• distributed system
• production system
• performance
• reliability
• internship

In this thesis, we discuss the design, implementation, and testing of solutions applied to enhance the performance and resiliency of the component handling the ingestion of events, called Intake. For performance, the so-called “protobuf batching” of events in the TCP Intakes managed to reduce the computational load and halve the size of the TCP fleet, all this being fully transparent to the customers. Regarding resiliency, the migration of an Intake’s dependency to its second version increased even more the decoupling between the two components, making Intake to operate safely in case of a dependency failure. In addition, we simulated how Intake would have reacted during a cloud provider network zonal outage. We investigated the discrepancies between the expected and the actual behavior, and we proposed some solutions for making Intake to react effectively to such a type of outage.