Thread: amount of WAL logs larger than expected

amount of WAL logs larger than expected

From
Mark Steben
Date:
Good afternoon

We run postgres 9.2.12  and run vacuum on a 540GB database.

I've attached a 'show all' for your reference.  With CHECkPOINT_SEGMENTS set at 128
and CHECKPOINT_COMPLETION_TARGET set at 0.9 I wouldn't expect the number of logs in pg_xlog to creep much over 400.  But when we run vacuum, the number can climb to over 5000 and threatens to blow out on space.  Is there something else I should be looking at causing this unexpected number of logs?

Our server is also master for a slony1.2.2.3 slave and also to a hot standby server.

Any insight /recommendations welcome.  Thank you
 

--
Mark Steben
 Database Administrator
@utoRevenue | Autobase 
  CRM division of Dominion Dealer Solutions 
95D Ashley Ave.
West Springfield, MA 01089

t: 413.327-3045
f: 413.383-9567

www.fb.com/DominionDealerSolutions
www.twitter.com/DominionDealer
 www.drivedominion.com





Attachment

Re: amount of WAL logs larger than expected

From
"Joshua D. Drake"
Date:
On 05/09/2016 11:23 AM, Mark Steben wrote:
> Good afternoon
>
> We run postgres 9.2.12  and run vacuum on a 540GB database.
>
> I've attached a 'show all' for your reference.  With CHECkPOINT_SEGMENTS
> set at 128
> and CHECKPOINT_COMPLETION_TARGET set at 0.9 I wouldn't expect the number
> of logs in pg_xlog to creep much over 400.  But when we run vacuum, the
> number can climb to over 5000 and threatens to blow out on space.  Is
> there something else I should be looking at causing this unexpected
> number of logs?
>
> Our server is also master for a slony1.2.2.3 slave and also to a hot
> standby server.

Are you archiving to your hot standby or streaming? If you are
archiving, you may not be returning a proper success code and thus
postgresql is keeping logs.

Also, you are well behind production (9.2.16) and there are
*significant* bug fixes in the release gaps between.

JD

--
Command Prompt, Inc.                  http://the.postgres.company/
                         +1-503-667-4564
PostgreSQL Centered full stack support, consulting and development.
Everyone appreciates your honesty, until you are honest with them.


Re: amount of WAL logs larger than expected

From
bricklen
Date:
On Mon, May 9, 2016 at 11:23 AM, Mark Steben <mark.steben@drivedominion.com> wrote:
We run postgres 9.2.12  and run vacuum on a 540GB database.

I've attached a 'show all' for your reference.  With CHECkPOINT_SEGMENTS set at 128
and CHECKPOINT_COMPLETION_TARGET set at 0.9 I wouldn't expect the number of logs in pg_xlog to creep much over 400.  But when we run vacuum, the number can climb to over 5000 and threatens to blow out on space.  Is there something else I should be looking at causing this unexpected number of logs?

Our server is also master for a slony1.2.2.3 slave and also to a hot standby server.


Is the hot standby in a different network or over the WAN? Have you checked for bandwidth saturation? Also, have a look at the directory the hot standby is receiving WALs at, check if they most recent ones have a current timestamp. If the WALs that are arriving have much older timestamps than what is being generated on the primary, that could indicate slow transfer.
For example, I had an issue recently where 4k WALs built up on the primary during a large ETL process. It took a few hours to ship those (compressed) WALs over the WAN to the replica's data centre.