The meaning of a conflict
Hi, the current Condor DRMAA implementation returns DRMAA_ERRNO_CONFLICTING_ATTRIBUTE_VALUES every time you try to set the same attribute a second time. Both the DRMAA 1.0 and the IDL-spec only declare that this error code is reserved for cases where the attribute value conflicts with earlier settings. The problem seems to be that there is no explicit list of job template attribute interdependencies. One obvious example is "JOIN_FILES", but there are more. Also the description of the error code is not clear enough. The Condor solution is the most easiest (and compliant !) one, but it breaks some test cases ;-) What is the strategy in the other implementations ? Do we want to have a more detailed description in the spec ? Regards, Peter.
On Sun, 16 Oct 2005, Peter Troeger wrote:
Hi,
the current Condor DRMAA implementation returns DRMAA_ERRNO_CONFLICTING_ATTRIBUTE_VALUES every time you try to set the same attribute a second time. Both the DRMAA 1.0 and the IDL-spec only declare that this error code is reserved for cases where the attribute value conflicts with earlier settings.
The problem seems to be that there is no explicit list of job template attribute interdependencies. One obvious example is "JOIN_FILES", but there are more. Also the description of the error code is not clear enough.
The Condor solution is the most easiest (and compliant !) one, but it breaks some test cases ;-) What is the strategy in the other implementations ? Do we want to have a more detailed description in the spec ?
I don't think so. To me it seems questionable whether more detailed diagnosis information provided within error codes will finally help to resolve problems. I would rather emphasize the meaning of the textual diagnosis information. Above you're saying test cases are broken due to Condor implementation. Which one is broken and why? Possibly we can change the test procedure to make it more tolerant? Regards, Andreas
participants (2)
-
Andreas Haas -
Peter Troeger