Re: partitioning - changing a slot's descriptor is expensive

On 2018/06/29 15:23, Amit Langote wrote:
> Instead of a single TupleTableSlot attached at partition_tuple_slot, we
> allocate an array of TupleTableSlot pointers of same length as the number
> of partitions, as you mentioned upthread.  We then call
> MakeTupleTableSlot() only if a partition needs it and pass it the
> partition's TupleDesc.  Allocated slots are remembered in a list.
> ExecDropSingleTupleTableSlot is then called on those allocated slots when
> the plan ends.  Note that the array of slots is not allocated if none of
> the partitions affected by a given query (or copy) needed to convert tuples.

Forgot to show effect the patch has (on my workstation, so numbers a bit
noisy).

create table p (a int, b int, c int) partition by range (a);
create table p1 (b int, a int, c int);
alter table p attach partition p1 for values from (1) to (maxvalue);

Note that every row will end up into p1 and will require conversion.

-- 2 million records
copy (select i, i+1, i+2 from generate_series(1, 2000000) i) to
'/tmp/data.csv' csv;


Un-patched:

truncate p;
copy p from '/tmp/data.csv' csv;
COPY 2000000
Time: 8521.308 ms (00:08.521)

truncate p;
copy p from '/tmp/data.csv' csv;
COPY 2000000
Time: 8160.741 ms (00:08.161)

truncate p;
copy p from '/tmp/data.csv' csv;
COPY 2000000
Time: 8389.925 ms (00:08.390)


Patched:

truncate p;
copy p from '/tmp/data.csv' csv;
COPY 2000000
Time: 7716.568 ms (00:07.717)

truncate p;
copy p from '/tmp/data.csv' csv;
COPY 2000000
Time: 7569.224 ms (00:07.569)

truncate p;
copy p from '/tmp/data.csv' csv;
COPY 2000000
Time: 7572.085 ms (00:07.572)

So, there is at least some speedup.

Thanks,
Amit