Re: Concurrency bug in UPDATE of partition-key

On Tue, Jun 5, 2018 at 8:03 PM, Amit Khandekar <amitdkhan(dot)pg(at)gmail(dot)com>
wrote:

> Attached is a rebased patch version. Also included it in the upcoming
> commitfest :
> https://commitfest.postgresql.org/18/1660/
>
>
Doesn't this belong to PostgreSQL 11 Open Items [1] or are you proposing it
as a feature enhancement for next version?

> In the rebased version, the new test cases are added in the existing
> isolation/specs/partition-key-update-1.spec test.
>
>
if (!tuple_deleted)
- return NULL;
+ {
+ /*
+ * epqslot will be typically NULL. But when ExecDelete() finds
+ * that another transaction has concurrently updated the same
+ * row, it re-fetches the row, skips the delete, and epqslot is
+ * set to the re-fetched tuple slot. In that case, we need to
+ * do all the checks again.
+ */
+ if (TupIsNull(epqslot))
+ return NULL;
+ else
+ {
+ slot = ExecFilterJunk(resultRelInfo->ri_junkFilter, epqslot);
+ tuple = ExecMaterializeSlot(slot);
+ goto lreplace;
+ }
+ }

I think this will allow before row delete triggers to be fired more than
once. Normally, if the EvalPlanQual testing generates a new tuple, we
don't fire the triggers again. Now, one possibility could be that we don't
skip the delete after a concurrent update and still allow inserts to use a
new tuple generated by EvalPlanQual testing of delete. However, I think we
need to perform rechecks for update to check if the modified tuple still
requires to be moved to new partition, right or do you have some other
reason in mind?

[1] - https://wiki.postgresql.org/wiki/PostgreSQL_11_Open_Items

--
With Regards,
Amit Kapila.
EnterpriseDB: http://www.enterprisedb.com