Bruce Momjian <bruce@momjian.us> writes:
> On Mon, Jul 25, 2016 at 12:34:45PM -0500, Jerry Sievers wrote:
>
>> Stephen Frost <sfrost@snowman.net> writes:
>>
>> > * Stephen Frost (sfrost@snowman.net) wrote:
>> >
>> >> > I have among others, a 5TB system w/about 40G of unlogged data.
>> >>
>> >> Honestly, sending an extra 40G out of 5TB (we're talking less than 1%
>> >> here...) might be worth it to keep the process simple.
>> >
>> > Of course, I realize right after sending this that you aren't sending
>> > the 5TB. :)
>>
>> Roger that! Really, the hard link pg-upgrade option and rsync method
>> for standbys is pure beauty :-)
>>
>> We have only the post-analyze and this unlogged data problem to deal
>> with.
>>
>> The warehouse is our only system with any substantial amount of unlogged
>> data.
>>
>> Reducing post-analyze delay was solved by me here with our groovy
>> threaded AnalyzMaTron. I can get a >5TB system done in about 45 minutes
>> w/20 threads.
>
> The only thing I can suggest is to document this limitation, if people
> think it is worth it.
+1
A couple sentences right near the rsync piece in pg_upgrade docs would
do it.
Caveat:UNNLogged data will be unnecessarily transfered via rsync and
then immediately discarded from standbys when the standbys are started
unless those tables are truncated on the master before the upgrade.
--
Jerry Sievers
e: jerry.sievers@comcast.net
p: 312.241.7800