tomcat-users mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From André Warnier ...@ice-sa.com>
Subject Re: [partially OT] Should a webapp close a DBCP datasource obtained through JNDI?
Date Fri, 25 Mar 2011 22:57:03 GMT
Mark Thomas wrote:
...

> 
> For 1, it is not necessary. Stopping the app makes it eligible for GC.
> However, Tomcat 7.0.11 onwards does stop it (see [1]) since this forces
> the database connections to be closed immediately rather than waiting
> for GC. Since this is viewed as a 'nice to have' rather than a bug (it
> is not a memory leak) then it won't be back-ported to earlier versions.
> 
I am wondering if this could not create the same phenomenon which I have seen with another

app a while ago : a number of connections, at the OS level, left in the CLOSE_WAIT state 
until the next GC.
Scenario :
- a Java "connection object" is created, which contains a socket (OS-level structure)
- the object is used, and then discarded, but without explicitly closing the embedded socket
- the object, now without reference to it within Java, sits on the Heap, waiting to be GC-ed
- but because it still has this unclosed socket in it, that socket still exists at the OS

level, and cannot be reclaimed by the OS

Slowly, such CLOSE_WAIT sockets accumulate, and in Linux at least, when there are a few 
hundreds of them, the system suddenly becomes irresponsive to TCP/IP in general (no new 
connections can be created for example).

Easy to diagnose : do this a few times and watch the output of
"netstat -pan | grep CLOSE_WAIT"
Then force a GC, and look again.

(Of course then, with a Tomcat version < 7.0.11)


---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@tomcat.apache.org
For additional commands, e-mail: users-help@tomcat.apache.org


Mime
View raw message