Обсуждение: pg_dump and pg_restore with multiple streams does Not seem to improve overall times

On May 1, 2015, at 9:16 AM, Susan K. McClure <smcclure@rice.edu> wrote:
>
> The DB in question is ~25GB.  The processor has 24 Cpus, 12 cores
>
> I have tried with "--jobs = 8, 12, and 20"  with little or no discernible improvements.

I would immediately suspect that you might be disk-bound instead of CPU-bound. What’s the setup of your storage system?

--
Scott Ribe
scott_ribe@elevated-dev.com
http://www.elevated-dev.com/
https://www.linkedin.com/in/scottribe/
(303) 722-0567 voice

Am 2015-05-01 17:49, schrieb Scott Marlowe:
> On Fri, May 1, 2015 at 9:16 AM, Susan K. McClure <smcclure@rice.edu
> [1]> wrote:
>
>> Running postgresql 9-4 on REHL 7 system. I am trying to speed up
>> pg_dump and pg_restore by
>> using a postgresql.conf with various performance options set, and
>> the --jobs option to force multiple
>> streams.  But various tests, with various "--jobs="  numbers only
>> achieve at most a 1 minute improvement
>> in elapsed time versus doing pg_dump or pg_restore with no "--jobs"
>> option and no postgresql.conf with performance
>> options.  Am I missing some key option(s) to improve things?? 
>>  
>> The DB in question is ~25GB.  The processor has 24 Cpus, 12 cores
>>  
>> I have tried with "--jobs = 8, 12, and 20"  with little or no
>> discernible improvements.
>
> So have you tried 2 jobs first? Id see how 1, 2, 3, 4 etc work. See
> if
> 2 is faster than 1, then 3 faster than 2 etc.
>
> Most of the time, unless youve got a really fast IO subsystem
> increasing the --jobs doesnt make a big difference as a lot of the
> work is sequential. Also on restores I think the extra jobs part only
> kicks in for index builds.

It does parallel COPY, too.

Jan

Am 2015-05-01 17:16, schrieb Susan K. McClure:

[...]
> My postgresql.conf has these options for (hopefully) pg_dump and
> pg_restore improvements:
>
> =================
> work_mem = 1GB # dump/restore Perf Value
> #maintenance_work_mem = 2048MB # min 1MB
> maintenance_work_mem = 1GB # dump/restore Perf Value
> #autovacuum_work_mem = -1 # min 1MB, or -1 to use
> maintenance_work_mem
>
> ...
> fsync = off # dump/restore Perf Valuecheckpoint_segments = 60 #
> dump/restore Perf Value
> checkpoint_warning = 60s # dump/restore Perf Value
> ......
> autovacuum = off # dump/restore Perf Value..
> =============================
>
> Any thoughts on what else I might try to improve things ?
[...]

It ist more interesting to know how your database schema is. How many
"large" tables do you have. What is e.g. the size of your 10,20,30
largest tables?

Regards

Jan

My understanding of parallel dump performance is that it only makes a
difference when you have a large number of DBs (thousands if not tens of
thousands). We performed similar testing using 9.3.x and found little
performance gains using -j (with 100+ tables). See Bruce Momjian’s post :
http://momjian.us/main/blogs/pgblog/2012.html

Mel Llaguno • Staff Engineer – Team Lead
Office: +1.403.264.9717 x3310
www.coverity.com <http://www.coverity.com/> • Twitter: @coverity


On 5/4/15, 6:17 AM, "Jan Lentfer" <Jan.Lentfer@web.de> wrote:

>Am 2015-05-01 17:16, schrieb Susan K. McClure:
>
>[...]
>> My postgresql.conf has these options for (hopefully) pg_dump and
>> pg_restore improvements:
>>
>> =================
>> work_mem = 1GB # dump/restore Perf Value
>> #maintenance_work_mem = 2048MB # min 1MB
>> maintenance_work_mem = 1GB # dump/restore Perf Value
>> #autovacuum_work_mem = -1 # min 1MB, or -1 to use
>> maintenance_work_mem
>>
>> ...
>> fsync = off # dump/restore Perf Valuecheckpoint_segments = 60 #
>> dump/restore Perf Value
>> checkpoint_warning = 60s # dump/restore Perf Value
>> ......
>> autovacuum = off # dump/restore Perf Value..
>> =============================
>>
>> Any thoughts on what else I might try to improve things ?
>[...]
>
>It ist more interesting to know how your database schema is. How many
>"large" tables do you have. What is e.g. the size of your 10,20,30
>largest tables?
>
>Regards
>
>Jan
>
>
>
>
>-- 
>Sent via pgsql-admin mailing list (pgsql-admin@postgresql.org)
>To make changes to your subscription:
>http://www.postgresql.org/mailpref/pgsql-admin

Am 2015-05-04 18:19, schrieb Mel Llaguno:
> My understanding of parallel dump performance is that it only makes a
> difference when you have a large number of DBs (thousands if not tens
> of
> thousands). We performed similar testing using 9.3.x and found little
> performance gains using -j (with 100+ tables). See Bruce Momjian’s
> post :
> http://momjian.us/main/blogs/pgblog/2012.html

I don't know about parallel pg_dump as we use -Fc and pg_dump can't do
that in parallel (afaik). For dumping I have wrapped pg_dump in a shell
script to dump several databases in parallel.
But for pg_restore -j option does make a big difference, at least when
you have a lot of larger tables and indexes.

Regards,

Jan

-----Original Message-----
From: pgsql-admin-owner@postgresql.org [mailto:pgsql-admin-owner@postgresql.org] On Behalf Of Jan Lentfer
Sent: Monday, May 04, 2015 12:42 PM
To: Mel Llaguno
Cc: pgsql-admin@postgresql.org
Subject: Re: [ADMIN] pg_dump and pg_restore with multiple streams does Not seem to improve overall times

I don't know about parallel pg_dump as we use -Fc and pg_dump can't do that in parallel (afaik). For dumping I have
wrappedpg_dump in a shell script to dump several databases in parallel.
 
But for pg_restore -j option does make a big difference, at least when you have a lot of larger tables and indexes.

Regards,

Jan


Combination of -Fc  and -j definitely works in pg_dump.  And there is nothing in the docs that states otherwise.

Regards,
Igor Neyman