add cache metrics

[matsumana]

Hello.

With reference to #87, I added metrics the following.

Please review.

# HELP elasticsearch_indices_query_cache_cache_size Query cache cache size
# TYPE elasticsearch_indices_query_cache_cache_size gauge
elasticsearch_indices_query_cache_cache_size{cluster="test"} 28869
# HELP elasticsearch_indices_query_cache_cache_count Query cache cache count
# TYPE elasticsearch_indices_query_cache_cache_count gauge
elasticsearch_indices_query_cache_cache_count{cluster="test"} 28925
# HELP elasticsearch_indices_query_cache_total Query cache total count
# TYPE elasticsearch_indices_query_cache_total counter
elasticsearch_indices_query_cache_total{cluster="test"} 4.916801e+06
# HELP elasticsearch_indices_query_cache_count Query cache count
# TYPE elasticsearch_indices_query_cache_count counter
elasticsearch_indices_query_cache_count{category="hit",cluster="test"} 1.958521e+06
elasticsearch_indices_query_cache_count{category="miss",cluster="test"} 2.95828e+06
# HELP elasticsearch_indices_request_cache_count Request cache count
# TYPE elasticsearch_indices_request_cache_count counter
elasticsearch_indices_request_cache_count{category="hit",cluster="test"} 75749
elasticsearch_indices_request_cache_count{category="miss",cluster="test"} 65144

[zwopir]

Hi Matsumana,

thanks for you contribution. I added some comments to the code, mainly asking you to stick to the pattern we are already using. The additional type nodeMetricVec and its usage makes the code really hard to read.

[zwopir]

a don't think "category" describes the label correctly. I would then prefer "cache"

[zwopir]

"hit" and "miss" are the labelvalues for defaultCacheCategoryLabels, not labels. So it really confuses the code reader if they appear in a slice together with labels

[zwopir]

I don't see a reason adding a new type for metrics. The type nodeMetric already covers the usecase. Please use it. It is also pretty unclear what the dimensions of LabelValues func(...) [][]string represent. Same for Values func(...) []float64. A metric label can't be a slice

[zwopir]

NewCounterVec must not be called in the Collect func implementing prometheus.Collector. Throughout the collector we are using the pattern to create a new const metric. (prometheus.MustNewConstMetric). It perfectly fits to our use case.

[zwopir]

this metric returns false results regarding the labels: on our cluster we get:

# TYPE elasticsearch_indices_query_cache_count counter
elasticsearch_indices_query_cache_count{category="hit",cluster="cluster",host="host",name="name"} 3.1548049e+07
elasticsearch_indices_query_cache_count{category="miss",cluster="cluster",host="host",name="name"} 4.9529376e+07

(the label values for host and name are not (!) redacted)

[matsumana]

@zwopir Thank you for the comments.
I fixed.
Would you please review again?

[zwopir]

yes, that looks good - Thanks again for your contribution!