[ANNOUNCE] supervision (framework) 0.12.0_beta

From: tokiclover <tokiclover_at_gmail.com>
Date: Wed, 20 Jul 2016 16:22:08 +0200

Hello fellow supervision users,

Just annoucing a *new* milestone release which depart from the old...
more on this later.

What's changed besides the version number scheme?

A new init-system was added since 0.12.0_alpha to be able to completely
bring up a system from.. scratch. So, a few things has to change in the
service scheduling side to be able to do so with ease while keeping the
parallel start/stop inherent to supervision suites. So, a *very* simple
algorithm was added to push forward and safely the old *uncomplete*
scheme which was relying on a single pass of the new algorithm, meaning,
giving priority to needed or used service to simplify. This is now done
by a simple recursive algorithm.

But the hard stuff was to put together services and init-system bits to
be able to boot. And this not that simple, unless an over simplification
is done for this to happen. -- Say, void Linux for example, uses a
simple shell script to bring up dmcrypt LUKS and LVM2 before actually
mounting the filesystems. Simple, certainly. I even saw a simple `mount
/' in other over simplified sollutions.

Now, here the init-system is being taken by services like in OpenRC; so
that, having dmcrypt LUKS, device-mapper (LVM2), (software and fake ATA)
RAID, ZFS, or a single of those would pretty much work the same.

Enough with the details.

A test bed here with a Intel(R) quad core (8 threads) and good spin
drive boots like:

  * first, I have device-mapper/LVM2 on top of dmcrypt LUKS;
  * second, ZFS on top of dmcrypt LUKS;
  * third, no long filesystem checks for those tests;
  * ZRAM devices to setup for /(var/)tmp (there is a service for this);
  * ...sorry forgot RAID...;
  * system init (stage-0) completes in less than a seconde;
  * system boot (stage-1) (set devices etc.) completes in ~8 secondes;
  * (stage-2) completes in a second (thanks to a supervision suite,
    runit for this test);

That's 10 second for a common, well minor the ZFS, dmcrypt, device-mapper,
system to bring up the system.

Now shutdown... 19 secondes! Yes it's the double of the boot up. Now 10
secondes are used to try to get device-mapper and dmcrypt down... There
is an optional logfile which pretty much details on the culpirit here.
But hey, this is beta! There is certainly room for shutdown

Let's stop here, the rest is over there[0].

[0]: https://github.com/tokiclover/supervision
Received on Wed Jul 20 2016 - 14:22:08 UTC

This archive was generated by hypermail 2.3.0 : Sun May 09 2021 - 19:44:19 UTC