Custom dns policy makes sdp-dnsmasq crash

[thomascellerier]

Given a pod including the following spec:

     dnsConfig:
       nameservers:
       - 10.97.2.20
       - 10.97.2.23
     dnsPolicy: None

Results in the sdp-dnsmasq sidecar failing to start:

Unable to get the kube dns service!. Dying now!

We should at least detect the issue and emit a relevant error.

[lepiolet]

Right now we detect it but the error is really far from relevant :D

Since we don't support that option (I don't really know if there is a user case for it) we can just let the injector to fail the validation of the POD in that case.

[thomascellerier]

Yes let's just make it clear the dnsPolicy: None is not supported right now.
We should be explicit about which dnsPolicy we support.

[lepiolet]

#61 This PR made the injector to override the dnsConfig always into something like this

   dnsConfig:
       nameservers:
       - 127.0.0.1
       searches:
       - svc.cluster.local
       - cluster.local
     dnsPolicy: None```

where searches can be configured.

The injector discovers the DNS service in the cluster and pass it to sdp-dnsmasq container to configure the DNS to be compatible with the sdp-client

[anoncam]

what does your Corefile look like?

[anoncam]

Have you tried adding something like:

spec:
  hostname: 
  subdomain: 
  containers:
  - image: # image ref
    name: sdp # name reference to corresponding service `metadata.name`
---
# associated service
apiVersion: v1
kind: Service
metadata:
  name: sdp # or something similar
spec:
  selector:
    name: sdp
  clusterIP: None
  ports:
  - name: foo # Actually, no port is needed.
    port: 1234
    targetPort: 1234

using approach outlined here

[anoncam]

FIX:: can't use reserved port...ie: 53.

use 50053 or something that ur non-root user can open without needing privs ... or got back to running as root.