The current replication process is strictly sequential with parallelization only between multiple shards. Every event runs the cycle send -> wait for response that can slow the whole process significantly especially with high latency on read.
Rework the process to push the updates without waiting for response immediately and read the responses asynchronously in separate task to react on possible errors.
The current replication process is strictly sequential with parallelization only between multiple shards. Every event runs the cycle
send -> wait for responsethat can slow the whole process significantly especially with high latency on read.Rework the process to push the updates without waiting for response immediately and read the responses asynchronously in separate task to react on possible errors.