Datasets

中文：DataSet

Datasets describe telemetry or knowledge collections that can be attached to entities. They do not store runtime rows themselves; they describe the shape and query semantics of data held in storage systems.

Dataset Kinds

Kind	Use
metric_set	Time-series or metric-style signals.
log_set	Log records and structured log fields.
trace_set	Trace/span data.
event_set	Event streams or change records.
profile_set	Profiling data.
runbook_set	Operational playbooks or remediation knowledge.

MetricSet

A MetricSet defines labels and metrics. For example, a service metric set can describe request metrics keyed by a stable service identifier.

kind: metric_set
metadata:
  name: "devops.metric.devops.service"
  domain: devops
spec:
  labels:
    keys:
      - name: service_id
        type: string
  metrics:
    - name: request_count
      aggregator: sum

LogSet And TraceSet

LogSet and TraceSet definitions play the same modeling role for logs and traces:

• They name the dataset.
• They describe important fields.
• They can be linked to EntitySets.
• They can be routed to storage by StorageLink.

Dataset Binding

Datasets become useful when both links exist:

Design Rules

• Keep dataset names stable and semantic.
• Put entity-to-telemetry matching rules in data_link.fields_mapping.
• Put physical storage information in storage definitions and StorageLinks.
• Use one dataset when the labels, metrics, and storage behavior belong together.
• Split datasets when they have different storage, refresh, or ownership behavior.

Query Examples

List metric sets:

go run ./cmd/umctl --addr http://localhost:8080 query run demo ".umodel with(kind='metric_set') | sort name | limit 20"

Inspect datasets and storage links in the Web UI Explorer after importing a model pack that includes telemetry definitions.