MDC looks like the way to go.  Thanks for that info, David.

        Erik

On Dec 17, 2008, at 3:13 PM, Smiley, David W. wrote:

You bet it does; that's the point!

Under the covers, I believe it's simply a thread-local hashmap. Nothing is stored in the loggers the code is using. We just need to be careful to remove the variable from MDC when we're done.

~ David


On 12/17/08 3:09 PM, "Ryan McKinley" <ryan...@gmail.com> wrote:

but does this work in a multi-threaded environment?

if multiple requests are coming in on multiple threads, would it still
be accurate?  Perhaps that depends on the underlying implementation?

adding the core to the MDC within a RequestHandler context seems
reasonable and minimally invasive.

ryan


On Dec 17, 2008, at 2:24 PM, Smiley, David W. wrote:

I propose that MDC or NDC be used instead.  I prefer MDC.  I've
written some server-side multi-threaded code where each Thread would
run a job and I wanted the job name in the logs.

http://www.slf4j.org/api/org/slf4j/MDC.html

At some early point when Solr receives a request, you simply store a
name-value pair which would be "core" and the name of the core for
us.  In a finally block, remove it.  In a log4j config file, in the
ConversionPattern, put this:  %X{core}   and this will expand to the
core name.

Some groovy snippet I wrote:
final MDC_KEY = "OraSeqId"
MDC.put(MDC_KEY, seq.id as String)//must be removed; see finally

//in-finally
MDC.remove(MDC_KEY)

~ David Smiley

On 12/17/08 2:17 PM, "Erik Hatcher" <e...@ehatchersolutions.com>
wrote:



On Dec 17, 2008, at 12:24 PM, Ryan McKinley wrote:

I'm not sure I understand...

are you suggesting that rather then configuring our logger like this:
static Logger log = LoggerFactory.getLogger(SolrCore.class);

We get the Logger everytime we use it with something like:
Logger log = LoggerFactory.getLogger(classname+":"+core.getName() );

That seems like a lot of overhead for marginal gain...

I'm not up on the logging tips and tricks best practices these days,
but yeah, I was suggesting something like that.  There's no rule that
says logging keys be FQCNs.... that's a decent convention for most
cases with a package structure that is well organized and filterable.
In this case having the core name in there as a prefix makes a lot of
sense to me.

We could provide a LoggerUtils.getLogger(core, clazz) or something
like to keep it DRY and consistent.

      Erik





Reply via email to