[
https://issues.apache.org/jira/browse/UIMA-2790?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Richard Eckart de Castilho updated UIMA-2790:
---------------------------------------------
Component/s: Core Java Framework
> Change Index Merge for aggregates to exclude remote delegate index
> specifications
> ---------------------------------------------------------------------------------
>
> Key: UIMA-2790
> URL: https://issues.apache.org/jira/browse/UIMA-2790
> Project: UIMA
> Issue Type: Improvement
> Components: Core Java Framework
> Affects Versions: 2.4.0SDK
> Reporter: Marshall Schor
> Priority: Minor
>
> The general approach for UIMA aggregates includes merging the meta
> information of all the delegates to create a common CAS specification. The
> CAS Specification includes a type system, and an index specification, among
> other things. If the delegates are local (running in the same JVM and
> sharing the CAS's index repositories) then the index specifications from the
> delegates need to be "merged" (a logical union operation) so that an
> add-to-indexes operation can add to all of the indexes in that repository
> that might be used by a delegate. However, for remote delegates, any "adding
> to index" operations are done by the remote. Omitting their index
> definitions from the aggregate merge set makes no difference to the operation
> in this case (other than to potentially reduce the indexes that are
> maintained by client running the aggregate. So it is a potential performance
> improvement to omit these.
> I asked Adam if he recalled why the design was this way,and he reminded me
> that a long time ago, we didn't have "automatic" bag indexes for every type,
> like we do now. In that older design, if the indexes were not merged, then
> the creation of a type which had no index would get forgotten if it was added
> to the indexes, as there was no index in which to remember it. Now that this
> is not the case, there's no logical reason (that I can think of) why remote
> delegate index specifications (including their type priorities) need to be
> "merged" with a client aggregate.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
