Helmut Grohne writes ("upstart should support systemd-style socket activation"):
> I talked to James Hunt about this during DebConf 13 and also asked at
> least Steve Langasek about the rationale behind the socket activation
> interface that upstart uses. At that point, we were unable to understand
> the technical reasons and agreed that I should attempt to contact Scott
> James Remnant to ask him for his thoughts on this. Sadly he never
> replied even though I included at least two of his email addresses. We
> further agreed that the condition of timeout would be met with upstart
> actually implementing systemd's interface and James Hunt doing the work.
> The missing piece currently seems to be a bug report for getting the
> process going.Just to be clear, here is a sketch of the changes which I think should be made in upstart: 1. A new option in job configuration files which controls at the very least item (3) below and perhaps item (2). 2. Use consecutive fds starting at 3 for UPSTART_FDS. There would be no harm in doing this always, but it only needs to happen if we are trying to emulate systemd's socket activation protocol. 3. Also pass two new environment variables: LISTEN_FDS shoudld be set to the number of fds in UPSTART_FDS, and LISTEN_PID should be set to the pid of the process which is going to become the daemon. This must be conditional on whether the process is expecting it, because otherwise other processes might inherit the new environment variables. Thanks, Ian. -- To UNSUBSCRIBE, email to [email protected] with a subject of "unsubscribe". Trouble? Contact [email protected]
