From 5878b2263a6c38ac6ca786e663ca3db97fb9ca2c Mon Sep 17 00:00:00 2001
From: afenner <andrew.fenner@est.tech>
Date: Mon, 27 Jul 2020 14:53:19 +0100
Subject: [PATCH 1/1] 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>
---
 .../playbooks/roles/install/tasks/main.yaml   |  9 ++++++
 .../templates/ping-tunnel-workaround.yaml.j2  | 29 +++++++++++++++++++
 .../install/vars/offline-deployment.yaml      |  1 +
 .../roles/install/vars/online-deployment.yaml |  1 +
 vars/kubernetes.yaml                          |  1 +
 5 files changed, 41 insertions(+)
 create mode 100644 apps/ceph/kubespray/playbooks/roles/install/templates/ping-tunnel-workaround.yaml.j2

diff --git a/apps/ceph/kubespray/playbooks/roles/install/tasks/main.yaml b/apps/ceph/kubespray/playbooks/roles/install/tasks/main.yaml
index b977018..90fb05b 100644
--- 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
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
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 9a4c206..aea92a7 100644
--- 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 21a9bb7..76caf86 100644
--- 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 fed310c..f533d2e 100644
--- a/vars/kubernetes.yaml
+++ b/vars/kubernetes.yaml
@@ -80,6 +80,7 @@ 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"
-- 
2.25.1