When we get the error - Timeout building Daffodil on "Host (RHEL6 x64 12 cpu 96GB)" the parsers then fail to build correctly.
I think this is a timing issue. When the CLI needs to be rebuilt, it does that first and then rebuilds the parsers. If this is taking too long, the VM is stopped and put back in the queue to try again. When it tries that machine again, it shows that the CLI has already been built, so it does not try to rebuild the parsers. Any parsers that weren't built the first time around, will not exist until the CLI needs to be rebuilt again, or they are created manually.
I'm increasing the timeout for now but this needs a real solution as well. The harness also needs to be updated to send an error message when a parser fails to build. Currently the error we get is for running a test that is trying to use a parser that doesn't exist.