Metrics

Improving performance of S4 applications requires metrics that can be analyzed.

Some metrics suggested in S4-65 are:

• event rate
• average/median event processing time
• event queue sizes
• number of PE instances
• number of processed events ( maybe within a sliding window)
• exceptions during event processing

We'd also add things like:

• shedding stats
• memory usage
• CPU load
• checkpointing: effective checkpoints, rejected ones
• etc...

Ideally, we'd be able to have different kinds of metrics, including historical ones.

We also want to be able to expose those metrics through JMX, and probably through other monitoring systems such as nagios.

A library that could be useful here is codahale's metrics http://metrics.codahale.com/ , but there are alternatives that could be evaluated as well.