October 2023 Summaries

Companies using tiered or usage-based pricing models must safeguard their profit margins against abusive overuse, particularly in sectors like AI and DevOps that rely on costly resources. Overage can occur due to surpassing plan caps, exhausting resource quotes, or significant usage spikes from DDoS attacks or software bugs. To prevent this, companies should implement proactive measures such as customer dashboards for real-time monitoring, user alerts, and billing thresholds. Usage gating and overage charges can also be introduced to secure margins. Challenges in implementation include handling large volumes of usage data, enforcing limits for latency-sensitive applications, and forecasting with seasonal fluctuations.

The text discusses the development of a robust solution for transferring events from Kafka to ClickHouse without duplicates, which is crucial for accurate billing in OpenMeter. Initially, ksqlDB and Kafka Connect were used, but scalability issues led to moving deduplication logic out of the API layer and building a custom Go Kafka Consumer. The new consumer ensures consistent deduplication, exactly-once inserts into ClickHouse, event validation, and horizontal scaling through partitions. It also improves processing efficiency by handling invalid events and managing data volumes effectively.

The text discusses the development of OpenMeter, an application designed to help engineers collect and process large volumes of usage events for accurate billing and product use cases. A significant challenge faced during this process was generating sample data at a scale that mimics real-world scenarios. To overcome this, Benthos, a robust stream processing tool, was discovered. Benthos is highlighted as a powerful tool in the event processing domain due to its ability to ingest data from various inputs, apply custom transformations using Bloblang (a data mapping language), and send it to multiple outputs. The text provides an example of how Benthos can be used to generate sample usage events and format them into CloudEvents before sending them to the OpenMeter API.

The text discusses the importance of aligning cloud cost management with business context by fostering collaboration between engineering and finance teams. It emphasizes that focusing solely on reducing individual resource costs may not yield sustainable results in the long run. Instead, understanding how cloud spending relates to revenue, products, and customers is crucial. The author shares his experience at Stripe, where finance and engineering teams joined forces to better understand cloud spending through a data-driven strategy. The text explores the concept of "cost per transaction" as a way to translate cloud costs into business-relevant terms. It also introduces rate cards, which set prices for units that internal services or teams offer, and can be built upon one another. The author suggests keeping rate cards fixed for a year and re-calculating them only when significant architectural changes occur. The text concludes by discussing the concept of Cloud Unit Economics, which focuses on profit maximization through objective measurements. It highlights the growing focus on the FinOps Foundation, dedicated to advancing individuals practicing cloud financial management through best practices.