llama1's picture
Migrated from GitHub
b80fc11 verified
import * as fs from "node:fs/promises";
import { execAsync, sleep } from "../utils/process/execAsync";
import { execAsyncRemote } from "../utils/process/execAsync";
interface GPUInfo {
driverInstalled: boolean;
driverVersion?: string;
gpuModel?: string;
runtimeInstalled: boolean;
runtimeConfigured: boolean;
cudaSupport: boolean;
cudaVersion?: string;
memoryInfo?: string;
availableGPUs: number;
swarmEnabled: boolean;
gpuResources: number;
}
export async function checkGPUStatus(serverId?: string): Promise<GPUInfo> {
try {
const [driverInfo, runtimeInfo, swarmInfo, gpuInfo, cudaInfo] =
await Promise.all([
checkGpuDriver(serverId),
checkRuntime(serverId),
checkSwarmResources(serverId),
checkGpuInfo(serverId),
checkCudaSupport(serverId),
]);
return {
...driverInfo,
...runtimeInfo,
...swarmInfo,
...gpuInfo,
...cudaInfo,
};
} catch (_error) {
return {
driverInstalled: false,
driverVersion: undefined,
runtimeInstalled: false,
runtimeConfigured: false,
cudaSupport: false,
cudaVersion: undefined,
gpuModel: undefined,
memoryInfo: undefined,
availableGPUs: 0,
swarmEnabled: false,
gpuResources: 0,
};
}
}
const checkGpuDriver = async (serverId?: string) => {
let driverVersion: string | undefined;
let driverInstalled = false;
let availableGPUs = 0;
try {
const driverCommand =
"nvidia-smi --query-gpu=driver_version --format=csv,noheader";
const { stdout: nvidiaSmi } = serverId
? await execAsyncRemote(serverId, driverCommand)
: await execAsync(driverCommand);
driverVersion = nvidiaSmi.trim();
if (driverVersion) {
driverInstalled = true;
const countCommand =
"nvidia-smi --query-gpu=gpu_name --format=csv,noheader | wc -l";
const { stdout: gpuCount } = serverId
? await execAsyncRemote(serverId, countCommand)
: await execAsync(countCommand);
availableGPUs = Number.parseInt(gpuCount.trim(), 10);
}
} catch (error) {
console.debug("GPU driver check:", error);
}
return { driverVersion, driverInstalled, availableGPUs };
};
const checkRuntime = async (serverId?: string) => {
let runtimeInstalled = false;
let runtimeConfigured = false;
try {
// First check: Is nvidia-container-runtime installed?
const checkBinaryCommand = "command -v nvidia-container-runtime";
try {
const { stdout } = serverId
? await execAsyncRemote(serverId, checkBinaryCommand)
: await execAsync(checkBinaryCommand);
runtimeInstalled = !!stdout.trim();
} catch (error) {
console.debug("Runtime binary check:", error);
}
// Second check: Is it configured in Docker?
try {
const runtimeCommand = 'docker info --format "{{json .Runtimes}}"';
const { stdout: runtimeInfo } = serverId
? await execAsyncRemote(serverId, runtimeCommand)
: await execAsync(runtimeCommand);
const defaultCommand = 'docker info --format "{{.DefaultRuntime}}"';
const { stdout: defaultRuntime } = serverId
? await execAsyncRemote(serverId, defaultCommand)
: await execAsync(defaultCommand);
const runtimes = JSON.parse(runtimeInfo);
const hasNvidiaRuntime = "nvidia" in runtimes;
const isDefaultRuntime = defaultRuntime.trim() === "nvidia";
// Only set runtimeConfigured if both conditions are met
runtimeConfigured = hasNvidiaRuntime && isDefaultRuntime;
} catch (error) {
console.debug("Runtime configuration check:", error);
}
} catch (error) {
console.debug("Runtime check:", error);
}
return { runtimeInstalled, runtimeConfigured };
};
const checkSwarmResources = async (serverId?: string) => {
let swarmEnabled = false;
let gpuResources = 0;
try {
const nodeCommand =
"docker node inspect self --format '{{json .Description.Resources.GenericResources}}'";
const { stdout: resources } = serverId
? await execAsyncRemote(serverId, nodeCommand)
: await execAsync(nodeCommand);
if (resources && resources !== "null") {
const genericResources = JSON.parse(resources);
for (const resource of genericResources) {
if (
resource.DiscreteResourceSpec &&
(resource.DiscreteResourceSpec.Kind === "GPU" ||
resource.DiscreteResourceSpec.Kind === "gpu")
) {
gpuResources = resource.DiscreteResourceSpec.Value;
swarmEnabled = true;
break;
}
}
}
} catch (error) {
console.debug("Swarm resource check:", error);
}
return { swarmEnabled, gpuResources };
};
const checkGpuInfo = async (serverId?: string) => {
let gpuModel: string | undefined;
let memoryInfo: string | undefined;
try {
const gpuInfoCommand =
"nvidia-smi --query-gpu=gpu_name,memory.total --format=csv,noheader";
const { stdout: gpuInfo } = serverId
? await execAsyncRemote(serverId, gpuInfoCommand)
: await execAsync(gpuInfoCommand);
[gpuModel, memoryInfo] = gpuInfo.split(",").map((s) => s.trim());
} catch (error) {
console.debug("GPU info check:", error);
}
return { gpuModel, memoryInfo };
};
