RE: Parallel INSERT (INTO ... SELECT ...)

From: Tang, Haiying <tanghy(dot)fnst(at)cn(dot)fujitsu(dot)com>
> Execute EXPLAIN on Patched:
> Insert on public.test_part (cost=0.00..15.00 rows=0 width=0) (actual
> time=44.139..44.140 rows=0 loops=1)
> Buffers: shared hit=1005 read=1000 dirtied=3000 written=2000
> -> Seq Scan on public.test_data1 (cost=0.00..15.00 rows=1000
> width=8) (actual time=0.007..0.201 rows=1000 loops=1)
> Output: test_data1.a, test_data1.b
> Buffers: shared hit=5

> Execute EXPLAIN on non-Patched:
> Insert on public.test_part (cost=0.00..15.00 rows=0 width=0) (actual
> time=72.656..72.657 rows=0 loops=1)
> Buffers: shared hit=22075 read=1000 dirtied=3000 written=2000
> -> Seq Scan on public.test_data1 (cost=0.00..15.00 rows=1000
> width=8) (actual time=0.010..0.175 rows=1000 loops=1)
> Output: test_data1.a, test_data1.b
> Buffers: shared hit=5

I don't know if this is related to this issue, but I felt "shared hit=5" for Seq Scan is strange. This test case reads 1,000 rows from 1,000 partitions, one row per partition, so I thought the shared hit should be 1,000 in Seq Scan. I wonder if the 1,000 is included in Insert node?

Regards
Takayuki Tsunakawa