Re: POC: make mxidoff 64 bits

Heikki Linnakangas <hlinnaka@iki.fi>

From: Heikki Linnakangas <hlinnaka@iki.fi>
To: Maxim Orlov <orlovmg@gmail.com>
Cc: wenhui qiu <qiuwenhuifx@gmail.com>, Alexander Korotkov <aekorotkov@gmail.com>, Postgres hackers <pgsql-hackers@lists.postgresql.org>
Date: 2025-01-01T22:12:36Z
Lists: pgsql-hackers

Commits

Same data as JSON: GET /api/v1/messages/:b64id/commits the thread's linked commits as JSON, with link sources. API reference →
  1. Fix partial read handling in pg_upgrade's multixact conversion

  2. Increase timeout in multixid_conversion upgrade test

  3. Improve sanity checks on multixid members length

  4. Clarify comment on multixid offset wraparound check

  5. Never store 0 as the nextMXact

  6. Add runtime checks for bogus multixact offsets

  7. Widen MultiXactOffset to 64 bits

  8. Move pg_multixact SLRU page format definitions to a separate header

  9. Convert confusing macros in multixact.c to static inline functions

  10. Index SLRUs by 64-bit integers rather than by 32-bit integers

  11. Cope with possible failure of the oldest MultiXact to exist.

On 27/12/2024 19:09, Maxim Orlov wrote:
> 
> On Wed, 18 Dec 2024 at 13:21, Heikki Linnakangas <hlinnaka@iki.fi 
> <mailto:hlinnaka@iki.fi>> wrote:
>     Does the pg_upgrade code work though, if you have that buggy situation
>     where oldestOffsetKnown == false ?

...

>      >
>      >               if (!TransactionIdIsValid(*xactptr))
>      >               {
>      >                       /* Corner case 3: we must be looking at
>     unused slot zero */
>      >                       Assert(offset == 0);
>      >                       continue;
>      >               }
> 
>     After upgrade, this corner case 3 would *not* happen on offset == 0. So
>     looks like we're still missing test coverage for this upgrade corner
>     case.
> 
> Am I understanding correctly that you want to have a test corresponding 
> to the buggy 9.3 and 9.4 era versions?

No, those were two different things. I think there might be two things 
wrong here:

1. I suspect pg_upgrade might not correctly handle the situation where 
oldestOffsetKnown==false, and

2. The above assertion in "corner case 3" would not hold. It seems that 
we don't have a test case for it, or it would've hit the assertion.


Now that I think about it, yes, a test case for 1. would be good too. 
But I was talking about 2.

> Do you think we could imitate this scenario on a current master branch 
> like that:
> 1) generate a couple of offsets segments for the first table;
> 2) generate more segments for a second table;
> 3) drop first table;
> 4) stop pg cluster;
> 5) remove pg_multixact/offsets/0000
> 6) upgrade?

I don't remember off the top of my head.

It might be best to just refuse the upgrade if oldestOffsetKnown==false. 
It's a very ancient corner case. It seems reasonable to require you to 
upgrade to a newer minor version and run VACUUM before upgrading. IIRC 
that sets oldestOffsetKnown.

-- 
Heikki Linnakangas
Neon (https://neon.tech)