[ 
https://issues.apache.org/jira/browse/FLINK-6352?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16365137#comment-16365137
 ] 

ASF GitHub Bot commented on FLINK-6352:
---------------------------------------

Github user tzulitai commented on the issue:

    https://github.com/apache/flink/pull/5282
  
    @aljoscha regarding the potential flakiness of the test you mentioned:
    I think the test will be stable, as long as the recorded timestamp of the 
second run is larger than the first run. We can add a loop (with max retries) 
for the test topic generation, until that condition is met.
    
    For the verification side (reading from Kafka), we'll essentially also be 
relying on Kafka to correctly return corrrect offsets for a given timestamp, 
but that is the case for almost all Kafka ITCases.
    
    Am I missing any other potential flakiness aspects of this?


> FlinkKafkaConsumer should support to use timestamp to set up start offset
> -------------------------------------------------------------------------
>
>                 Key: FLINK-6352
>                 URL: https://issues.apache.org/jira/browse/FLINK-6352
>             Project: Flink
>          Issue Type: Improvement
>          Components: Kafka Connector
>            Reporter: Fang Yong
>            Assignee: Fang Yong
>            Priority: Blocker
>             Fix For: 1.5.0
>
>
>     Currently "auto.offset.reset" is used to initialize the start offset of 
> FlinkKafkaConsumer, and the value should be earliest/latest/none. This method 
> can only let the job comsume the beginning or the most recent data, but can 
> not specify the specific offset of Kafka began to consume. 
>     So, there should be a configuration item (such as 
> "flink.source.start.time" and the format is "yyyy-MM-dd HH:mm:ss") that 
> allows user to configure the initial offset of Kafka. The action of 
> "flink.source.start.time" is as follows:
> 1) job start from checkpoint / savepoint
>   a> offset of partition can be restored from checkpoint/savepoint,  
> "flink.source.start.time" will be ignored.
>   b> there's no checkpoint/savepoint for the partition (For example, this 
> partition is newly increased), the "flink.kafka.start.time" will be used to 
> initialize the offset of the partition    
> 2) job has no checkpoint / savepoint, the "flink.source.start.time" is used 
> to initialize the offset of the kafka
>   a> the "flink.source.start.time" is valid, use it to set the offset of kafka
>   b> the "flink.source.start.time" is out-of-range, the same as it does 
> currently with no initial offset, get kafka's current offset and start reading



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

Reply via email to