On 05/04/2016 01:28 AM, Klaus P. wrote:
> In order to improve cpu and disk utilization, I am testing parallel
> queries.
>
> The approach with dblink_send_query() and dblink_get_result() works in
> my proof-of-concept. Runtime of my reference query was reduced from 55
> seconds to ~20seconds using 4 parallel connections. Not what I had hoped
> but certainly a significant improvement.
>
> My approach is fairly easy:
>
> db_link_send_query('conn1', 'statement based on partitioning field');
> db_link_send_query('conn2', 'statement based on partitioning field');
>
> ...
>
> SELECT
> dblink_get_result('conn1')
> UNION ALL
> dblink_get_result('conn2')
>
> ...
>
>
> However, using fdw foreign data wrappers, I was not able to run any
> query in parallel. I came across this presentation
> http://de.slideshare.net/EnterpriseDB/foreign-data-wrappers-and-you-with-postgres
> where it says on page 12 "parallel FDW access". Is there any example
> available on how to run FDW queries in parallel?
>
> My approach with FDW is similar to the dblink example above:
>
> SELECT * FROM myFdwTable WHERE (clause based on partitioning field)
> UNION ALL
> SELECT * FROM myFdwTable WHERE (clause based on partitioning field)
> ...
>
> My experience is however that the statements are carried out sequentially.
I am not sure of the status of parallel query in FDW, but for those that
are some more information would be helpful:
1) You are using postgres_fdw, correct?
2) What version of the FDW are you using?
3) What version(s) of Postgres are you connecting from/to?
>
> Thanks
>
> Klaus Pieper
--
Adrian Klaver
adrian.klaver@aklaver.com
