Observability Platform · View 19 of 25 · Operations

Deployment Architecture

Three regions that collect and store locally, and one account deliberately outside the failure domain.

Editable source SVG draw.io All views

Each region collects and stores locally and replicates only a reduced set — aggregates and SLO series — so a region loss costs detail for that region and nothing else.
The self-telemetry stack runs in a separate account and region on managed services, so it shares no EKS control plane, no account limit and no scaling event with the platform it watches.
Ingest is reachable only through PrivateLink from workload accounts. There is no public ingest path for in-house telemetry.

The tail sampler gets its own memory-optimised node group sized at 5× burst headroom; the ruler gets its own node group with no shared capacity with query.
MSK spans three AZs with 72 h retention — the single most consequential capacity number in the deployment.

The self-telemetry account is a second bill for a stack that does almost nothing on a good day, and is the most likely thing to be cut. Without it the circular dependency in view 01 is unanswered.
Cross-region query of the full corpus is deliberately not offered. An investigation spanning regions gets aggregates from the far region and must be opened there for detail.