Feature Request: reload after error

[valgarf]

Even after the subprocess crashes, uvicorn should still reload, i.e. restart the process after the next file change. Maybe with an additional command line flag (--reload-after-error).

Not having to restart the server after every syntax error would help during development.

[tomchristie]

Slightly depends on what you mean.
We recommend using Gunicorn (or another process monitor) in production, as that'll handle hard process restarts.

Incidentally, how is the subprocess crashing? Python exceptions from the child application should just end up logging as errors. Is it something more serious than that?

[valgarf]

It reliably crashes from syntax errors in my code. This would be a pure development feature, as the whole 'reload after file change' logic should have nothing to do with production code in any way.

When the subprocess crashes, the reloading after file changes is stopped, you have an explicit check process.is_alive() in your loop.

In case anyone else finds this discussion, I monkeypatched it for my use case:

from uvicorn.supervisors.statreload import HANDLED_SIGNALS, StatReload

def statreload_run(self, target, *args, **kwargs):
    pid = os.getpid()
    logger = self.config.logger_instance

    logger.info("Started reloader process [{}]".format(pid))

    for sig in HANDLED_SIGNALS:
        signal.signal(sig, self.handle_exit)

    def get_subprocess():
        spawn = multiprocessing.get_context("spawn")
        try:
            fileno = sys.stdin.fileno()
        except OSError:
            fileno = None

        return spawn.Process(
            target=self.handle_fds, args=(target, fileno), kwargs=kwargs
        )

    process = get_subprocess()
    process.start()

    # Changes are in this loop. No check if the process is alive in the loop
    while not self.should_exit:
        time.sleep(0.3)
        if self.should_restart():
            self.clear()
            if process.is_alive():
                os.kill(process.pid, signal.SIGTERM)
                process.join()

            process = get_subprocess()
            process.start()
            self.reload_count += 1

    logger.info("Stopping reloader process [{}]".format(pid))

StatReload.run = statreload_run

[tomchristie]

Ah right, I was misunderstanding slightly - I though you were talking about runtime errors in a deployed version, rather than syntax errors during development (which yes, will just quit the server).

[maxfurman]

Can this issue be reopened? I am using uvicorn --reload for development and it is a hassle to restart the server every time I accidentally save a syntax error.

[tomchristie]

Sure thing, let's put it on the list.

[gnat]

+1 for this.

[anand870]

Any update on this?

[tomchristie]

Folks, if there's an update on a ticket, there'll be an update on the ticket. Please don't do issue-spam.

[rafalp]

It appears that this issue no longer exists as of Uvicorn 0.11.

[kungfoome]

This is still an issue for me Running uvicorn 0.11.8 with CPython 3.8.5 on Windows

[nareto]

Still an issue for me in 0.16.0. This makes it unusable for me because I like to have autosave enabled in my code editor, so as I write code it restarts and crashes.

[Kludex]

Still an issue for me in 0.16.0. This makes it unusable for me because I like to have autosave enabled in my code editor, so as I write code it restarts and crashes.

It's not the same issue, but a fix is already provided on #1313