[rt-users] large query results in User Lock on Apache-Session

Brian Kerr kerrboy at gmail.com
Mon Jul 24 09:40:36 EDT 2006


Hi,

We are currently running RT 3.4.5 on mysql 4.1.18-log with
fcgi+apache2, which was previously migrated from 3.0.7.

Here are relevant fcgi config directives in our configuration:
FastCgiIpcDir /u2/apache/logs/fastcgi
FastCgiServer /u2/rt3/bin/mason_handler.fcgi -appConnTimeout 305
-init-start-delay 20 -idle-timeout 120 -processes 8


Generally speaking, the system runs very smoothly.  I have discovered
through our slow-query logging and testing that if a "large" query is
run, example: query for every ticket in queue: test which returns
36,000 results with Rows per page: Unlimited(the default), the
database query finishes and the results are shown in the browser(takes
a long time due to the amount of tickets).

If I try to do anything before the page is fully finished loading,
however, such as click "home" or anything else, the session hangs and
these selects start piling up, requiring an apache restart to fix:

|    1 | rt_user | localhost | rt3  | Query   |   38 | User lock |
SELECT GET_LOCK('Apache-Session-9f37563f3608b5bca122c81829a57f9b',
3600) |

These build up only for the session that ran the original "large"
query, other activity continues to function normally.

|    1 | rt_user | localhost | rt3  | Query   |   90 | User lock |
SELECT GET_LOCK('Apache-Session-9f37563f3608b5bca122c81829a57f9b',
3600) |
|    4 | rt_user | localhost | rt3  | Query   |   18 | User lock |
SELECT GET_LOCK('Apache-Session-9f37563f3608b5bca122c81829a57f9b',
3600) |
|    8 | rt_user | localhost | rt3  | Query   |   35 | User lock |
SELECT GET_LOCK('Apache-Session-9f37563f3608b5bca122c81829a57f9b',
3600) |
| 7093 | rt_user | localhost | rt3  | Query   |   29 | User lock |
SELECT GET_LOCK('Apache-Session-9f37563f3608b5bca122c81829a57f9b',
3600) |

After the GET_LOCK has been running for 120 seconds an ISE is
displayed(due to the 120second fcgi idle-timeout directive), but
further attempts to access the system still result in select
get_lock's piling up.

Is there any way to fix this behavior, or prevent it?

Thanks in advance,
Brian



More information about the rt-users mailing list