Description of problem: The EC2E feature does not successfully complete on RHEL4. The job is translated and executed in EC2 correctly, but the finalize hook doesn't find the job attribute to denote that the job completed successfully so it forces the job to be re-routed. Version-Release number of selected component (if applicable): 7.2.0-2 How reproducible: Steps to Reproduce: 1. 2. 3. Actual results: Expected results: Additional info:
This is caused by the condor hooks pipe handling combined with RHEL4's pipe buffer size of 4k. With condor's implementation, it is only possible to pass a kernel's pipe buffer size of data to the child hook and on pre-2.6.11 kernels the pipe buffer size of 4k (2.6.11+ is pipe buffer size is 65k).
Pipe handling was re-worked to remove artificial limitations. Fixed in the next build post: condor-7.2.2-0.9
Is this still a bug? The last version of python-boto, required by the new EC2E API[1], requires python 2.5, with a best-effort support for python 2.4. Do we support this version of boto on RHEL 4 (python 2.3), or is this functionality not supported anymore on RHEL 4? [1] https://rhn.redhat.com/errata/RHEA-2009-1187.html [2] http://code.google.com/p/boto/
There was a bug in condor that prevented the hooks from working with EC2E. That has been fixed, but the new python-boto package hasn't been tested on RHEL4. The author states he attempts to keep python2.4 compatibility, but doesn't guarantee it nor test against it. We had to make changes to the 1.0a version for EC2E to work and it is possible such issues exist in the new boto. It was decided to leave RHEL4 unsupported for EC2E at this time due to uncertainties with the boto package.
Moving back to ASSIGNED state to make sure that we do the proper messaging and whatever else might be necessary to retire an already shipping product.