[Metricbeat] Don't require instance_name when there's a single perfmon counter instance

[andrewkroh]

• Beat: Metricbeat
• Module: windows
• Metricset: perfmon
• Version: 6.2.2
• Operating System: Windows
• Steps to Reproduce:
  • Monitor a counter that does not have multiple instances. For example, query: '\UDPv4\Datagrams Sent/sec'.
  • Don't set the instance_name configuration option.
  • Start Metricbeat and you will get the error of: Exiting: 1 error: 1 error: initialization failed: query doesn't contain an instance name. In this case you have to define 'instance_name'

If a counter does not have multiple instances then the instance_label and instance_name options don't need to be used or required. The general structure for a counter path is: \\PerfObject( ParentInstance/ObjectInstance#InstanceIndex )\\Counter. And it seems that the instance section is optional.

So my proposal is to not require instance_name when the counter path doesn't include any "instance" information. This will likely lead to wanting to be able to group related counter data into a single event, like all counters from the UDPv4 perf object into a single event.

[crossan007]

I think I'm seeing this issue with the following as well:

    - instance_label: memory
      measurement_label: memory.committed.pct
      query: "\\Memory\\% Committed Bytes In Use"

It seems like #4944 might be related - in that, I get the same errors with the non-instance_named queries on that report; despite the reporter seeming to get data from the queries

[crossan007]

Update - I was completely off-base in my assessment of my environment exhibiting this issue.

I did have some confusion about the behavior of instance_name; however, the issue I was facing stemmed from multiple tweaks to my perfmon.counters collection in windows.yml after having already added perfmon documents to the ES index; thereby causing the index pattern to not fit the incoming perfmon data. This caused new perfmon data to fail to parse and be inserted to the index. The symptoms were that the metricbeat.log file showed everything as "fine", but no data was showing up in kibana. My "ah-ha" moment came from examining my logstash logs.