Thread: Logical replication support for generic wal record
Hi All,
I am writing a postgres extension which writes only generic wal record, but this wal is not recognized by logical replication decoder. I have a basic understanding of how logical replication(COPY command for initial sync, wal replica for final sync) works, can you please tell us a way to support this?
Thanks,
Natarajan.R
Natarajan.R
On Mon, Aug 22, 2022 at 11:59 AM Natarajan R <nataraj3098@gmail.com> wrote: > > Hi All, > > I am writing a postgres extension which writes only generic wal record, but this wal is not recognized by logical replicationdecoder. I have a basic understanding of how logical replication(COPY command for initial sync, wal replica forfinal sync) works, can you please tell us a way to support this? "Generic" resource manager doesn't have a decoding API, see [1], which means that the generic WAL records will not get decoded. Can you be more specific about the use-case? Why use only "Generic" type WAL records? Why not use "LogicalMessage" type WAL records if you want your WAL records to be decoded? [1] https://github.com/postgres/postgres/blob/master/src/include/access/rmgrlist.h#L48 -- Bharath Rupireddy RDS Open Source Databases: https://aws.amazon.com/rds/postgresql/
On Mon, Aug 22, 2022 at 11:59 AM Natarajan R <nataraj3098@gmail.com> wrote: > > Hi All, > > I am writing a postgres extension which writes only generic wal record, but this wal is not recognized by logical replicationdecoder. I have a basic understanding of how logical replication(COPY command for initial sync, wal replica forfinal sync) works, can you please tell us a way to support this? > Did you try with a custom WAL resource manager [1][2]? [1] - https://www.postgresql.org/docs/devel/custom-rmgr.html [2] - https://git.postgresql.org/gitweb/?p=postgresql.git;a=commitdiff;h=5c279a6d350205cc98f91fb8e1d3e4442a6b25d1 -- With Regards, Amit Kapila.
On Mon, 22 Aug 2022 at 12:16, Bharath Rupireddy <bharath.rupireddyforpostgres@gmail.com> wrote:
On Mon, Aug 22, 2022 at 11:59 AM Natarajan R <nataraj3098@gmail.com> wrote:
>
> Hi All,
>
> I am writing a postgres extension which writes only generic wal record, but this wal is not recognized by logical replication decoder. I have a basic understanding of how logical replication(COPY command for initial sync, wal replica for final sync) works, can you please tell us a way to support this?
"Generic" resource manager doesn't have a decoding API, see [1], which
means that the generic WAL records will not get decoded.
Can you be more specific about the use-case? Why use only "Generic"
type WAL records? Why not use "LogicalMessage" type WAL records if you
want your WAL records to be decoded?
I am writing an extension which implements postgres table access method interface[1] with master-slave architecture, with the help of doc[1] i decided to go with generic_wal to achieve crash_safety and also for streaming replication. It seems like generic_wal couldn't help with logical replication..
But, I don't have knowledge on "LogicalMessage" Resource Manager, need to explore about it.
But, I don't have knowledge on "LogicalMessage" Resource Manager, need to explore about it.
