openstack_sd chosing random network-interface

[alfarme]

Bug Report

What did you do?

I set up openstack_sd with the below config.

What did you expect to see?
openstack_sd should discover nodes and use consistently use the same network-interface (10.x.x.x) on all servers.

What did you see instead? Under which circumstances?
openstack_sd successfully added all available nodes, but the chose network interface seems arbitrary. it also changes on each discovery run (every 60 seconds)

out of 44 nodes, between 6 and 10 get "re-discovered" with a new private_ip every 60 seconds.

** examples **

before:

address="10.150.x.x:9100"
__meta_openstack_instance_flavor="1310a6bc-3b91-47d0-ad50-9a50b9cf743f"
__meta_openstack_instance_id="1cda4a6c-fc4a-4de5-9cc5-13bba908333f"
__meta_openstack_instance_name="xxx"
__meta_openstack_instance_status="ACTIVE"
__meta_openstack_private_ip="10.150.x.x"
metrics_path="/metrics"
scheme="http"
job="openstack_dc1"

after:

(marked in yellow are the two servers from the 'before' picture)

address="100.100.x.x:9100"
__meta_openstack_instance_flavor="1310a6bc-3b91-47d0-ad50-9a50b9cf743f"
__meta_openstack_instance_id="1cda4a6c-fc4a-4de5-9cc5-13bba908333f"
__meta_openstack_instance_name="xxx"
__meta_openstack_instance_status="ACTIVE"
__meta_openstack_private_ip="100.100.x.x"
metrics_path="/metrics"
scheme="http"
job="openstack_dc1"

Environment

• System information:

Linux 4.4.70 x86_64`

• Prometheus version:

Version 1.8.1
Revision 3a7c51a
Branch HEAD
BuildUser root@ab78fb101775
BuildDate 20171023-15:50:57
GoVersion go1.9.1

• Prometheus configuration file: (relevant part)

- job_name: 'openstack_dc1'
    scrape_interval: 1m
    scrape_timeout: 10s
    metrics_path: /metrics
    scheme: http
    tls_config:
     insecure_skip_verify: true
    openstack_sd_configs:
    - identity_endpoint: 'https://xxx:5000/v2.0'
      username: <secret>
      userid: <secret>
      password: '<secret>'
      project_name: xxx
      project_id: xxx
      domain_name: default
      domain_id: default
      role: instance
      region: RegionOne
      refresh_interval: 1m
      port: 9100