Bug#190110: libc6: Sub-processes flaking out.

[mah@everybody.org (Mark A. Hershberger)]

GOTO Masanori <gotom@debian.or.jp> writes:

> (1) Why do you think it's glibc bug?  I don't still understand your
>     point.

Random breakage when I'm pretty sure it is not a hardware problem.
Random breakage after upgrading glibc.

>> (2) Randomly breakage is normally hardware bug.  Well there are some
>     possibilities of software bug in gcc/glibc/binutils/kernel and
>     debian tools like debuild.

Which is what I suspect.

> (3) If this bug is glibc, then almost debian guys should be caused
>     this problem, but I have no experiments of this kind of bug.

All I know is that the machine passes memtest86 with flying colors and
a clean install on a different disk seems to work.

> This seems very bad behavior.  Did you try to test memory on your
> machine?  memtest86 is a good tool to check it.

I've used memtest86 on other machines to find problems.  This time
when I ran it, there were no problems.

>  #147183 says:
>
>> The fact that running tar (on anything) before running dpkg
>> makes a difference indicates clearly that the problem was
>> some sort of system flakiness.  
>
> This is very bad state.  If you have this problem, the filesystem will
> be totally unusable from my experiment with bad CPU which is weak of
> temperature-resistant under high load.

If you read my bug report, you would notice that I was able to do a
clean install and proceed without problems -- performing the same
tasks under the same load conditions.

I've performed many tests to verify that this is not a hardware
issue.  Further, if it were a hardware issue, it is strange that two
separate machines would show the same behavior at the same time
(i.e. after upgrading glibc) and that the buggy behavior disappears
after a clean install.

Note that mounting the old disk in the rebuilt system and using it for
a chrooted environment fails to reproduce the problem.  So, I thought,
"perhaps a corrupted kernel" and used dpkg --root=/mnt to install a
new kernel on the old partition.  Yet, the problem is still there if I
boot from the old partition after doing this.

Note that I'm reasonably sure that the configuations of the new
install and old install the same.  I used jablicate to generate a
package list.

Mark.

-- 
You are a mystery as deep as the sea; the more I search, the more
I find, and the more I find the more I search for you.
	    -- St. Catherine of Siena