Skip to content

fix(inference): reap managed vLLM processes #7221

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
jyaunches merged 2 commits into from
Jul 20, 2026
Merged

fix(inference): reap managed vLLM processes #7221

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
4ed9f2b
fix(inference): reap managed vllm processes
senthilr-nv Jul 20, 2026
c7c41ae
Merge branch 'main' into codex/vllm-init-restart
senthilr-nv Jul 20, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
18 changes: 16 additions & 2 deletions src/lib/inference/vllm.test.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -285,6 +285,7 @@ describe("vLLM profile detection", () => {
const args = buildVllmRunArgs(runtime, ultra!, flags, {} as NodeJS.ProcessEnv);
expect(args).toEqual([
"--pull=never",
"--init",
"--restart",
"unless-stopped",
"--gpus",
Expand Down Expand Up @@ -446,11 +447,18 @@ describe("vLLM image pull", () => {
});

describe("vLLM run command", () => {
it("adds Docker init so restarts can reap the server process (#7219)", () => {
const profile = detectVllmProfile({ platform: "spark", type: "nvidia" });
expect(profile).not.toBeNull();
const args = buildVllmRunArgs(profile!, profile!.defaultModel, profile!.dockerRunFlags);
expect(args.slice(0, 4)).toEqual(["--pull=never", "--init", "--restart", "unless-stopped"]);
});

it("adds --restart unless-stopped so the container survives a host reboot (#4886)", () => {
const profile = detectVllmProfile({ platform: "spark", type: "nvidia" });
expect(profile).not.toBeNull();
const args = buildVllmRunArgs(profile!, profile!.defaultModel, profile!.dockerRunFlags);
expect(args.slice(0, 3)).toEqual(["--pull=never", "--restart", "unless-stopped"]);
expect(args).toEqual(expect.arrayContaining(["--restart", "unless-stopped"]));
expect(args).toContain("--name");
expect(args[args.indexOf("--name") + 1]).toBe(profile!.containerName);
expect(args).toEqual(
Expand Down Expand Up @@ -997,7 +1005,13 @@ describe("installVllm model resolution", () => {
{ env?: Record<string, string> },
];
expect(args).toEqual(
expect.arrayContaining(["--pull=never", "--restart", "unless-stopped", profile.image]),
expect.arrayContaining([
"--pull=never",
"--init",
"--restart",
"unless-stopped",
profile.image,
]),
);
expect(args).not.toContain("HF_TOKEN");
expect(args.join(" ")).not.toContain("hf_test");
Expand Down
10 changes: 6 additions & 4 deletions src/lib/inference/vllm.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -497,10 +497,11 @@ function validateDockerArgs(args: readonly string[], label: string): string[] {
}

// Build the `docker run` argv for the long-lived vLLM inference container.
// Exported for testing. `--restart unless-stopped` makes the container come
// back after a host reboot or Docker daemon restart (#4886); without a restart
// policy the container stays down after a reboot and `nemoclaw inference get`
// fails until a full `nemoclaw onboard --fresh --gpu` recreates it.
// Exported for testing. `--init` forwards signals and reaps child processes so
// Docker can stop and restart the long-lived server cleanly. `--restart
// unless-stopped` brings it back after a host reboot or Docker daemon restart
// (#4886); without a restart policy the container stays down after a reboot and
// `nemoclaw inference get` fails until onboarding recreates it.
export function buildVllmRunArgs(
profile: VllmProfile,
model: VllmModelDef,
Expand All @@ -513,6 +514,7 @@ export function buildVllmRunArgs(
const safeRunFlags = validateDockerArgs(runFlags, "vLLM docker run flags");
return [
"--pull=never",
"--init",
"--restart",
"unless-stopped",
...safeRunFlags,
Expand Down
Loading