I’m looking at this and have a couple of concerns. The script has
HADOOP_ROOT_LOGGER=${HADOOP_ROOT_LOGGER:-${HADOOP_LOGLEVEL},console}
HADOOP_DAEMON_ROOT_LOGGER=${HADOOP_DAEMON_ROOT_LOGGER:-${HADOOP_LOGLEVEL},RFA}
HADOOP_SECURITY_LOGGER=${HADOOP_SECURITY_LOGGER:-INFO,NullAppender}
So it seems you need this for more than just the root logger.
Second, you are asking us to accept “level, appender” as the value and under
the covers split the
value and assign the level to the level attribute and the appender name to the
AppenderRef.
While this certainly can be done it seems like it would be just as easy to do
the split in the script
and create two different variables for the logging configuration to pick up.
Is there a reason that
you can’t do this - for example perhaps users have hacked your scripts and now
you can’t
change them without breaking them? If so, was this “supported”?
Ralph
> On Jan 9, 2022, at 8:01 AM, 张铎(Duo Zhang) <[email protected]> wrote:
>
> I brought this up in the incubator mailing list, and was suggested to
> report directly to the log4j community.
>
> https://issues.apache.org/jira/browse/HADOOP-16206
>
> In hadoop we started to try migrating to log4j2 long ago, but it is not
> easy. For now, one of the most blocker issues is the lack of support of
> 'log4j.rootLogger=INFO,Console' grammar. Notice that we do not want to stay
> on the bridge api, we want to fully migrate to log4j2 finally, so
> supporting the above grammar in log4j1 bridge is not enough.
>
> This is the main log4j configuration file for hadoop
>
> https://github.com/apache/hadoop/blob/trunk/hadoop-common-project/hadoop-common/src/main/conf/log4j.properties
>
> We have this in the file
>
> # Define the root logger to the system property "hadoop.root.logger".
>> log4j.rootLogger=${hadoop.root.logger}
>
>
> So our users could simply pass -Dhadoop.root.logger to control the level
> and appender.
>
> https://github.com/apache/hadoop/blob/39efbc6b6fe53abed15f5639edcbaaa2a9dda6d2/hadoop-common-project/hadoop-common/src/main/bin/hadoop-functions.sh#L904
>
> Here we use an environment variable in our start up scripts. And I believe
> there are lots of other hadoop deployment systems which will have their own
> start scripts.
>
> So in general, it is just impossible for us to drop the
> -Dhadoop.root.logger way of configuring our logging system.
>
> So here I want to ask if it is possible for log4j2 to still support the
> 'log4j.rootLogger=INFO,Console' grammar. I'm not saying you must add it
> back with no change, we just need a way to configure the level and appeners
> at once, so something like
> 'log4j2.rootLogger.levelAndAppender=INFO,Console' is also OK.
>
> Thanks.
