I seem to have a blind spot when searching the systemd docs for the
(very common?) pattern of managing the life cycle of resources which
have init, renew, quit phases. Here is an example of using upnp to
configure your home modem/router (IGD) with a NAT entry:
Description="Configure the IGD's NAT for VOIP"
Type=oneshot
ExecStart="upnpc -a ..."
ExecStop="upnpc -d ..."
RemainActiveAfterExit=yes
Now it is easy to use systemd to start and stop the service. However,
upnp also requires renewing the NAT entry periodically. The semantically
correct way would be to create a "partOf" timer which will start and
stop along with the service:
Description="BAD TIMER"
PartOf=configure-the-igd.service
OnUnitActiveSec = "1hr";
The ExecStart script is fine for renewal, but unfortunately
RemainActiveAfterExit prevents timers from recurring.
The answer seems to be to create a second service for renewal. This
second unit has no RemainActiveAfterExit so it will repeat:
Description="Renew the IGD's NAT for VOIP"
Type=oneshot
ExecStart="upnpc -a ..."
Now the timer is part of the first service, but triggers the second:
Description="BETTER TIMER"
PartOf=configure-the-igd.service
OnUnitActiveSec = "1hr";
Unit=renew-the-igd.service
My question: Is it really necessary to use two service units for this
pattern, or am I missing some obvious common pattern that everyone else
knows about?
Warning: That was untested code above, and my unit file syntax might be
slightly wrong, since I generally use nix expressions to create unit files.
--
Anthony Carrico
