1 |
On Wed, 2006-04-05 at 00:13 +0100, Stuart Herbert wrote: |
2 |
> On Tue, 2006-04-04 at 22:35 +0100, Roy Marples wrote: |
3 |
> > As more and more init scripts "stopping" can trigger other services to |
4 |
> > restart, it becomes very desirable for this not to happen. So I propose for |
5 |
> > the next baselayout release (1.12.0_pre17) to default start-stop-daemon calls |
6 |
> > to SIGINT for stop commands instead of the current SIGTERM. My testing on my |
7 |
> > boxes has no adverse effects so far. |
8 |
> > |
9 |
> > So ...... thoughts? Good or bad idea? Reasons and explanations welcome :) |
10 |
> |
11 |
> From a standards point of view, SIGINT is strictly meant to be an |
12 |
> interupt from the keyboard, and SIGTERM is there to notify the process |
13 |
> that it should stop. |
14 |
> |
15 |
> I'm uneasy about going against accepted, standardised, and decades-old |
16 |
> UNIX behaviour at this level. |
17 |
> |
18 |
> Why are bind et al getting into the state that they do? If you attach a |
19 |
> debugger to the running processes, what state are they in? Why does |
20 |
> stopping dhcpd using SIGINT et al prevent that? What is dhcpd doing in |
21 |
> its signal handler? |
22 |
> |
23 |
> That seems to be the real issue that needs investigating and fixing. I |
24 |
> feel that changing the behaviour of start-stop-daemon is masking the |
25 |
> symptom, rather than fixing the bug. |
26 |
> |
27 |
|
28 |
I would have to agree - it sounds more like dhcpcd is doing something |
29 |
bad when it receives stop and runs resolvconf. So rather figure out |
30 |
what it does wrong, and if really critical, at least just make its SSD |
31 |
call use SIGINT and not SIGTERM then doing it tree wide - until you |
32 |
figured out what is wrong that is. |
33 |
|
34 |
|
35 |
-- |
36 |
Martin Schlemmer |