From: Fujii Masao <masao.fujii@oss.nttdata.com>
To: pgsql-hackers@postgresql.org
Subject: A bug in LWLOCK_STATS
Date: Wed, 5 Feb 2020 14:43:49 +0900
Message-ID: <26359edb-798a-568f-d93a-6aafac49752d@oss.nttdata.com> (raw)
Hi,
When I compiled PostgreSQL with -DLWLOCK_STATS and tried to check
the statistics of light-weight locks, I observed that more than one
statistics entries were output *for the same backend process and
the same lwlock*. For example, I got the following four statistics
when I checked how the process with PID 81141 processed ProcArrayLock.
This is strange, and IMO only one statistics should be output for
the same backend process and lwlock.
$ grep "PID 81141 lwlock ProcArrayLock" data/log/postgresql-2020-02-05_141842.log
PID 81141 lwlock ProcArrayLock 0x111e87780: shacq 4000 exacq 0 blk 0 spindelay 0 dequeue self 0
PID 81141 lwlock ProcArrayLock 0x111e87780: shacq 2 exacq 0 blk 0 spindelay 0 dequeue self 0
PID 81141 lwlock ProcArrayLock 0x111e87780: shacq 6001 exacq 1 blk 0 spindelay 0 dequeue self 0
PID 81141 lwlock ProcArrayLock 0x111e87780: shacq 5 exacq 1 blk 0 spindelay 0 dequeue self 0
The cause of this issue is that the key variable used for lwlock hash
table was not fully initialized. The key consists of two fields and
they are initialized as follows. But the following 4 bytes allocated
for the alignment was not initialized. So even if the same key was
specified, hash_search(HASH_ENTER) could not find the existing
entry for that key and created new one.
key.tranche = lock->tranche;
key.instance = lock;
Attached is the patch fixing this issue by initializing the key
variable with zero. In the patched version, I confirmed that only one
statistics is output for the same process and the same lwlock.
Also this patch would reduce the volume of lwlock statistics
very much.
This issue was introduced by commit 3761fe3c20. So the patch needs
to be back-patch to v10.
Regards,
--
Fujii Masao
NTT DATA CORPORATION
Advanced Platform Technology Group
Research and Development Headquarters
Attachments:lwlock_stats.patch (text/plain, 480B) downloaddiff --git a/src/backend/storage/lmgr/lwlock.c b/src/backend/storage/lmgr/lwlock.cindex d07ce609d4..4c14e51c67 100644--- a/src/backend/storage/lmgr/lwlock.c+++ b/src/backend/storage/lmgr/lwlock.c@@ -310,6 +310,7 @@ get_lwlock_stats_entry(LWLock *lock)
return &lwlock_stats_dummy;
/* Fetch or create the entry. */
+ MemSet(&key, 0, sizeof(key));
key.tranche = lock->tranche;
key.instance = lock;
lwstats = hash_search(lwlock_stats_htab, &key, HASH_ENTER, &found);
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 from your mail client; the link fills in
the headers below: reply via email
Some clients, webmail especially, drop In-Reply-To and References from
a link, and the reply then starts a new thread: use the mbox below.
To: pgsql-hackers@postgresql.org
Cc: masao.fujii@oss.nttdata.com
Subject: Re: A bug in LWLOCK_STATS
In-Reply-To: <26359edb-798a-568f-d93a-6aafac49752d@oss.nttdata.com>
References: <26359edb-798a-568f-d93a-6aafac49752d@oss.nttdata.com>
* 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