Resolution:
When there is a problem with a grid library on the engines (not properly downloaded or corrupted or missing) the task is not consider to have been tried (hence no need for a retry) and therefore (with default config) the error is propagated to the client and the client cancels the session.

But from 4.2 update 24, we've included the ability for this to be a candidate for retries and the proper way to enable this will be as follows viz:

1) update the engine configuration to have the parameter "-Dcom.livecluster.engine.TaskExecutor.retryOnLibraryLoadFailure=true" in the command-line arguments box of the Engine JVM section.

2) change the default setting of AutoCancel.LIBRARY_LOAD to AutoCancel.NEVER from the service type registry definition

3) enable engineblacklisting to prevent same engine from picking up the failed task again (optional depending on the service requirement and the nature of the library load failures)

Kindly note that retries will not be enabled for load library failures unless the steps above are followed so the original behaviour is preserved unless you explicitly change it.