I reran the tests with the 3.10 VM distributed with Debian Lenny. They pass, though it took 3 hrs and 22 minutes. This was on the machine with the previously failing test. So the test now passes on both systems where I've tried it, using 3.10VM and 3.9 image with fixes.
So, either the VM change made the difference, or the failure is erratic, or the load on the system was different. I prefer the first explanation :)
Should it take so long?
Is there an easy way to skip the HA tests? Really all I need is the client and server in one image scenario.
A few other small points: for all versions I took from SM, I got a message saying roughly "This has not been published. Do you want to accept it?". Also, when the magma test images start there are messages like "all methods are recompiling." Finally, the transcript on the magmaTestConductor image is not big enough to hold the entire transcript.
So now I can move on to using magma! Thanks. Ross
On Fri, 2009-03-27 at 19:15 -0700, Ross Boylan wrote:
It looks as if my earlier report, sent from within the image, didn't make it out. It was in HA Unit Test 5 and had UUID being nil. I proceeded, but it got hung up further in the same unit test. The server2 process was using all the CPU and not repainting the screen. I killed everything in shut it all down.
The machine with the failure is the one I want to run on. It's fairly capable: P4 (3.0Ghz), 2G RAM. There were other things running, some of them taking up lots of memory. This was 32 bit, 3.9 based VM, and 3.9 image from Debian.
On Fri, 2009-03-27 at 17:03 -0700, Ross Boylan wrote:
The more capable machine completed the test suite without error in 43 minutes. That machine is also 64 bit.
Unfortunately, that's not the machine I want to run on. On the machine for which I reported the earlier error, I hit proceed rather than closing the debugger. It continues to chug along--no further errors.
I see now the error was raised in the very last frame, as a result of a UUID being nil.