Admin options adjustments (different compute devices support) (#267)

Minor fixes, adjusted Admin settings UI and backend part with optional compute device (NVIDIA/ADM/CPU) daemon configuration. --------- Signed-off-by: Andrey Borysenko <[email protected]>
nextcloud · Apr 9, 2024 · 9fb115b · 9fb115b
1 parent 5acfac7
commit 9fb115b
Show file tree

Hide file tree

Showing 18 changed files with 321 additions and 138 deletions.
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -5,6 +5,12 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](http://keepachangelog.com/)
 and this project adheres to [Semantic Versioning](http://semver.org/).
 
+## [2.5.0 - 2024-04-xx]
+
+### Added
+
+- Different compute device configuration for Daemon (NVIDIA, AMD, CPU)
+
 ## [2.4.0 - 2024-04-04]
 
 ### Added

diff --git a/appinfo/info.xml b/appinfo/info.xml
@@ -43,7 +43,7 @@ to join us in shaping a more versatile, stable, and secure app landscape.
 *Your insights, suggestions, and contributions are invaluable to us.*
 
 	]]></description>
-	<version>2.4.0</version>
+	<version>2.5.0</version>
 	<licence>agpl</licence>
 	<author mail="[email protected]" homepage="https://github.com/andrey18106">Andrey Borysenko</author>
 	<author mail="[email protected]" homepage="https://github.com/bigcat88">Alexander Piskun</author>
@@ -72,6 +72,7 @@ to join us in shaping a more versatile, stable, and secure app landscape.
 		<install>
 			<step>OCA\AppAPI\Migration\DataInitializationStep</step>
 			<step>OCA\AppAPI\Migration\DaemonUpdateV2RepairStep</step>
+			<step>OCA\AppAPI\Migration\DaemonUpdateGPUSRepairStep</step>
 		</install>
 	</repair-steps>
 	<commands>

diff --git a/docs/CreationOfDeployDaemon.rst b/docs/CreationOfDeployDaemon.rst
@@ -32,7 +32,7 @@ Register
 
 Register Deploy Daemon (DaemonConfig).
 
-Command: ``app_api:daemon:register [--net NET] [--gpu] [--] <name> <display-name> <accepts-deploy-id> <protocol> <host> <nextcloud_url>``
+Command: ``app_api:daemon:register [--net NET] [--haproxy_password HAPROXY_PASSWORD] [--compute_device COMPUTE_DEVICE] [--set-default] [--] <name> <display-name> <accepts-deploy-id> <protocol> <host> <nextcloud_url>``
 
 Arguments
 *********
@@ -49,7 +49,7 @@ Options
 
 	* ``--net [network-name]``  - ``[required]`` network name to bind docker container to (default: ``host``)
 	* ``--haproxy_password HAPROXY_PASSWORD`` - ``[optional]`` password for AppAPI Docker Socket Proxy
-	* ``--gpu GPU`` - ``[optional]`` GPU device to expose to the daemon (e.g. ``/dev/dri``)
+	* ``--compute_device GPU`` - ``[optional]`` GPU device to expose to the daemon (e.g. ``cpu|cuda|rocm``, default: ``cpu``)
 	* ``--set-default`` - ``[optional]`` set created daemon as default for ExApps installation
 
 DeployConfig
@@ -64,7 +64,10 @@ ExApp container.
 		"net": "host",
 		"nextcloud_url": "https://nextcloud.local",
 		"haproxy_password": "some_secure_password",
-		"gpus": true,
+		"computeDevice": {
+			"id": "cuda",
+			"name": "CUDA (NVIDIA)",
+		},
 	}
 
 DeployConfig options
@@ -73,7 +76,7 @@ DeployConfig options
 	* ``net`` **[required]** - network name to bind docker container to (default: ``host``)
 	* ``nextcloud_url`` **[required]** - Nextcloud URL (e.g. ``https://nextcloud.local``)
 	* ``haproxy_password`` *[optional]* - password for AppAPI Docker Socket Proxy
-	* ``gpus`` *[optional]* - GPU device to attach to the daemon (e.g. ``/dev/dri``)
+	* ``computeDevice`` *[optional]* - Compute device to attach to the daemon (e.g. ``{ "id": "cuda", "label": "CUDA (NVIDIA)" }``)
 
 Unregister
 ----------

diff --git a/lib/Command/Daemon/RegisterDaemon.php b/lib/Command/Daemon/RegisterDaemon.php
@@ -38,11 +38,12 @@ protected function configure(): void {
 		$this->addOption('net', null, InputOption::VALUE_REQUIRED, 'DeployConfig, the name of the docker network to attach App to');
 		$this->addOption('haproxy_password', null, InputOption::VALUE_REQUIRED, 'AppAPI Docker Socket Proxy password for HAProxy Basic auth');
 
