RPS On average 4-5 thousand RPS, peaked at 12k But it is important to understand that the main load was on Kafka. Since the architecture is asynchronous, for each ad creation, my orchestrator generated a chain of 5–7 events to different check services. In total, we processed a stream of up to 30k messages per second on average.

In the described architecture with asynchronous processing via Kafka, the load is distributed among services, allowing the system to scale and handle high volumes of messages.

Example:

• Average load — 4-5 thousand requests per second (RPS).
• Peak load reaches 12 thousand RPS.
• Each request generates a chain of 5–7 events, which are published to Kafka for different services.
• As a result, the total message flow reaches 30 thousand per second.

This approach helps to offload the main service, increase fault tolerance, and flexibly scale event processing depending on the load.