Parent and architecture

• Implementation epic: #6
• Architecture: #1
• Cloud bootstrap: #12
• Depends on: #17, live resources from #12

Outcome

Make the low-traffic cloud-dev environment observable and prove that actionable failures notify the configured operator.

Estimate

• 3 SP / 2 engineering days
• Risk: medium

Scope

• Define alert thresholds and evaluation windows suitable for sparse, bursty development traffic.
• Create alerts for:
  • Cloudflare Worker exceptions/5xx and Yandex handoff failures;
  • Yandex Workflow failed executions;
  • API Gateway/control API 5xx and latency;
  • trigger errors, retry growth and YMQ DLQ messages;
  • Serverless Container errors, timeouts and concurrency saturation;
  • YDB throttling, RU/storage and transaction errors;
  • Object Storage capacity;
  • terminal Telegram delivery failures.
• Use the Yandex Monitoring API/console where the Terraform provider has no alert resource.
• Keep contact channels and private recipient data outside the public repository.
• Add a redacted alert inventory/runbook to the repository.
• Execute one controlled test alert and record delivery evidence privately.

Non-goals

• Changing business-flow retry semantics.
• Full production SLOs or paging policy.
• Payload logging or message-content dashboards.
• Automated cost shutdown; the existing folder budget remains notification-only.

Verification

• Every alert has an owner, metric/query, threshold, window and recovery condition.
• Alert IDs are recorded in the private deployment record.
• A controlled injected failure reaches the configured notification channel.
• Recovery clears the alert.
• Normal low-volume idle periods do not page.

Acceptance criteria

• Terraform apply is no longer treated as proof of monitoring readiness.
• The cloud-dev operator can distinguish ingress, workflow, control, queue, worker, database and delivery failures.
• Public documentation contains procedures and redacted identifiers only.
• No email address, token, Telegram identity or payload is committed or posted publicly.