This triggered CI jobs which resulted in unexpected failures which were not observable in development environments:
The Windows VM on the CI worker host where this work was being carried out was displaying subpar performance. On some job runs the VM stalled, unable to finish its provisioning process, and had to aborted. Once the host was restarted the test that was failing passed once but failed in subsequent attempts:
15:34:29.288: jq: Test concluded - Module "gpii.tests.windows.killProcess" Test name "Testing Killing Processes": 4/4 passed - PASS
Further troubleshooting led us to believe that disabling 3D acceleration led to a more responsive VM that could be provisioned as expected and resulted in the gpii.tests.windows.killProcess test passing. Searches performed using VirtualBox logs excerpts showed other performance issues related to 3D acceleration https://www.virtualbox.org/ticket/10879#comment:2
I would like to suggest that we disable 3D acceleration (and increase the amount of allocated video memory) for all VMs used by GPII projects:
- http://pastie.org/10884233 Example changes for the gpii/windows Vagrantfile
The current CI worker host is a Mac Mini with 16 GB of RAM, onboard Intel GPU, and running Xfce on Fedora 22. There was a discussion at the last APCP face-to-face event held in Toronto about using server grade hardware for worker hosts. The replacements will be Dell C2100 servers with 96 GB RAM and dual processors which will provide greater capacity for CI workloads. The replacement servers have older and less capable video cards (Integrated AST2050 with 8MB VGA memory) than the current Mac Mini and should not be relied upon for 3D acceleration anyway.