Horton
Menu

Re: eliminate xl_heap_visible to reduce WAL (and eventually set VM on-access)

Melanie Plageman <melanieplageman@gmail.com>

View thread
From: Melanie Plageman <melanieplageman@gmail.com>
To: David Rowley <dgrowleyml@gmail.com>
Cc: Tomas Vondra <tomas@vondra.me>, Andres Freund <andres@anarazel.de>, Kirill Reshke <reshkekirill@gmail.com>, Chao Li <li.evan.chao@gmail.com>, Andrey Borodin <x4mmm@yandex-team.ru>, Xuneng Zhou <xunengzhou@gmail.com>, Robert Haas <robertmhaas@gmail.com>, PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>, Heikki Linnakangas <hlinnaka@iki.fi>
Date: 2026-03-27T19:17:45Z
Lists: pgsql-hackers

Attachments

On Thu, Mar 26, 2026 at 7:10 PM David Rowley <dgrowleyml@gmail.com> wrote:
>
> I was looking at v46-0001. With:

Thanks for taking a look!

> +++ b/src/include/nodes/plannodes.h
> + Bitmapset  *modifiedRelids;
> +
>
> This doesn't really mention anything about leaf partitions not being
> mentioned for INSERT queries. You did mention it in standard_planner()
> here:
>
> I'm also wondering about having this combined field. If you were to
> have a Bitmapset field that mirrors "List *resultRelations;", then
> have another:
>
> /* a list of PlanRowMark's */
> List   *rowMarks;
>
> + /* Relids which have rowMarks */
> + Bitmapset *rowMarkRelids;
>
> I think they're more likely to be useful for other purposes, and I
> think the only pain that it causes you is that you have to call
> bms_is_member() twice in ScanRelIsReadOnly().

Yea, outside of the insert into leaf partitions case, I thought of
another, perhaps even more compelling reason the combined field might
be a bit confusing:
Take a table t1 and a table t2. If you do
SELECT * FROM t1 JOIN t2 ON t1.id  = t2.id FOR UPDATE of t1;
t1 will get a ROW_MARK_EXCLUSIVE and t2 would get a ROW_MARK_REFERENCE
(that's just how preprocess_rowmarks() works).
That means modifiedRelids would contain t2, even though t2 is not
being locked for update. For the purposes of setting the VM, it's
totally fine that we are more conservative than we need to be and
don't consider setting it when scanning t2. But for the purposes of
modifiedRelids, it's a bit confusing that t2 is in there.

But we can't just exclude ROW_MARK_REFERENCE from modifiedRelids
because we rely on ROW_MARK_REFERENCE to avoid setting the VM for a
table we are updating or deleting from when it is mentioned more than
once in the query (e.g. UPDATE foo SET x = 1 FROM foo f2 WHERE foo.id
= f2.id).

So, for that reason and because of the missing leaf partitions for
inserts, I think making quick reference bitmapsets would be better.
I've done this in attached v47.

I've also removed the asserts in ExecInsert/Update/Delete because they
are a bit tautological now.

My one remaining question is whether the two new bitmapsets
(rowMarkRelids and resultRelationRelids) should move from the
PlannedStmt to the EState. They are determined at plan time and never
modified during execution. However, I do notice there are other EState
members that seem like just a copy of info from the PlannedStmt that
isn't modified during execution (e.g. es_rteperminfos/permInfos).
However, putting them in the EState increases the work required to get
them to parallel workers and to the child estate for EPQs. I would
prefer to keep it in the PlannedStmt but am worried that breaks
convention.

> Then, as a follow-up, maybe we could consider removing
> PlannedStmt.resultRelations.  (The deprecated)
> ExecRelationIsTargetRelation() could use the new Bitmapset, which
> would be more efficient.

Yea, I like this and think it makes sense. Done in v47.

> I did quickly look over the remaining patches. I wondered if you might
> want to add a new ScanOption SO_NONE = 0, or SO_EMPTY_FLAGS. It might
> make the places where you're passing zero directly easier to read?

That makes sense to me. Done in v47.

- Melanie

Commits

Same data as JSON: GET /api/v1/messages/:b64id/commits the thread's linked commits as JSON, with link sources. API reference →

  1. Stabilize btree_gist test against on-access VM setting

  2. Stabilize plancache test against on-access VM setting

  3. Make local buffers pin limit more conservative

  4. Update FSM during prune/freeze replay even if freespace is zero

  5. Pass down information on table modification to scan nodes

  6. Thread flags through begin-scan APIs

  7. Remove PlannedStmt->resultRelations in favor of resultRelationRelids

  8. Make it cheap to check if a relation is modified by a query

  9. Keep newest live XID up-to-date even if page not all-visible

  10. Add pruning fast path for all-visible and all-frozen pages

  11. Detect and fix visibility map corruption in more cases

  12. Save vmbuffer in heap-specific scan descriptors for on-access pruning

  13. Avoid BufferGetPage() calls in heap_update()

  14. Initialize missing fields in CreateExecutorState()

  15. Prefix PruneState->all_{visible,frozen} with set_

  16. Add PageGetPruneXid() helper

  17. Move commonly used context into PruneState and simplify helpers

  18. Remove table_scan_analyze_next_tuple unneeded parameter OldestXmin

  19. Simplify visibility check in heap_page_would_be_all_visible()

  20. Rename LVRelState VM-related logging counters

  21. Fix flakiness in the pg_visibility VM-only vacuum test by using a temporary table.

  22. Refactor lazy_scan_prune() VM clear logic into helper

  23. Eliminate use of cached VM value in lazy_scan_prune()

  24. Combine visibilitymap_set() cases in lazy_scan_prune()

  25. Fix const qualification in prune_freeze_setup()

  26. Simplify vacuum visibility assertion

  27. Split heap_page_prune_and_freeze() into helpers

  28. Assert that cutoffs are provided if freezing will be attempted

  29. Split PruneFreezeParams initializers to one field per line

  30. Refactor heap_page_prune_and_freeze() parameters into a struct

  31. Make heap_page_is_all_visible independent of LVRelState

  32. Inline TransactionIdFollows/Precedes[OrEquals]()

  33. Add helper for freeze determination to heap_page_prune_and_freeze

  34. Bump XLOG_PAGE_MAGIC after xl_heap_prune change

  35. Correct prune WAL record opcode name in comment

  36. Add error codes when vacuum discovers VM corruption

  37. Remove unused xl_heap_prune member, reason

  38. Remove unneeded VM pin from VM replay

  39. Add assert and log message to visibilitymap_set

  40. Stamp HEAD as 19devel.

  41. Add error codes to some corruption log messages