From 5ad4fc0857f47a82f0702e9061c7e1769a6cc204 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Martin=20Andr=C3=A9?= <m.andre@redhat.com>
Date: Tue, 12 May 2026 14:29:36 +0200
Subject: [PATCH] OCPBUGS-85457: Add init container for Manila node daemonset

The Manila CSI driver node plugin fatally exits on startup when the NFS
CSI plugin socket (/var/lib/kubelet/plugins/csi-nfsplugin/csi.sock) is
not yet available. After node disruptions, both DaemonSets restart
concurrently and the Manila driver crashes because upstream ProbeForever
only retries on DeadlineExceeded, not on Unavailable (connection
refused). This causes excessive container restarts (>3), failing the CI
invariant test.

Add a wait-for-nfs-plugin init container to the Manila node DaemonSet
that polls for the NFS plugin socket before allowing the csi-driver
container to start. This ensures proper startup ordering between the
two DaemonSets without requiring changes to the upstream driver.
---
 .../generated/hypershift/node.yaml            | 22 +++++++++++++++++++
 .../generated/standalone/node.yaml            | 22 +++++++++++++++++++
 .../patches/node_add_driver.yaml              | 22 +++++++++++++++++++
 3 files changed, 66 insertions(+)

diff --git a/assets/overlays/openstack-manila/generated/hypershift/node.yaml b/assets/overlays/openstack-manila/generated/hypershift/node.yaml
index d4e48ab14..2931a8080 100644
--- a/assets/overlays/openstack-manila/generated/hypershift/node.yaml
+++ b/assets/overlays/openstack-manila/generated/hypershift/node.yaml
@@ -158,6 +158,28 @@ spec:
           name: registration-dir
       dnsPolicy: ClusterFirstWithHostNet
       hostNetwork: true
+      initContainers:
+      - command:
+        - sh
+        - -c
+        - |
+          until [ -S /var/lib/kubelet/plugins/csi-nfsplugin/csi.sock ]; do
+            echo "Waiting for NFS CSI plugin socket..."
+            sleep 2
+          done
+        image: ${DRIVER_IMAGE}
+        name: wait-for-nfs-plugin
+        resources:
+          requests:
+            cpu: 10m
+            memory: 50Mi
+        securityContext:
+          privileged: false
+          readOnlyRootFilesystem: true
+        terminationMessagePolicy: FallbackToLogsOnError
+        volumeMounts:
+        - mountPath: /var/lib/kubelet/plugins/csi-nfsplugin
+          name: fwd-plugin-dir
       nodeSelector:
         kubernetes.io/os: linux
       priorityClassName: system-node-critical
diff --git a/assets/overlays/openstack-manila/generated/standalone/node.yaml b/assets/overlays/openstack-manila/generated/standalone/node.yaml
index d4e48ab14..2931a8080 100644
--- a/assets/overlays/openstack-manila/generated/standalone/node.yaml
+++ b/assets/overlays/openstack-manila/generated/standalone/node.yaml
@@ -158,6 +158,28 @@ spec:
           name: registration-dir
       dnsPolicy: ClusterFirstWithHostNet
       hostNetwork: true
+      initContainers:
+      - command:
+        - sh
+        - -c
+        - |
+          until [ -S /var/lib/kubelet/plugins/csi-nfsplugin/csi.sock ]; do
+            echo "Waiting for NFS CSI plugin socket..."
+            sleep 2
+          done
+        image: ${DRIVER_IMAGE}
+        name: wait-for-nfs-plugin
+        resources:
+          requests:
+            cpu: 10m
+            memory: 50Mi
+        securityContext:
+          privileged: false
+          readOnlyRootFilesystem: true
+        terminationMessagePolicy: FallbackToLogsOnError
+        volumeMounts:
+        - mountPath: /var/lib/kubelet/plugins/csi-nfsplugin
+          name: fwd-plugin-dir
       nodeSelector:
         kubernetes.io/os: linux
       priorityClassName: system-node-critical
diff --git a/assets/overlays/openstack-manila/patches/node_add_driver.yaml b/assets/overlays/openstack-manila/patches/node_add_driver.yaml
index ecff0460f..854bed3c1 100644
--- a/assets/overlays/openstack-manila/patches/node_add_driver.yaml
+++ b/assets/overlays/openstack-manila/patches/node_add_driver.yaml
@@ -90,6 +90,28 @@ spec:
               cpu: 10m
               memory: 50Mi
           terminationMessagePolicy: FallbackToLogsOnError
+      initContainers:
+        - name: wait-for-nfs-plugin
+          image: ${DRIVER_IMAGE}
+          command:
+            - sh
+            - -c
+            - |
+              until [ -S /var/lib/kubelet/plugins/csi-nfsplugin/csi.sock ]; do
+                echo "Waiting for NFS CSI plugin socket..."
+                sleep 2
+              done
+          securityContext:
+            readOnlyRootFilesystem: true
+            privileged: false
+          resources:
+            requests:
+              cpu: 10m
+              memory: 50Mi
+          terminationMessagePolicy: FallbackToLogsOnError
+          volumeMounts:
+            - name: fwd-plugin-dir
+              mountPath: /var/lib/kubelet/plugins/csi-nfsplugin
       volumes:
         - name: cloud-credentials
           secret: