Re: Speeding up INSERTs and UPDATEs to partitioned tables

On 2018/08/21 14:44, David Rowley wrote:
> On 3 August 2018 at 17:58, Amit Langote <Langote_Amit_f8(at)lab(dot)ntt(dot)co(dot)jp> wrote:
>> On 2018/07/31 16:03, David Rowley wrote:
>>> Maybe we can do that as a follow-on patch.
>>
>> We probably could, but I think it would be a good idea get rid of *all*
>> redundant allocations due to tuple routing in one patch, if that's the
>> mission of this thread and the patch anyway.
>
> I started looking at this patch today and I now agree that it should
> be included in the main patch.

Great, thanks.

> I changed a few things with the patch. For example, the map access
> macros you'd defined were not in CamelCase.

In the updated patch:

+#define PartitionTupRoutingGetToParentMap(p, i) \
+#define PartitionTupRoutingGetToChildMap(p, i) \

If the "Get" could be replaced by "Child" and "Parent", respectively,
they'd sound more meaningful, imho.

> I also fixed a bug where
> the child to parent map was not being initialised when on conflict
> transition capture was required. I added a test which was crashing the
> backend but fixed the code so it works correctly.

Oops, I guess you mean my omission of checking if
mtstate->mt_oc_transition_capture is non-NULL in ExecInitRoutingInfo.

Thanks for fixing it and adding the test case.

> I also got rid of
> the child_parent_map_not_required array since we now no longer need
> it. The code now always initialises the maps in cases where they're
> going to be required.

Yes, thought I had removed the field in my patch, but looks like I had
just removed the comment about it.

> I've attached a v3 version of your patch and also v6 of the main patch
> which includes the v3 patch.

I've looked at v6 and spotted some minor typos.

+ * ResultRelInfo for, before we go making one, we check for a
pre-made one

s/making/make/g

+ /* If nobody else set the per-subplan array of maps, do so ouselves. */

I guess I'm the one to blame here for misspelling "ourselves".

Since the above two are minor issues, fixed them myself in the attached
updated version; didn't touch the macro though.

Do you agree to setting this patch to "Ready for Committer" in the
September CF?

Thanks,
Amit