1 |
On Mon, Jul 9, 2012 at 3:05 AM, <heroxbd@×××××.com> wrote: |
2 |
> Luca Barbato <lu_zero@g.o> writes: |
3 |
>>> What we've discussed in the beginning, such as event-driven init, |
4 |
>>> periodical events, process monitoring and crash restart are still on |
5 |
>>> the todo list. |
6 |
>> |
7 |
>> That's great, do you feel confident you'll be able to get all of this |
8 |
>> done? |
9 |
> |
10 |
> I feel these are not technically difficult. But the policies count, |
11 |
> besides the debates that if we really need these fancy features for an |
12 |
> init system. My current feeling (or planning) is that just to make dirty |
13 |
> ones with simple scripts to see if our community (debian is more similar |
14 |
> to us than fedora) really like the things. The rule of thumb is to |
15 |
> always make them optional, hopefully independent, components. |
16 |
> |
17 |
|
18 |
As a certified member of the peanut gallery I can testify that process |
19 |
monitoring and restarting would be a very nice feature to have. |
20 |
Having this sort of capability in a chroot/prefix would probably |
21 |
create a bunch of possibilities. Tools for doing this exist, but they |
22 |
are weak. |
23 |
|
24 |
I switched a VM over to systemd because it had an unstable daemon and |
25 |
I wanted to try out this feature. Since systemd places each daemon in |
26 |
a cgroup it is able to fairly effectively monitor what is going on |
27 |
with them. That might be something to keep in mind if you move ahead |
28 |
with this. |
29 |
|
30 |
I don't use a laptop with Gentoo so event-driven init is less useful |
31 |
personally, but I imagine that if I had one it would be very nice to |
32 |
have. Again, I'd look at what upstart/systemd are doing to avoid |
33 |
re-inventing the wheel here. |
34 |
|
35 |
Keep in mind that process restarting is actually a very standard |
36 |
feature in init - we just rarely actually run daemons directly from |
37 |
init. On the occasion that I've actually stuck something in inittab |
38 |
init is very diligent in propping it back up. The challenge will be |
39 |
things like hung processes that don't actually die - but we can start |
40 |
with the simple case. |
41 |
|
42 |
Rich |