On Wed, Aug 24, 2022 at 5:12 PM Natarajan R <nataraj3098@gmail.com> wrote: > > > On Mon, 22 Aug 2022 at 12:16, Bharath Rupireddy <bharath.rupireddyforpostgres@gmail.com> wrote: >> >> On Mon, Aug 22, 2022 at 11:59 AM Natarajan R <nataraj3098@gmail.com> wrote: >> > >> > Hi All, >> > >> > I am writing a postgres extension which writes only generic wal record, but this wal is not recognized by logical replicationdecoder. I have a basic understanding of how logical replication(COPY command for initial sync, wal replica forfinal sync) works, can you please tell us a way to support this? >> >> "Generic" resource manager doesn't have a decoding API, see [1], which >> means that the generic WAL records will not get decoded. >> >> Can you be more specific about the use-case? Why use only "Generic" >> type WAL records? Why not use "LogicalMessage" type WAL records if you >> want your WAL records to be decoded? >> > I am writing an extension which implements postgres table access method interface[1] with master-slave architecture, withthe help of doc[1] i decided to go with generic_wal to achieve crash_safety and also for streaming replication. It seemslike generic_wal couldn't help with logical replication.. > But, I don't have knowledge on "LogicalMessage" Resource Manager, need to explore about it. > > > [1] https://www.postgresql.org/docs/current/tableam.html I think the 'Custom WAL Resource Managers' feature would serve the exact same purpose (as also pointed out by Amit upthread), you may want to explore that feature [1]. Here's a sample extension using that feature [2], for a different purpose though, but helps to understand the usage of custom WAL rmgrs. I notice that the docs [3] don't mention the feature in the right place, IMO, it can be improved to refer to custom-rmgr.sgml page, cc-ing Jeff Davis for his thoughts. This would help developers quickly try the feature out and saves time. [1] https://git.postgresql.org/gitweb/?p=postgresql.git;a=commitdiff;h=5c279a6d350205cc98f91fb8e1d3e4442a6b25d1 [2] https://github.com/BRupireddy/pg_synthesize_wal [2] https://www.postgresql.org/docs/devel/tableam.html "For crash safety, an AM can use postgres' WAL, or a custom implementation. If WAL is chosen, either Generic WAL Records can be used, or a new type of WAL records can be implemented. Generic WAL Records are easy, but imply higher WAL volume. Implementation of a new type of WAL record currently requires modifications to core code (specifically, src/include/access/rmgrlist.h)." -- Bharath Rupireddy RDS Open Source Databases: https://aws.amazon.com/rds/postgresql/
Thanks, I'll check it out.
On Wed, 24 Aug 2022 at 18:00, Bharath Rupireddy <bharath.rupireddyforpostgres@gmail.com> wrote:
On Wed, Aug 24, 2022 at 5:12 PM Natarajan R <nataraj3098@gmail.com> wrote:
>
>
> On Mon, 22 Aug 2022 at 12:16, Bharath Rupireddy <bharath.rupireddyforpostgres@gmail.com> wrote:
>>
>> On Mon, Aug 22, 2022 at 11:59 AM Natarajan R <nataraj3098@gmail.com> wrote:
>> >
>> > Hi All,
>> >
>> > I am writing a postgres extension which writes only generic wal record, but this wal is not recognized by logical replication decoder. I have a basic understanding of how logical replication(COPY command for initial sync, wal replica for final sync) works, can you please tell us a way to support this?
>>
>> "Generic" resource manager doesn't have a decoding API, see [1], which
>> means that the generic WAL records will not get decoded.
>>
>> Can you be more specific about the use-case? Why use only "Generic"
>> type WAL records? Why not use "LogicalMessage" type WAL records if you
>> want your WAL records to be decoded?
>>
> I am writing an extension which implements postgres table access method interface[1] with master-slave architecture, with the help of doc[1] i decided to go with generic_wal to achieve crash_safety and also for streaming replication. It seems like generic_wal couldn't help with logical replication..
> But, I don't have knowledge on "LogicalMessage" Resource Manager, need to explore about it.
>
>
> [1] https://www.postgresql.org/docs/current/tableam.html
I think the 'Custom WAL Resource Managers' feature would serve the
exact same purpose (as also pointed out by Amit upthread), you may
want to explore that feature [1]. Here's a sample extension using that
feature [2], for a different purpose though, but helps to understand
the usage of custom WAL rmgrs.
I notice that the docs [3] don't mention the feature in the right
place, IMO, it can be improved to refer to custom-rmgr.sgml page,
cc-ing Jeff Davis for his thoughts. This would help developers quickly
try the feature out and saves time.
[1] https://git.postgresql.org/gitweb/?p=postgresql.git;a=commitdiff;h=5c279a6d350205cc98f91fb8e1d3e4442a6b25d1
[2] https://github.com/BRupireddy/pg_synthesize_wal
[2] https://www.postgresql.org/docs/devel/tableam.html
"For crash safety, an AM can use postgres' WAL, or a custom
implementation. If WAL is chosen, either Generic WAL Records can be
used, or a new type of WAL records can be implemented. Generic WAL
Records are easy, but imply higher WAL volume. Implementation of a new
type of WAL record currently requires modifications to core code
(specifically, src/include/access/rmgrlist.h)."
--
Bharath Rupireddy
RDS Open Source Databases: https://aws.amazon.com/rds/postgresql/