Horton

Re: [WIP]Vertical Clustered Index (columnar store extension) - take2

Timur Magomedov <t.magomedov@postgrespro.ru>

View thread
From: Timur Magomedov <t.magomedov@postgrespro.ru>
To: Peter Smith <smithpb2250@gmail.com>, "pgsql-hackers@lists.postgresql.org" <pgsql-hackers@lists.postgresql.org>
Cc: Tomas Vondra <tomas@vondra.me>, "Aya Iwata (Fujitsu)" <iwata.aya@fujitsu.com>, Japin Li <japinli@hotmail.com>, shveta malik <shveta.malik@gmail.com>
Date: 2025-09-17T15:15:00Z
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. Mark ItemPointer arguments as const throughout

  2. Reorganize GUC structs

  3. Eliminate XLOG_HEAP2_VISIBLE from vacuum phase III

  4. Update various forward declarations to use typedef

  5. Pathify RHS unique-ification for semijoin planning

  6. Revert "Don't lock partitions pruned by initial pruning"

  7. For inplace update, send nontransactional invalidations.

  8. Remove nearly-unused SizeOfIptrData macro.

Attachments

Hello Peter!


I've found (using valgrind) some cases of reading random garbage after
allocated memory. Investigation showed this was caused by converting
some nodes to VciScanState* even if they have smaller size allocated
originally. So accessing VciScanState fields was accessing memory after
palloc'ed memory which could be used by any other allocation.

I think converting to VciScanState* is only valid for nodes with tag 
T_CustomScanState so here is a patch that adds assertions for this:
0001-Assert-corrrect-node-tags-when-casting-to-VciScanSta.patch

VCI v23 passes the tests with this patch applied.

There are queries that fail unfortunately. I've added one of them to
bugs.sql:
0002-Reproducer-of-invalid-cast-to-VciScanState.patch
Node with tag 420 (T_NestLoopState) is cast to VciScanState* that fails
newly added assertion.

I'm not sure where to look next to fix this. Looking forward for you
comments and ideas.

-- 
Regards,
Timur Magomedov