-		$this->addOption('gpu', null, InputOption::VALUE_NONE, 'Enable support of GPUs for containers');
+		$this->addOption('compute_device', null, InputOption::VALUE_REQUIRED, 'Compute device for GPU support (cpu|cuda|rocm)');
 
 		$this->addOption('set-default', null, InputOption::VALUE_NONE, 'Set DaemonConfig as default');
 
 		$this->addUsage('local_docker "Docker local" "docker-install" "http" "/var/run/docker.sock" "http://nextcloud.local" --net=nextcloud');
+		$this->addUsage('local_docker "Docker local" "docker-install" "http" "/var/run/docker.sock" "http://nextcloud.local" --net=nextcloud --set-default --compute_device=cuda');
 	}
 
 	protected function execute(InputInterface $input, OutputInterface $output): int {
@@ -57,7 +58,7 @@ protected function execute(InputInterface $input, OutputInterface $output): int
 			'net' => $input->getOption('net') ?? 'host',
 			'nextcloud_url' => $nextcloudUrl,
 			'haproxy_password' => $input->getOption('haproxy_password') ?? '',
-			'gpu' => $input->getOption('gpu') ?? false,
+			'computeDevice' => $this->buildComputeDevice($input->getOption('compute_device') ?? 'cpu'),
 		];
 
 		if (($protocol !== 'http') && ($protocol !== 'https')) {
@@ -95,4 +96,26 @@ protected function execute(InputInterface $input, OutputInterface $output): int
 		$output->writeln('Daemon successfully registered.');
 		return 0;
 	}
+
+	private function buildComputeDevice(string $computeDevice): array {
+		switch ($computeDevice) {
+			case 'cpu':
+				return [
+					'id' => 'cpu',
+					'label' => 'CPU',
+				];
+			case 'cuda':
+				return [
+					'id' => 'cuda',
+					'label' => 'CUDA (NVIDIA)',
+				];
+			case 'rocm':
+				return [
+					'id' => 'rocm',
+					'label' => 'ROCm (AMD)',
+				];
+			default:
+				throw new \InvalidArgumentException('Invalid compute device value.');
+		}
+	}
 }
diff --git a/lib/DeployActions/AIODockerActions.php b/lib/DeployActions/AIODockerActions.php
@@ -14,7 +14,6 @@
  */
 class AIODockerActions {
 	public const AIO_DAEMON_CONFIG_NAME = 'docker_aio';
-	public const AIO_DAEMON_CONFIG_NAME_GPU = 'docker_aio_gpu';
 	public const AIO_DOCKER_SOCKET_PROXY_HOST = 'nextcloud-aio-docker-socket-proxy:2375';
 
 	public function __construct(
@@ -46,13 +45,12 @@ public function registerAIODaemonConfig(): ?DaemonConfig {
 			'net' => 'nextcloud-aio', // using the same host as default network for Nextcloud AIO containers
 			'nextcloud_url' => 'https://' . getenv('NC_DOMAIN'),
 			'haproxy_password' => null,
-			'gpu' => false,
+			'computeDevice' => [
+				'id' => 'cpu',
+				'label' => 'CPU',
+			],
 		];
 
-		if ($this->isGPUsEnabled()) {
-			$this->registerAIODaemonConfigWithGPU();
-		}
-
 		$daemonConfigParams = [
 			'name' => self::AIO_DAEMON_CONFIG_NAME,
 			'display_name' => 'AIO Docker Socket Proxy',
@@ -68,44 +66,4 @@ public function registerAIODaemonConfig(): ?DaemonConfig {
 		}
 		return $daemonConfig;
 	}
-
-	/**
-	 * Registers DaemonConfig with default params to use AIO Docker Socket Proxy with GPU
-	 */
-	private function registerAIODaemonConfigWithGPU(): ?DaemonConfig {
-		$daemonConfigWithGPU = $this->daemonConfigService->getDaemonConfigByName(self::AIO_DAEMON_CONFIG_NAME_GPU);
-		if ($daemonConfigWithGPU !== null) {
-			return $daemonConfigWithGPU;
-		}
-
-		$deployConfig = [
-			'net' => 'nextcloud-aio', // using the same host as default network for Nextcloud AIO containers
-			'nextcloud_url' => 'https://' . getenv('NC_DOMAIN'),
-			'haproxy_password' => null,
-			'gpu' => true,
-		];
-
-		$daemonConfigParams = [
-			'name' => self::AIO_DAEMON_CONFIG_NAME_GPU,
-			'display_name' => 'AIO Docker Socket Proxy with GPU',
-			'accepts_deploy_id' => 'docker-install',
-			'protocol' => 'http',
-			'host' => self::AIO_DOCKER_SOCKET_PROXY_HOST,
-			'deploy_config' => $deployConfig,
-		];
-
-		return $this->daemonConfigService->registerDaemonConfig($daemonConfigParams);
-	}
-
-	/**
-	 * Check if /dev/dri folder mounted to the container.
-	 * In AIO this means that NEXTCLOUD_ENABLE_DRI_DEVICE=true
-	 */
-	private function isGPUsEnabled(): bool {
-		$devDri = '/dev/dri';
-		if (is_dir($devDri)) {
-			return true;
-		}
-		return false;
-	}
 }
diff --git a/lib/DeployActions/DockerActions.php b/lib/DeployActions/DockerActions.php
@@ -125,11 +125,20 @@ public function createContainer(string $dockerUrl, array $imageParams, array $pa
 			$containerParams['NetworkingConfig'] = $networkingConfig;
 		}
 
