sd_notify(3) and sd_watchdog_enabled(3) client functionality implemented in Python 3 for writing Python daemons
$ pip install systemd-watchdog
or
$ git clone ...
$ make install
import systemd_watchdog
import time
wd = systemd_watchdog.watchdog()
if not wd.is_enabled:
# Then it's probably not running is systemd with watchdog enabled
raise Exception("Watchdog not enabled")
# Report a status message
wd.status("Starting my service...")
time.sleep(3)
# Report that the program init is complete
wd.ready()
wd.status("Init is complete...")
wd.notify()
time.sleep(3)
# Compute time between notifications
timeout_half_sec = int(float(wd.timeout) / 2e6) # Convert us->s and half that
wd.status("Sleeping and then notify x 3")
time.sleep(timeout_half_sec)
wd.notify()
time.sleep(timeout_half_sec)
wd.notify()
time.sleep(timeout_half_sec)
wd.notify()
wd.status("Spamming loop - should only see 2-3 notifications")
t = float(0)
while t <= 4*timeout_half_sec:
time.sleep(0.05)
wd.ping()
t += 0.05
# Report an error to the service manager
wd.notify_error("An irrecoverable error occured!")
# The service manager will probably kill the program here
time.sleep(3)In one terminal:
socat unix-recv:/tmp/tempo.sock -
In another terminal:
NOTIFY_SOCKET=/tmp/tempo.sock WATCHDOG_USEC=5000000 python example.py
Expected output (in first terminal):
STATUS=Starting my service...
READY=1
STATUS=Init is complete...
WATCHDOG=1
STATUS=Sleeping and then notify x 3
WATCHDOG=1
WATCHDOG=1
WATCHDOG=1
STATUS=Spamming loop - should only see 2-3 notifications
WATCHDOG=1
WATCHDOG=1
WATCHDOG=1
STATUS=An irrecoverable error occured!
WATCHDOG=trigger
The only method required for the simplest implementation; combines notify_due with notify() to only send "alive" notifications at reasonable intervals.
Returns boolean indicating if a message was sent or not.
Alias for ping() if you prefer heartbeat terminology.
Report ready service state, i.e. completed init (only needed with Type=notify).
Send a service status message.
Boolean property stating whether watchdog capability is enabled.
Property reporting the number of microseconds (int) before process will be killed.
It is recommended that you call notify() once roughly half of this interval has passed (see notify_due).
Property that is the same as timeout but presented as datetime.timedelta for easier manipulation.
Boolean property indicating more than half of the watchdog interval has passed since last update.
Report a healthy service state. Other calls, e.g. status() do not reset the watchdog.
Report an error to the watchdog manager. This program will likely be killed upon receipt.
If msg is provided, it will be reported as a status message prior to the error.
Aaron D. Marasco May 2020
- Forked from the sd-notify project https://github.com/stigok/sd-notify
- Additional contributors can be found in GitHub repository history
See LICENSE file