Skip to content

Latest commit

 

History

History
76 lines (55 loc) · 2.6 KB

daemon.md

File metadata and controls

76 lines (55 loc) · 2.6 KB
Raw

pystemd.daemon

This just provides a python interface to some methods in sd-daemon.h, most important, it provides access to sd_notify in a pythonic way (what ever that means to you). This is built more in the spirit of: if you already have pystemd installed, why not just use it to sd_notify and other stuff.

Since we are there we also provide 2 other things: access to the listen_fd passed to your process during socket activation and the starting FD for those sockets. Just one warning, the LISTEN_FDS_START is not really exposed in sd-daemon.h but that has not changed and is probably not going to.

Usage

How to notify? Imagine that this is your service:

import sys
import time
import pystemd.daemon

time.sleep(10) # preparation work
pystemd.daemon.notify(False, ready=1, status="starting to do some work")
for i in range(10):
  time.sleep(1) # some hard work
  pystemd.daemon.notify(False, status=f"Complete: {(i+1)*100}% of work...")

pystemd.daemon.notify(False, 'STOPPING=1')
# this line can also be `pystemd.daemon.notify(False, stopping=1)`

sys.exit(0)

and your service unit is

[Unit]
Description="my notify example"

[Service]
Type=notify
ExecStart=/usr/bin/python /path/to/my/file.py

Then from the moment it starts to the first pystemd.daemon.notify your script is in activating state. When you send pystemd.daemon.notify(False, ready=1) you signal systemd that your script is ready and that the service has been loaded. This also signals systemd that other units that are in the same transaction and depend on your unit can be started.

status calls are meant to provide human readable output on the state of your service. They are not meant for logging or to store complex structures, but just to provide feedback for a human querying the state of your service.

The following calls are equivalent: pystemd.daemon.notify(False, ready=1) and pystemd.daemon.notify(False, "READY=1")

Options of notify can be found in systemd project at: https://github.com/systemd/systemd/blob/master/src/systemd/sd-daemon.h#L173-L232

The first parameter is meant to signal if we should remove the notify information from the service. For example, you can do pystemd.daemon.notify(True, ready=1) if you don't intend to speak to the systemd notify socket ever again.

Watchdog

If you set WatchdogSec=X in your unit, you can use watchdog functionality. To get the number of usec between pooling, that is X*10^6 where X is defined in WatchdogSec=X. you can get it with pystemd.daemon.watchdog_enabled, and if you want to ping systemd, you do pystemd.daemon.notify(False, watchdog=1).