Reinplementation for MR #251

[bz007]

Use CREATE TABLE as SELECT allowing parallel plans if it is possible.

[bz007]

I've reimplemented MR #251. The only problem is with TABLESPACE tests, it is not good.

[levinsv]

Hi!
I applied this patch and performed clustering on a large table of 88 million rows.

env PGOPTIONS="-c work_mem=100MB -c enable_indexscan=off" pg_repack -t past.history__0191_2404 -T 600 -o def,inv,dt -s ts_compress -S

This query plan:

SQL QUERY: CREATE TABLE repack.table_6378674 WITH (fillfactor=95, autovacuum_analyze_threshold=10000, autovacuum_vacuum_scale_factor=0.01, oids = false)  TABLESPACE ts_compress AS SELECT inv,def,dt,quality,value_num,value_dt,value_char FROM ONLY past.history__0191_2404 ORDER BY def,inv,dt
Gather Merge (Current loop: running time=10848.435 actual rows=0, loop number=1)
  Workers Planned: 2
  Workers Launched: 2
  ->  Sort (Current loop: running time=10846.615 actual rows=0, loop number=1)
        Sort Key: def, inv, dt
        ->  Parallel Seq Scan on history__0191_2404 (Current loop: actual time=1.236..7897.329 rows=22189018, loop number=1)
              Buffers: shared hit=30687 read=142746 dirtied=19995
              I/O Timings: shared/local read=2626.642

This top output:

    PID USER      PR  NI    VIRT    RES    SHR S  %CPU  %MEM     TIME+ COMMAND
2938492 postgres  20   0   44.0g   2.6g  10824 R 100.0   2.7   0:30.09 postgres: asu: parallel worker for PID 2938490
2938493 postgres  20   0   44.0g   2.5g  10564 R 100.0   2.6   0:30.17 postgres: asu: parallel worker for PID 2938490
2938490 postgres  20   0   44.0g   2.6g  15000 R  99.7   2.7   0:30.16 postgres: asu: postgres asu [local] CREATE TABLE AS

Great patch. Increases operating speed by 3 times.
I hope the developers will accept it.
The benefit of the patch is obvious.

[levinsv]

Hi!
What should I do to accept this request?