tomcat-users mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From marko lugarič <marko.luga...@gmail.com>
Subject Re: Tomcat connection pool "bleeding" under heavy load
Date Wed, 20 Nov 2013 13:14:26 GMT
Hello

The tests are run in the following manner: around 5 min to start the server
and connect all clients. Then I can apply load from clients: on the graph
I applied only half load for first seven minutes an then full load but this
is
not required - I can go with full load after all clients connect. So once
full load is
applied usually in less then 5 minutes pool has no connections (it) - then
I can
leave it for half hour (tested today) and nothing changes (0 empty and zero
active connections). But i didn't have logAbandonded set - I can test this
too
if you think there will be difference. In the usual test when i saw that
there is no
connections I turned off server (2 - 5 minutes later).



On Wed, Nov 20, 2013 at 2:35 AM, Christopher Schultz <
chris@christopherschultz.net> wrote:

> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA256
>
> Marko,
>
> On 11/19/13, 10:16 AM, marko lugarič wrote:
> > About using useEquals="false" property I guess we defined when we
> > started using this pool. After reading the documentation i guess it
> > is better to leave it out (it has no effect on the test - i tested
> > it).
>
> Okay. Let me know if you discover a reason to set that option again.
>
> > We dont have long running queries because there is not much data
> > and all operations are fast (the longest one are ranging from 1 to
> > 2 seconds: i used slowqueryinterceptor and sql server profiler).
> > The test always starts with empty database.
>
> Ok.
>
> > I never got any abandoned messages in the stdout or catalina.out
> > (logabandoned set to true) - I put everything that is not from our
> > application and is at least on INFO level to catalina.out. I added
> > org.apache.tomcat.jdbc.pool package to logs on TRACE level and the
> > only line that is repeating (5 times) is: 2013-11-19 15:25:24,764
> > DEBUG [main] org.apache.tomcat.jdbc.pool.PooledConnection
> > connectUsingDriver(): Instantiating driver using class:
> > com.microsoft.sqlserver.jdbc.SQLServerDriver
> > [url=jdbc:sqlserver://*****;databaseName=chiTestDB]
>
> That will be filling the pool. It's clear that the pool is emptying
> and never re-filling, or you'd get more of the above messages.
>
> > Dont know if there is really that little logging in connection
> > pool. In that case debugging looks like the only option?
>
> Perhaps, but I think you might be able to get more information from
> the pool.
>
> > About C3P0 pool:we are under the impression that
> > org.apache.tomcat.jdbc.pool is the only way to go (production
> > quality) after reading The Tomcat JDBC Connection Pool page (after
> > your comment i guess this is not true and commons-dbcp is
> > production quality alternative?).
>
> IMO the Tomcat-pool documentation overstates the awfulness of the
> dbcp-based pool. We use it for production and it works just fine.
> We've never had a problem with the pool that we didn't cause by our
> own bad webapp code (e.g. not returning connections) or due to
> long-running queries triggering the "abandoned" alarms. OTOH, our site
> does not run an enormously-high transaction rate, either.
>
> > So since this did not work for us we started looking for
> > alternative and C3P0 pooped up in connection pool comparisons or in
> > some stackoverflow thread.
>
> Just note that C3P0 is self-described "beta" software.
>
> So, back to the problem: how long does your test run? Do you allow the
> server to cool-down after your tests? I'm wondering if you are
> shutting-down the server before it has a chance to complain about the
> abandoned connections.
>
> I've never used Tomcat-pool, so I'm not exactly sure what to expect,
> but the abandoned stuff should probably work exactly as described in
> the documentation, and roughly equivalently to the dbcp-based pool.
>
> - -chris
> -----BEGIN PGP SIGNATURE-----
> Version: GnuPG v1.4.15 (Darwin)
> Comment: GPGTools - http://gpgtools.org
> Comment: Using GnuPG with Thunderbird - http://www.enigmail.net/
>
> iQIcBAEBCAAGBQJSjBHIAAoJEBzwKT+lPKRYQ4AQALKwD6/Cko6t7di8oXZS8sKO
> pdHc2AQhrPbg95stygZiBm8lH4ds7RhwslD7JcWafSUfln/smLJO0gaV+M3PNDzX
> gQs/zqZyHtEW4vD2oIiL+CY/I0kbAON3tRbXFBjfJKikeINiIrTYGvm8h2dkhq4X
> WLijKrsh/qy96RkmUYPd0o1RsmraYCeGdSNpA96vycgrb89YHiJj07BkinKdhxTQ
> 903huox7ZspSGR0bl/+zEchHKj5AO9D9QELas6Z3qGwd+O7A5rxcX5YHsEXkNX6z
> mTwvAHzobdLXtu7LS6i+td46hFd0QurutnhrOAIZpoHNz7kItRZGDxYB296A5lY/
> xdOzGXKMMaTM854hvLuSPy6zkU2S235+KEA6sVi9HpNtP42yoPbdJfM3thHcKrxB
> Zm/8z6DrMtHFXBRbCFE7PGds9tymCvZx3/n/Zv5a6jm2iSEOLcSSKGnvQG663029
> l2XEGQyZufo+YsY9kQN+zxwSdbBl381PC8YXB+Dh2VR+LYaEuJqpxIkJjR+gIdiT
> JhXm4qSAYVsPDesG7GwXMwY9oFfpVvOzvv8KbH1AM/GaiLBLclTmHWZ8xw+alD7i
> HlR0M9gJ2dw6x6hYbv+wguBksqLxhLFOe+j6rekd7esyMP71nZh4TT4rf+r4XrGe
> 5RkJAULb0vE+ipAgZT/P
> =Vjf7
> -----END PGP SIGNATURE-----
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: users-unsubscribe@tomcat.apache.org
> For additional commands, e-mail: users-help@tomcat.apache.org
>
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message