Re: Parallel INSERT (INTO ... SELECT ...)

On Fri, Feb 19, 2021 at 6:56 AM Greg Nancarrow <gregn4422(at)gmail(dot)com> wrote:
>
> Posting a new version of the patches, with the following updates:
>

I am not happy with the below code changes, I think we need a better
way to deal with this.

@@ -313,19 +314,35 @@ standard_planner(Query *parse, const char
*query_string, int cursorOptions,
glob->transientPlan = false;
glob->dependsOnRole = false;

+ if (IsModifySupportedInParallelMode(parse->commandType) &&
+ !parse->hasModifyingCTE)
+ {
+ /*
+ * FIXME
+ * There is a known bug in the query rewriter: re-written queries with
+ * a modifying CTE may not have the "hasModifyingCTE" flag set. When
+ * that bug is fixed, this temporary fix must be removed.
+ *
+ * Note that here we've made a fix for this problem only for a
+ * supported-in-parallel-mode table-modification statement (i.e.
+ * INSERT), but this bug exists for SELECT too.
+ */
+ parse->hasModifyingCTE = query_has_modifying_cte(parse);
+ }
+

I understand that this is an existing bug but I am not happy with this
workaround. I feel it is better to check for modifyingCTE in
max_parallel_hazard_walker. See attached, this is atop
v18-0001-Enable-parallel-SELECT-for-INSERT-INTO-.-SELECT.

--
With Regards,
Amit Kapila.