Lokesh Lingarajan created HUDI-6724:
---------------------------------------
Summary: Initializing prevInstance to
HoodieTimeline.INIT_INSTANT_TS to avoid partial reading of first commit
Key: HUDI-6724
URL: https://issues.apache.org/jira/browse/HUDI-6724
Project: Apache Hudi
Issue Type: Bug
Reporter: Lokesh Lingarajan
Since object based incr jobs now have batching with in the commit, we can
end-up in a situation for the first commit where prevInstance is same as
startInstance according to existing code for batches within the first commit.
In this scenario when we incremental query rows > prevInstance, we will skip
the first commit as startInstance is also pointing to the same commit.
This is due to defaulting prevInstance to startInstance in
generateQueryInfo API.
Fix is to have this default to HoodieTimeline.INIT_INSTANT_TS so batching can
continue on the first commit
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
