Originally posted by intelfx
View Post
Announcement
Collapse
No announcement yet.
Debian Developers Take To Voting Over Init System Diversity
Collapse
X
-
- 4 likes
-
Originally posted by phoronix_anon View Post
Isn't it needlessly costly to increase the size of containers by many megabytes when a simple init like init.d or supervisord provide identical functionality with smaller attack surface and footprint? Every byte should be audited in a container especially during development when pushing and pulling remote repos.
Leaving aside nitpicks like neither init.d nor supervisord being actual inits, I will attempt to explain where you are wrong, again.
Docker-style containers do not use any of the traditional init systems at all (not systemd, not upstart, not sysvinit, not openrc, not runit nor anything else from the list that's up for discussion in Debian today). Most of the time, the application process itself is PID 1. Otherwise (in a very minor share of cases), if the application in container is badly written and can't be bothered to reap its own children, a stub init process like tini may be involved — nobody is talking about these.
LXC-style containers, on the other hand, are what you may call "thin VMs", complete with multiple processes running at once and everything else you expect from a GNU/Linux installation, requiring supervision, activation, logging and sometimes interactive control, so "a simple init like init.d or supervisord" won't be providing identical functionality. Anyways, the question of "which init to use in a full-fledged [Debian] GNU/Linux installation" has long been decided in favor of systemd.
So, no, there are no use-cases when "a simple init like init.d or supervisord" will suffice. In one case they are already several orders of magnitude too heavy, and in the other case they are not enough.Last edited by intelfx; 07 December 2019, 02:08 AM.
- 9 likes
Comment
-
Originally posted by microcode View PostThere is no good reason to impose the cost of maintaining non-systemd init scripts on every Debian package maintainer, to satisfy a group of people who appear to have few or no valid practical concerns.
- 1 like
Comment
-
Originally posted by intelfx View PostSo, no, there are no use-cases when "a simple init like init.d or supervisord" will suffice. In one case they are already several orders of magnitude too heavy, and in the other case they are not enough.
- 2 likes
Comment
-
Originally posted by intelfx View PostI'm not using either distribution. But I doubt the accuracy of your statement.
Maybe diversity won't kill you too..
You should experiment.. and maybe you will find out that what somebody told you here on phoronix after all..is true.
- 2 likes
Comment
-
In my Opinion Debian should advance the Idea of a Standard API to treat Services..
LSB Already have the LSB Headers, it could be not enough, maybe needs to be worked out.. but something Standard, and not deciding in favor of the Init A or the Init B..
Propose a standard API, the InitSystems around, should follow.
Debian users will win with that..
- 2 likes
Comment
-
Originally posted by intelfx View PostIn case I was too vague, that was an invitation to share a proof of your claim.
You can prove it yourself.. install both images and see.. its free software
- 2 likes
Comment
Comment