Dashboards
Cadence provides Grafana dashboards for metrics across the cluster, and Cadence Web for inspecting individual workflows.
Grafana dashboards
The main Cadence repository includes an overview dashboard, one dashboard for each server service, and one for SDK workers. One way to use them is to start with Cadence-Server for overall cluster health, then open the service dashboard for the area that looks wrong. Each dashboard can be filtered by domain, and most also let you choose the latency percentile.
| Dashboard | Shows |
|---|---|
| Cadence-Server | Frontend requests by domain, API latency by operation, service restarts, shard movement, running server versions, and timer and transfer task processing |
| Cadence-Frontend | Requests, errors, and latency for each Frontend API |
| Cadence-History | API latency, decisions, caches, shard movement, task lag, and timer processing |
| Cadence Matching | API latency, task matching, and backlog per task list |
| Cadence-Persistence | Database requests by operation, domain, and shard, and errors and latency by operation |
| Cadence-Archival | Archival requests and workflow cleanup |
| Cadence-Client | Workflow, activity, and poller metrics from SDK workers, plus a few related server metrics |
The dashboards do not include alerts. Cluster monitoring suggests monitors and thresholds for the key panels.
The default Docker Compose setup starts Prometheus and Grafana with these dashboards loaded. For Kubernetes, see the Grafana Helm setup guide.
Requirements
The dashboards need:
- Cadence services exposing Prometheus metrics. See the setup steps in Cluster monitoring.
- Prometheus scraping every Cadence service. On Kubernetes, the Grafana Helm setup guide configures this with the chart's
serviceMonitor. - SDK workers also exporting metrics, for the Cadence-Client dashboard. Cluster monitoring covers the Go and Java SDKs.
Cadence Web
Cadence Web is the project's browser UI for domains, workflows, and task lists. When a Grafana dashboard shows a problem in a domain, Cadence Web can be used to find the affected workflows.
In Cadence Web you can:
- Search workflows in a domain, including query-based search when advanced visibility is enabled.
- Open a workflow to see its event history, pending activities, and payloads.
- Check which workers are polling a task list.
- Use Workflow Diagnostics, an opt-in feature, to list common problems found in a workflow's history.
Docker Compose includes Cadence Web, and the Helm chart can deploy it. See UX and UI for a list of its other features and the cadence-web README for opt-in features.