On 9/23/2010 7:25 AM, Jonathan Costers wrote:
2010/9/23 Patricia Shanahan<[email protected]>
I'll set up another VirtualBox Ubuntu and use the Sun JDK to build and run
in it. If the test ever fails I'll know the JDK is not the issue. If it does
not fail, the best I'll be able to do is estimate the posterior probability
of a failure rate due to similar probability of failure to the observed rate
on OpenJDK, given 0 observed failures on N runs.
One question, how much RAM are you assigning to your VMs?
1 GB per VirtualBox. I also tell them there are 8 processors, so they
can use all the hardware threads there are on my computer.
Also, it may be interesting to note is that on Hudson, when using the ubuntu
executors, the QA test suite consistently passes.
On the solaris executors however, we consistently get one failing test.
That is why you are seeing failed and passed builds intermittently being
reported by Hudson.
Would anybody oppose to split the QA build again:
- one build only running on solaris executors (this one should consistently
fail, for now)
- one build only running on ubuntu executors (this one should consistently
pass)
- both are mutually exclusive, so if one is running, the other must wait (to
not bring the Apache infrastructure to its knees ...)
- eventually, we can look at running the QA suite on a windows(!) executor
as well ...<ducks>
I would like to see MS-Windows added as soon as possible. It is a rather
popular operating system. I recently saw a major regression running QA
on WindowsXP, so that I am currently limited to my VirtualBox environments.
Patricia