[erlang-questions] Send VM metrics to Grafana
Fri Feb 24 18:18:54 CET 2017
On 2/20/17 11:26 PM, Jack Tang wrote:
> We try to integrate the metrics to our grafana monitoring by adding
> infuxdb consumer. Thanks Siraaj.
The pleasure is all mine, Jack. Let me know if you have any
> BTW: Did somebody collect the crash metrics, like the count of crashed
> process, the restart count of the process and etc?
I haven't come across it before, but it sounds like a great idea.
Off the top of my head I can't think of a general and direct way to
collect that without instrumenting a supervisor (either by patching
stdlib or user app).
The only thing that comes to mind right now is snapshotting PIDs at each
collection and tracking the duration of each uptime, then we can compute
whatever statistical summaries on current uptimes (e.g. if something
starts to crash too-often/too-soon it'll reduce the expected/normal min,
med, avg, etc.). There're certainly problems with this approach, since
it'd be just an approximation, but it doesn't sound terribly
unreasonable to me right now.
> On Sat, Feb 11, 2017 at 2:01 AM, Siraaj Khandkar <
> <mailto:>> wrote:
> My biased preference is:
> It's concern is only with sampling VM metrics and shipping them off
> to arbitrary destinations (Folsomite was Graphite-only, the last
> time I checked). Graphite and StatsD consumer modules are included.
> StatsD is the one we use in prod now.
> It's been running in prod for 1.5 years now, and even though it
> wasn't originally the main motivation, my favorite feature
> turned-out to be the sampling of individual processes and ETS
> tables, which was a life-saver in tracking down multiple memory
> leaks in the legacy system I inherited.
> On 2/9/17 4:06 AM, Frank Muller wrote:
> Hi everyone,
> Does anyone succeeded to push VM metrics to Grafana
> Here I'm looking for a simple integration with minimal
> dependencies if
> Thank you.
> erlang-questions mailing list
> Jack Tang
More information about the erlang-questions