Horton

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

Wong, Yi Wen <yiwong@amazon.com>

View thread View on postgr.es
From: "Wong, Yi Wen" <yiwong@amazon.com>
To: Peter Geoghegan <pg@bowt.ie>, "Wood, Dan" <hexpert@amazon.com>
Cc: Michael Paquier <michael.paquier@gmail.com>, Alvaro Herrera <alvherre@2ndquadrant.com>, PostgreSQL Hackers <pgsql-hackers@postgresql.org>
Date: 2017-10-04T03:43:00Z
Lists: pgsql-hackers 
My interpretation of README.HOT is the check is just to ensure the chain is continuous; in which case the condition should be:

>                 if (TransactionIdIsValid(priorXmax) &&
>                         !TransactionIdEquals(priorXmax, HeapTupleHeaderGetRawXmin(htup)))
>                         break;

So the difference is GetRawXmin vs GetXmin, because otherwise we get the FreezeId instead of the Xmin when the transaction happened

The interesting consequence of changing that is the prune seems to get the entire chain altogether with Dan's repro... I've run it a couple of times and have consistently gotten the following page

 lp | t_ctid | lp_off | lp_flags | t_infomask | t_infomask2 
----+--------+--------+----------+------------+-------------
  1 | (0,1)  |   8152 |        1 |       2818 |           3
  2 |        |      7 |        2 |            |            
  3 |        |      0 |        0 |            |            
  4 |        |      0 |        0 |            |            
  5 |        |      0 |        0 |            |            
  6 |        |      0 |        0 |            |            
  7 | (0,7)  |   8112 |        1 |      11010 |       32771
(7 rows)

I've made this change to conditions in both heap_prune_chain and heap_get_root_tuples and this seems to cause things to pass my smoke tests.

I'll look into this deeper tomorrow.

Yi Wen
________________________________________
From: Peter Geoghegan <pg@bowt.ie>
Sent: Tuesday, October 3, 2017 6:19 PM
To: Wood, Dan
Cc: Michael Paquier; Alvaro Herrera; PostgreSQL Hackers; Wong, Yi Wen
Subject: Re: [HACKERS] [COMMITTERS] pgsql: Fix freezing of a dead HOT-updated tuple

On Tue, Oct 3, 2017 at 6:09 PM, Wood, Dan <hexpert@amazon.com> wrote:
> I’ve just started looking at this again after a few weeks break.

>                 if (TransactionIdIsValid(priorXmax) &&
>                         !TransactionIdEquals(priorXmax, HeapTupleHeaderGetXmin(htup)))
>                         break;

> We need to understand why these TXID equal checks exist.  Can we differentiate the cases they are protecting against with the two exceptions I’ve found?

I haven't read your remarks here in full, since I'm about to stop
working for the day, but I will point out that
src/backend/access/heap/README.HOT says a fair amount about this,
under "Abort Cases".


--
Peter Geoghegan

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.