Hi, guys. As described in https://pulsar.apache.org/docs/next/concepts-replication/#aggregation-replication, there is an aggregation replication mode. Yet in the doc of https://pulsar.apache.org/docs/next/administration-geo/, there is no instruction of setting the aggregation mode. So my question is how to set the aggregation replication mode ? Thank you.

there is no special "aggregration replication mode". The documentation is not very clear and doesn't provide an example of such configuration.

Per my understanding, normal replication provides two-way ability, how one-way (aggregation) can be done via configuration ?

I have been using geo-replication without a shared metadata store. At least in that case, replication is configured point-to-point for each one-way direction. Yes, there are some gotchas with partitioned topics. I think one solution is to disable topic auto creation. I believe that the replicator will attempt to create the topic in the remote cluster when replication starts the first time. In addition, setting up authorization and TLS isn't very straightforward.

Agreed. It seems I have done exactly the same thing. And I shall continue on that track. Thank you very much for your help.

I found out the

replicated subscription status

seems not working.

@huangyanshuo what doesn't work there? Replicated subscriptions have some limitations. the discussion in https://github.com/apache/pulsar/discussions/22315 contains some details

I misunderstood the mechanism of replicated subscriptions. I originally thought that if I consume topic

t1

with subscription name

s

in cluster

z1

, the subscription

s

of topic

t1

in cluster

z2

won’t consume the same message and there would be a subscription created in the `z2`'s topic. An illustration would be:

sub s1 -> topic t1 (in cluster z1)
sub s1 -> topic t1 (in cluster z2)
s1 in z1 & s1 in z2 shouldn't consume the same message

But after reading your excellent post, and I quote

The main limitation of replicated subscription is that only the "mark delete" position is replicated

, I think the behavior observed is quite normal and expected. Thank you again!