[
https://issues.apache.org/jira/browse/LOG4J2-1713?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15726263#comment-15726263
]
Steve Cohen commented on LOG4J2-1713:
-------------------------------------
I no longer believe in this solution. It is incorrect and I would not want
anyone to follow me down this path. There are two problems with it.
1. It doesn't work for Stateless Session Beans. That's probably because it
relies on non-final static Objects, which is verboten under EJB spec. See
http://www.oracle.com/technetwork/java/restrictions-142267.html#static_fields.
2. While I was able to fix this problem by making the LoggerContext static
final and the Logger non-static non-final, the Oracle doc put me in mind of the
fact that I was only testing on a standalone server. The minute we leave this
pleasant sandbox for clustered servers with the EJBs able to be distributed
elsewhere, the more problematic it looked. The complexities of a static final
LoggerContext in a distributed object are not something I want to think about,
much less be responsible for.
I have sadly become convinced that container-managed Objects should not hold
LoggerContext members at all, especially if they might be distributed to other
machines as in a cluster. In such a situation, it may be best to use the
Container-provided logging system. Unfortunately, with JBoss as currently
implemented this does not allow log4j2.
> Allow for more general serialization of Log4j2 configurations
> -------------------------------------------------------------
>
> Key: LOG4J2-1713
> URL: https://issues.apache.org/jira/browse/LOG4J2-1713
> Project: Log4j 2
> Issue Type: New Feature
> Components: Configurators
> Affects Versions: 2.7
> Reporter: Steve Cohen
>
> We want to implement the following system:
> We would like to write an external program that reads several Log4j
> Configuration Files, combines the configurations, and then outputs the
> combined configuration to a new Log4j configuration file. This file can then
> simply be dropped on a running log4j instance on a server, and cause an
> update of the running configuration.
> Existing APIs do not support this use case. There is nothing that supports
> the serialization to XML of a loaded configuration. There is the new
> ConfigurationBuilder.writeToXml() method in 2.7, but that's on
> ConfigurationBuilder, not Configuration. Nor is it possible to take a
> Configuration, and get a ConfigurationBuilder from it. Another way it could
> work is having some sort of ConfigurationBuilder that accepted parameters of
> type Logger, Appender, etc. These would enable Loggers and Appenders and
> other Log4j2 objects to be copied from an existing configuration to a new one
> being built. But this doesn't exist either. One would have to drill down
> into each component, extract the necessary data, and add it to a builder.
> In other words (H/T to Gary Gregory)
> c1 = load config from XML file 1 (but do not apply the c1 configuration)
> c2 = load config from XML file 2 (but do not apply the c2 configuration)
> c3 = c1 + c2 (but do not apply the c3 configuration) write c3 to disk
> In case you wonder, there is actually a use case behind this:
> Imagine a Servlet web-app that, depending on request parameters, invokes one
> of a number of possible EJBs (they could also be non-EJB POJOs, but for the
> purposes of this discussion, we assume EJBs), in order to produce a response.
> This is not a shopping cart but a back end system. We do NOT wish to deploy
> these into a single EAR file but want separate deployment of each component,
> and each component to have a separate logging configuration, deployable at
> the same time as the component. Since Log4j allows only one configuration
> context per class-loader, the ideal scheme would be there can only be one
> configuration file. The only way to update it non-programatically is to drop
> a new configuration in the correct location. In order to produce this, we
> would like the separate logging configs deployed to some directory. A
> separate program would read in all of these and add the loggers and appenders
> to a new master configuration which would then be written out to disk and
> copied to the proper location. The usual change mechanism would then load
> the new configuration. The current configuration of the running system would
> always match what is in this master configuration file, which is not the case
> with programmatic configuration.
> Without something like this, how is it possible to run multiple EJBs out of a
> web-app that are separately deployable and have separately manageable logging
> configurations?
> One could manually of course parse the individual files on the XML level (or
> JSON, YAML, whatever), combine them, and serialize the output. But since
> log4j knows its own object model better than any xml parser, it makes sense
> to have this capability within log4j.
> And yes, I've thought of the fact that property-name, logger-name,
> appender-name collisions are possible and could cause trouble. I would be
> prepared to live with real restrictions that prevent this. I don't think
> there is any need to support concatenation of random config files that do not
> prevent such collisions.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: log4j-dev-unsubscr...@logging.apache.org
For additional commands, e-mail: log4j-dev-h...@logging.apache.org
