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.
Hi Ross, congratulations on getting through the test suite. I'm not sure what happened when your server2 image locked up (that should never happen) but it sounds like one of those low-level problems. I have been using the 3.10-4 VM and recommend it for anyone running Magma. It seems to be the most stable VM, the tests pass for me every time with no hang-ups.
3 hours is a long time. My dual-core laptop gets through them in less than half that time. One thing to note is that each extra core should dramatically improve the run-times of the test-suite because, during the HA tests, it wants up to five cores. The HA tests can be disabled by commenting out the code of that test.
Magma doesnt' recompile anything, what I suspect you saw was that it was unloading the Server package from the two client-player images. This is done to test there are no packaging issues; i.e., a method in the server package that should be in the client package would be exposed by unloading the server packagtes, which is what the test-suite does at the very beginning.
Cheers, Chris
On Sat, Mar 28, 2009 at 10:57 AM, Ross Boylan RossBoylan@stanfordalumni.org wrote:
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.
magma@lists.squeakfoundation.org
-
Chris Muller -
Ross Boylan