tibbe / ekg-statsd Goto Github PK

commercialhaskell/stackage#2678

commercialhaskell/stackage#2670

#13 would fix this.

Needs base upper bound bump. See commercialhaskell/stackage#5365

When sending metrics to statsd it would be useful to allow prefixing all metrics. This way you can differentiate between metrics from different processes.

Distributions are currently silently dropped. They could be published as a count and gauges.

We are using forkStatsd to start collecting stats, and later on stopping the stats collection by calling killThread on the statsdThreadId, as indicated in the documentation. However, the implementation of forkStatsd rethrows all exceptions coming from the forked thread to the thread that called forkStatsd in the first place, including any ThreadKilled exceptions. This was confusing to say the least!

Is this the expected behaviour, or am I missing a better way of doing this? It feels a bit clumsy to control the Statsd by throwing an exception that you then have to remember to catch, and you have no guarantee that the exception you catch is the one you threw in the first place. This also feels inconsistent with a normal forkIO which swallows ThreadKilled (and Blocked*) exceptions so you don't see them in the main thread.

One needs to compile with -with-rtsopts=-T in order for GC metrics to be pushed to statsd.

Update cabal upper bounds:

• base <4.9
• time <1.6