stats

[xiang90]

collect various stats and report them (syslog; file; statshat, etc.)

[xiang90]

https://github.com/rcrowley/go-metrics

[xiang90]

from #960
Inspired by #915:

current etcd counts the total time for each AppendEntry RPC, and it varies a lot when leader changes frequently.

Personally i cannot infer root cause from it:

RPC time includes both the time for network transmission, state machine handling and disk I/O.

minimum, maximum and variance cannot give me the whole view of RPC. Maybe it is Normal distribution, or it could be binomial distribution.

More statistics could help us to figure out which part could be the problem for etcd cluster.

[xiang90]

we end up with http://prometheus.io/