Skip site navigation (1) Skip section navigation (2)

Re: ERROR: canceling query due to user request

From: "Kevin Grittner" <Kevin(dot)Grittner(at)wicourts(dot)gov>
To: <tgl(at)sss(dot)pgh(dot)pa(dot)us>
Cc: <pgsql-admin(at)postgresql(dot)org>
Subject: Re: ERROR: canceling query due to user request
Date: 2005-09-13 20:14:37
Message-ID: s326eceb.025@gwmta.wicourts.gov (view raw or flat)
Thread:
Lists: pgsql-adminpgsql-jdbc
Thanks, Tom.  An excellent suggestion.  (This 50 year old dog has
today learned a new trick.)
 
There is good news and bad news.  The good news is that I found
the cause, and we can keep this from happening with a change on
our end.  The bad news is that I think it also points to a backend bug,
although not as serious as the potential (hypothetical) one I was
asking about earlier.
 
For the record (and the benefit of anyone with similar problems who
may search the archives), I got the connection pool set up, and found
the pids for the connection processes like this:
 
# ps aux|grep postgres
postgres 18307  0.0  0.0   3052  1100 pts/0    S    Sep09   0:00 su postgres
postgres 18308  0.0  0.0   3408  1828 pts/0    S+   Sep09   0:00 bash
postgres 15463  0.0  0.1  89508  3852 pts/0    S    10:09   0:05 /usr/local/pgsql/bin/postmaster -D /var/pgsql/data
postgres 15466  0.0  4.0  89652 83004 pts/0    S    10:09   0:00 postgres: writer process
postgres 15467  0.0  0.1   7052  2796 pts/0    S    10:09   0:02 postgres: stats buffer process
postgres 15468  0.0  0.0   6388  1996 pts/0    S    10:09   0:03 postgres: stats collector process
postgres   926  0.0  0.3  90484  7576 pts/0    S    14:34   0:00 postgres: dtr dtr 165.219.88.77(4436) idle
postgres   927  0.0  0.2  89956  4684 pts/0    S    14:34   0:00 postgres: dtr dtr 165.219.88.77(4437) idle
postgres   928  3.0  1.1  90404 23764 pts/0    S    14:34   0:07 postgres: dtr dtr 165.219.88.77(4438) idle
postgres   929  2.7  1.1  90468 23260 pts/0    S    14:34   0:07 postgres: dtr dtr 165.219.88.77(4439) idle
root       935  0.0  0.0   1796   648 pts/2    S+   14:38   0:00 grep postgres
 
I then established an strace session for each connection like this:
 
strace -tt -o strace.926 -p 926
 
Other flags may have been useful, but strace is new to me, so I took
the route I was sure I understood.  I ran until we got an error, which
involved running through 7,278 transactions.  I used Ctrl+C to stop
each strace and searched the results for SIGINT.  There were 1,799
of them.  They always came in a set of lines like this:
 
13:59:19.625498 recv(7, "P\0\0\0Y\0SELECT lcmtr.\"relationName"..., 8192, 0) = 125
13:59:19.625976 _llseek(32, 0, [0], SEEK_END) = 0
13:59:19.626057 _llseek(33, 0, [8192], SEEK_END) = 0
13:59:19.626159 _llseek(32, 0, [0], SEEK_END) = 0
13:59:19.626257 send(7, "1\0\0\0\0042\0\0\0\4T\0\0\0D\0\2relationName\0\0\0"..., 97, 0) = 97
13:59:19.626352 recv(7, 0x82b1000, 8192, 0) = ? ERESTARTSYS (To be restarted)
13:59:19.628477 --- SIGINT (Interrupt) @ 0 (0) ---
13:59:19.628559 sigreturn()             = ? (mask now [])
 
The SELECT statement was easy to find, and it became clear that a
programmer had code which was incorrectly canceling a JDBC
Statement after reaching the end of the (empty) ResultSet.  One
time out of 1,799 this was causing the error we were seeing on
the subsequent commit, which strikes me as a bug.
 
We've already changed the offending code to avoid the invocation
of the Statement.cancel method.  Not that it merits high priority, but
it might make sense for PostgreSQL to behave more consistently on
a commit when a statement within the database transaction has been
canceled.  There currently is a race condition where if the commit
comes fast enough after the Statement.cancel, it receives the error
which is the subject of this thread.
 
-Kevin
 
 
>>> Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> 09/13/05 1:18 PM >>>
"Kevin Grittner" <Kevin(dot)Grittner(at)wicourts(dot)gov> writes:
> I'm having a really hard time coming up with theories about the cause
> or things to check.

Have you tried strace'ing the backend process to see if you can see a
signal being delivered to it?

			regards, tom lane


Responses

pgsql-admin by date

Next:From: Ezra TaylorDate: 2005-09-13 21:25:48
Subject: Re: How to improve Postgres performance
Previous:From: Marcin GiedzDate: 2005-09-13 19:49:43
Subject: Re: plperl again but different problem

pgsql-jdbc by date

Next:From: Kevin GrittnerDate: 2005-09-13 20:33:59
Subject: Re: java.sql.SQLException: ERROR: canceling query due
Previous:From: Oliver JowettDate: 2005-09-13 11:12:03
Subject: Re: Strange behavoir of batches

Privacy Policy | About PostgreSQL
Copyright © 1996-2014 The PostgreSQL Global Development Group