Re: Optimize pg_dump schema-only

[senor <frio_cervesa@xxxxxxxxxxx>]

I know from a previous post to the community that pg_dump --schema-only is single threaded and the --jobs option cannot benefit pg_dump in my case (single DB, Single schema, 100000+ tables). Using pg_upgrade with the --link option is very fast except for the pg_dump portion of the upgrade which takes days.

I think I am looking for advice on how to optimize the postgres process for best performance of a single thread making whatever queries pg_dump --schema-only does. Obviously, that's not a common concern for DB tuning.  At the moment, all I'm reasonably sure about is that a reduction in the number of tables should reduce the time needed but that is not the reality I'm faced with.

Thanks,
Senor

On 4/28/2019 14:08, Ron wrote:

On 4/28/19 3:21 PM, senor wrote:

Hi All,

I'm looking for advice for optimizing the pg_dump portion of "pg_upgrade
--link". Since this schema only dump can't take advantage of parallel
processing with jobs I'm looking for any preparation or configuration
settings that can improve speed.

9.2 to 9.6
CentOS 6/64bit
512GB

I see only one CPU of 32 doing anything and it's often at 100%. Disk IO
is minimal. Memory use varies but always plenty to spare.

"pg_dump --schema-only" is single-threaded.