[logback-user] Separating the logging-output of libraries in tomcat/shared

[Ceki Gülcü]

You are right. ContextBasedDiscriminator will not help you because the context 
name is retreived from the LoggingEvent which ultimately gets its value from the 
calling logger. If the calling logger is attached to the wrong context, 
ContextBasedDiscriminator will give you the wrong context name. Sorry for 
misleading you.

However, I've written a new discriminator which should work assuming you have 
installed ContextJNDISelector as explained in

  http://logback.qos.ch/manual/loggingSeparation.html

JNDIBasedContextDiscriminator can be found at

 
http://github.com/ceki/logback/blob/90e50c/logback-classic/src/main/java/ch/qos/logback/classic/sift/JNDIBasedContextDiscriminator.java

   or the same in tiny-speak

   http://tinyurl.com/yb6pdj2

Getting values from JNDIBasedContextDiscriminator, SiftingAppender will get the 
correct context name and things should work much better. Give it a try.



On 01/03/2010 9:38 AM, Achim Wiedemann wrote:
> Hey Ceki,
>
> thanks for your fast reply. I tested your tip with the SiftingAppender,
> but couldn't see how this gets me further compared to the
> logging-separation described in Ch8 of the manual (separation via
> JNDIContextSelector). The problem is still the same: The context of the
> logger, once initialized, remains the same in shared libraries. Since
> SiftingAppender has to sieve based on some criteria (in my case the
> context), it also can't solve this issue.
>
> So I see, as you already stated, it's not possible to get this scenario
> to work. However, thanks for your time and all your work.
>
>
> greetz,
> achim
>
>
> Ceki Gulcu schrieb:
>>
>> Hello Achim,
>>
>> Thank you for your message.
>>
>>
>> The only way to solve this issue transparently and perfectly would be
>> to introduce another level of indirection inside loggers so that each
>> logger-shell somehow delegate work to an inner logger attached to the
>> appropriate context. This is quite a difficult problem technically and
>> would generate significant computational overhead. I don't see the
>> problem fixed via such indirection any time soon. Read, as a logback
>> user, don't expect this problem to be solved without a little work on
>> your side.
>>
>> As far as I can see, you have two options:
>>
>> 1) move shared classes whose loggers you care about to the web-app
>> (unshare them)
>>
>> If 1) is not an option, there is:
>>
>> 2) Use a single shared logback.xml file with separation done by
>> SiftingAppender [1] using the context as separation criteria [2]. In
>> conjunction with prudent mode of FileAppender, you will be able to
>> separate logging by context.
>>
>> Let me know if the above does not make sense.
>>
>>
>> [1] http://logback.qos.ch/manual/appenders.html#SiftingAppender
>> [2]
>> http://logback.qos.ch/xref/ch/qos/logback/classic/sift/ContextBasedDiscriminator.html
>>
>>
>> On 26.02.2010 08:21, Achim Wiedemann wrote:
>>> Hey everybody,
>>>
>>> just wanted to port our application from commons-logging / log4j to
>>> slf4j / logback. Unfortunately I stumbled upon some problems when
>>> separating the logging-output from different webapps (using the
>>> JNDIContextSelector). As long as you only use logging from classes
>>> inside the webapp, everything works fine. But as soon as you've got
>>> classes in tomcat/shared which use logging, you run into trouble: All
>>> logging output of the shared classes (e.g. Spring) are directed to the
>>> logger-context of the webapp which first loads the (shared) class.
>>>
>>> Example: Let's say you've got webbapp [A], webbapp [B], and Spring in
>>> tomcat/shared. Now both webapps use Spring. If webapp [B] first accesses
>>> any classes of Spring, all output of Spring will be directed to the
>>> log-file of [B], no matter from which webapp Spring is used. This is
>>> because static Logger references are used in Spring (which is
>>> commonplace) and are initialized with the context of webapp [B].
>>>
>>> The issue is already described very well in the logback JIRA:
>>> http://jira.qos.ch/browse/LBCLASSIC-166
>>>
>>> I wondered if anyone can make a suggestion how to deal with this. Any
>>> advice, tricks or magic is welcome. ;)
>>>
>>>
>>> Thanks a lot,
>>> Achim
>> _______________________________________________
>> Logback-user mailing list
>> Logback-user at qos.ch
>> http://qos.ch/mailman/listinfo/logback-user
> _______________________________________________
> Logback-user mailing list
> Logback-user at qos.ch
> http://qos.ch/mailman/listinfo/logback-user