submariner: Restart CoreDNS after lighthouse configures clusterset.local

subctl join patches the Corefile, but CoreDNS reload defaults to ~30s.
Until then clusterset.local misses the lighthouse forward and NXDOMAIN
is cached, delaying the first connectivity test after cluster start.
Restart CoreDNS once lighthouse is ready so the forward applies immediately.

Restarting CoreDNS adds 2 seconds delay to submariner start, but the
test takes now 20 seconds instead of 75 seconds. The actual connectivity
check takes now 0.1s instead of 54s (500x times faster).

Example run:

    2026-08-07 04:33:49,876 DEBUG   [hub/0] Restarting CoreDNS on cluster 'dr1'
    2026-08-07 04:33:49,937 DEBUG   [hub/0] deployment.apps/coredns restarted
    2026-08-07 04:33:49,938 DEBUG   [hub/0] Restarting CoreDNS on cluster 'dr2'
    2026-08-07 04:33:50,011 DEBUG   [hub/0] deployment.apps/coredns restarted
    2026-08-07 04:33:50,012 DEBUG   [hub/0] Waiting for CoreDNS on cluster 'dr1'
    2026-08-07 04:33:50,066 DEBUG   [hub/0] Waiting for deployment "coredns" rollout to finish: 0 of 1 updated replicas are available...
    2026-08-07 04:33:51,255 DEBUG   [hub/0] deployment "coredns" successfully rolled out
    2026-08-07 04:33:51,257 DEBUG   [hub/0] Waiting for CoreDNS on cluster 'dr2'
    2026-08-07 04:33:51,333 DEBUG   [hub/0] Waiting for deployment "coredns" rollout to finish: 0 of 1 updated replicas are available...
    2026-08-07 04:33:51,654 DEBUG   [hub/0] deployment "coredns" successfully rolled out
    ...
    2026-08-07 04:34:06,531 DEBUG   [hub/0] Accessing 'nginx.ns1.svc.clusterset.local.' in cluster 'dr2'
    2026-08-07 04:34:06,635 DEBUG   [hub/0] Access completed in 0.104 seconds
    2026-08-07 04:34:06,635 DEBUG   [hub/0] Accessing 'nginx.ns2.svc.clusterset.local.' in cluster 'dr1'
    2026-08-07 04:34:06,730 DEBUG   [hub/0] Access completed in 0.095 seconds

Assisted-by: Cursor/Cursor Grok 4.5 High
Signed-off-by: Nir Soffer <nsoffer@redhat.com>

Author: nirs

test/drenv/addons/submariner/start.py

@@ -35,6 +35,12 @@ def start(broker, *clusters):
     for cluster in clusters:
         wait_for_cluster(cluster)
 
+    for cluster in clusters:
+        restart_coredns(cluster)
+
+    for cluster in clusters:
+        wait_for_coredns(cluster)
+
 
 def deploy_broker(broker):
     print(f"Waiting until broker '{broker}' is ready")
@@ -106,6 +112,33 @@ def wait_for_cluster(cluster):
     wait_for_deployments(cluster, CLUSTER_DEPLOYMENTS, NAMESPACE)
 
 
+def restart_coredns(cluster):
+    """
+    Reload cluster CoreDNS after lighthouse has patched the Corefile.
+
+    CoreDNS reload defaults to ~30s; until then clusterset.local queries miss
+    the lighthouse forward and NXDOMAIN is cached.
+    """
+    print(f"Restarting CoreDNS on cluster '{cluster}'")
+    kubectl.rollout(
+        "restart",
+        "deployment/coredns",
+        "--namespace=kube-system",
+        context=cluster,
+    )
+
+
+def wait_for_coredns(cluster):
+    print(f"Waiting for CoreDNS on cluster '{cluster}'")
+    kubectl.rollout(
+        "status",
+        "deployment/coredns",
+        "--namespace=kube-system",
+        timeout=60,
+        context=cluster,
+    )
+
+
 def wait_for_deployments(cluster, names, namespace):
     for name in names:
         deployment = f"deploy/{name}"