is Node.js. StatsD uses UDP. It’s asynchronous and painless. Backends are pluggable. /statsd 8 Thursday, November 8, 12 Etsy’s work was inspired by Flickr. standard backends: graphite, console, reapeter available backends: librato, cloudwatch, mongodb, ...
Python. /statsd/pattern 9 Thursday, November 8, 12 Original, original (Cal Henderson) — something in Perl at Flickr. StatsD as a pattern to collect metrics.
12 Really simple, no PhD required. Counter: grows absolute over time (number of total downloads) Gauge: for current values (e.g. “currently online”) Timings: how long something took There is more in master.
12 We wanted a hosted service (less management). Read a couple blog posts about scaling Graphite which scared me off. Librato has less aggregation and very little transformation but it’s been rock solid.
Github: https://github.com/etsy/statsd npm install statsd-librato-backend /statsd/install 12 Thursday, November 8, 12 Install Node.js through: homebrew, aptitude, Windows installer cd into statsd directory to npm install
{ email: 'email_on_librato', token: 'librato_token', source: 'where_metrics_are_coming_from' } } /statsd/configure 13 Thursday, November 8, 12 Watch me type! The minimum is email and token. port: port of the statsd process librato.email/.token: credentials from librato librato.source: e.g. your cluster (application, whatever)
an excerpt of stuff we collect. And every time there’s a problem, we add to it. https://metrics.librato.com/pricing 50 metrics every 60 seconds = ~5.00 USD/month flush every 30 = ~8.50 USD a month