From: Andres Freund <andres@anarazel.de>
To: Michael Paquier <michael@paquier.xyz>
Cc: Julien Rouhaud <rjuju123@gmail.com>
Cc: Justin Pryzby <pryzby@telsasoft.com>
Cc: Masahiko Sawada <masahiko.sawada@2ndquadrant.com>
Cc: Robert Haas <robertmhaas@gmail.com>
Cc: PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>
Cc: Masahiko Sawada <sawada.mshk@gmail.com>
Subject: Re: Online checksums verification in the backend
Date: Thu, 29 Oct 2020 11:17:29 -0700
Message-ID: <20201029181729.2nrub47u7yqncsv7@alap3.anarazel.de> (raw)
In-Reply-To: <20201028050852.GF28445@paquier.xyz>
References: <CAOBaU_bsMjyqpwd5H=U9VkiQQAGaiak2ouD-DRO7849Drj5crw@mail.gmail.com>
<20201019023918.GB9612@paquier.xyz>
<CAOBaU_YZTowoJJ+NzsBieV0K5OvpOS+dAMD1cmz25hxtVZpQqA@mail.gmail.com>
<20201019075248.GD9612@paquier.xyz>
<20201023072832.GE5180@paquier.xyz>
<CAOBaU_bznSonMWqF+N3A8ZOcXx=G4ovwvfgmyES9rmGh5+rV5Q@mail.gmail.com>
<20201023090630.GA26347@paquier.xyz>
<20201027070746.GD28445@paquier.xyz>
<CAOBaU_bM2tCRAOpQyTSaMLT--=YLKGsyZ-tzWTn25kGbaqrevA@mail.gmail.com>
<20201028050852.GF28445@paquier.xyz>
Hi,
On 2020-10-28 14:08:52 +0900, Michael Paquier wrote:
> Thanks for confirming. I have gone through the whole set today,
> splitted the thing into two commits and applied them. We had
> buildfarm member florican complain about a mistake in one of the
> GetDatum() calls that I took care of already, and there is nothing
> else on my radar.
The code does IO while holding the buffer mapping lock. That seems
*entirely* unacceptable to me. That basically locks 1/128 of shared
buffers against concurrent mapping changes, while reading data that is
likely not to be on disk? Seriously?
/* see if the block is in the buffer pool or not */
LWLockAcquire(partLock, LW_SHARED);
buf_id = BufTableLookup(&buf_tag, buf_hash);
if (buf_id >= 0)
{
uint32 buf_state;
/*
* Found it. Now, retrieve its state to know what to do with it, and
* release the pin immediately. We do so to limit overhead as much as
* possible. We keep the shared LWLock on the target buffer mapping
* partition for now, so this buffer cannot be evicted, and we acquire
* an I/O Lock on the buffer as we may need to read its contents from
* disk.
*/
bufdesc = GetBufferDescriptor(buf_id);
LWLockAcquire(BufferDescriptorGetIOLock(bufdesc), LW_SHARED);
buf_state = LockBufHdr(bufdesc);
UnlockBufHdr(bufdesc, buf_state);
/* If the page is dirty or invalid, skip it */
if ((buf_state & BM_DIRTY) != 0 || (buf_state & BM_TAG_VALID) == 0)
{
LWLockRelease(BufferDescriptorGetIOLock(bufdesc));
LWLockRelease(partLock);
return true;
}
/* Read the buffer from disk, with the I/O lock still held */
smgrread(smgr, forknum, blkno, buffer);
LWLockRelease(BufferDescriptorGetIOLock(bufdesc));
}
else
{
/*
* Simply read the buffer. There's no risk of modification on it as
* we are holding the buffer pool partition mapping lock.
*/
smgrread(smgr, forknum, blkno, buffer);
}
The justification in the in-shared-buffers case seems to completely
mis-judge costs too:
* Found it. Now, retrieve its state to know what to do with it, and
* release the pin immediately. We do so to limit overhead as much as
* possible. We keep the shared LWLock on the target buffer mapping
* partition for now, so this buffer cannot be evicted, and we acquire
* an I/O Lock on the buffer as we may need to read its contents from
* disk.
a pin is cheap. Holding the partition lock is not.
Also, using char[BLCKSZ] as a buffer isn't ok. This should use
PGAlignedBlock:
/*
* Use this, not "char buf[BLCKSZ]", to declare a field or local variable
* holding a page buffer, if that page might be accessed as a page and not
* just a string of bytes. Otherwise the variable might be under-aligned,
* causing problems on alignment-picky hardware. (In some places, we use
* this to declare buffers even though we only pass them to read() and
* write(), because copying to/from aligned buffers is usually faster than
* using unaligned buffers.) We include both "double" and "int64" in the
* union to ensure that the compiler knows the value must be MAXALIGN'ed
* (cf. configure's computation of MAXIMUM_ALIGNOF).
*/
typedef union PGAlignedBlock
I think this needs to be quickly reworked or reverted.
Greetings,
Andres Freund
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Reply to all the recipients using the --to and --cc options:
reply via email
To: pgsql-hackers@postgresql.org
Cc: andres@anarazel.de, michael@paquier.xyz, rjuju123@gmail.com, pryzby@telsasoft.com, masahiko.sawada@2ndquadrant.com, robertmhaas@gmail.com, pgsql-hackers@lists.postgresql.org, sawada.mshk@gmail.com
Subject: Re: Online checksums verification in the backend
In-Reply-To: <20201029181729.2nrub47u7yqncsv7@alap3.anarazel.de>
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
This inbox is served by DDX for PostgreSQL; see mirroring instructions
for how to clone and mirror all data and code used for this inbox