We missed a big part of the fact that the internal retries of the ConnectionProvider also call on close.
For example, JdbcDbWriter#closeQuitely is called during internal retries JdbcSinkTask#put. This can cause NPEs as we return a null connection back when a new connection is requested since we have set isRunning to false.
This regression wasn't caught by 5.0.x tests. It failed further down on the pint merge on newer branches.
What?
Reverts confluentinc/kafka-connect-jdbc#1206
Why?
We missed a big part of the fact that the internal retries of the ConnectionProvider also call on
close
.For example, JdbcDbWriter#closeQuitely is called during internal retries JdbcSinkTask#put. This can cause NPEs as we return a null connection back when a new connection is requested since we have set
isRunning
to false.This regression wasn't caught by 5.0.x tests. It failed further down on the pint merge on newer branches.