Operations overview
Operations groups SAP health into 13 domains. Each domain answers a concrete operational question and combines current cards, series, breakdowns, tables, and thresholds defined by the canonical catalog.
The catalog also defines sections, order, and compositions. A composition brings related signals into one panel—for example, current state and trend, or per-instance and global variants—without changing their queries, units, or identity. Desktop uses a 12-column grid, tablet reorganizes it into 6 columns, and mobile stacks the panels.
| Domain | Main question |
|---|---|
| ICM / Gateway | Are inbound HTTP/RFC channels healthy? |
| Dispatcher | Where and how long do requests wait before processing? |
| Work Process | Are free processes scarce, is one type saturated, or are executions long? |
| Job Monitoring | Was the work completed, and where is batch impact concentrated? |
| Locks Enqueue | Is enqueue saturated, waiting on locks, or unavailable? |
| Performance | Are response time, CPU, memory, or DB degrading service? |
| Availability | Are instances available? |
| DB | Does the database show impact signals? |
| Filesystem | Is there host space risk? |
| Sessions | How are users and sessions distributed? |
| Web Services | Are monitored web services degraded? |
| Certificates | Are certificates expired, close to expiry, or showing relevant renewal changes? |
| RFC & Integrations | Are tRFC, aRFC, qRFC, or bgRFC failures affecting destinations and queues? |
Reading method
- Identify whether the domain is OK, WARN, or KO.
- If it shows
UNKNOWN, check collection before interpreting health. - Compare the current sample with the trend from the same
metricId. - Open Events for system messages in the same time window.
- Open Jobs when the domain can be affected by batch workload.
- Use Metrics for precise time-series analysis and AI for a guided summary.
When one signal in a composed panel is UNKNOWN or returns an error, the other
valid signals remain visible and the panel reports a partial result. SPOT does
not invent zero values to complete the visualization.
Keeping context
Keep the same SAP source, time range, and host when moving between domains. If any global filter changes, re-check the header before comparing charts. Cards and trends come from the same canonical identifier: for a historical range, the card uses the latest fresh sample inside the window while the chart keeps all points. Freshness is measured against the selected end; if the final gap exceeds three collection cycles, the card shows UNKNOWN. For a window that reaches the present, the reference remains the current time.
For an active incident, start with 15 or 30 minutes and use refresh when you need to confirm whether the state is changing. For post-incident analysis, expand the range and look for recurrence: same hour, same host, same job, same queue, or same event.
For domains 02–05, Landing retains functional health metrics while Operations provides detailed diagnostics:
- 02 Dispatcher: queues and timing by server or instance.
- 03 Work Process: global capacity, composition by type, free processes, dumps, and long executions.
- 04 Job Monitoring: functional job outcome, per-server batch signals, runs, and events.
- 05 Locks Enqueue: utilization, requests, and lock wait time.
- 12 Certificates: inventory, expiry, changes, and fingerprint history.
- 13 RFC & Integrations: failures, affected queues or destinations, age, and recoveries.
Analysis: thresholds, ML, and forecast
The global Analysis menu starts disabled. Enable static thresholds, Expected range, Forecast, or ML score points independently when needed. The 0–100 range limits visible ML points, and the compact legend explains colors using the effective thresholds.
Composed panels preserve Analysis, Expand, and the variant selector when one is available. Metric Explorer does not apply these Operations compositions: it requests every metric as an individual time series.
For compositions with a selector, Operations requests only the active variant; changing it runs the corresponding query. Expected range and ML score points request a baseline only while either layer is visible, and Forecast is requested only when its layer is enabled.
Use these layers as supporting context. If a domain is OK but leaves its expected range, it may be an early deviation. If it is WARN/KO and also breaks the range, prioritize Events, Jobs, or Metrics. An Alerts rule may evaluate ML; visualization alone sends no notification. See Machine Learning.