Prometheus plugin: incorrect counter retrieval for summary and histogram types

[tobischo]

Bug report

System info:

Telegraf 0.13.1

Steps to reproduce:

1. Setup a service with prometheus /metrics endpoint
2. Load data metrics endpoint of that service

Expected behavior:

When parsing the output anything with a _count should be properly parsed into the count field of the telegraf.Metric

Actual behavior:

The count value stays 0 and is sent with 0 in every dataset.
e.g.

http_request_duration_seconds,code=200,host=localhost,method=get,path=/metrics,url=http://localhost:4007/metrics 0.5=0.003071558,0.9=0.003609577,0.99=0.003609577,count=0,sum=0.009393333 1465218854910415916

Additional info:

There are actually tests for this, but they test it in the same wrong way, namely by expecting 0.
See for test data:

telegraf/plugins/inputs/prometheus/parser_test.go

Line 31 in bac1c22

http_request_duration_microseconds_count{handler="prometheus"} 9

See for (invalid) expectation:

telegraf/plugins/inputs/prometheus/parser_test.go

Line 141 in bac1c22

"count": 0.0,

The issue appears to be caused by selecting the wrong prometheus metrics type when filling the telegraf.Metric

See

telegraf/plugins/inputs/prometheus/parser.go

Line 77 in bac1c22

fields["count"] = float64(m.GetHistogram().GetSampleCount())

where actually the wrong type is selected for count but not for sum.

[sparrc]

closed by #1337