-		if (isset($params['gpu']) && filter_var($params['gpu'], FILTER_VALIDATE_BOOLEAN)) {
-			if (isset($params['deviceRequests'])) {
-				$containerParams['HostConfig']['DeviceRequests'] = $params['deviceRequests'];
-			} else {
-				$containerParams['HostConfig']['DeviceRequests'] = $this->buildDefaultGPUDeviceRequests();
+		if (isset($params['computeDevice'])) {
+			if ($params['computeDevice']['id'] === 'cuda') {
+				if (isset($params['deviceRequests'])) {
+					$containerParams['HostConfig']['DeviceRequests'] = $params['deviceRequests'];
+				} else {
+					$containerParams['HostConfig']['DeviceRequests'] = $this->buildDefaultGPUDeviceRequests();
+				}
+			}
+			if ($params['computeDevice']['id'] === 'rocm') {
+				if (isset($params['devices'])) {
+					$containerParams['HostConfig']['Devices'] = $params['devices'];
+				} else {
+					$containerParams['HostConfig']['Devices'] = $this->buildDevicesParams(['/dev/kfd', '/dev/dri']);
+				}
 			}
 		}
 
@@ -346,10 +355,15 @@ public function buildDeployParams(DaemonConfig $daemonConfig, array $appInfo): a
 		$externalApp = $appInfo['external-app'];
 		$deployConfig = $daemonConfig->getDeployConfig();
 
-		if (isset($deployConfig['gpu']) && filter_var($deployConfig['gpu'], FILTER_VALIDATE_BOOLEAN)) {
-			$deviceRequests = $this->buildDefaultGPUDeviceRequests();
+		if (isset($deployConfig['computeDevice'])) {
+			if ($deployConfig['computeDevice']['id'] === 'cuda') {
+				$deviceRequests = $this->buildDefaultGPUDeviceRequests();
+			} elseif ($deployConfig['computeDevice']['id'] === 'rocm') {
+				$devices = $this->buildDevicesParams(['/dev/kfd', '/dev/dri']);
+			}
 		} else {
 			$deviceRequests = [];
+			$devices = [];
 		}
 		$storage = $this->buildDefaultExAppVolume($appId)[0]['Target'];
 
@@ -375,8 +389,9 @@ public function buildDeployParams(DaemonConfig $daemonConfig, array $appInfo): a
 			'port' => $appInfo['port'],
 			'net' => $deployConfig['net'] ?? 'host',
 			'env' => $envs,
+			'computeDevice' => $deployConfig['computeDevice'] ?? null,
+			'devices' => $devices,
 			'deviceRequests' => $deviceRequests,
-			'gpu' => count($deviceRequests) > 0,
 		];
 
 		return [
@@ -398,10 +413,14 @@ public function buildDeployEnvs(array $params, array $deployConfig): array {
 			sprintf('NEXTCLOUD_URL=%s', $deployConfig['nextcloud_url'] ?? str_replace('https', 'http', $this->urlGenerator->getAbsoluteURL(''))),
 		];
 
+		// Always set COMPUTE_DEVICE=cpu|cuda|rocm
+		$autoEnvs[] = sprintf('COMPUTE_DEVICE=%s', $deployConfig['computeDevice']['id']);
 		// Add required GPU runtime envs if daemon configured to use GPU
-		if (isset($deployConfig['gpu']) && filter_var($deployConfig['gpu'], FILTER_VALIDATE_BOOLEAN)) {
-			$autoEnvs[] = sprintf('NVIDIA_VISIBLE_DEVICES=%s', 'all');
-			$autoEnvs[] = sprintf('NVIDIA_DRIVER_CAPABILITIES=%s', 'compute,utility');
+		if (isset($deployConfig['computeDevice'])) {
+			if ($deployConfig['computeDevice']['id'] === 'cuda') {
+				$autoEnvs[] = sprintf('NVIDIA_VISIBLE_DEVICES=%s', 'all');
+				$autoEnvs[] = sprintf('NVIDIA_DRIVER_CAPABILITIES=%s', 'compute,utility');
+			}
 		}
 		return $autoEnvs;
 	}
@@ -518,8 +537,6 @@ private function isGPUAvailable(): bool {
 
