[Pkg-sysvinit-devel] sysv-rc: Multiple init execution (fixed?)

[Michael]

Dear Petter,

I think this is (hopefully) the last mail you will get from in this repsect :) at least, if there aren't any further requests from you.

I wrote 'fixed?' because i can not tell how it got resolved. Suddenly, it works again as expected.
Of course, i did many things before to find the reason; but then, after it worked again, i reverted all these changes one by one, and it still works.

First of all, and most important, it turned out that bootlogd does not work when /var is a mounted partition (as in my case).

I moved /var to a separate partition about 1 year ago, when i figured that separate partitions would speed up the harddisk by parallel r/w, and avoid fragmentation in the first place. This was done at my laptop first, where disk always was the bottleneck. I set up some more partitions (you got my fstab), and it all worked flawless. So i migrated this setup to my PC too. I also modified mount paramters like 'noatime, nodiratime' and move /tmp to shm.
As i said, it all worked fine, and was quite fast.

(Now, with /var/run and /var/lock moved to shm by debian, my somewhat rude /var tweak is no more necessary anyway.)

I'll attach some bootlogs, you can see there were old 'boot' files on the mounted partition, all new files were written to the unmounted /var which still was on the root FS, but later got replaced by the mount.

So, i removed that line in /fstab and copied the recent stuff from the /var partition over to the root FS.

Now before you think this is the reason for the double-init: Probably it's not so easy.
Because after i did this, it still was broken like before. And vice versa: When it got 'solved' magically, i reverted to the original /var-as-mount layout, and it still was 'fixed'.

However, maybe there was some influence of files in /var/lib ? Are there any such files accessed before mountall.sh ? Then they would have been the 1 year old version, never got updated.
But, as i said, even with the new versions copied over, and shut down /var partition, it still was broken (for several reboots).

Then i did some other things. For example, i edited the keyboardrequest line in inittab, to see if i can have a 'sleep 3' when pressing ALT+UpArrow. (See attached file). It did not work (maybe because the USB keyboard connected to a hub gets recognized too slowly by udev, anyway USB is somewhat flakey here on this cheap upgrade kit mainboard). However, i did not change anything else. Can just writing inittab to disk make a difference ? (Encoding trouble ??)

Then at some point (when i added stable to sources to install kernel 2.6.26, which did not solve the problem) there was an update for the latest 'unstable' kernel 2.6.31-22.

I then launched 2.6.26 and next 2.6.30, to follow my checklist, but it turned out to make no difference (init still broken), *but* of course the previous kernel update triggered a new initrd which means, update-grub was called and then grub was written on-disk (into the MBR) again. At least, as i understand it. I know for sure there was no grub related package update.

Can just re-writing grub on disk make a difference (with no package update) ? It's at least fact that the whole bug appeared after the update to grub2, although i did not consider this to be related until now.

Strange it got 'fixed' only after i launched the latest (updated) kernel. It also went fine now with the other kernel versions, which previously were also displaying that bug, even after the update-grub  run. Can just running a different initrd lead to something re-configured differently in /var ?

Anyway. I'm somehow finished, after 8 hours of tracking this, and as i said, i will not dive further into it until there's a reason. But you like to see some specific file of the old root filesystem /var which could have been used at booting (before /var got mounted), i have a full backup. 

The reason why i write 'fixed?' is that things that disappear magically, have the ugly feature to re-appear magically again...i really would have preferred to track the final reason for this bug. As it looks, i have to be satisfied with it being gone.

Oh well. I forgot my laptop. It's still untweaked, the bug is still there. 
Hmpf :P
I'll look for it some other day!

Kind regards, 

Michael





-------------- next part --------------
A non-text attachment was scrubbed...
Name: 2009-12-12-fixed.tar.bz2
Type: application/x-bzip
Size: 60940 bytes
Desc: not available
URL: <http://lists.alioth.debian.org/pipermail/pkg-sysvinit-devel/attachments/20091212/be224338/attachment-0001.bin>