Feature(new_metrics): go-collector supports to collect and expose the metrics to Prometheus

[empiredan]

1. Prometheus metrics of old framework

In the old framework, all the metrics have the identical metric labels. The only difference is that for those metrics which have no attributes of app and partition, their values will be left empty as following example:

zion_profiler_RPC_RRDB_RRDB_GET_qps{cluster="pegasus_cluster",host_name="abc.xyz",pegasus_job="replica",port="34601",service="pegasus",app="",partition=""} 0.000000

For those metrics which have attributes of app and partition their values will certainly left non-empty:

replica_app_pegasus_get_qps{cluster="pegasus_cluster",host_name="abc.xyz",pegasus_job="replica",port="34801",service="pegasus",app="4",partition="2"} 0.000000

2. The Design of Prometheus metrics of new framework

Referring to the old framework and the features of new metrics system, the new Prometheus labels could be designed as below:

Label name	Description	Metric types	Entities
cluster	cluster name	-	-
role	role name, i.e. meta/replica	-	-
host	hostname of the server	-	-
port	the port of the role instance	-	-
entity	such as server/table/replica ...	-	-
table	table id	-	for table/partition/replica entity
partition	partition id	-	for partition/replica entity
p	the percentile, for example the value may be "90", "95", "999"	only for Percentile	-
task	the task name	-	only for profiler entity
dir	the disk directory	-	only for disk entity
queue	the queue name	-	only for queue entity
policy	the policy name	-	only for backup_policy entity
tracer	the latency_tracer description	-	only for latency_tracer entity
start	the starting point of latency_trace	-	only for latency_tracer entity
end	the end point of latency_tracer	-	only for latency_tracer entity