From: Tom Lane <tgl@sss.pgh.pa.us>
To: Kyotaro Horiguchi <horikyota.ntt@gmail.com>
cc: bossartn@amazon.com, michael@paquier.xyz, andres@anarazel.de,
        sulamul@gmail.com, pgsql-hackers@postgresql.org
Subject: Re: inefficient loop in StandbyReleaseLockList()
In-reply-to: <20211102.114313.713245704579079227.horikyota.ntt@gmail.com>
References: <1138901.1635782315@sss.pgh.pa.us>
 <C84534A6-2E66-47FF-B193-EBB566544D49@amazon.com>
 <1311319.1635804078@sss.pgh.pa.us>
 <20211102.114313.713245704579079227.horikyota.ntt@gmail.com>
Comments: In-reply-to Kyotaro Horiguchi <horikyota.ntt@gmail.com>
	message dated "Tue, 02 Nov 2021 11:43:13 +0900"
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-ID: <1376383.1635862908.1@sss.pgh.pa.us>
Date: Tue, 02 Nov 2021 10:21:48 -0400
Message-ID: <1376384.1635862908@sss.pgh.pa.us>
Archived-At: 
 <https://www.postgresql.org/message-id/1376384.1635862908%40sss.pgh.pa.us>
Precedence: bulk

Kyotaro Horiguchi <horikyota.ntt@gmail.com> writes:
> At Mon, 01 Nov 2021 18:01:18 -0400, Tom Lane <tgl@sss.pgh.pa.us> wrote in 
>> So what I did in the attached is add a "canceled" flag to
>> PendingUnlinkEntry, which lets us deal with canceled or finished
>> entries without having to delete them from the list right away.
>> Then we only need to physically clean up the list once per
>> SyncPostCheckpoint call.

> We don't loop over so many canceled elements usually so I think it
> works well. However, shouldn't we consider canceled before checking
> cycle_ctr?

Good point.  I was thinking that it's best to break out of the loop
at the first opportunity.  But if the first few entries with the next
cycle_ctr value are canceled, it's best to advance over them in the
current SyncPostCheckpoint call.  It saves nothing to postpone that
work to later, and indeed adds a few cycles by leaving more data to
be copied by list_delete_first_n.  Will change it.

> I feel that we might need to wipe_mem for the memmove case as well
> (together with list_delete_nth_cell) but that is another thing even if
> that's correct.

Hm.  I made this function by copying-and-modifying list_delete_nth_cell,
so if there's something wrong there then this code inherited it.  But
I don't think it's wrong.  The wipe_mem business is only intended to
be used when enabling expensive debug options.

> Otherwise it looks good to me.

Thanks for looking!

			regards, tom lane