[opencms-dev] Problem for (re)starting OpenCms

[Brendon Price]

Thanks Niklas,

This helps to define the actual problem. We have since seen that even with
the increased file descriptors, the deadlock still occurs, just far less
frequently. 

Regards
Brendon 

-----Original Message-----
From: Niklas Eklund
To: opencms-dev at opencms.org
Sent: 14/07/2003 11:31 p.m.
Subject: Re: [opencms-dev] Problem for (re)starting OpenCms

Brendon et al,

I've had the same, or similar maybe, problem (with Oracle 9) as well and

traced it to some sort of deadlock in the java.sql.DriverManager. The 
system frooze with a couple of threads waiting in 
DriverManager.getConnection() and another in DriverManager.getDrivers().

The call to getDrivers() is done by the Oracle driver in it's connect() 
method()...
After searching (for "DriverManager") at Sun's Java bugparade it was 
clear that there is a problem with the DriverManager class and that Sun 
treats this as "Not a bug", and the only workaround seems to be to use 
the new DataSource paradigm instead.
Since the pool already seems to be thrown out in the development branch,

this will be resolved in the future. I ended up replacing the calls (all

239 of them ;) to DriverManager.getConnection() with another static 
method call that uses a DataSource type pool configured in tomcat. This 
seems to work for us, but haven't yet been thorougly tested...

Regards,

  Niklas

Brendon Price wrote:
> Hi all,
> 
> We had also encountered this problem previously and had spent several
days
> trying to resolve the issue. In the end we added extensive debug code
to the
> connection pool code of opencms. It appeared that the code hangs on
the
> re-establishment of the connections in the connection pool. This is
the
> point at which it times out the connections and re-estabilshes them.
We
> could get the product to break every 2 minutes by reducing the timeout
of
> connections down to 2 minutes.
> 
> In the end it was it appeared that the code fails within the JDBC
connection
> creation. In the end we found an article that suggested we increase
the
> number of file handles available to the Oracle 9i process (our DB).
Once we
> did this our problem disappeared, however I think there is something
else at
> work here that means we may have just hidden the symptoms of the real
> problem.
> 
> We made a changes to the
> /etc/sysctl.conf
> file and increased the number of file handles by adding the line
> 
> fs.file-max=32768
> 
> After a restart of the whole machine, the problem did not occur again,
even
> with our previous test case (that failed every 2 minutes).
> 
> I hope this helps.
> 
> Regards
> Brendon
> 
> 
> -----Original Message-----
> From: Nicolas Delsaux
> To: opencms-dev at opencms.org
> Sent: 11/07/2003 9:40 p.m.
> Subject: Re: [opencms-dev] Problem for (re)starting OpenCms
> 
> On Fri, 11 Jul 2003 10:23:54 +0200 (CEST), Joachim Arrasz
> <info at arrasz.de> 
> wrote:
> 
> 
>>Hy Nicolas,
>>
>>what was your solution to this crazy thing?
>>How we can configurate Opencms to do that NOT?
> 
> 
> I've not yet found any kind of solution; In fact, the problem is
> currently 
> being corrected by a Tomcat restart each night.
> But, if anybody can have any idea on that problem, I'm willing to
apply
> it 
> as soon as possible.
> 
>>thanx in advance
>>
>>Achim
> 
> 
> 


_______________________________________________
This mail is send to you from the opencms-dev mailing list
To change your list options, or to unsubscribe from the list, please
visit
http://mail.opencms.org/mailman/listinfo/opencms-dev