Horton

Re: Parallel Append subplan order instability on aye-aye

Andres Freund <andres@anarazel.de>

View thread View on postgr.es
From: Andres Freund <andres@anarazel.de>
To: Tom Lane <tgl@sss.pgh.pa.us>
Cc: Thomas Munro <thomas.munro@gmail.com>, David Rowley <david.rowley@2ndquadrant.com>, pgsql-hackers <pgsql-hackers@postgresql.org>
Date: 2019-07-16T19:23:44Z
Lists: pgsql-hackers 
Hi,

On 2019-07-15 21:12:32 -0400, Tom Lane wrote:
> But I bet that these tables forming
> an inheritance hierarchy (with multiple inheritance even) does
> have something to do with it somehow, because if this were a
> generic VACUUM bug surely we'd be seeing it elsewhere.

It's possible that it's hidden in other cases, because of

void
table_block_relation_estimate_size(Relation rel, int32 *attr_widths,
								   BlockNumber *pages, double *tuples,
								   double *allvisfrac,
								   Size overhead_bytes_per_tuple,
								   Size usable_bytes_per_page)
...
	 * If the table has inheritance children, we don't apply this heuristic.
	 * Totally empty parent tables are quite common, so we should be willing
	 * to believe that they are empty.
	 */
	if (curpages < 10 &&
		relpages == 0 &&
		!rel->rd_rel->relhassubclass)
		curpages = 10;

which'd not make us actually take a relpages=0 into account for tables
without inheritance.  A lot of these tables never get 10+ pages long, so
the heuristic would always apply...

Greetings,

Andres Freund

Commits

  1. ANALYZE a_star and its children to avoid plan instability in tests.

  2. Finish reverting "Insert temporary debugging output in regression tests."

  3. Partially revert "Insert temporary debugging output in regression tests."

  4. Insert temporary debugging output in regression tests.

  5. Improve the heuristic for ordering child paths of a parallel append.