Mobile/Testing/03 06 13: Difference between revisions
Jump to navigation
Jump to search
(→IT) |
|||
| (7 intermediate revisions by 5 users not shown) | |||
| Line 6: | Line 6: | ||
** Are running Android Fennec x86 unittests on emulator still a Q1 priority given the state of automation? | ** Are running Android Fennec x86 unittests on emulator still a Q1 priority given the state of automation? | ||
* [wlach] - run pgo vs non pgo on developer builds inside of eideticker to measure imporovements. | * [wlach] - run pgo vs non pgo on developer builds inside of eideticker to measure imporovements. | ||
** done - measured no difference between pgo vs. non pgo for eideticker startup test | |||
* [dividehex/jmaher] - coordinate on temperature and device monitoring | * [dividehex/jmaher] - coordinate on temperature and device monitoring | ||
** done - a q2 goal for dividehex to collect this via logd/collectd on all the foopies, an experiment for this quarter | ** done - a q2 goal for dividehex to collect this via logd/collectd on all the foopies, an experiment for this quarter | ||
| Line 11: | Line 12: | ||
= Status reports = | = Status reports = | ||
== Dev team == | == Dev team == | ||
* gbrown and glandium refining pgo procedure: several minor issues identified recently | |||
* glandium is going to have a look at running the emulator on aws | |||
== Rel Eng == | == Rel Eng == | ||
* (Callek) handed Android PGO work to to glandium now that we are not longer working on the emulator, now working on Android debug tests and xpcshell tests | |||
* (kmoir) landed split robocop (rc) into 2 chunks {{bug|843107}}. Working on mozpool/mozharness and looking at sources of failure for bad pandas {{bug|843659}} | |||
== IT == | == IT == | ||
| Line 29: | Line 35: | ||
== x86 automation == | == x86 automation == | ||
* [bc] Having problems making progress. Reaching out for help. | |||
== Autophone == | == Autophone == | ||
* Cleared the s1s2 db and rerunning results with new setup in Virginia; Montreal to follow. | |||
** Running unit tests on 5 phones destroyed local network in Montreal; setting up a separate network for autophone. | |||
** Power fluctuations took out a bunch of phones in Virginia. | |||
* Investigation into whether we need to run 20 iterations of s1s2 turned up interesting result: throbberstart/stop times *increase* with each iteration. | |||
** Currently investigating whether this is due to not clearing the profile. If not, we should probably just run one iteration, since taking the mean of a monotonically increasing series doesn't help anything. Regardless this bears further investigation as to why this is happening. | |||
== Eideticker == | == Eideticker == | ||
* still prototyping new capture solution for non-HDMI devices using point grey solutions camera. need to use their sdk, which is kinda baroque and buggy, but should have sometime to show later this week. | |||
* have a new startup test which doesn't use a clean profile. should i add that to the dashboard? | |||
= Round Table = | = Round Table = | ||
= Action Items = | = Action Items = | ||
* [gbrown & jmaher & ted] investigate replacing libc and libdvm with custom compiled versions we have symbols for | |||
** ted is having trouble sync'ing the repos for the revision we run | |||
Latest revision as of 18:34, 6 March 2013
Previous Action Items
- [gbrown & jmaher & ted] investigate replacing libc and libdvm with custom compiled versions we have symbols for
- ted is having trouble sync'ing the repos for the revision we run
- [blassey] review RelEng mobile priorities
- Top failures related to mobile automation (as mentioned by the Sheriff's): https://etherpad.mozilla.org/xDFAvmdwhi
- Are running Android Fennec x86 unittests on emulator still a Q1 priority given the state of automation?
- [wlach] - run pgo vs non pgo on developer builds inside of eideticker to measure imporovements.
- done - measured no difference between pgo vs. non pgo for eideticker startup test
- [dividehex/jmaher] - coordinate on temperature and device monitoring
- done - a q2 goal for dividehex to collect this via logd/collectd on all the foopies, an experiment for this quarter
Status reports
Dev team
- gbrown and glandium refining pgo procedure: several minor issues identified recently
- glandium is going to have a look at running the emulator on aws
Rel Eng
- (Callek) handed Android PGO work to to glandium now that we are not longer working on the emulator, now working on Android debug tests and xpcshell tests
- (kmoir) landed split robocop (rc) into 2 chunks bug 843107. Working on mozpool/mozharness and looking at sources of failure for bad pandas bug 843659
IT
- Working on a higher density chassis bug 777393 - Parts are in for a prototype
A Team
- continuing to look into reboots during the tests
- highest suspect is temperatures, experimenting with temperature monitoring on a foopy/panda set: [results]
- other possibilities are kernel panics or OOM
- ???
- robocop is in 2 chunks now
x86 automation
- [bc] Having problems making progress. Reaching out for help.
Autophone
- Cleared the s1s2 db and rerunning results with new setup in Virginia; Montreal to follow.
- Running unit tests on 5 phones destroyed local network in Montreal; setting up a separate network for autophone.
- Power fluctuations took out a bunch of phones in Virginia.
- Investigation into whether we need to run 20 iterations of s1s2 turned up interesting result: throbberstart/stop times *increase* with each iteration.
- Currently investigating whether this is due to not clearing the profile. If not, we should probably just run one iteration, since taking the mean of a monotonically increasing series doesn't help anything. Regardless this bears further investigation as to why this is happening.
Eideticker
- still prototyping new capture solution for non-HDMI devices using point grey solutions camera. need to use their sdk, which is kinda baroque and buggy, but should have sometime to show later this week.
- have a new startup test which doesn't use a clean profile. should i add that to the dashboard?
Round Table
Action Items
- [gbrown & jmaher & ted] investigate replacing libc and libdvm with custom compiled versions we have symbols for
- ted is having trouble sync'ing the repos for the revision we run