1 |
"P.V.Anthony" <pvantony@×××××××××××.sg> posted |
2 |
49721DB1.6080200@×××××××××××.sg, excerpted below, on Sun, 18 Jan 2009 |
3 |
02:04:33 +0800: |
4 |
|
5 |
> On the motherboard the watchdog is de-activated. The watchdog timer is |
6 |
> compiled into the kernel. |
7 |
> |
8 |
> During the bootup, when the server does a file system check, the server |
9 |
> reboots without fulling booting up. |
10 |
> |
11 |
> The cause of the reboot is the watchdog in the kernel. |
12 |
> |
13 |
> Guessing that if the watchdog daemon was to start much earlier, it will |
14 |
> prevent the kernel from rebooting. That will also allow more time to do |
15 |
> a file system check. Currently to get the server to boot up, the setting |
16 |
> of "0 0" is used so that it will not do a filesystem check. |
17 |
> |
18 |
> How to make the watchdog daemon start much earlier? |
19 |
|
20 |
I'd check the documentation for your kernel watchdog driver. My guess is |
21 |
that there's a kernel command-line option to disable it at boot, and |
22 |
something (a sysfs file or the like) you can prod later, to tell it to |
23 |
start, which you could then add to the watchdog daemon service script if |
24 |
there's not already provision for it. |
25 |
|
26 |
Alternatively, you could of course compile the driver as a module and not |
27 |
have it load until later, but the kernel folks are pretty good about |
28 |
making it such that it can be compiled in if desired -- there's command- |
29 |
line options and etc such that preventing early-activation or otherwise |
30 |
setting module options can be done from the command line instead, if it's |
31 |
built-in. If that's /not/ the case here, and the documentation does not |
32 |
specifically /say/ that bit isn't working yet (which may be the case with |
33 |
an experimental driver or one still specifically under development), I'd |
34 |
consider not having a way to disable it thru boot and enable it later a |
35 |
bug, and I expect the kernel folks would as well. |
36 |
|
37 |
FWIW, I too have the fsck service deactivated here, as you do, by setting |
38 |
the fsck order column in fstab to 0. On reiserfs, which is what I run |
39 |
here for pretty much everything, it's not really needed anyway, since |
40 |
reiserfs automatically does pretty much everything it could do in a |
41 |
normal fsck at mount in any case, and if it's beyond that, it's time for |
42 |
--rebuild-tree. But I've not had to run that in /ages/, I think since |
43 |
the kernel got and activated by default data=ordered for reiserfs, and |
44 |
likely since I've been on Gentoo as well. (IDR which came first, |
45 |
reiserfs data=ordered or my switch to Gentoo, but I sure do remember |
46 |
looking forward to each of them! =:^) But, while that eliminates the |
47 |
(unnecessary for reiserfs) fsck time, because reiserfs can take a bit to |
48 |
replay the journal if the system crashed or the filesystems were |
49 |
otherwise not unmounted properly, you may end up with the same problem |
50 |
under those conditions. |
51 |
|
52 |
But, if you have no strong feelings against reiserfs (some do, but as I |
53 |
said I've been very happy with it since data=ordered became the default, |
54 |
YMMV), that might be another thing you could try. At least you wouldn't |
55 |
have to feel so bad about missing the fscks then, and if you happen to be |
56 |
already running it, hey! =:^) |
57 |
|
58 |
-- |
59 |
Duncan - List replies preferred. No HTML msgs. |
60 |
"Every nonfree program has a lord, a master -- |
61 |
and if you use the program, he is your master." Richard Stallman |