Engine operations dashboard — spec
One dashboard per business process (definition key), not per engine. Data
sources: the six probe signals (runtime) + two history queries. All counts are
totals over indexed queries — never row scans.
Row 1 — “is anything stuck?” (runtime, red/green)
| Panel |
Query |
Alert |
| Dead letters |
management/deadletter-jobs total |
> 0 → page on-call eng |
| Overdue timer jobs |
management/timer-jobs?dueBefore=now |
> 10 sustained 10 min → page on-call eng |
| Oldest open instance |
oldest startTime, runtime instances |
> designed lifetime × 1.5 → notify process owner |
Row 2 — “is the human side keeping up?” (runtime)
| Panel |
Query |
Alert |
| Pool depth per candidate group |
query/tasks unassigned per group |
> per-team SLA depth → ops lead |
| Oldest unclaimed item per pool |
same query, sort dueDate/createTime asc |
> claim SLA → ops lead |
| Overdue tasks |
query/tasks dueBefore=now |
> SLA budget → ops lead |
Row 3 — “how is the process performing?” (history, trends)
| Panel |
Query |
Alert |
| Applications per day (started/completed) |
historic instances by start/end date |
started rate −X% WoW → product owner |
| Median time-to-decision |
historic ACTINST: start → decision activity end |
creeping trend, review weekly |
| Outcome mix |
historic variable decision distribution |
auto-approve share shift after any DMN deploy (Phase 5 impact replay, continuous) |
| Per-step wait times |
ACTINST durations for user tasks |
identifies the bottleneck step for staffing |
Routing
- Row 1 → engineering on-call (these are system failures).
- Row 2 → ops leads (staffing/process, not code).
- Row 3 → product/process owner (weekly review, not paging).
Non-goals
CPU, heap, and HTTP latency belong to the platform dashboard. DB health belongs to
lesson 05’s runbook. This dashboard answers exactly one question: is the
business process moving?
