Telemetry is only useful if operators can find and compare it during pressure.

Naming conventions reduce the translation tax between teams, dashboards, alerts, and incident notes.

Log Fields

Useful common fields:

timestamp
level
service
environment
cluster
namespace
pod
trace_id
request_id
user_safe_error_code

Do not put secrets or personal data in logs. Redaction should be a platform expectation, not an afterthought.

Metric Names

Metric names should describe the measured thing and unit.

Examples:

http_request_duration_seconds
http_requests_total
queue_depth
volume_attach_failures_total

Labels should support aggregation without exploding cardinality.

Avoid labels for unbounded values such as user IDs, raw paths, request IDs, or pod UIDs unless there is a deliberate high-cardinality system for them.

Ownership Labels

Telemetry should carry ownership context:

  • service.
  • team.
  • environment.
  • region.
  • cluster.

This makes alerts routable and dashboards filterable.

Acceptance Criteria

  • Logs can be searched by service, environment, and request correlation ID.
  • Metric units are clear.
  • Labels do not create uncontrolled cardinality.
  • Sensitive data is not logged.
  • Alert labels route to the right owner.