[jira] [Commented] (KAFKA-2285) Logging trait obfuscates call site information

["Grant Henke (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/KAFKA-2285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14591990#comment-14591990 ] 

Grant Henke commented on KAFKA-2285:
------------------------------------

It looks like it's where the call to _logger.*_ is that really matters. However, the Kafka Logging trait guards every call with
_if (logger.isLevelEnabled())_  so moving the calls out to the code would be messy or potentially impact performance.

The scala-logging (https://github.com/typesafehub/scala-logging) library could be nice solution. It uses a macro at compile time to turn calls like:
_logger.debug(s"Some $expensive message!")_ into _if (logger.isDebugEnabled) logger.debug(s"Some $expensive message!")_

And preserves the call site since _logger._* is called in the class which mixes in the trait.

If adding a dependency is an issue the library is small enough that we could implement our own version on similar principles.

> Logging trait obfuscates call site information
> ----------------------------------------------
>
>                 Key: KAFKA-2285
>                 URL: https://issues.apache.org/jira/browse/KAFKA-2285
>             Project: Kafka
>          Issue Type: Improvement
>          Components: core
>    Affects Versions: 0.8.2.0
>            Reporter: E. Sammer
>
> Using a logging trait, as many components in the codebase do, destroys call site information in logging message making debugging certain kinds of failures annoying in production systems. Most messages end up look like:
> {code}
> 2015-06-18 07:41:11,550 (kafka-request-handler-0) [WARN - kafka.utils.Logging$class.warn(Logging.scala:83)] Partition [events,1] on broker 1: No checkpointed highwatermark is found for partition [events,1]
> {code}
> I think the mental overhead of issuing the standard incantation of {{private static final Logger logger = LoggerFactory.get(Foo.class)}} (or the even shorter Scala equivalent) for each class is outweighed by the operational overhead of mapping strings back to their original call sites. This is an easy win improve the traceability of complex failures in production deployments.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)