new from systemd: portable services

[RS Wood]

From the «but can I theme it?» department:
Title: Systemd introduces "Portable Services"
Author: mrpg
Date: Tue, 29 May 2018 21:10:00 -0400
Link: https://soylentnews.org/article.pl?sid=18/05/30/0147216&from=rss

mechanicjay[1] writes:

Systemd Introduces "Portable Services" Functionality, Similar To Containers[2]

Lennart is at it again, making complicated things that nobody asked for.

The past several months Lennart Poettering has been working on a "portable
services" concept and that big ticket new feature has now landed in Systemd.
Portable services are akin to containers but different.

[...] A portable service is ultimately just an OS tree, either inside of a
directory tree, or inside a raw disk image containing a Linux file system.
This tree is called the "image". It can be "attached" or "detached" from the
system. When "attached" specific systemd units from the image are made
available on the host system, then behaving pretty much exactly like locally
installed system services. When "detached" these units are removed again from
the host, leaving no artifacts around (except maybe messages they might have
logged).

[...] The primary focus use-case of "portable services" is to extend the host
system with encapsulated extensions, but provide almost full integration with
the rest of the system, though possibly restricted by effective security
knobs. This focus includes system extensions otherwise sometimes called
"super-privileged containers".

------------------------------------------------------------------------------

Original Submission[3]

Read more of this story[4] at SoylentNews.

Links:
[1]: http://{mechanicjay} {at} {soylentnews.org}/ (link)
[2]: https://www.phoronix.com/scan.php?page=news_item&px=Systemd-Portable-Services (link)
[3]: http://soylentnews.org/submit.pl?op=viewsub&subid=26962 (link)
[4]: https://soylentnews.org/article.pl?sid=18/05/30/0147216&from=rss (link)

[Andy Burns]

Lennart Poettering has been working on [...]

Oh god what now?

almost full integration with the rest of the system

"almost" that's great ....

[RS Wood]

Post by Andy Burns

Lennart Poettering has been working on [...]

Oh god what now?

almost full integration with the rest of the system

"almost" that's great ....

Yeah, there may be a use case for what he's proposing, but I am truly not
seeing it at the moment. Having just tried to fix a syslog issue on a
systemd box I'm not feeling very charitable. Everything just worked, why
fuck with it?

[Marko Rauhamaa]

Post by RS Wood
Yeah, there may be a use case for what he's proposing, but I am truly
not seeing it at the moment. Having just tried to fix a syslog issue
on a systemd box I'm not feeling very charitable. Everything just
worked, why fuck with it?

I understand your frustration. Systemd has been a bloodless coup d'état.

However, things did *not* just work before systemd.

With systemd, at least the trains run on time.


Marko

[RS Wood]

On Fri, 01 Jun 2018 08:22:01 +0300

Post by Marko Rauhamaa

Post by RS Wood
Yeah, there may be a use case for what he's proposing, but I am truly
not seeing it at the moment. Having just tried to fix a syslog issue
on a systemd box I'm not feeling very charitable. Everything just
worked, why fuck with it?

I understand your frustration. Systemd has been a bloodless coup d'état.
However, things did *not* just work before systemd.

Yes, I've heard that. For my simple cases though, it worked just fine.

My machines were a bit slow to boot up, but that never bothered me.

[Marko Rauhamaa]

Post by RS Wood
On Fri, 01 Jun 2018 08:22:01 +0300

Post by Marko Rauhamaa
However, things did *not* just work before systemd.

Yes, I've heard that. For my simple cases though, it worked just fine.
My machines were a bit slow to boot up, but that never bothered me.

Things that didn't work:

* I started a service but the service wasn't really up immediately
after the start command returned. With careful daemon startup logic,
you could have had the parent exit only after the service was up, but
that principle wasn't followed rigorously.

* SIGHUP was often used to make the daemon reload its configuration.
Signals are very tricky to handle correctly (and often weren't). Most
importantly, though, you have no reliable way of knowing when the new
settings were taken into use.

* A PID file was often used as an indication of an already running
service. However, the file might have been left behind by a previous
unclean shutdown. Killing PID might have victimized a random innocent
bystander process instead.

These issues have better support in systemd. I'm not saying systemd
is doing a good job, but it's addressing real issues.


Marko

[Jerry Peters]

Post by Marko Rauhamaa

Post by RS Wood
On Fri, 01 Jun 2018 08:22:01 +0300

Post by Marko Rauhamaa
However, things did *not* just work before systemd.

Yes, I've heard that. For my simple cases though, it worked just fine.
My machines were a bit slow to boot up, but that never bothered me.

* I started a service but the service wasn't really up immediately
after the start command returned. With careful daemon startup logic,
you could have had the parent exit only after the service was up, but
that principle wasn't followed rigorously.
* SIGHUP was often used to make the daemon reload its configuration.
Signals are very tricky to handle correctly (and often weren't). Most
importantly, though, you have no reliable way of knowing when the new
settings were taken into use.
* A PID file was often used as an indication of an already running
service. However, the file might have been left behind by a previous
unclean shutdown. Killing PID might have victimized a random innocent
bystander process instead.
These issues have better support in systemd. I'm not saying systemd
is doing a good job, but it's addressing real issues.
Marko

If that's all it did people wouldn't be complaining, it's the
metastatizing into areas other than system startup that's irritating.
Things like its own DNS server. Systemd just keeps expanding its
scope.

[Dan Purgert]

Post by Jerry Peters
[...]
If that's all it did people wouldn't be complaining, it's the
metastatizing into areas other than system startup that's irritating.
Things like its own DNS server. Systemd just keeps expanding its
scope.

Not to mention that it doesn't do any of those things particularly well
either. If I knew enough to write an init-system, I think 'systeme'
would be a fitting name.

Tagline - "Because Systems A through D have all failed." :)