Caizhi Weng created FLINK-18558:
-----------------------------------
Summary: Introduce collect iterator with at least once semantics
and exactly once semantics without fault tolerance
Key: FLINK-18558
URL: https://issues.apache.org/jira/browse/FLINK-18558
Project: Flink
Issue Type: Improvement
Components: API / DataStream, Table SQL / API
Reporter: Caizhi Weng
Fix For: 1.12.0
Currently {{TableResult#collect}} and {{DataStreamUtils#collect}} can only
produce results if users explicitly enable checkpoint for infinite streaming
jobs. It would be strange to require the users to do so if they just want to
take a look at their data.
We should introduce collect iterator with at least once semantics and exactly
once semantics without fault tolerance. When calling the {{collect}} method, we
should automatically pick an iterator for the user:
* If the user does not explicitly enable a checkpoint, we use exactly once
iterator without fault tolerance. That is to say, the iterator will throw
exception once the job restarts.
* If the user explicitly enables an exactly once checkpoint, we use the current
implementation of collect iterator.
* If the user explicitly enables an at least once checkpoint, we use the at
least once iterator. That is to say, the iterator ignores both checkpoint
information and job restarts.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
