Chronoverse uses Kafka for durable asynchronous workflows and preserves ordering within each partition.
Topics
| Topic | Primary traffic |
|---|---|
workflows | Workflow build and lifecycle events |
jobs | Dispatch and execution work |
job_logs | Retained stdout and stderr events |
analytics | Workflow, job, and log analytics events |
Topic auto-creation is disabled. The Compose initialization job creates or expands the required partition counts.
Partition runner
Each assigned partition is processed independently. This avoids a slow record in one partition blocking unrelated partitions while retaining per-partition order.
Commit policy
Handlers classify outcomes so the runner can decide whether to commit, retry with backoff, or treat an error as terminal. Offsets are not advanced simply because a handler returned.
Rebalances and cancellation
Partition work is tied to assignment lifetime. Cancellation stops processing cleanly so records are not acknowledged after ownership changes.
Capacity
Useful consumer parallelism is bounded by topic partitions. Increasing replicas beyond partitions adds standby capacity, not additional throughput.