metrics.md

title
Metrics

S4 continuously collects runtime statistics. Let's see how to access these and add custom ones.

Why?

S4 aims at processing large quantities of events with low latency. In order to achieve this goal, a key requirement is to be able to monitor system internals at runtime.

How?

For that purpose, we include a system for gathering statistics about various parts of the S4 platform.

We rely on the metrics library, which offers an efficient way to gather such information and relies on statistical techniques to minimize memory consumption.

What?

By default, S4 instruments queues, caches, checkpointing, event reception and emission and statistics are available for all of these components.

You can also monitor your own PEs. Simply add new probes (Meter, Gauge, etc..) and report interesting updates to them. There is nothing else to do, these custom metrics will be reported along with the S4 metrics, as explained next.

Where?

By default, metrics are exposed by each node through JMX.

The s4.metrics.config parameter enables periodic dumps of aggregated statistics to the console or to files in csv format. This parameter is specified as a node or application parameter ¹, and must match the following regular expression:

(csv:.+|console):(\d+):(DAYS|HOURS|MICROSECONDS|MILLISECONDS|MINUTES|NANOSECONDS|SECONDS)

Examples:

# dump metrics to csv files to /path/to/directory every 10 seconds
# (recommendation: use a different and clean directory for each of the nodes)
csv:/path/to/directory:10:SECONDS

# dump metrics to the console every minute
console:1:MINUTES

Reporting to Ganglia or Graphite is not provided out of the box with S4, but it's quite easy to add. You simply have to add the corresponding dependencies to your project and enable reporting to these systems during the initialization of your application. See the metrics documentation for more information.

Footnotes

1. csv-based metrics logging are usually specified as a node configuration, and must point to a clean directory, different for each node. Otherwise the metrics logging system will try to create multiple new files with the same name, and depending on your setup, that may not be possible. ↩