[jira] [Created] (TEZ-4571) Shared fetch enabled fetches all partitions on task 0000s directly to disk for non-broadcast edges

[Jonathan Turner Eagles (Jira)]

Jonathan Turner Eagles created TEZ-4571:
-------------------------------------------

             Summary: Shared fetch enabled fetches all partitions on task 0000s 
directly to disk for non-broadcast edges
                 Key: TEZ-4571
                 URL: https://issues.apache.org/jira/browse/TEZ-4571
             Project: Apache Tez
          Issue Type: Improvement
            Reporter: Jonathan Turner Eagles


I think the idea of shared fetch is to fetch once per node for broadcast input. 
However, the enabled in the fetcher doesn't check the edge type but only that 
1) shared fetch is enabled and 2) task for the vertex is 0000. For broadcast 
edge this is correct perhaps, but for non-broadcast edges, all partitions are 
fetched to disk without possibility of sharing



--
This message was sent by Atlassian Jira
(v8.20.10#820010)