Observability & Alerting

Observability in Axis is opt-in but batteries-included: turn it on and you get CloudWatch alarms, a dedicated alerting SNS topic, a subscriber adapter Lambda that posts to Slack or Teams, and an optional dashboard — all synth-pure and least-privilege.

What AtfObservability gives you

• Alarms — synth-pure metric-by-dimension alarms with per-category toggles and overridable defaults. Target "all" or a subset; relative p99 duration alarms; a 1-minute period floor; and an additionalAlarms escape hatch for anything bespoke.
• A dedicated alert topic — a single SNS topic for alerts (or reuse an existing topic’s ARN from SSM).
• A Slack/Teams adapter — one subscriber Lambda per configured provider, reading its webhook from an SSM SecureString at runtime (never in synth, env, or plaintext). It fails fast at synth if no provider is configured and degrades gracefully at runtime if the secret is missing.
• A dashboard — opt-in.

The ObservabilityStack

ObservabilityStack composes the construct and publishes only the alert topic ARN to SSM. Wire alarms against resources in other stacks by name (the DLQ-depth + lambda-error alarms in example-rest-full are the canonical pair).

A common first alert

A DLQ-depth alarm on your busiest queue plus a lambda-error alarm on your API handler catches the two failure modes that hurt most, for almost nothing.

Where to go next

• AtfObservability for the full toggle surface.
• The Alerting Adapter for how the runtime dispatch works.
• Budgets & Cost — budget alerts reuse the same alert topic.