Horton

Re: Table AM Interface Enhancements

Melanie Plageman <melanieplageman@gmail.com>

View thread View on postgr.es
From: Melanie Plageman <melanieplageman@gmail.com>
To: Andres Freund <andres@anarazel.de>
Cc: Alexander Korotkov <aekorotkov@gmail.com>, Robert Haas <robertmhaas@gmail.com>, Pavel Borisov <pashkin.elfe@gmail.com>, Japin Li <japinli@hotmail.com>, Mark Dilger <mark.dilger@enterprisedb.com>, PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>, Michael Paquier <michael@paquier.xyz>, Heikki Linnakangas <hlinnaka@iki.fi>, Thomas Munro <thomas.munro@gmail.com>, Nazir Bilal Yavuz <byavuz81@gmail.com>
Date: 2024-04-11T16:19:09Z
Lists: pgsql-hackers 
On Wed, Apr 10, 2024 at 5:21 PM Andres Freund <andres@anarazel.de> wrote:
>
> Hi,
>
> On 2024-04-10 16:50:44 -0400, Melanie Plageman wrote:
> > This brings up a question about the prefetching. We never had to have
> > this discussion for sequential scan streaming read because it didn't
> > (and still doesn't) do prefetching. But, if we push the streaming read
> > code down into the heap AM layer, it will be doing the prefetching.
> > So, do we remove the prefetching from acquire_sample_rows() and expect
> > other table AMs to implement it themselves or use the streaming read
> > API?
>
> The prefetching added to acquire_sample_rows was quite narrowly tailored to
> something heap-like - it pretty much required that block numbers to be 1:1
> with the actual physical on-disk location for the specific AM.  So I think
> it's pretty much required for this to be pushed down.
>
> Using a read stream is a few lines for something like this, so I'm not worried
> about it. I guess we could have a default implementation for block based AMs,
> similar what we have around table_block_parallelscan_*, but not sure it's
> worth doing that, the complexity is much lower than in the
> table_block_parallelscan_ case.

This makes sense.

I am working on pushing streaming ANALYZE into heap AM code, and I ran
into a few roadblocks.

If we want ANALYZE to make the ReadStream object in heap_beginscan()
(like the read stream implementation of heap sequential and TID range
scans do), I don't see any way around changing the scan_begin table AM
callback to take a BufferAccessStrategy at the least (and perhaps also
the BlockSamplerData).

read_stream_begin_relation() doesn't just save the
BufferAccessStrategy in the ReadStream, it uses it to set various
other things in the ReadStream object. callback_private_data (which in
ANALYZE's case is the BlockSamplerData) is simply saved in the
ReadStream, so it could be set later, but that doesn't sound very
clean to me.

As such, it seems like a cleaner alternative would be to add a table
AM callback for creating a read stream object that takes the
parameters of read_stream_begin_relation(). But, perhaps it is a bit
late for such additions.

It also opens us up to the question of whether or not sequential scan
should use such a callback instead of making the read stream object in
heap_beginscan().

I am happy to write a patch that does any of the above. But, I want to
raise these questions, because perhaps I am simply missing an obvious
alternative solution.

- Melanie

Commits

  1. Remove extra comment at TableAmRoutine.scan_analyze_next_block

  2. revert: Generalize relation analyze in table AM interface

  3. Revert: Allow table AM to store complex data structures in rd_amcache

  4. Revert: Allow table AM tuple_insert() method to return the different slot

  5. Revert: Allow locking updated tuples in tuple_update() and tuple_delete()

  6. Revert: Let table AM insertion methods control index insertion

  7. Revert: Custom reloptions for table AM

  8. Provide a way block-level table AMs could re-use acquire_sample_rows()

  9. Custom reloptions for table AM

  10. Use streaming I/O in ANALYZE.

  11. Revert "Custom reloptions for table AM"

  12. Let table AM insertion methods control index insertion

  13. Generalize relation analyze in table AM interface

  14. Improve error message for tts_(virtual|minimal)_is_current_xact_tuple

  15. Add comments on some MinimalTupleSlots methods usage

  16. Add TupleTableSlotOps.is_current_xact_tuple() method

  17. Allow table AM tuple_insert() method to return the different slot

  18. Allow table AM to store complex data structures in rd_amcache