Wellington Friends in testing vs joyride-2301
martin.langhoff at gmail.com
Sat Aug 16 02:50:05 EDT 2008
The test efforts were divided in 2 tracks:
- generating a lot of network traffic in an attempt to repro the
- run through parts of the smoketest revisiting bugs found last week
We didn't have time to explore other areas of the smoketest as we
found plenty to keep us busy.
Network lockups - none
For the network traffic test, I associated 2 XOs to a XS with an AA
and generated wifi traffic with
while true; do wget http://localkernelmirror/path/to/kernel.tar.gz
; rm kernel* ; done;
wget reported between ~400KB/s on both (probably bottlenecked on a
slow switch - the nz kernel mirror is a dozen meters away). We did not
see any glitch at all. The night before, 6 XOs had updated from 2270
to 2301 using olpc-update with not issues. And when we started the
tests I quickly ran the activities update on all 6 XOs. In short, we
failed to see any anomaly -- the RF env is very lightly loaded in this
place. A few faint beacons can be seen, and there are 2 other strong
signals - neither in the same channel.
Smoketests and revisiting bugs
In general the system worked well, though we didn't get through the
whole smoketest. I am focussing here on the notably bad issues :-) -
which I tagged 'blocker?:8.2.0'. We tried to explore these deeper, get
logs and narrow down on steps-to-repro. Feel free to ask for more info
and retests on each bug.
- Measure is still broken #7771 - this one looks like a trivial
dependency is missing. Add the python-numeric rpm to the build...
- Record is still broken #7887 - but more consistenly than before.
Logs attached showing the last acrions before the segfaults. If you
can tell me where the coredump gets stored, I'll attach one too...
We got a lot of almost consistent *hard* lockups related to sound.
TamTamJam doesn't start, and Acoustic Measure (Distance) locks us
badly when using the sound. This is a lot more consistent than in
joyride-2270 (where we had seen 1 machine lock up consistently on
TamTamMini, and nothing else), although still never 100%. Acoustic
Measure is the most consistent reproducer of the problem. See #7996
Trying to revisit #7985 (Journal does not always show mounted USB
drives) and using USB drives to pass around activity logs we came
across a new and very nasty bug: when you unmount a USB drive from the
Journal, it doesn't really unmount it. We nuked a couple of
FAT-formatted USB drives (losing all data) to this bug, because FAT is
particularly unhappy about unclean unmounts. Filed as #7993, I'd
characterise this as a high-priority dataloss blocker. This build eats
USB drives with gusto.
(Luckily, we didn't lose any data. Thanks for asking :-) )
Credits to Alastair Munro and Tabitha Roder for the bug hunting.
martin.langhoff at gmail.com
martin at laptop.org -- School Server Architect
- ask interesting questions
- don't get distracted with shiny stuff - working code first
More information about the Devel