 	/**
 	 * Return default GPU device requests for container.
-	 * For now only NVIDIA GPUs supported.
-	 * TODO: Add support for other GPU vendors
 	 */
 	private function buildDefaultGPUDeviceRequests(): array {
 		return [

diff --git a/lib/Migration/DaemonUpdateGPUSRepairStep.php b/lib/Migration/DaemonUpdateGPUSRepairStep.php
@@ -0,0 +1,72 @@
+<?php
+
+declare(strict_types=1);
+
+namespace OCA\AppAPI\Migration;
+
+use OCA\AppAPI\Db\DaemonConfig;
+use OCA\AppAPI\Db\DaemonConfigMapper;
+use OCP\DB\Exception;
+use OCP\Migration\IOutput;
+use OCP\Migration\IRepairStep;
+use Psr\Log\LoggerInterface;
+
+class DaemonUpdateGPUSRepairStep implements IRepairStep {
+	public function __construct(
+		private readonly DaemonConfigMapper $daemonConfigMapper,
+		private readonly LoggerInterface $logger,
+	) {
+	}
+
+	public function getName(): string {
+		return 'AppAPI Daemons configuration GPU params update';
+	}
+
+	public function run(IOutput $output): void {
+		$daemons = $this->daemonConfigMapper->findAll();
+		$daemonsUpdated = 0;
+		// Update manual-install daemons
+		/** @var DaemonConfig $daemon */
+		foreach ($daemons as $daemon) {
+			$daemonsUpdated += $this->updateDaemonConfiguration($daemon);
+		}
+		$output->info(sprintf('Daemons configuration GPU params updated: %s', $daemonsUpdated));
+	}
+
+	private function updateDaemonConfiguration(DaemonConfig $daemonConfig): int {
+		$updated = false;
+
+		$deployConfig = $daemonConfig->getDeployConfig();
+		if (isset($deployConfig['gpu'])) {
+			if (filter_var($deployConfig['gpu'], FILTER_VALIDATE_BOOLEAN)) {
+				$deployConfig['computeDevice'] = [
+					'id' => 'cuda',
+					'label' => 'CUDA (NVIDIA)',
+				];
+			} else {
+				$deployConfig['computeDevice'] = [
+					'id' => 'cpu',
+					'label' => 'CPU',
+				];
+			}
+			unset($deployConfig['gpu']);
+			$daemonConfig->setDeployConfig($deployConfig);
+			$updated = true;
+		}
+
+		if ($updated) {
+			try {
+				$this->daemonConfigMapper->update($daemonConfig);
+				return 1;
+			} catch (Exception $e) {
+				$this->logger->error(
+					sprintf('Failed to update Daemon config (%s: %s)',
+						$daemonConfig->getAcceptsDeployId(), $daemonConfig->getName()),
+					['exception' => $e]
+				);
+				return 0;
+			}
+		}
+		return 0;
+	}
+}
diff --git a/package.json b/package.json
@@ -29,7 +29,8 @@
 		"lint": "eslint --ext .js,.vue src",
 		"lint:fix": "eslint --ext .js,.vue src --fix",
 		"stylelint": "stylelint src/**/*.vue src/**/*.scss src/**/*.css",
-		"stylelint:fix": "stylelint src/**/*.vue src/**/*.scss src/**/*.css --fix"
+		"stylelint:fix": "stylelint src/**/*.vue src/**/*.scss src/**/*.css --fix",
+		"serve": "NODE_ENV=development webpack serve --allowed-hosts all --config webpack.js"
 	},
 	"browserslist": [
 		"extends @nextcloud/browserslist-config"

diff --git a/screenshots/app_api_1.png b/screenshots/app_api_1.png
diff --git a/screenshots/app_api_2.png b/screenshots/app_api_2.png
diff --git a/screenshots/app_api_3.png b/screenshots/app_api_3.png
diff --git a/screenshots/app_api_4.png b/screenshots/app_api_4.png
diff --git a/src/components/AdminSettings.vue b/src/components/AdminSettings.vue
@@ -39,6 +39,7 @@
 				:options="['no', 'always', 'unless-stopped']"
 				:placeholder="t('app_api', 'ExApp container restart policy')"
 				:aria-label="t('app_api', 'ExApp container restart policy')"
+				:aria-label-combobox="t('app_api', 'ExApp container restart policy')"
 				@input="onInput" />
 		</NcSettingsSection>
 	</div>