Cronflower: Stop Chaining Cron Jobs and Run a DAG

Engineering teams frequently encounter the same architectural ceiling as systems scale. Setting one cron job to charge orders at 02:00 and another to ship them at 02:15 relies entirely on an arbitrary fifteen-minute buffer, assuming the first task finishes on time. There are no actual dependencies, no shared data validation, and no reliable way to verify whether step two executed because step one succeeded or simply because the clock moved forward. This approach exposes the fundamental limitation of standard cron: it schedules isolated jobs rather than orchestrating an interconnected execution flow.

To overcome this, an open-source distributed scheduler for Spring Boot named cronflower provides a comprehensive solution. It splits into two core components: cronsmith, the distributed scheduler, and cronflow, the DAG orchestrator. With cronflow, developers declare steps and their interdependencies as a graph, allowing the cluster to execute the workflow node by node while passing typed data between them and recording every execution instance.

Stock photo for illustration only, not from the actual event

Architecturally, a DAG is defined as a Spring bean utilizing the @Dag annotation for naming and @DagNode methods for individual steps, with the 'to' list establishing graph edges. Nodes exchange data by returning a Map of named channel writes, reading upstream outputs through DagState. When multiple processes write concurrently—such as three scorers writing to a score channel simultaneously—hand-rolled locking mechanisms are replaced by channel reducers. Each @Channel defines how concurrent writes merge, enabling downstream nodes to read combined values such as a summed score of 90, a max weight of 40, or concatenated factors without risking mutable state conflicts.

"That's the wall plain cron hits: it schedules single jobs, it can't orchestrate a flow of them."

Paganini2008

The framework also incorporates several advanced structural capabilities: