Hi Folks, We have a table with replicated_storage= true. During rolling upgrades, we see broker loads getting cancelled/stuck for sometime. What we see in the code is Starrocks does sequential replication where it replicates to one node secondary, waits for response and so on. Check ReplicateChannel::async_segment code path. Ideally the request should be fanned out to secondaries and then a quorum should be established. Is there a specific reason Starrocks does the replication serially one secondary at a time?