Yep. That's correct.
Cool that you are writing a book! :)
-Matthias
On 6/9/17 7:07 PM, john cheng wrote:
> Tks Matthias, I add some offset log in ProcessorStateManager and
> RocksDBStore.
> The behaviour is just like you explaination.
>
> This is restart instance1's log. The two active task re
Tks Matthias, I add some offset log in ProcessorStateManager and
RocksDBStore.
The behaviour is just like you explaination.
This is restart instance1's log. The two active task really doing replay
work from position 2 of checkpoint file
[06:55,061] createStreamTask 0_0, partitions: [streams-wc-in
Your observation is completely correct and this is also correct behavior.
Note, that instance1 and instance2 both also do have a local RocksDB
instance that holds the state. The checkpoint file basically tells
streams, what prefix of the changelog topic is already in RocksDB.
As Streams loads (no
I have two app instance, input topic has 2 partitions, each instance config
one thread and one replicas.
also, instance1's state-store is /tmp/kafka-streams, instance2's
state-store is /tmp/kafka-streams2.
now I do this experiment to study checkpointin kafka streams (0.10.0.0).
1. start instance1,
