runit: runsv(8) incorrect regarding control/[dx]

From: Lorenzo <plorenzo_at_disroot.org>
Date: Tue, 15 Feb 2022 02:26:08 +0100

Hi all,

I'm maintaining runit in Debian and I have a bug report [1] on the
customized control of runsv.
The standard behavior appears to be that first, files in
service/control/ are checked and then, only if they don't exists or
return nonzero runsv proceeds to send the appropriate signal.
However with control/[dx] the manpage gets more confused and it's not
obvious that the code actually does [2]

[A]
  * check for control/t , possibly overriding SIGTERM
  * SIGTERM (not if control/t returned zero)
  * SIGCONT (control/c is disregarded)
  * check and run control/[dx]
    (does the return code of [dx] have any effect here?)

The reporter asks to change the code so that it behave consistently
with the standard mechanism, it could look something like

[B]
  * check and run control/[dx] , possibly overriding SIGTERM
  * check for control/t , possibly overriding SIGTERM (not if
    control/[dx] returned zero)
  * SIGTERM (not if control/[dxt] returned zero)
  * SIGCONT

I can't think of a reasonable use case for [B], except that is
possible to have a different override for SIGTERM with t and dx
commands.. not even sure is a good thing to have this inconsistent
behavior.
On the other hand I'm not sure of what was the intended use for [dx]
in [A]: d or x scripts are like a "special finish file" that is run
before the actual finish file, but only when the requested status of
the service is WANT_DOWN or WANT_EXIT, with the return code that has no
effect? ..

Any example/opinion on why [A] is better than [B] or vice versa?

Best,
Lorenzo


[1] https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=983726
[2] https://github.com/vulk-archive/runit/blob/master/src/runsv.c#L246
 
Received on Tue Feb 15 2022 - 02:26:08 CET

This archive was generated by hypermail 2.4.0 : Tue Feb 15 2022 - 02:26:49 CET