easy answer except to
> truncate the unlogged tables before the upgrade.
Yup, as I suspected.
My data warehouse team *rather* not have to dirive the unlogged data
again since it takes time... and I rather not needlessly replicate it
during upgrade (er, because that takes time.)
At my company, if we choose to not avoid replicating unlogged data *and*
run for a short interval in production without a standby, snapshot the
halted master and take the rsync off of that meanwhile the master is
back online.
I'll figure something out.
Thanks!
--
Jerry Sievers
e: jerry.sievers@comcast.net
p: 312.241.7800