|
[
Permlink
| « Hide
]
Steve Loughran added a comment - 14/May/08 04:52 PM (BST)
This is almost but not quite complete. A key improvement is to do this asynchronously and have the components lifecycle match that of the job
fileRequired Attribute should be in Job, not submitter, and used in the validation logic of JobImpl to skip the filename attribute if false
The current design is no good. We should not have a separate "job" and JobSubmission. The JobSubmission should contain all information about the Job, and take a cluster parameter to point to a specific cluster configuration to use/overwrite
moving to Cluster to point to cluster; no Job reference used any more. Need to look at how the URI gets resolved; if it is in the cluster it should be picked up there and then.
thats the attribute "mapred.job.tracker"
-if a cluster is set, it could be picked up from there -to allow jobs to be submitted to remote clusters, we should not require cluster to be defined, and instead fallback to a local attribute. The current method (used by SubmitterImpl and ClusterStatusImpl does not look at the cluster, but it does take a remote ref, which could be a CD. We will need to require live Prim instances with sfResolve working to support XML-file driven settings. what is checked in will now break hudson. I am aware of that fact :)
|
||||||||||||||||||||||||||||||||||||||||||||||
 |
|
Sub-task
Resolved
Major
