Merge pull request #328 from PierreBeucher/infra-fixes

Various infra fixes and small improvements

Author: PierreBeucher

ansible/roles/nvidia-driver/tasks/main.yml

@@ -7,6 +7,7 @@
   apt:
     name: "linux-modules-extra-{{ ansible_kernel }}"
     state: present
+    update_cache: true
 
 - name: unload nouveau
   modprobe:

ansible/roles/wolf/defaults/main.yml

@@ -17,4 +17,8 @@ wolf_project_dir: "{{ ansible_user_dir }}/wolf"
 # Where Wolf container data are stored
 wolf_data_dir: "/var/lib/cloudypad/wolf"
 
-wolf_legacy_data_dir: "/etc/wolf"
+wolf_legacy_data_dir: "/etc/wolf"
+
+# Whether to pull app images in parallel for faster subsequent startup when using Wolf
+# Disabled by default to avoid pulling images when not needed
+wolf_app_preheat_images_pull: false

ansible/roles/wolf/tasks/main.yml

@@ -7,6 +7,7 @@
   apt:
     name: "linux-modules-extra-{{ ansible_kernel }}"
     state: present
+    update_cache: true
 
 - name: Ensure uinput module enabled
   community.general.modprobe:
@@ -138,11 +139,13 @@
     files: docker-compose.nvidia.yml
 
 - name: Copy Docker Compose preheat pull file
+  when: wolf_app_preheat_images_pull
   ansible.builtin.template:
     src: docker-compose.app-preheat.yml
     dest: "{{ wolf_project_dir }}/docker-compose.app-preheat.yml"
 
 - name: Pull images using preheat Compose file
+  when: wolf_app_preheat_images_pull
   community.docker.docker_compose_v2_pull:
     project_src: /tmp
     files: "{{ wolf_project_dir }}/docker-compose.app-preheat.yml"

containers/sunshine/overlay/cloudy/bin/setup-nvidia-driver.sh

@@ -13,7 +13,7 @@
 
 set -e
 
-NVIDIA_DATA_DIR=$CLOUDYPAD_DATA_DIR/nvidia/
+NVIDIA_DATA_DIR="$CLOUDYPAD_DATA_DIR/nvidia"
 mkdir -p $NVIDIA_DATA_DIR
 
 if [ -z "${NVIDIA_DRIVER_VERSION}" ]; then
@@ -31,13 +31,20 @@ NVIDIA_INSTALLER="$NVIDIA_DATA_DIR/nvidia-${NVIDIA_DRIVER_TYPE}-${NVIDIA_DRIVER_
 # Create a marker file in XDG_RUNTIME_DIR to indicate that the driver is installed
 # This is used to avoid reinstalling the driver on container restart
 # But driver should be installed in freshly created container
+# As runtime dir is emptied when container is recreated
 NVIDIA_INSTALL_MARKER="$XDG_RUNTIME_DIR/nvidia-driver-${NVIDIA_DRIVER_TYPE}-${NVIDIA_DRIVER_VERSION}.installed"
 
+
 echo "Checking Nvidia driver ${NVIDIA_DRIVER_VERSION} (${NVIDIA_DRIVER_TYPE}) component installation..."
 
-# Check if the file already exists
-if [ ! -f "$NVIDIA_INSTALLER" ]; then
-    echo "NVIDIA driver version ${NVIDIA_DRIVER_VERSION} (type: '${NVIDIA_DRIVER_TYPE}') not found. Downloading..."
+# Check if installer file is not downloaded yet
+# Create a marker file in NVIDIA_DATA_DIR to indicate that the driver has been downloaded successfully
+# Use marker file to avoid checking if plain .run file exists as sometime file may be partially downloaded 
+# but failed in the middle, causing a corrupted file and/or a file with incorrect permissions
+NVIDIA_DOWNLOAD_MARKER="$NVIDIA_DATA_DIR/nvidia-driver-${NVIDIA_DRIVER_TYPE}-${NVIDIA_DRIVER_VERSION}.downloaded"
+
+if [ ! -f "$NVIDIA_DOWNLOAD_MARKER" ]; then
+    echo "NVIDIA driver version ${NVIDIA_DRIVER_VERSION} (type: '${NVIDIA_DRIVER_TYPE}') not downloaded yet. Downloading..."
 
     echo "Removing old NVIDIA driver installers..."
     find "$NVIDIA_DATA_DIR" -type f -name "nvidia-*.run" -exec rm -f {} +
@@ -54,8 +61,11 @@ if [ ! -f "$NVIDIA_INSTALLER" ]; then
     chmod +x "$NVIDIA_INSTALLER"
 
     echo "Downloaded $NVIDIA_INSTALLER"
+
+    # Create a marker file to indicate that the driver has been downloaded successfully
+    touch "$NVIDIA_DOWNLOAD_MARKER"
 else
-    echo "NVIDIA driver installer file already exists: $NVIDIA_INSTALLER"
+    echo "NVIDIA driver installer file already downloaded: $NVIDIA_INSTALLER ($NVIDIA_DOWNLOAD_MARKER file exists)"
 fi
 
 # Check if the installation marker file exists

containers/sunshine/overlay/cloudy/bin/stop-container-user.sh

@@ -1,4 +1,4 @@
 #!/usr/bin/env bash
 
 # Can be used by user to stop session
-gnome-terminal -- bash -c "stop-container-exit-prompt.sh" 
+xterm stop-container-exit-prompt.sh

containers/sunshine/overlay/cloudy/conf/xfce4-default/panel/launcher-5/exit.desktop

