site key generation is deterministic, you can do the same
>> thing on both streams, and join on the composite key.
>>
>> You’d want to cache the mapping from the real key to the synthetic value,
>> to avoid doing this calculation for every record.
>>
>> If that sounds
teful join.
>
> If it’s something like a left outer join without any state TTL or need to
> keep both sides in state, then it’s pretty easy.
>
> — Ken
>
> PS - it’s pretty easy to figure out a “-xxx” value to append to a topic
> name to get the hashCode() result you need.
>
Hello,
My team has a Flink streaming job that does a stateful join across two high
throughput kafka topics. This results in a large amount of data ser/de and
shuffling (about 1gb/s for context). We're running into a bottleneck on
this shuffling step. We've attempted to optimize our flink configura