Add workaround for tunnel connectivity

This patch creates a DeamonSet (pod on each node) that pings all the ipip tunnels of all the other nodes.
It should be removed once we find why the tunnels aren't open without pinging

Signed-off-by: afenner <andrew.fenner@est.tech>
Change-Id: Ic4e8241aa2a014daa0d299186d07615ee79030a2
Signed-off-by: afenner <andrew.fenner@est.tech>

diff --git a/apps/ceph/kubespray/playbooks/roles/install/tasks/main.yaml b/apps/ceph/kubespray/playbooks/roles/install/tasks/main.yaml
index b977018025287c576a429692c0f62c11c0269e37..2ff419116e112bb77941cfb8947780da88bb575d 100644 (file)
--- a/apps/ceph/kubespray/playbooks/roles/install/tasks/main.yaml
+++ b/apps/ceph/kubespray/playbooks/roles/install/tasks/main.yaml
@@ -76,6 +76,15 @@
   loop_control:
     loop_var: config_file
 
+- name: Implement Workaround for connectivity problem - ping all tunnels
+  k8s:
+    state: present
+    definition: "{{ lookup('template', config_file) }}"
+  with_items:
+    - ping-tunnel-workaround.yaml.j2
+  loop_control:
+    loop_var: config_file
+
 - name: Wait until OPERATOR pod is available
   k8s_facts:
     kind: Pod
@@ -88,7 +97,7 @@
   until:
     - rook_mgr_status.resources is defined
     - rook_mgr_status.resources
-  retries: 20
+  retries: 40
   delay: 5
 
 - name: Create rook cluster
@@ -111,7 +120,7 @@
   register: rook_cluster_status
   until:
     - rook_cluster_status.resources
-  retries: 10
+  retries: 20
   delay: 5
 
 - name: Wait until MGR pods are available
@@ -126,7 +135,7 @@
   until:
     - rook_mgr_status.resources is defined
     - rook_mgr_status.resources
-  retries: 30
+  retries: 40
   delay: 10
 
 - name: Wait until OSD pods are available
@@ -141,7 +150,7 @@
   until:
     - rook_osd_status.resources is defined
     - rook_osd_status.resources
-  retries: 30
+  retries: 60
   delay: 10
 
 - name: Create rook block storage

diff --git a/apps/ceph/kubespray/playbooks/roles/install/templates/ping-tunnel-workaround.yaml.j2 b/apps/ceph/kubespray/playbooks/roles/install/templates/ping-tunnel-workaround.yaml.j2
new file mode 100644 (file)
index 0000000..089f729
--- /dev/null
+++ b/apps/ceph/kubespray/playbooks/roles/install/templates/ping-tunnel-workaround.yaml.j2
@@ -0,0 +1,29 @@
+apiVersion: apps/v1
+kind: DaemonSet
+metadata:
+  name: pingtunnelworkaround
+  namespace: "{{ rook_namespace }}"
+  labels:
+    app: pingtunnelworkaround
+spec:
+  selector:
+    matchLabels:
+      app: pingtunnelworkaround
+  template:
+    metadata:
+      labels:
+        app: pingtunnelworkaround
+    spec:
+      tolerations:
+      # this toleration is to have the daemonset runnable on master nodes
+      # remove it if your masters can't run pods
+      - key: node-role.kubernetes.io/master
+        effect: NoSchedule
+      nodeSelector:
+        kubernetes.io/os: linux
+      hostNetwork: true
+      containers:
+      - name: busybox
+        image: {{ busybox_repository }}/library/busybox:{{ busybox_version }}
+        command: ["/bin/sh"]
+        args:  ["-c", "while true ; do ip route | grep tunl0 | awk -F/ '{print $1}' | xargs -n 1 ping -c 1; sleep 5; done"]

diff --git a/apps/ceph/kubespray/playbooks/roles/install/vars/offline-deployment.yaml b/apps/ceph/kubespray/playbooks/roles/install/vars/offline-deployment.yaml
index 9a4c206085b34d3060ff4dfa6d44e2df0fa79cd4..aea92a7ef5ff361924543e8a8c843e58fb550b9c 100644 (file)
--- a/apps/ceph/kubespray/playbooks/roles/install/vars/offline-deployment.yaml
+++ b/apps/ceph/kubespray/playbooks/roles/install/vars/offline-deployment.yaml
@@ -19,6 +19,7 @@
 
 ceph_repository: "{{ server_fqdn }}/ceph/ceph"
 rook_repository: "{{ server_fqdn }}/rook/ceph"
+busybox_repository: "{{ server_fqdn }}/library/busybox"
 cephcsi_repository: "{{ server_fqdn }}/cephcsi/cephcsi"
 csi_node_driver_registrar_repository: "{{ server_fqdn }}/k8scsi/csi-node-driver-registrar"
 csi_provisioner_repository: "{{ server_fqdn }}/k8scsi/csi-provisioner"

diff --git a/apps/ceph/kubespray/playbooks/roles/install/vars/online-deployment.yaml b/apps/ceph/kubespray/playbooks/roles/install/vars/online-deployment.yaml
index 21a9bb7c3b29819d6a31bfb4c013c5a65b02c4c3..76caf867d91b21301c4ccb67cfb226d13c48f79a 100644 (file)
--- a/apps/ceph/kubespray/playbooks/roles/install/vars/online-deployment.yaml
+++ b/apps/ceph/kubespray/playbooks/roles/install/vars/online-deployment.yaml
@@ -19,6 +19,7 @@
 
 ceph_repository: "docker.io/ceph/ceph"
 rook_repository: "rook/ceph"
+busybox_repository: "docker.io/library/busybox"
 cephcsi_repository: "quay.io/cephcsi/cephcsi"
 csi_node_driver_registrar_repository: "quay.io/k8scsi/csi-node-driver-registrar"
 csi_provisioner_repository: "quay.io/k8scsi/csi-provisioner"

diff --git a/vars/kubernetes.yaml b/vars/kubernetes.yaml
index fa13a643d7c3d9f9d2f522a3134c677b6476ec74..a5d1a2fa1a268a3264a53f196d242b09f9640c1c 100644 (file)
--- a/vars/kubernetes.yaml
+++ b/vars/kubernetes.yaml
@@ -69,6 +69,7 @@ kubectl_version: "{{ kubernetes_version }}"
 # Kubernetes: Versions of rook, ceph and their dependencies
 # -------------------------------------------------------------------------------
 rook_version: "v1.1.2"
+busybox_version: "1.32.0"
 ceph_version: "v14.2.4-20190917"
 cephcsi_version: "v1.2.1"
 csi_node_driver_registrar_version: "v1.1.0"