November 2024 Summaries

Kafka, while popular for distributed streaming, is not optimized for traditional queuing tasks due to its resource-intensive nature and lack of native support for core queuing features like concurrency control, error handling, and reliable message ordering. Inngest offers a modern alternative to Kafka with built-in queuing functionality that's simple, reliable, and resource-efficient. It provides high-level abstractions for building sophisticated workflows without needing to construct additional capabilities from scratch. Inngest simplifies workflows by providing built-in reliability and scalability without the complexity of traditional queuing systems, allowing teams to focus on building reliable applications faster and without the maintenance burdens of traditional queuing and workflow solutions.

Inngest, a developer-first platform for handling orchestration at scale, has introduced two new AI-native features: AgentKit and step.ai. AgentKit is designed to develop, test, and ship complex AI workflows to production, while step.ai brings the same world-class developer experience from Inngest's existing product to AI-native applications. These features aim to simplify the process of building production-ready AI workflows by handling AI model calls with built-in observability and support for agentic workflows out of the box.

During October 2024, multiple infrastructure upgrades were made to the queueing system at Inngest, which unfortunately resulted in incidents causing degraded performance but no data loss. The issues included degraded performance on October 4th due to a new design of individual queue processing that became less efficient under high load with extreme custom key cardinality; execution and cron delays on October 9th after an improvement and refactor to the cron service; function run status update issues on October 11th due to unreliable data publishing within NATS; delayed execution for some function runs on October 18th due to increased load on the primary queue shard's CPU; sporadic delayed event matching on October 22nd; processing issues with step.waitForEvent and cancelOn operations on October 24th due to unoptimized expressions being processed at high volumes; and degraded UI and DB performance on October 30th due to a disk issue within the Postgres RDS instance. Despite these incidents, improvements were made in various areas such as increased capacity for event matching, optimizing expression handling, adjusting worker configurations, implementing dynamic scaling based on resource utilization metrics, moving reads to a read replica, improving local caching for perf, and applying maintenance to the primary Postgres instance.

Inngest has released its Function Run Search feature, allowing users to search across all their function runs using advanced queries filtering on events and function run properties. This feature greatly improves observability and debugging capabilities for developers. The search bar lets users search and filter runs based on event properties and function run output. It also enables quicker issue resolution, an easier way to cross-find runs from application logs, and a simple way to answer complex questions by combining event and output properties. Searching returns results in milliseconds, even with hundreds of millions of daily runs. The Function Run Search feature is built using the CEL expression engine and has been shaped with the help of beta testers.