On Tue, Apr 23, 2019 at 07:09:00PM -0400, Gunther wrote:
On 4/23/2019 16:43, Justin Pryzby wrote:
It wrote 40GB tempfiles - perhaps you can increase work_mem now to improve the
query time.
I now upped my shared_buffers back from 1 to 2GB and work_mem from 4 to
16MB. Need to set vm.overcom
On 4/23/2019 16:43, Justin Pryzby wrote:
It wrote 40GB tempfiles - perhaps you can increase work_mem now to improve the
query time.
I now upped my shared_buffers back from 1 to 2GB and work_mem from 4 to
16MB. Need to set vm.overcommit_ratio from 50 to 75 (percent, with
vm.overcommit_memory =
On Tue, Apr 23, 2019 at 04:37:50PM -0400, Gunther wrote:
On 4/21/2019 23:09, Tomas Vondra wrote:
What I think might work better is the attached v2 of the patch, with a
single top-level condition, comparing the combined memory usage
(spaceUsed + BufFile) against spaceAllowed. But it
On Tue, Apr 23, 2019 at 03:43:48PM -0500, Justin Pryzby wrote:
On Tue, Apr 23, 2019 at 04:37:50PM -0400, Gunther wrote:
On 4/21/2019 23:09, Tomas Vondra wrote:
>What I think might work better is the attached v2 of the patch, with a
Thanks for this, and I am trying this now.
...
Aand, it's
On Tue, Apr 23, 2019 at 04:37:50PM -0400, Gunther wrote:
> On 4/21/2019 23:09, Tomas Vondra wrote:
> >What I think might work better is the attached v2 of the patch, with a
> Thanks for this, and I am trying this now.
...
> Aand, it's a winner!
>
> Unique (cost=5551524.36..5554207.33 rows=3461
On 4/21/2019 23:09, Tomas Vondra wrote:
What I think might work better is the attached v2 of the patch, with a
single top-level condition, comparing the combined memory usage
(spaceUsed + BufFile) against spaceAllowed. But it also tweaks
spaceAllowed once the size needed for BufFile gets over wor
