RE: row filtering for logical replication

On Wed, Nov 24, 2021 1:46 PM Amit Kapila <amit(dot)kapila16(at)gmail(dot)com> wrote:
> On Wed, Nov 24, 2021 at 6:51 AM houzj(dot)fnst(at)fujitsu(dot)com
> <houzj(dot)fnst(at)fujitsu(dot)com> wrote:
> >
> > On Tues, Nov 23, 2021 6:16 PM Amit Kapila <amit(dot)kapila16(at)gmail(dot)com> wrote:
> > > On Tue, Nov 23, 2021 at 1:29 PM houzj(dot)fnst(at)fujitsu(dot)com
> > > <houzj(dot)fnst(at)fujitsu(dot)com> wrote:
> > > >
> > > > On Tues, Nov 23, 2021 2:27 PM vignesh C <vignesh21(at)gmail(dot)com> wrote:
> > > > > On Thu, Nov 18, 2021 at 7:04 AM Peter Smith
> > > > > <smithpb2250(at)gmail(dot)com>
> > > > > wrote:
> > > > > >
> > > > > > PSA new set of v40* patches.
> > > > >
> > > > > Few comments:
> > > > > 1) When a table is added to the publication, replica identity is
> > > > > checked. But while modifying the publish action to include
> > > > > delete/update, replica identity is not checked for the existing
> > > > > tables. I felt it should be checked for the existing tables too.
> > > >
> > > > In addition to this, I think we might also need some check to
> > > > prevent user from changing the REPLICA IDENTITY index which is used in
> > > > the filter expression.
> > > >
> > > > I was thinking is it possible do the check related to REPLICA
> > > > IDENTITY in function CheckCmdReplicaIdentity() or In
> > > > GetRelationPublicationActions(). If we move the REPLICA IDENTITY
> > > > check to this function, it would be consistent with the existing
> > > > behavior about the check related to REPLICA IDENTITY(see the
> > > > comments in CheckCmdReplicaIdentity) and seems can cover all the cases
> > > > mentioned above.
> > >
> > > Yeah, adding the replica identity check in CheckCmdReplicaIdentity()
> > > would cover all the above cases but I think that would put a premium
> > > on each update/delete operation. I think traversing the expression
> > > tree (it could be multiple traversals if the relation is part of
> > > multiple publications) during each update/delete would be costly.
> > > Don't you think so?
> >
> > Yes, I agreed that traversing the expression every time would be costly.
> >
> > I thought maybe we can cache the columns used in row filter or cache
> > only the a
> > flag(can_update|delete) in the relcache. I think every operation that
> > affect the row-filter or replica-identity will invalidate the relcache
> > and the cost of check seems acceptable with the cache.
> >
>
> I think if we can cache this information especially as a bool flag then that should
> probably be better.

When researching and writing a top-up patch about this.
I found a possible issue which I'd like to confirm first.

It's possible the table is published in two publications A and B, publication A
only publish "insert" , publication B publish "update". When UPDATE, both row
filter in A and B will be executed. Is this behavior expected?

For example:
---- Publication
create table tbl1 (a int primary key, b int);
create publication A for table tbl1 where (b<2) with(publish='insert');
create publication B for table tbl1 where (a>1) with(publish='update');

---- Subscription
create table tbl1 (a int primary key);
CREATE SUBSCRIPTION sub CONNECTION 'dbname=postgres host=localhost
port=10000' PUBLICATION A,B;

---- Publication
update tbl1 set a = 2;

The publication can be created, and when UPDATE, the rowfilter in A (b<2) will
also been executed but the column in it is not part of replica identity.
(I am not against this behavior just confirm)

Best regards,
Hou zj