We are starting to look at building a new architecture for Nutch. One
of the key pieces is removing the current plugin architecture in favor
of a dependency injection framework (i.e. spring).
The problem with this is certain hadoop components (mapper, reducer,
maprunnable) would need to be able to be created by the IoC container
and have their dependencies injected. I put up a basic patch that
allows mappers and reducers to be created using spring, HADOOP-3261, but
it feels cludgy. I think a better solution would be to have some type
of interface or factory that allows changing out how new instances of
mapper, reducer, maprunnables, and possibly other components are
created. Then specific implementations could be configured in the xml
configuration files. I was wondering if anybody has worked on something
like this or has any thoughts on the subject?
Dennis
