Horton

Re: [HACKERS] [COMMITTERS] pgsql: Fix freezing of a dead HOT-updated tuple

Alvaro Herrera <alvherre@2ndquadrant.com>

View thread View on postgr.es
From: Alvaro Herrera <alvherre@2ndQuadrant.com>
To: Andres Freund <andres@anarazel.de>
Cc: "Wood, Dan" <hexpert@amazon.com>, Peter Geoghegan <pg@bowt.ie>, pgsql-committers@postgresql.org, "Wong, Yi Wen" <yiwong@amazon.com>, Michael Paquier <michael.paquier@gmail.com>, Robert Haas <robertmhaas@gmail.com>, Tom Lane <tgl@sss.pgh.pa.us>, pgsql-hackers@postgresql.org
Date: 2017-12-06T16:21:15Z
Lists: pgsql-hackers

Attachments

I think you've done a stellar job of identifying what the actual problem
was.  I like the new (simpler) coding of that portion of
HeapTupleSatisfiesVacuum.

freeze-the-dead is not listed in isolation_schedule; an easy fix.
I confirm that the test crashes with an assertion failure without the
code fix, and that it doesn't with it.

I think the comparison to OldestXmin should be reversed:

			if (!TransactionIdPrecedes(xmax, OldestXmin))
				return HEAPTUPLE_RECENTLY_DEAD;

			return HEAPTUPLE_DEAD;

This way, an xmax that has exactly the OldestXmin value will return
RECENTLY_DEAD rather DEAD, which seems reasonable to me (since
OldestXmin value itself is supposed to be still possibly visible to
somebody).  Also, this way it is consistent with the other comparison to
OldestXmin at the bottom of the function.  There is no reason for the
"else" or the extra braces.

Put together, I propose the attached delta for 0001.

Your commit message does a poor job of acknowledging prior work on
diagnosing the problem starting from Dan's initial test case and patch.

I haven't looked at your 0002 yet.

-- 
Álvaro Herrera                https://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services

Commits

  1. Backport addition of rs_old_rel to rewriteheap's state.

  2. Perform a lot more sanity checks when freezing tuples.

  3. Revert bogus fixes of HOT-freezing bug

  4. Fix traversal of half-frozen update chains

  5. Fix freezing of a dead HOT-updated tuple

  6. During index build, check and elog (not just Assert) for broken HOT chain.

  7. Fix WAL replay of locking an updated tuple

  8. Change the way we mark tuples as frozen.