On 6/2/23 19:58, Adrian Klaver wrote:
> On 6/2/23 17:44, Ron wrote:
>> Ran into this when upgrading from 13.11 to 15.3...
>>
>> The pg_restore phase failed with "ERROR: out of shared memory", and
>> recommended that I increase max_locks_per_transaction. Doing so let the
>> process run to completion.
>>
>> It took 12.5 minutes to upgrade a 13GB instance. Soon after, I upgraded
>> a 78GB cluster, and it only took 3.1 minutes.
>
> Where/how did you measure those sizes?
Does it really matter?
START_SECS=$(date +"%s")
pg_upgrade ...
FINISH_SECS=$(date +"%s")
ET=`echo "scale=2;(${FINISH_SECS} - ${START_SECS})/60" | bc`
date +"%F %T pg_upgrade finished. Elapsed time: ${ET} minutes."
(Text copied between air-gapped computers, so there might be errors.)
>>
>> (Both are VMs (same number of CPUs and RAM) connected to the same SAN.)
>>
>> A "pg_dump --schema-only" of the two systems shows that the
>> small-but-slow schema is 5.9M lines.
>
> Anything special you are doing in this cluster to create all those lines?
I do nothing; the schema is provided by the vendor.
>
> What is the line count for the other instance?
227K rows.
>>
>> Is this to be expected of such a huge schema?
>>
--
Born in Arizona, moved to Babylonia.