Den Cluster Harness / Project note
Inspectable dispatch for private cluster jobs
Den Cluster Harness defines a bounded workflow for dispatching replicated Docker Swarm jobs, capturing task state and logs, summarizing outcomes, and using a separate helper to transfer and compare image IDs without claiming an exercised infrastructure result.
- Our role
- Job dispatch and image distribution tooling
- Stage
- Internal tool
- Category
- Tools & automation
Cluster job workflow
Specify The Job
Define the replicated job, resources, and timeout; dispatcher restart is hardcoded to none
Check And Dispatch
Refuse replacement, enforce limits, and create the service
Capture Job Evidence
Record task state, inspection, logs, status, failure, and elapsed time
Report Cleanup Result
Attempt removal and record whether cleanup succeeds or fails
Workflow illustration from source, not live UI, showing job specification, guarded dispatch, evidence capture, cleanup reporting, and separate image identity verification.
Make Jobs Inspectable
Private cluster jobs need more than a service creation command. We designed a workflow that makes replacement checks, resource limits, task behavior, runtime evidence, and cleanup outcomes visible so a job result can be reviewed without treating infrastructure assumptions as proof.
Bound Dispatch And Evidence
The dispatcher checks for an existing service name and refuses replacement before creation. CPU and memory limits, one replica per node, a bounded timeout, and hardcoded no restart constrain the job. It captures task state, inspection, logs, and a JSON summary of status, failure, elapsed time, and cleanup.
Results Follow The Job
The harness turns dispatch into an inspectable record, including what failed, how long it took, and whether cleanup succeeded or failed. A separate image helper compares manager and worker image IDs before dispatch, supporting a concrete distribution check.
Guarded dispatch design
The source-backed design pairs replacement checks and configured resource limits with structured execution evidence, cleanup reporting, and a separate manager to worker image identity comparison.