ElasticSearch uses logger names like "node", "plugins", "discovery" etc. and there are a lot of loggers. When it is run as a standalone server, there won't be problem as all the code are supposed to be ElasticSearch code.

However, when ES is used in a separated Java application as client, the logger mess up with my application's log. I use log4j.xml and I have to add a log of custom loggers entries because ES's ain't under its own package.

My suggestion:

1. I suppose your objective is to make the server log easier to read. I suggest to use full class name as logger name, and provide a default log4j layout pattern that doesn't show the full class name. Log4j's pattern is quite flexible.
   http://logging.apache.org/log4j/1.2/apidocs/org/apache/log4j/PatternLayout.html
   For example, for the category name "a.b.c" the pattern %c{2} will output "b.c".

For the majority of cases, %c{1} should work. Except some cases such as "cluster.service" that %c{1} won't show a meaningful logger name.

1. a workaround is to use a System property to define a prefix for the logger. So when ES is run in a client environment, we can set a property like -Des.logger.prefix=com.elasticsearch. and then we can use a single log4j category to capture all the logs.