https://livekit.io logo
Join Slack
Powered by
how can I check if the job is shutting down? I'd l...
# ask-ai
p
how can I check if the job is shutting down? I'd like to freeze the agent in case the shutdown hook has been called
t
Thanks for reaching out, Mario. I'm an AI assistant and I'm here to help. Give me a moment while I research this topic.
Thanks — I’m responding to your question about how to detect when a job is shutting down so you can freeze the agent if the shutdown hook has been called. There isn’t a built-in “is_shutting_down” check exposed by the job framework. The recommended approach is to register a shutdown callback and set your own shutdown flag there, then have your agent check that flag and stop processing when it’s set. Below are two concise approaches you can use. 1) Shutdown flag in the entrypoint (functional style)
Copy code
@server.rtc_session(agent_name="my-agent")
async def entrypoint(ctx: JobContext):
    # create a mutable flag
    shutdown_flag = {"is_shutting_down": False}

    async def shutdown_hook():
        shutdown_flag["is_shutting_down"] = True
        <http://logger.info|logger.info>("Shutdown initiated - freezing agent")
        # short cleanup here

    ctx.add_shutdown_callback(shutdown_hook)

    # pass flag into whatever logic needs to observe it
    while not shutdown_flag["is_shutting_down"]:
        # normal agent work
        await do_work_step()

    # optional finalization after flag set
2) Track state in your Agent class (object-oriented)
Copy code
class MyAgent(Agent):
    def __init__(self):
        super().__init__()
        self.is_shutting_down = False

    def freeze_agent(self):
        self.is_shutting_down = True
        # cancel timers, stop accepting new tasks, etc.

    async def run(self):
        while not self.is_shutting_down:
            await self.process_next()

@server.rtc_session(agent_name="my-agent")
async def entrypoint(ctx: JobContext):
    agent = MyAgent()

    async def shutdown_hook():
        agent.freeze_agent()
        <http://logger.info|logger.info>("Shutdown initiated - freezing agent")
        # short cleanup

    ctx.add_shutdown_callback(shutdown_hook)
    await agent.run()
Important operational notes • Shutdown hooks should complete quickly: the framework waits 10 seconds by default before forcefully terminating the process. You can adjust this using 
shutdown_process_timeout
in server options. • When the process receives SIGTERM / SIGINT, the server uses 
drain_timeout
(defaults to 30 minutes) to allow active jobs to finish. If you’d like, I can: • Show an example that cancels specific async tasks or listeners when freezing the agent. • Provide an example that integrates the flag into your existing session/AgentSession flow. Which would help you most? Sources: Job lifecycle | LiveKit Documentation | Server options | LiveKit Documentation
p
what about the job state?
t
Thanks — you’re asking about the “job state” (beyond the shutdown hook) and how to tell whether a job is running, shutting down, or closed. Short answer: there is no exposed API that returns a named job state (like get_job_state()). The reliable ways to detect lifecycle transitions are the lifecycle hooks and session events you can register for. What you can use right now • Shutdown hooks: Register a callback with 
ctx.add_shutdown_callback(...)
. That callback runs when the job is being terminated, so setting a flag there lets your agent know the job is shutting down and freeze behavior immediately. • Session events: Use session events such as 
session.on("user_state_changed")
to observe participant activity; when all participants leave or the session is ended, the job will move toward termination. • Local state tracking: Since there’s no built-in state query, maintain your own job-state variable (e.g., 
is_shutting_down
or 
job_state = "running"|"shutting_down"|"closed"
) and update it in your shutdown hook and other lifecycle points. Have your agent logic check that variable before starting work or between work items. Important operational notes • Shutdown hooks should finish quickly: the framework waits 10 seconds by default before forcefully terminating the process. You can change that with 
shutdown_process_timeout
. • When a SIGTERM/SIGINT arrives, the server will wait up to 
drain_timeout
(default 30 minutes) for jobs to finish. If you’d like, I can: • Show a short example that exposes a 
job_state
variable and updates it in shutdown hooks and session events. • Show how to cancel running async tasks or listeners when the shutdown flag is set. Which example would you prefer? Sources: Job lifecycle | LiveKit Documentation | How to detect and handle silence during calls | LiveKit
3 Views
Open in Slack
PreviousNext
Powered by