ci: improve collect-log script

I was debugging helm tests, I noticed, the operator namespace
content is empty and operator is created in same namespace as
other pods. Also, let's collect the 'kube-system' namespace logs
in multus test only as that is the only test where we need to debug
cluster networking and require kube-system logs.

Signed-off-by: subhamkrai <srai@redhat.com>

tests/scripts/collect-logs.sh

@@ -18,7 +18,6 @@ $CEPH_CMD osd dump >"${LOG_DIR}"/ceph-osd-dump.txt
 $CEPH_CMD report >"${LOG_DIR}"/ceph-report.txt
 
 NAMESPACES=("$CLUSTER_NAMESPACE")
-NAMESPACES+=("$KUBE_SYSTEM_NAMESPACE")
 if [[ "$OPERATOR_NAMESPACE" != "$CLUSTER_NAMESPACE" ]]; then
   NAMESPACES+=("$OPERATOR_NAMESPACE")
 fi
@@ -27,6 +26,11 @@ if [ "$1" == "multi-cluster-mirroring" ]; then
   NAMESPACES+=("rook-ceph-secondary")
 fi
 
+# Add kube-system namespace for multus test only as we need to debug network in multus test
+if [ "$1" == "canary-multus" ]; then
+  NAMESPACES+=("$KUBE_SYSTEM_NAMESPACE")
+fi
+
 for NAMESPACE in "${NAMESPACES[@]}"; do
   # each namespace is a sub-directory for easier debugging
   NS_DIR="${LOG_DIR}"/namespace-"${NAMESPACE}"