On Mon, May 15, 2017 at 7:36 PM, Haribabu Kommi
<kommi.haribabu@gmail.com> wrote:
>
> The wait of the workers to send tuples is may be
> because of less number of workers. So increasing
> the parallel workers may improve the performance.
Yes, for the cases where a significant amount of work is still
pending, but it might hurt the performance where the work is about to
complete but not yet completed.
@@ -346,6 +346,38 @@ gather_readnext(GatherState *gatherstate) if (nvisited >= gatherstate->nreaders) { /*
+ * As we are going to wait for the workers to send tuples, this may be
+ * possible because of not sufficient workers that are planned?
+ * Does the gather have all the required parallel workers? If not try to get
+ * some more workers (only when all the previously allocated workers are still
+ * doing the job) before we wait, this will further increase the performance
+ * of the query as planned.
+ */
You might want to do similar changes for gather_merge_readnext.
--
Regards,
Dilip Kumar
EnterpriseDB: http://www.enterprisedb.com
