Fix coredns checks

The readiness probe and helm test currently rely on
the return code of `dig`, which doesn't fail for
NXDOMAIN, which means they are not checking that
the names are actually resolved. This moves to using
`host` instead which does check this.

This also removes the checks for kubernetes etcd
domain names, since that doesn't get deployed until
after coredns.

Change-Id: I0b459f52663c936ed4b8b216614c5b4824a0713f

charts/coredns/templates/bin/_probe.py.tpl

@@ -17,12 +17,24 @@ class httpHandler(BaseHTTPRequestHandler):
         failed = False
         res = requests.get("http://127.0.0.1:{}/health".format(args.check_port))
         if res.status_code >= 400:
+          print('Failed /health check, status code = : {}'.format(res.status_code))
           failed = True
-        res = subprocess.run(
+
-                ["dig", "+time=2", "+tries=1", "@127.0.0.1", "-f", args.filename],
+        with open(args.filename, 'r') as fh:
-                stdout=subprocess.DEVNULL)
+          for host in fh.read().splitlines():
-        if res.returncode != 0:
+            # ignore blank lines
-          failed = True
+            if not host:
+              continue
+            res = subprocess.run(
+              ["host", "-W=2", "-R=1", host, "127.0.0.1"],
+              stdout=subprocess.PIPE,
+              stderr=subprocess.STDOUT)
+            if res.returncode != 0:
+              print('Failed to resolve host: "{}"'.format(host))
+              print(res.stdout)
+              failed = True
+              break
+
         if failed:
           print('Check failed')
           self.send_response(500)

charts/coredns/templates/configmap-list.yaml

@@ -4,5 +4,5 @@ kind: ConfigMap
 metadata:
   name: {{ .Values.service.name }}-list
 data:
-  names_to_resolve: |
+  names_to_resolve: |-
 {{ tuple "etc/_list.tpl" . | include "helm-toolkit.utils.template" | indent 4 }}

charts/coredns/templates/etc/_list.tpl

@@ -1,3 +1,3 @@
-{{- range .Values.conf.test.names_to_resolve }}
+{{- range .Values.conf.test.names_to_resolve -}}
 {{ . }}
-{{- end }}
+{{ end -}}

charts/coredns/templates/pod-test.yaml

@@ -40,16 +40,26 @@ spec:
         - -c
         - |
           SUCCESS=1
-          {{- range .Values.conf.test.names_to_resolve }}
+          while read host; do
-          if dig {{ . }}; then
+            if [ -n "$host" ]; then
-            echo "Successfully resolved {{ . }}"
+              if host "$host"; then
-          else
+                echo "Successfully resolved: \"$host\""
-            echo "Failed to resolve {{ . }}"
+              else
-            SUCCESS=0
+                echo "Failed to resolve: \"$host\""
-          fi
+                SUCCESS=0
-          {{- end }}
+              fi
+            fi
+          done < /tmp/etc/names_to_resolve
           if [ "$SUCCESS" != "1" ]; then
             echo "Test failed to resolve all names."
             exit 1
           fi
+      volumeMounts:
+        - name: dns-names
+          mountPath: /tmp/etc
+  volumes:
+    - name: dns-names
+      configMap:
+        name: {{ $envAll.Values.service.name }}-list
+        defaultMode: 0555
 {{- end }}

examples/basic/armada-resources.yaml

@@ -563,7 +563,6 @@ data:
       test:
         names_to_resolve:
           - calico-etcd.kube-system.svc.cluster.local
-          - kubernetes-etcd.kube-system.svc.cluster.local
           - kubernetes.default.svc.cluster.local
     images:
       tags:

examples/complete/armada-resources.yaml

@@ -584,7 +584,6 @@ data:
       test:
         names_to_resolve:
           - calico-etcd.kube-system.svc.cluster.local
-          - kubernetes-etcd.kube-system.svc.cluster.local
           - kubernetes.default.svc.cluster.local
 
     images:

examples/containerd/armada-resources.yaml

@@ -438,7 +438,6 @@ data:
           - att.com
           - calico-etcd.kube-system.svc.cluster.local
           - google.com
-          - kubernetes-etcd.kube-system.svc.cluster.local
           - kubernetes.default.svc.cluster.local
 
     images:

examples/gate/armada-resources.yaml

@@ -459,7 +459,6 @@ data:
           - att.com
           - calico-etcd.kube-system.svc.cluster.local
           - google.com
-          - kubernetes-etcd.kube-system.svc.cluster.local
           - kubernetes.default.svc.cluster.local
 
     images: