Long-running work¶

sb.run() waits until the command finishes. The default timeout is 120 seconds; the ceiling is 24 hours. The SDK talks to POST /v1/machines/{id}/execs under the hood, so a proxy cannot kill the job by dropping the HTTP request.

from sistemo import Sandbox

with Sandbox(stack="python") as sb:
    r = sb.run("pip install -r requirements.txt && pytest", timeout=3600)
    print(r.exit_code, r.stdout)

import { Sandbox } from "@sistemo/sdk";

const sb = await Sandbox.create({ stack: "node" });
try {
  const r = await sb.run("npm ci && npm test", 3600);
  console.log(r.exitCode, r.stdout);
} finally {
  await sb.close();
}

timeout is enforced inside the machine. When it fires the process is killed (exit 124).

cmd & inside run() does not detach. The job is the process; backgrounding inside the script is how you lose it.

Stream or reconnect¶

If you need logs as they happen, or to survive a client crash, use the same job with a handle:

job = sb.start("npm ci && npm run build", timeout=3600, exec_id="build-1")
for chunk in job.stream():
    print(chunk, end="")
print(job.wait().exit_code)

# later
job = sb.job("build-1")

REST: POST /v1/machines/{id}/execs (you mint exec_id), then GET status / output, DELETE to cancel. See Exec.

Related¶

• Exec
• Files — run() output is 32 MiB per stream, with truncated