Re: Improve eviction algorithm in ReorderBuffer

From: Jeff Davis <pgsql(at)j-davis(dot)com>
To: Masahiko Sawada <sawada(dot)mshk(at)gmail(dot)com>, Amit Kapila <amit(dot)kapila16(at)gmail(dot)com>
Cc: vignesh C <vignesh21(at)gmail(dot)com>, Peter Smith <smithpb2250(at)gmail(dot)com>, Tomas Vondra <tomas(dot)vondra(at)enterprisedb(dot)com>, "Hayato Kuroda (Fujitsu)" <kuroda(dot)hayato(at)fujitsu(dot)com>, Shubham Khanna <khannashubham1197(at)gmail(dot)com>, Dilip Kumar <dilipbalaut(at)gmail(dot)com>, PostgreSQL Hackers <pgsql-hackers(at)lists(dot)postgresql(dot)org>
Subject: Re: Improve eviction algorithm in ReorderBuffer
Date: 2024-04-03 08:45:55
Message-ID: 12747c15811d94efcc5cda72d6b35c80d7bf3443.camel@j-davis.com
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

On Mon, 2024-04-01 at 12:42 +0900, Masahiko Sawada wrote:
> While reviewing the patches, I realized the comment of
> binearyheap_allocate() should also be updated. So I've attached the
> new patches.

In sift_{up|down}, each loop iteration calls set_node(), and each call
to set_node does a hash lookup. I didn't measure it, but that feels
wasteful.

I don't even think you really need the hash table. The key to the hash
table is a pointer, so it's not really doing anything that couldn't be
done more efficiently by just following the pointer.

I suggest that you add a "heap_index" field to ReorderBufferTXN that
would point to the index into the heap's array (the same as
bh_nodeidx_entry.index in your patch). Each time an element moves
within the heap array, just follow the pointer to the ReorderBufferTXN
object and update the heap_index -- no hash lookup required.

That's not easy to do with the current binaryheap API. But a binary
heap is not a terribly complex structure, so you can just do an inline
implementation of it where sift_{up|down} know to update the heap_index
field of the ReorderBufferTXN.

Regards,
Jeff Davis

In response to

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message Jakub Wartak 2024-04-03 08:58:16 Re: Doc limitation update proposal: include out-of-line OID usage per TOAST-ed columns
Previous Message Alexander Korotkov 2024-04-03 08:42:11 Re: [HACKERS] make async slave to wait for lsn to be replayed