I find our current use of apache commons while depending on specific
features of log4j awkward.
I think we are optimizing for the wrong things here. We are going to
want server logs from MR and HDFS to be consistent and manageable by
other components of hadoop. This removes most of the value in user
configurability of log formal or control of log output methods
(outside of options provided by hadoop). For this reason, I'd be in
favor of using the simplest, lowest common denominator system the
supports the features we require.
It would be interesting to enumerate what we need from a log system.
Does j.u.l meet our needs? If so, why not use it? The last time we
had this conversation we were reviewing the java 1.4 log APIs and
these were too rudimentary, but maybe java has caught up?
E14
On Sep 28, 2007, at 3:53 PM, Michael Stack wrote:
Thanks for the considered response Torsten.
a) Thanks for the pointer on log4j. I'll take a look.
b) java.util.logging has always worked just fine in my experience.
c) Less dependencies and if the extra jars were removed, it would
imply
less indirection when logging, is always better, no?
d) Agreed (Of note, Hadoop isn't yet 1.0).
e) For j.u.l the appropriate JMX beans are part of the (Sun) JDK
and its
just a matter of setting a flag starting the JVM to make the bean
available but yes, this is a minor point (especially if only the
Sun JVM
supports this).
f) Pardon me, Torsten, are you saying a logging facade is
inappropriate
(because it is not a library or framework)? Perhaps you are saying
the
opposite?
Thanks,
St.Ack
Torsten Curdt wrote:
>
> On 28.09.2007, at 19:23, Michael Stack wrote:
>
>> Reading the below (old) discussion of JCL vs SLF4J made me want to
>> ask if there is anyone out there who actually makes use of the fact
>> that logging goes via the commons-logging intermediary? If its not
>> being used, why not cut to the chase and use log4j directly or what
>> seems to be just as capable, native java.util.logging.
>
> I seriously had enough of this logging debates at other projects
so I
> will just provide another few cents and then shut up :)
>
> a) Everyone suggesting log4j must have never looked at the code :)
> b) I only know people complaining about j.u.l (but never used it
myself)
> c) Not sure why it would make sense to save two jars on project like
> hadoop (where you usually are not short of space in the dimension a
> few KB on a machine)
> d) Changing logging (facades) is not fun. Make sure to stick with
one
> ones 1.0 is out
> e) Changing log levels via JMX also works quite straight forward
with
> other logging implementations (it's just that setting up JMX
properly
> is not)
> f) Logging facades are more useful for libraries/frameworks
>
> cheers
> --
> Torsten
