Re: Implementing Incremental View Maintenance

On Thu, Feb 3, 2022 at 8:50 AM Yugo NAGATA <nagata(at)sraoss(dot)co(dot)jp> wrote:

> On Thu, 3 Feb 2022 08:48:00 -0800
> Zhihong Yu <zyu(at)yugabyte(dot)com> wrote:
>
> > On Thu, Feb 3, 2022 at 8:28 AM Yugo NAGATA <nagata(at)sraoss(dot)co(dot)jp> wrote:
> >
> > > Hi,
> > >
> > > On Thu, 13 Jan 2022 18:23:42 +0800
> > > Julien Rouhaud <rjuju123(at)gmail(dot)com> wrote:
> > >
> > > > Hi,
> > > >
> > > > On Thu, Nov 25, 2021 at 04:37:10PM +0900, Yugo NAGATA wrote:
> > > > > On Wed, 24 Nov 2021 04:31:25 +0000
> > > > > "r(dot)takahashi_2(at)fujitsu(dot)com" <r(dot)takahashi_2(at)fujitsu(dot)com> wrote:
> > > > >
> > > > > >
> > > > > > I checked the same procedure on v24 patch.
> > > > > > But following error occurs instead of the original error.
> > > > > >
> > > > > > ERROR: relation "ivm_t_index" already exists
> > > > >
> > > > > Thank you for pointing out it!
> > > > >
> > > > > Hmmm, an index is created when IMMV is defined, so CREAE INDEX
> called
> > > > > after this would fail... Maybe, we should not create any index
> > > automatically
> > > > > if IMMV is created WITH NO DATA.
> > > > >
> > > > > I'll fix it after some investigation.
> > > >
> > > > Are you still investigating on that problem? Also, the patchset
> doesn't
> > > apply
> > > > anymore:
> > >
> > > I attached the updated and rebased patch set.
> > >
> > > I fixed to not create a unique index when an IMMV is created WITH NO
> DATA.
> > > Instead, the index is created by REFRESH WITH DATA only when the same
> one
> > > is not created yet.
> > >
> > > Also, I fixed the documentation to describe that foreign tables and
> > > partitioned
> > > tables are not supported according with Takahashi-san's suggestion.
> > >
> > > > There isn't any answer to your following email summarizing the
> feature
> > > yet, so
> > > > I'm not sure what should be the status of this patch, as there's no
> ideal
> > > > category for that. For now I'll change the patch to Waiting on
> Author
> > > on the
> > > > cf app, feel free to switch it back to Needs Review if you think it's
> > > more
> > > > suitable, at least for the design discussion need.
> > >
> > > I changed the status to Needs Review.
> > >
> > >
> > > Hi,
> > Did you intend to attach updated patch ?
> >
> > I don't seem to find any.
>
> Oops, I attached. Thanks!
>
> Hi,
For CreateIndexOnIMMV():

+ ereport(NOTICE,
+ (errmsg("could not create an index on materialized view
\"%s\" automatically",
...
+ return;
+ }

Should the return type be changed to bool so that the caller knows whether
the index creation succeeds ?
If index creation is unsuccessful, should the call
to CreateIvmTriggersOnBaseTables() be skipped ?

For check_ivm_restriction_walker():

+ break;
+ expression_tree_walker(node, check_ivm_restriction_walker,
NULL);
+ break;

Something is missing between the break and expression_tree_walker().

Cheers