Horton

Re: GiST VACUUM

x4mmm@yandex-team.ru

View thread View on postgr.es
From: Andrey Borodin <x4mmm@yandex-team.ru>
To: Heikki Linnakangas <hlinnaka@iki.fi>
Cc: Michael Paquier <michael@paquier.xyz>, Thomas Munro <thomas.munro@enterprisedb.com>, Robert Haas <robertmhaas@gmail.com>, pgsql-hackers <pgsql-hackers@postgresql.org>, Костя Кузнецов <chapaev28@ya.ru>
Date: 2019-03-21T16:06:41Z
Lists: pgsql-hackers 

> 21 марта 2019 г., в 2:30, Heikki Linnakangas <hlinnaka@iki.fi> написал(а):
> one remaining issue that needs to be fixed:
> 
> During Hot Standby, the B-tree code writes a WAL reord, when a deleted 
> page is recycled, to prevent the deletion from being replayed too early 
> in the hot standby. See _bt_getbuf() and btree_xlog_reuse_page(). I 
> think we need to do something similar in GiST.
> 
> I'll try fixing that tomorrow, unless you beat me to it. Making the 
> changes is pretty straightforward, but it's a bit cumbersome to test.

I've tried to deal with it and stuck... I think we should make B-tree WAL record for this shared, remove BlockNumber and other unused stuff, leaving just xid and db oid.
And reuse this record for B-tree, GiST and GIN (yeah, it is not checking for that conflict).

Though, I'm not sure it is important for GIN. Scariest thing that can happen: it will return same tid twice. But it is doing bitmap scan, you cannot return same bit twice...

Eventually, hash, spgist and others will have same thing too.

Best regards, Andrey Borodin.

Commits

  1. Use full 64-bit XID for checking if a deleted GiST page is old enough.

  2. Refactor checks for deleted GiST pages.

  3. Delete empty pages during GiST VACUUM.

  4. Scan GiST indexes in physical order during VACUUM.

  5. Prevent GIN deleted pages from being reclaimed too early