@@ -2,7 +2,7 @@
 Version=1.0
 Type=Application
 Name=Exit to host
-Comment=comment
+Comment=Exit to host
 Exec=/cloudy/bin/stop-container-user.sh
 Icon=system-shutdown
 Path=

docs/src/contributing/contribution-guide.md

@@ -442,6 +442,10 @@ Instance is reachable via `192.168.56.43`.
 To pair, connect to the Sunshine web UI and use Moonlight manually:
 
 ```sh
+# A TLS certificate warning is shown as certificate is self-signed
+# login: sunshine
+# password: @!/:,?!*#'€`_\µ$="foo
+#   or 'sunshine'
 https://192.168.56.43:47990/
 ```
 

src/cli/updater.ts

@@ -53,10 +53,6 @@ export class InteractiveInstanceUpdater<ST extends InstanceStateV1, A extends Up
 
         const manager = await this.providerClient.getInstanceManager(instanceName)
 
-        // ensure instance is started for deployment
-        // as deployment will not necessarily start instance if it's stopped
-        await manager.start()
-
         await manager.deploy()
     }
 

src/core/manager.ts

@@ -259,6 +259,14 @@ const ALWAYS_ANSIBLE_ADDITIONAL_ARGS = ['-e', '\'ansible_ssh_common_args="-o Str
  * 
  * The concrete instance type is not known by this class: a per-provider factory is used 
  * to build each sub-managers.
+ * 
+ * Each action function (stop, start, configure, provision, etc.) is a wrapper with retry around
+ * low-level action functions (doStart, doStop, doConfigure, doProvision, etc.).
+ * - a top-level action like deploy() may rely on other top-level actions like configure()
+ *   or call underlying low-level actions like doConfigure wrapped in a retry().
+ *   They may also container logic like retry pattern, starting instance before configuring, etc.
+ * - a low-level action like doConfigure() must NOT call another low-level NOR top-level action.
+ *   These functions should only strictly do the required action without any mre logic.
  */
 export class GenericInstanceManager<ST extends InstanceStateV1> implements InstanceManager {
 
@@ -295,9 +303,23 @@ export class GenericInstanceManager<ST extends InstanceStateV1> implements Insta
             [currentState.configuration.input.ansible.additionalArgs] : undefined
 
         await this.addEvent(InstanceEventEnum.ConfigurationBegin)
+
+        // before configuring, check instance is running as it's possible user would call configure() directly without starting the instance first
+        // start it if needed
+        const currentStatus = await this.getInstanceStatus()
+        if(currentStatus.serverStatus !== ServerRunningStatus.Running){
+
+            this.logger.debug(`About to run configuration for instance ${this.name()} but it's not running, starting it first...`)
+
+            await this.doWithRetry(async () => {
+                await this.doStart({ wait: true })
+            }, 'Pre-configure start', opts)
+        }
+
         await this.doWithRetry(async () => {
             await this.doConfigure(configurationAnsibleAdditionalArgs)
         }, 'Configuration', opts)
+        
         await this.addEvent(InstanceEventEnum.ConfigurationEnd)
     }
 
@@ -331,6 +353,15 @@ export class GenericInstanceManager<ST extends InstanceStateV1> implements Insta
 
         this.logger.debug(`Starting instance ${this.name()}`)
 
+        // if deleteInstanceServerOnStop is enabled, instance server may have been deleted on last instance stop
+        // so we need to re-provision and re-configure instance using specific Ansible args
+        await this.doWithRetry(async () => {
+            if(this.args.options?.deleteInstanceServerOnStop?.enabled){
+                await this.doProvision(opts)
+                await this.doConfigure(this.args.options.deleteInstanceServerOnStop.postStartReconfigurationAnsibleAdditionalArgs)
+            }
+        }, 'Pre-start reconfiguration', opts)
+
         await this.addEvent(InstanceEventEnum.StartBegin)
         await this.doWithRetry(async () => {
             await this.doStart(opts)
@@ -401,6 +432,7 @@ export class GenericInstanceManager<ST extends InstanceStateV1> implements Insta
         const configurator = await this.buildConfigurator({
             additionalAnsibleArgs: ALWAYS_ANSIBLE_ADDITIONAL_ARGS.concat(additionalAnsibleArgs ?? [])
         })
+
         const output = await configurator.configure()
 
         this.logger.debug(`Configuration output for instance ${this.name()}: ${JSON.stringify(output)}`)
@@ -431,13 +463,6 @@ export class GenericInstanceManager<ST extends InstanceStateV1> implements Insta
      */
     async doStart(opts?: StartOptions): Promise<void> {
 
-        // if deleteInstanceServerOnStop is enabled, instance server may have been deleted on last instance stop
-        // so we need to re-provision and re-configure instance using specific Ansible args
-        if(this.args.options?.deleteInstanceServerOnStop?.enabled){
-            await this.doProvision(opts)
-            await this.doConfigure(this.args.options.deleteInstanceServerOnStop.postStartReconfigurationAnsibleAdditionalArgs)
-        }
-
         // always start instance as provisioning and configuring may not start the server for all providers
         // and startOptions logic is ported by runner
         const runner = await this.buildRunner()

src/core/moonlight/pairer/sunshine.ts

@@ -28,13 +28,11 @@ export class SunshineMoonlightPairer extends AbstractMoonlightPairer implements
     }
 
     private buildSshClient(): SSHClient {
+        this.logger.debug(`Building SSH client with args: ${JSON.stringify(this.args.ssh)}`)
         return new SSHClient(this.args.ssh);
     }
 
-    protected async doPair(pin: string) {
-
-        const ssh = this.buildSshClient()
-    
+    protected async doPair(pin: string) {    
         // try to pair for 2 min by punshing through Sunshine API with our pin
         // Using plain curl + SSH command as Sunshine Web UI is not reachable directly
         // A simple but effective enough method