Metrics

[SamBarker]

This PR adds the metrics outlined in proposal 47.

Note there are a couple of name changes in the implementation compared to the proposal:

Proposal	Implementation
io.strimzi.kafka.quotas:type=LocalThrottleFactor,name=ThrottleFactor	io.strimzi.kafka.quotas:type=ThrottleFactor,name=ThrottleFactor
io.strimzi.kafka.quotas:type=LocalThrottleFactor,name=FallbackThrottleFactorApplied	io.strimzi.kafka.quotas:type=ThrottleFactor,name=FallbackThrottleFactorApplied
io.strimzi.kafka.quotas:type=LocalThrottleFactor,name=LimitViolated	io.strimzi.kafka.quotas:type=ThrottleFactor,name=LimitViolated
io.strimzi.kafka.quotas:type=ClusterVolumeSouce,name=ActiveBrokers	io.strimzi.kafka.quotas:type=VolumeSouce,name=ActiveBrokers
io.strimzi.kafka.quotas:type=ClusterVolumeSouce,name=ActiveLogDirs	io.strimzi.kafka.quotas:type=VolumeSouce,name=ActiveLogDirs

In short LocalThrottleFactor -> ThrottleFactor and ClusterVolumeSource -> VolumeSource as the local and cluster distinctions no longer make sense (I don't think they actually did when merged the proposal but no one picked up on the fact we had lost the distinction).

The Metrics

Name	Metric Type	Meaning	Type	Tags
ThrottleFactor	Gauge	The current factor applied by the plug-in [0..1]	ThrottleFactor	observingBrokerId
FallbackThrottleFactorApplied	Counter	The number of times the plug-in has transitioned to using the fall back factor	ThrottleFactor	observingBrokerId
LimitViolated	Counter	A count of the number logDirs which violate the configured limit	ThrottleFactor	observingBrokerId
ActiveBrokers	Gauge	The current number of brokers returned by the describeCluster rpc	VolumeSource	observingBrokerId
ActiveLogDirs	Gauge	The number of logDirs returned by the describeLogDirs RPC	VolumeSource	observingBrokerId
available_bytes	Gauge	The number of available bytes returned by the describeLogDirs RPC	VolumeSource	[observingBrokerId, remoteBrokerId, logDir]
consumed_bytes	Gauge	The number of consumed bytes returned by the describeLogDirs RPC	VolumeSource	[observingBrokerId, remoteBrokerId, logDir]

Tag definitions

Tag	Definition
observingBrokerId	The BrokerId of the broker node executing the plug-in
remoteBrokerId	The BrokerId of the broker node hosting the logDir
logDir	The path to the specific logDir as returned by the describeLogDirs RPC

[robobario]

Instead of having the static metricName called from a few spots could we instead pass in some class with the metrics bits on it? Like new QuotaPluginMetrics(defaultTags) and call methods on it like metrics.volumeSource().incrementConsumedBytes(xyz). Then we could avoid some of the defaultTags fields and hopefully contain the metricName logic.

edit: looking at it, that's bit of a hefty refactor now with not much gain, happy with it as is and we could revisit later

[tombentley]

Thanks @SamBarker, there are a few nits, but this pretty good over all.

[ppatierno]

LGTM.

[tombentley]

A few more nits on the code. The other thing we should do is update the README to mention the availability and meaning of the metrics (you can probably copy and paste some of what you wrote from the PR)

[showuon]

LGTM! Nice tests!

[tombentley]

Overall this LGTM, thanks.

[SamBarker]

@ppatierno @tombentley I don't believe there is anything outstanding on this PR, if I'm right could one of you kindly hit the merge button? Otherwise please let me know what else needs looking at?