Enterprise Metadata Management System  ·  View 18 of 22  ·  Operations

Observability and Operations

What is measured, and which signals are allowed to wake somebody up.

Editable source SVG draw.io All views
Emit
Emit
Collect
Collect
Store
Store
Consume
Consume
Act
Act
Service health
Service health
OpenTelemetry SDK
OpenTelemetry SDK
OTel Collector
OTel Collector
Prometheus
Prometheus
SLO dashboard
99.9% read path
SLO dashboard...
Page on burn rate
Page on burn rate
Harvest health
Harvest health
Connector metrics
rows · duration
Connector metrics...
OTel Collector
OTel Collector
Prometheus
Prometheus
Freshness by source
age of last run
Freshness by source...
Retry, then quarantine
Retry, then quarantine
Projection lag
Projection lag
Change log offsets
Change log offsets
Lag exporter
Lag exporter
Prometheus
Prometheus
Index vs store lag
Index vs store lag
Show staleness banner
Show staleness banner
Metadata coverage
Metadata coverage
Nightly coverage job
Nightly coverage job
Batch export
Batch export
Aspect Store
Aspect Store
Domain scorecard
owned · described
Domain scorecard...
Steward campaign
Steward campaign
Access & audit
Access & audit
Audit events
Audit events
Log shipper
Log shipper
OpenSearch
7-year archive
OpenSearch...
Access review
quarterly
Access review...
Revoke and investigate
Revoke and investigate
Observability and Operations
Observability and Operations
Application we own
Application we own
Interface / broker
Interface / broker
Data store
Data store
Security / platform
Security / platform
Queue / topic
Queue / topic
Coverage is treated as a production signal, not a report: a domain whose ownership drops below target raises work, not just a number.
Coverage is treated as a production signal, not a report: a domain whose ownership drops below target raises work, not just a number.
v 1.0 · owner SRE · date 2026-08
v 1.0 · owner SRE · date 2026-08
Text is not SVG - cannot display

Decisions

  • Coverage is treated as a production signal, not a quarterly report. A domain whose ownership drops below target generates stewardship work automatically.
  • Projection lag is exposed to users as a staleness banner rather than hidden. A visible lag is a known state; a hidden one is a support ticket about wrong data.
  • Only service health and audit anomalies page. Harvest failures retry and quarantine first — a connector outage is not a 3 a.m. problem.

The four signals that matter

  • Read-path SLO burn rate against 99.9%.
  • Freshness per source: age of the last successful harvest.
  • Projection lag: change-log offset behind the aspect store, alert above 60 seconds.
  • Coverage and certification by domain — the adoption signal.

Left out

  • Runbooks, alert routing and on-call rota are operational documents, not architecture. The view names the act, not the procedure.