Re: pg_upgrade bug found!

Kevin Grittner wrote:
> Bruce Momjian <bruce@momjian.us> wrote:
>  
> > all we need to do is set those hint bits before the clog gets
> > remove, so maybe just a SELECT * would do the trick!
>  
> Does that mean that those experiencing the problem are failing to do
> the vacuumdb run which is recommended in the pg_upgrade instructions?

You know, I looked at that, but I don't think that is going to save me. 
:-(   It says:

	Upgrade complete
	----------------
	| Optimizer statistics are not transferred by pg_upgrade
	| so consider running:
	|       vacuumdb --all --analyze-only
	| on the newly-upgraded cluster.
	
	| Running this script will delete the old cluster's data files:
	|       /usr/var/local/pgdev/pgfoundry/pg_migrator/pg_migrator/delete_old_cluster.sh

We recommend 'vacuumdb --all --analyze-only' which I assume only samples
random pages and does not set all the hint bits.  In fact, you can't
even analyze TOAST tables:

	test=> ANALYZE pg_toast.pg_toast_3596;
	WARNING:  skipping "pg_toast_3596" --- cannot analyze non-tables or
	special system tables
	ANALYZE

but you can SELECT from them:

	 chunk_id | chunk_seq | chunk_data
	----------+-----------+------------
	(0 rows)

Also, if we force VACUUM FREEZE on the toast tables we would have no
need to advance their relfrozenxids because all the xids would be fixed.

-- 
  Bruce Momjian  <bruce@momjian.us>        http://momjian.us
  EnterpriseDB                             http://enterprisedb.com

  + It's impossible for everything to be true. +