[
https://issues.apache.org/jira/browse/BEAM-4457?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17030161#comment-17030161
]
Brian Hulette commented on BEAM-4457:
-------------------------------------
This is a cool idea! Ideally it would integrate with [~kirillkozlov]'s
projection push-down work as well (CC [~apilloud]), but I guess that's a long
way off since push-down relies on the SQL optimizer.
Would the runner be responsible for doing this?
> Analyze FieldAccessDescriptors and drop fields that are never accessed
> ----------------------------------------------------------------------
>
> Key: BEAM-4457
> URL: https://issues.apache.org/jira/browse/BEAM-4457
> Project: Beam
> Issue Type: Sub-task
> Components: io-java-gcp
> Reporter: Reuven Lax
> Assignee: Reuven Lax
> Priority: Major
>
> We can walk backwards through the graph, analyzing which fields are accessed.
> When we find paths where many fields are never accessed, we can insert a
> projection transform to drop those fields preemptively. This can save a lot
> of resources in the case where many fields in the input are never accessed.
> To do this, the FieldAccessDescriptor information must be added to the
> portability protos.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
