Re: Finding Broker/Connection leak

Hi Tino,

Tino Schöllhorn wrote:
> Hi,
> 
> I have a situation where we are running out of PersistenceBrokers (using 
> 1.0.1). We have a web application which so far runs fine - but somehow 
> some requests are causing the application not to close its PB.
> 
> We can see this when we are monitoring the used connections of mysql 
> which are slowly growing - even if we are using ConnectionPooling. These 
>  connection are on the state SLEEP for quite a long time and seem never 
> to be closed.
> 
> At some point the application runs out of PB (we are using size 100 of 
> the PB-Pool).
> 
> So it seems (quite obviously) that some PBs are not cleanly handled.
> 
> Do you have any advice for us finding those leaks?
> 

In current versions there is no PB leak detection. Such a feature will
be part of the next version (or try latest from SVN OJB_1_0_RELEASE branch).
It would be possible to modify your 1.0.1 version to support broker leak 
detection. Have a look in latest PBImpl (OJB_1_0_RELEASE branch) and 
port the detection part in method PBimpl#setClosed and PBImpl#finalize 
to your class.
http://svn.apache.org/viewvc/db/ojb/branches/OJB_1_0_RELEASE/src/java/org/apache/ojb/broker/core/PersistenceBrokerImpl.java?revision=398752&view=markup


> Does it do any good when we "tune" mysql that it discards connections 
> which are quite old (via property "max_timeout")? Is OJB discaring those 
> PBs whose connections have died ?

If you correctly use the PB instances (lookup PB instance when needed, 
close it after use) the connection pool should handle connections in a 
correct way.
But if you don't close PB instances after a query, the connection will 
still be associated with the PB instance, please see

http://www.mail-archive.com/ojb-user%40db.apache.org/msg15389.html

regards,
Armin

> 
> I am glad for any suggestions.
> 
> Tino
> 
> 
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: [email protected]
> For additional commands, e-mail: [email protected]
> 
>