Zombies when using noflushd

[Christian J. <chr...@et...>]

Hi Daniel and all

Have you noticed this bug report?:

http://sourceforge.net/tracker/index.php?func=detail&aid=695613&group_id=12257&atid=112257

Since I'm not sure anyone is reading the bug database, I'm posting to 
the list, and am sending Alex Clouter a copy.

The observations I have made are:

- I've seen problems with zombies on my ppc Laptop (Powerbook G3), 
regardless of ext2 or ext3, during the time I used noflushd. I 
switched noflushd off when I did upgrade my ext2 partition to ext3, 
and didn't have the following problems anymore, until I switched 
noflushd on again because of some conversation with Alex.

using noflushd,

- XMMS would frequently hang upon stopping and starting a new sound 
(i.e. after about 5-10 stop-and-go's), and needs to be restarted.
- Licq becomes unusable after about half an hour, and has about 
twenty zombies or so when this happens. It usually takes some time, 
during which a few zombies appear and get reaped again within 
seconds, but then suddenly new zombies start to appear en masse, 
about one every few seconds until licq issues popup warnings about 
failure to spawn new threads and then becomes unresponsive.
- the mysql server would sometimes show the same zombie problem, 
though it didn't make it unusable for me. I don't use mysql much on 
my laptop, so I don't know how severe the problem would be in real 
production use.
- NEdit would crash when it pops up a window asking the user for 
confirmation for something. Note that I compiled nedit with the real 
motiv not lesstif for stability and always worked reliably otherwise. 
(And the crashes were so bad that I migrated to xemacs in that time.. 
:) Later I discovered that it didn't crash anymore and (as I think) 
found out why.)

All of those disappear when removing noflushd from the sysv-init dir 
and rebooting the laptop. I've also never seen those problems on any 
of my ppc/x86 desktops. (Hm, though I even don't remember having seen 
any zombie problem even on one of my x86 desktops where I've run 
noflushd for some time. I would have to check that again. It's also 
interesting that a simple example threading program from Alex did not 
show zombies when I tried it out on my laptop with noflushd turned on 
again, though I might not have waited long enough (Alex says it takes 
43 minutes from the start of noflushd).)

(I've now worked about half a year without noflushd and have 
consequently never seen the problem again. I would like to downgrade 
to ext2 again and use it again for battery and noise savings, though.)

I don't know what noflushd is doing exactly. Maybe it just triggers 
some kernel bug.

Cheers
Christian.