We're developping a big Java application and starting to set up application monitoring, because we're detecting problems too late, and because parsing log files is boring.
We already set up a POC on Graphite and JMXTrans. It's working fine and we're starting to have some interesting graphs.
But I'm now wondering if we should use graphs for everything we'd like to trace. I think it would be fine to have messages somewhere, and basically, I'd like to go in log file only as a last chance to debug a problem. If I'm getting exceptions in production, I'd like to be warned automatically and know what's the problem is about.
I found Sentry which is a webapp that collects traces from clients, and logster, that parses log files and send lines matching some patterns to Graphite.
I think the philosophy of the first one is better. Basically, I'd like to dissociate what is traced from how it's going to be traced. We could use a logging facade like SLF4J, and then dispatch traces according to levels or some others rules (patterns in exception for instances) to log files, a NoSQLDB, dedicated tools like Sentry or Graphite, etc.
So basically,what are you using for application monitoring?
You may want to take a look at LogDigger. Instead monitoring logs and looking for errors and related log entries, you can get a notification containing request logs, POST parameters and other details.
Disclose: I worked on LogDigger so I may be biased.