On 29.04.20 09:18, Benoit Delbosc wrote:
>> 2) So in Kafka 2.0 we introduced poll(Duration) which would practically be
>> more strict in respecting the passed in timeout. It means, it could return
>> while we are still in the middle of a rebalance. At the same time we
>
On 29.04.20 09:18, Benoit Delbosc wrote:
> Hi
>
> On 28.04.20 21:05, Guozhang Wang wrote:
>> Thanks for the explanation Ben. They are very helpful.
>>
>> Just to clarify on the context here:
>>
>> 1) Before Kafka 2.0 the poll(long) call make sure that the r
ocation during an applicative checkpoint.
This should continue to work, assuming that after a `poll(Duration)`
call that doesn't trigger any ConsumerRebalanceListener callbacks,
`commitSync` must not raise a RebalanceInProgress exception. Am I right?
Thank you, I appreciate your very precise a
would suggest to improve the onPartitionsAssigned Javadoc or
poll(Duration) to make it clear, may be part of KAFKA-9882.
Regards
ben
>
> Guozhang
>
>
> On Mon, Apr 27, 2020 at 12:45 AM Benoit Delbosc wrote:
>
>> Hi,
>>
>> I am having a hard time un
Hi,
I am having a hard time understanding why a Consumer#commitSync can
randomly fail.
Not being able to commit creates duplicates, which I can understand in
failure cases:
- the poll interval is not respected max.poll.interval.ms (300s)
- a consumer is stuck or die -> no heart beat during sess