[ntp] Undefined behaviour with multiple instances

[masci]

Describe what happened:

If you list multiple instances in the config file of the ntp check, the check would be run once per instance with metrics and service checks posted without any tag, see https://github.com/DataDog/datadog-agent/blob/master/pkg/collector/corechecks/network/ntp.go#L146

Describe what you expected:

To be able to distinguish which value is coming from which NTP server.

[ls339]

Test
Configure 3 ntp instances in conf.d/ntp.d/check.yaml

init_config:

instances:
  - offset_threshold: 60
    host: 0.asia.pool.ntp.org

  - offset_threshold: 60
    host: 0.north-america.pool.ntp.org

  - offset_threshold: 60
    host: localhost # We expect this to fail  

Result
We can't distinguish between instances if more than one instance is configured in the ntp check. Metric ntp.offset and service check ntp.in_sync will be submitted once per instance run with only a host tag so we will not be able to differentiate between the metric value and the service check status.
It's not recommended to specify more than 1 instance in an ntp check because the service check ntp.in_sync may flap if one of the configured instances does not respond. In our test, we see that instance 2 has a status of 3 (unknown) causing our monitor to flap.

instance 0

Service Checks:
[
  {
    "check": "ntp.in_sync",
    "host_name": "ntpinvestigation",
    "timestamp": 1514998223,
    "status": 0,
    "message": "",
    "tags": null
  }
]

instance 1

Service Checks:
[
  {
    "check": "ntp.in_sync",
    "host_name": "ntpinvestigation",
    "timestamp": 1514998224,
    "status": 0,
    "message": "",
    "tags": null
  }
]

instance 2

Service Checks:
[
  {
    "check": "ntp.in_sync",
    "host_name": "ntpinvestigation",
    "timestamp": 1514998224,
    "status": 3,
    "message": "",
    "tags": null
  }
]