#8 Test failures on Debian system

[Niels Thykier]

When building libsysactivity on a Debian machine, the disk test (test_disk.c) now fails with a simple "IO error". I have not really debugged it.

Furthermore the process test (test_process.c) fails non-deterministically - sometimes it succeeds 3 times in a row, other times it fails (some times even repeatedly) due to the sys_time of the "process self" not increasing.

~Niels

[Carlos Olmedo Escobar]

Oops, i will need the full log of the IO error. Does it always happen?

The sys_time error is due to the kernel not refreshing the stats quickly enough so i will just remove the test.

[Niels Thykier]

I learned that the swap test fails on one of my machines, but it as I recall I did not bother putting any swap on it. The actual error is:

test_swap.c:98 ERROR: sa_get_swap(): No such device

As for the disk test, all I get is:

test_disk.c:109 ERROR: sa_open_disk(): Input/output error
2/5 Test #2: disk ...... *** Failed 0.00 sec

Finally, I noticed a new error in the CPU test (using a different machine this time). On this I get:
test_cpu.c:109 ERROR: Summatory of all cpu times is lower than 90.0: 0.000000

The full log is available at:
http://people.debian.org/~nthykier/libsysactivity.log

~Niels

[Carlos Olmedo Escobar]

Hi!

Ok, in [1] you can find a patch that hopefully fixes the swap problem.

About the disk failure, can your user open the /sys/block/ directory?

And the cpu test failure is confusing me. Can i have a copy of your /proc/stat file? (from the machine with 8 processors)

Cheers

[1] http://pastebin.com/raw.php?i=KfrU4Tz4

[Carlos Olmedo Escobar]

hi?

[Niels Thykier]

Hi,

Sorry for the late reply; the swap patch works file, so I will use that for now.

I suspect that the issue with disk failure is that /sys/ is not mounted in our build chroots, so I can just skip that.

Finally the stat file (and cpuinfo, should you need it) of my 8 cpu machine is available from[1] (and [2]).

~Niels

[1] http://people.debian.org/~nthykier/stat.txt

[2] http://people.debian.org/~nthykier/cpuinfo.txt

[Carlos Olmedo Escobar]

Ok, so the only one bug left is in the cpu and hopefully this patch [1] should fix it (actually it will report it but as a warning).

[1] http://libsysactivity.git.sourceforge.net/git/gitweb.cgi?p=libsysactivity/libsysactivity;a=blobdiff;f=test/test_cpu.c;h=69d9ceb188cfc4171cf4c9ecbfd96f967e7a90b4;hp=8057bff170b87052099a0724d55def905aa4861e;hb=558ac00fce6ec1ee548e91166a3a910e294e5ad6;hpb=e0d51cc1c41342c1e5d42112be89f6655ac33d37

[Niels Thykier]

Hi,

On my 8 CPU machine, the cpu test still fail[1]. As far as I can tell, the issue is that the test considers "100.000008" for "out of range". Though I suspect this is a rounding issue of double / floating point numbers itself[2] and not an issue in libsysactivity or my system.
Perhaps rounding it to 2 numbers after the decimal point (e.g. in this case to 100.00) would be in order to avoid false-positives.

~Niels

[1] http://people.debian.org/~nthykier/cpu-fail.log

[2] http://en.wikipedia.org/wiki/Floating_point#Accuracy_problems

[Niels Thykier]

Patch to give test_cpu a little error margin for floating point numbers

[Carlos Olmedo Escobar]

I hope this patch fixes the problem [1]. The floating point numbers will be rounded to the nearest integer. It should be enough if they are ok.

[1] http://pastebin.com/raw.php?i=9gxmiyt5

[Carlos Olmedo Escobar]

Oh, there is an error on the patch, please used this one [1].

[1] http://pastebin.com/raw.php?i=X4dnUuJ3