const checkCudaSupport = async (serverId?: string) => {
let cudaVersion: string | undefined;
let cudaSupport = false;
try {
const cudaCommand = 'nvidia-smi -q | grep "CUDA Version"';
const { stdout: cudaInfo } = serverId
? await execAsyncRemote(serverId, cudaCommand)
: await execAsync(cudaCommand);
const cudaMatch = cudaInfo.match(/CUDA Version\s*:\s*([\d\.]+)/);
cudaVersion = cudaMatch ? cudaMatch[1] : undefined;
cudaSupport = !!cudaVersion;
} catch (error) {
console.debug("CUDA support check:", error);
}
return { cudaVersion, cudaSupport };
};
export async function setupGPUSupport(serverId?: string): Promise<void> {
try {
// 1. Initial status check and validation
const initialStatus = await checkGPUStatus(serverId);
const shouldContinue = await validatePrerequisites(initialStatus);
if (!shouldContinue) return;
// 2. Get node ID
const nodeId = await getNodeId(serverId);
// 3. Create daemon configuration
const daemonConfig = createDaemonConfig(initialStatus.availableGPUs);
// 4. Setup server based on environment
if (serverId) {
await setupRemoteServer(serverId, daemonConfig);
} else {
await setupLocalServer(daemonConfig);
}
// 5. Wait for Docker restart
await sleep(10000);
// 6. Add GPU label
await addGpuLabel(nodeId, serverId);
// 7. Final verification
await sleep(5000);
await verifySetup(nodeId, serverId);
} catch (error) {
if (
error instanceof Error &&
error.message.includes("password is required")
) {
throw new Error(
"Sudo access required. Please run with appropriate permissions.",
);
}
throw error;
}
}
const validatePrerequisites = async (initialStatus: GPUInfo) => {
if (!initialStatus.driverInstalled) {
throw new Error(
"NVIDIA drivers not installed. Please install appropriate NVIDIA drivers first.",
);
}
if (!initialStatus.runtimeInstalled) {
throw new Error(
"NVIDIA Container Runtime not installed. Please install nvidia-container-runtime first.",
);
}
if (initialStatus.swarmEnabled && initialStatus.runtimeConfigured) {
return false;
}
return true;
};
const getNodeId = async (serverId?: string) => {
const nodeIdCommand = 'docker info --format "{{.Swarm.NodeID}}"';
const { stdout: nodeId } = serverId
? await execAsyncRemote(serverId, nodeIdCommand)
: await execAsync(nodeIdCommand);
const trimmedNodeId = nodeId.trim();
if (!trimmedNodeId) {
throw new Error("Setup Server before enabling GPU support");
}
return trimmedNodeId;
};
const createDaemonConfig = (availableGPUs: number) => ({
runtimes: {
nvidia: {
path: "nvidia-container-runtime",
runtimeArgs: [],
},
},
"default-runtime": "nvidia",
"node-generic-resources": [`GPU=${availableGPUs}`],
});
const setupRemoteServer = async (serverId: string, daemonConfig: any) => {
const setupCommands = [
"sudo -n true",
`echo '${JSON.stringify(daemonConfig, null, 2)}' | sudo tee /etc/docker/daemon.json`,
"sudo mkdir -p /etc/nvidia-container-runtime",
'sudo sed -i "/swarm-resource/d" /etc/nvidia-container-runtime/config.toml',
'echo "swarm-resource = \\"DOCKER_RESOURCE_GPU\\"" | sudo tee -a /etc/nvidia-container-runtime/config.toml',
"sudo systemctl daemon-reload",
"sudo systemctl restart docker",
].join(" && ");
await execAsyncRemote(serverId, setupCommands);
};
const setupLocalServer = async (daemonConfig: any) => {
const configFile = `/tmp/docker-daemon-${Date.now()}.json`;
await fs.writeFile(configFile, JSON.stringify(daemonConfig, null, 2));
const setupCommands = [
`sudo sh -c '
cp ${configFile} /etc/docker/daemon.json &&
mkdir -p /etc/nvidia-container-runtime &&
sed -i "/swarm-resource/d" /etc/nvidia-container-runtime/config.toml &&
echo "swarm-resource = \\"DOCKER_RESOURCE_GPU\\"" >> /etc/nvidia-container-runtime/config.toml &&
systemctl daemon-reload &&
systemctl restart docker
'`,
`rm ${configFile}`,
].join(" && ");
try {
await execAsync(setupCommands);
} catch (_error) {
throw new Error(
"Failed to configure GPU support. Please ensure you have sudo privileges and try again.",
);
}
};
const addGpuLabel = async (nodeId: string, serverId?: string) => {
const labelCommand = `docker node update --label-add gpu=true ${nodeId}`;
if (serverId) {
await execAsyncRemote(serverId, labelCommand);
} else {
await execAsync(labelCommand);
}
};
const verifySetup = async (nodeId: string, serverId?: string) => {
const finalStatus = await checkGPUStatus(serverId);
if (!finalStatus.swarmEnabled) {
const diagnosticCommands = [
`docker node inspect ${nodeId}`,
'nvidia-smi -a | grep "GPU UUID"',
"cat /etc/docker/daemon.json",
"cat /etc/nvidia-container-runtime/config.toml",
].join(" && ");
await (serverId
? execAsyncRemote(serverId, diagnosticCommands)
: execAsync(diagnosticCommands));
throw new Error("GPU support not detected in swarm after setup");
}
return finalStatus;
};