Thread: PG12: Any drawback of increasing wal_keep_segments

PG12: Any drawback of increasing wal_keep_segments

From
"Shukla, Pranjal"
Date:

Team,

Are there any disadvantages of increasing the “wal_keep_segments” to a higher number say, 500? Will it have any impact on performance of streaming replication, on primary or secondary servers?

 

Thanks & Regards

Pranjal Shukla

Re: PG12: Any drawback of increasing wal_keep_segments

From
Alvaro Herrera
Date:
On 2022-Mar-22, Shukla, Pranjal wrote:

> Team,
> Are there any disadvantages of increasing the “wal_keep_segments” to a
> higher number say, 500? Will it have any impact on performance of
> streaming replication, on primary or secondary servers?

No.  It just means WAL will occupy more disk space.  I've seen people go
even as high as 5000 with no issues.

-- 
Álvaro Herrera         PostgreSQL Developer  —  https://www.EnterpriseDB.com/



Re: PG12: Any drawback of increasing wal_keep_segments

From
Stephen Frost
Date:
Greetings,

* Alvaro Herrera (alvherre@alvh.no-ip.org) wrote:
> On 2022-Mar-22, Shukla, Pranjal wrote:
> > Are there any disadvantages of increasing the “wal_keep_segments” to a
> > higher number say, 500? Will it have any impact on performance of
> > streaming replication, on primary or secondary servers?
>
> No.  It just means WAL will occupy more disk space.  I've seen people go
> even as high as 5000 with no issues.

Yeah, though it makes the primary into essentially a WAL repository and,
really, you'd be better off having a dedicated repo that replicas can
pull from instead.  Consider that a replica might fall way behind and
then demand the primary send 5000 WAL segments to it.  The primary then
has to go pull that 80GB of data from disk and send it across the
network.  As to if that's an issue or not depends on the IOPS and
bandwidth available, of course, but it's not free.

Thanks,

Stephen

Attachment

Re: PG12: Any drawback of increasing wal_keep_segments

From
Marc
Date:

On 29 Mar 2022, at 17:17, Stephen Frost wrote:

Greetings,

On 2022-Mar-22, Shukla, Pranjal wrote:

Are there any disadvantages of increasing the “wal_keep_segments” to a
higher number say, 500? Will it have any impact on performance of
streaming replication, on primary or secondary servers?

No. It just means WAL will occupy more disk space. I've seen people go
even as high as 5000 with no issues.

Yeah, though it makes the primary into essentially a WAL repository and,
really, you'd be better off having a dedicated repo that replicas can
pull from instead. Consider that a replica might fall way behind and
then demand the primary send 5000 WAL segments to it. The primary then
has to go pull that 80GB of data from disk and send it across the
network. As to if that's an issue or not depends on the IOPS and
bandwidth available, of course, but it's not free.

Thanks,

Stephen

Hello Stephen,

How do you see a setup with a ‘a dedicated repo that replicas can pull from’ ?

Thanks in advance for the clarification.

Marc

Re: PG12: Any drawback of increasing wal_keep_segments

From
Stephen Frost
Date:
Greetings,

On Thu, Mar 31, 2022 at 12:58 Marc <postgres@arcict.com> wrote:

On 29 Mar 2022, at 17:17, Stephen Frost wrote:

Greetings,

On 2022-Mar-22, Shukla, Pranjal wrote:

Are there any disadvantages of increasing the “wal_keep_segments” to a
higher number say, 500? Will it have any impact on performance of
streaming replication, on primary or secondary servers?

No. It just means WAL will occupy more disk space. I've seen people go
even as high as 5000 with no issues.

Yeah, though it makes the primary into essentially a WAL repository and,
really, you'd be better off having a dedicated repo that replicas can
pull from instead. Consider that a replica might fall way behind and
then demand the primary send 5000 WAL segments to it. The primary then
has to go pull that 80GB of data from disk and send it across the
network. As to if that's an issue or not depends on the IOPS and
bandwidth available, of course, but it's not free.

Thanks,

Stephen

Hello Stephen,

How do you see a setup with a ‘a dedicated repo that replicas can pull from’ ?

Thanks in advance for the clarification.

I’d suggest checking out pgbackrest. There are other options out there but that’s my favorite (probably because I also am one of the folks involved in its development, full disclosure).

Thanks,

Stephen