Function lifecycleHot

services/cloud-agent-next/test/e2e/lifecycle.ts:175–247

(args: LifecycleArgs)

) {
  const { executeDocker, allowedDirectories = [] } = options;
  const familyGoneTimeoutMs = options.familyGoneTimeoutMs ?? 30_000;
  const current = (await listSandboxContainers(executeDocker)).find(
    container => container.name === sandbox.name
  );
  if (current && current.id !== sandbox.id)
    throw new Error(`Container identity changed for ${sandbox.name}`);
  let killed: string[];
  if (current && sessionId.startsWith('workspace_') && kiloSessionId !== undefined) {
    // Root presence is not exclusive ownership; the control-plane stop proves
    // the `exclusive` operation before it kills. The proof can fail merely
    // because the runtime was retired or replaced while the container was
    // winding down, so re-check the family before treating that as failure.
    try {
      killed = await stopOwnedControlPlaneSandbox(
        sandbox,
        kiloSessionId,
        executeDocker,
        allowedDirectories
      );
    } catch (error) {
      if (!(await waitForSandboxFamilyGone(sandbox, familyGoneTimeoutMs, executeDocker)))
        throw error;
      return [];
    }
  } else {
    if (current) {
      const owned = await sandboxOwnsSession(current.id, sessionId, kiloSessionId);
      if (!owned) throw new Error(`Cannot prove exclusive ownership of ${sandbox.name}`);
    }
    killed = await killSandboxFamily(sandbox, executeDocker);
  }
  if (!(await waitForSandboxFamilyGone(sandbox, familyGoneTimeoutMs, executeDocker))) {
    throw new Error(`Owned sandbox family ${sandbox.name} is still running after cleanup`);
  }
  return killed;
}

/**
 * A refused ownership proof is only re-checked this long for a sandbox that is
 * winding down on its own; the fault scenarios' 30 s default would stall every
 * scenario whose sandbox is not exclusive.
 */
const RECLAIM_FAMILY_GONE_TIMEOUT_MS = 5_000;
const RECLAIM_DOCKER_ATTEMPTS = 3;
const RECLAIM_DOCKER_RETRY_MS = 1_000;

export type ReclaimSession = { sessionId: string; kiloSessionId?: string };

export type SandboxReclaimReport = {
  /** Sandbox containers this call stopped. */
  stopped: string[];
  /** Why a session's sandbox was left alone. */
  failures: string[];
};

type LocatedSandbox = { container: SandboxContainer; directory: string | undefined };

/** The one primary sandbox a session provably owns, or `null` when none is running. */