Closed deemogsw closed 1 year ago
@deemogsw, thanks for the feature reporter. Do you have interest to create a pull request to contribute the feature?
@deemogsw, thanks for the feature reporter. Do you have interest to create a pull request to contribute the feature?
yeah,I have finished test case.After update the example code in README,I will push a branch
current: 1.Startmode is configured by ’consumer.offset.reset.to‘ This is an internal property to set the offset when consumer start. 2.Only three modes has been supported. 1.latest 2.earliest 3.timestamp When consumer.offset.reset.to = latest is set. if job restart from none checkpint,current offset will be reset to the maxOffset.You will lost the date before state of job turn to running. It has a lot of limits when job need whol stream data.
I'm working:
1.Add two mode:GROUP_OFFSETS and SPECIFIC_OFFSETS
2.Abandon ’consumer.offset.reset.to‘.Use setFunction.
When job is first running,it will consum from the latest offset. When job recover from checkPoint ,it will consum from the stored offset. When job has been modified,and restart with no state,it will consum from the group offset.
3.SPECIFIC_OFFSETS is a special case. It only be useful when job need recover from an unexpected error and none of checkpoint can touched.
4.Add metric currentFetchEventTimeLag:abs(fetchTime - storedTime) currentEmitEventTimeLag:abs(emitTime - storedTime) FLIP-33: Standardize Connector Metrics currentEmitEventTimeLag would be showed as delay in job deployments page at ali-cloud-product
It alse can be fetched in flink-web's metric page,as a judgment when the job delay.