I have an HTTP service (let's call it Foo) with systemd socket activation. I would like to support this setup:
- monitoring service tells Foo to stop (e.g. POST /stop)
- Foo finishes it's (potentially long) work then exits.
- monitoring service polls an HTTP endpoint on Foo (e.g. GET /are-you-up) waiting for it to exit.
At step 3 when the monitoring service hits /are-you-up, systemd's socket activation restarts the HTTP service!
I'd still like systemd to manage the socket (so that my service doesn't have to run/start as root). I'd also still like systemd to restart my service if it crashes (non-0 exit code).
Are either of these possible?
- systemd manages the socket but doesn't do socket activation?
- OR Foo service on shutdown tells systemd to close the socket / stop responding to it. I tried having the service run 'systemctl stop foo.socket' but it runs as an unprivileged user.
The big picture is that the service is being replaced. After it shuts down the (cloud) server will be deleted.
Best Answer
For anyone else stuck with this, what I ended up doing is shutting down the HTTP listener but keeping the process running. In pseudo code the Foo service does:
close(httpSocket) print("Stopped") sleep(forever)This means the monitoring service's
GET /are-you-upwill know the service is done, systemd won't restart it, and I still get all the other systemd goodness. In practice it has been working well for several months now.If there is a cleaner option please add your answer. Thanks!