1 |
On Thu, Feb 2, 2017 at 3:35 PM, james <garftd@×××××××.net> wrote: |
2 |
|
3 |
> |
4 |
> I think that unikernels are something everyone should be aware of |
5 |
> as they purport to be the latest trend in securing all sorts of systems. |
6 |
> (a brief read). |
7 |
> |
8 |
|
9 |
Not really for all sorts, more for servers. Otherwise I get it, and |
10 |
at this point now that I run almost everything in containers I tend to |
11 |
be more inclined to run different distros in those containers. |
12 |
|
13 |
> This is only the case because profiles are in general in a mess and there |
14 |
> are little in the way of conventions. What is so sacrosanct about upstream |
15 |
> for a truly embedded gentoo system or a gentoo based IoT device? |
16 |
|
17 |
Nothing, in that space. |
18 |
|
19 |
The problem is the new user experience. When somebody is new to |
20 |
Gentoo and not super-knowledgeable the first thing they're going to do |
21 |
is set up a desktop. Now, they might not call it a desktop. They |
22 |
might not even run X11 on it. But, they're basically falling into |
23 |
that desktop user experience where whatever they do install "just |
24 |
works" and is feature-complete. |
25 |
|
26 |
It is true that we also attract advanced users who are looking for |
27 |
something different. They have no issues getting any distro to dance |
28 |
for them, and they're picking Gentoo because it is best suited for |
29 |
their specific need. These users are much more likely to be |
30 |
interested in minimal configurations, embedded systems, the hardened |
31 |
profiles, and so on. |
32 |
|
33 |
However, the problem is that if we optimize mainly for the second |
34 |
group we basically lose the first group entirely, and I suspect that |
35 |
is overall going to be the bigger group. |
36 |
|
37 |
If what you want is a "unikernel profile" for Gentoo then you're going |
38 |
to be changing a LOT of assumptions. Forget openrc vs systemd, there |
39 |
is no reason to have any init implementation on the thing. Forget |
40 |
linux vs bsd, there is also no reason to have a kernel in a container. |
41 |
We don't need any editor because you're probably going to do any |
42 |
config file editing from outside of the container. And that @system |
43 |
set that has all that bootstrapping stuff is probably way overkill if |
44 |
all you ultimately need is a single package to work (and maybe not all |
45 |
of that package). Heck, your overall install approach also should be |
46 |
questioned. Rather than build your unikernel from inside its own |
47 |
container, you should be building from a more complete image and just |
48 |
installing the minimum RDEPENDs in the production container (as with |
49 |
catalyst or the chromiumos builds). And you probably wouldn't be |
50 |
upgrading such things in place either, you'd just be creating newer |
51 |
instances and cutting over from the old. |
52 |
|
53 |
I don't question that it would be great for Gentoo to support all of |
54 |
this stuff. I just think that we need to be careful not to destroy |
55 |
the experience of somebody who just wants a "typical" install in order |
56 |
to do it. Somebody who doesn't want to take the time to tweak how |
57 |
their java implementation works probably wants the default install to |
58 |
be something that meets the Oracle standard. Now, somebody who is |
59 |
into tailoring can look at their application and tweak the living |
60 |
daylights out of it, but that shouldn't be what you get when you run |
61 |
"emerge icedtea" or whatever. |
62 |
|
63 |
Sure, you could do all that with a profile, but the problem is: |
64 |
1. Maintainers aren't going to necessarily invest in that profile. |
65 |
2. New users won't necessarily use that new profile. |
66 |
|
67 |
And when those things doesn't happen users look at Gentoo as the OS |
68 |
that nothing works right on. |
69 |
|
70 |
-- |
71 |
Rich |