Animesh Nath

A Spring Java-based application for streaming messages from the upstream pub/sub-component. Most of the time, the legacy application had Out Of Memory errors because of the heavy volume and the high number of objects created. The size of each object was huge due to the complete DOM tree loaded for the XML message. As a result of the banking domain component, zero defects were expected as every message was needed, and throughput was expected to grow. We met all the above expectations during delivery by following a blue-green approach to delivery and a high level of testing and coding best practices using trade-offs on time over space and using SAX parsing technology.

The memory footprint was reduced by 80% and throughput by 10%.

The adaptive rate limiter helps to flexibly adapt the current rate limit of the flow of messages in the system concerning the current error count collected within an interval. If the error count increases, the rate limit is reduced to reduce the flow. If fewer errors exist, the rate increases until it reaches the max throughput threshold or gives more errors.

We have also written the rate limiters: heavy volume flow rate limiter (>1 message per second to 100 million messages per second), low volume flow rate limiter for delivering messages (less than one message per 10s). This helped reduce overall client-to-provider error rates (HTTP 429 or rate limit exceeds errors) by 50% by regulating the overall flow in the system, as per the exact rate.