Production ready?

[rtoma]

Go-carbon looks very clean and promising.

Before replacing an array of 8 carbon-cache daemons I was wondering if you believe go-carbon is ready. Do you (or anyone else) use go-carbon in a production setup? How many datapoints/sec does it handle? My $dayjob graphite cluster currently handles 45k points/sec per node, for some 2.6M metrics in 1.6T on flash.

Thank you so much.

[szibis]

Yes, I use it on production from August 2015.

Before carbon-cache on PyPy in 80 instances and 4x lower traffic ended on same setup as go-carbon now.

Up to 16 mln metrics in cluster with replication factor 2. About 35 mln updates per minute (update operations * point per update) with 5 instances each with one go-carbon - update time ~0.4ms. Each node generates ~50k IOPS on SSD drives and go-carbon consumes ~600-700% CPU. Metrics in go-carbon cache ~1.5 mln per node. Huge dirty pages used, buffers and VFS cache.

Metrics sizes from 1.5MB to 2.5MB and from 1s to 30s update times precision in whisper. All space used ~20TB of whisper data.

[rtoma]

Thanks for the feedback. I'm gonna start a trial.