Kubernetes and Infrastructure¶
Managed Kubernetes with Istio service mesh is the current-lean runtime
pattern (DR-009, open — ENG-01 §4, ENG01-21), reflecting the parent
PE company's shared platform stack — leaning this way specifically for
the operational tooling/runbook/support transfer it would give a small
team if confirmed, not because DragonPay's current scale demands
Kubernetes on its own. Azure AKS specifically is contingent on the
still-open Cloud decision and may need relabeling, not re-deciding, once
that resolves (ENG-01 §4). The current-lean IaC tool (Terraform —
open, ENG-01 §4/Decision Register ENG02-25) provisions all of it
(ENG-01 §7, ENG01-08); manual
production infrastructure changes are exceptional and controlled, not
the normal path. Every other "Azure"/"Terraform"/"AKS" mention below is
the same illustrative naming, not a separate decision.
Cluster Architecture¶
ENG01-21's deliverable (cluster topology, namespace conventions,
resource/quota baselines), provisioned through the current-lean IaC tool
(Terraform — open, ENG-01 §4/Decision Register ENG02-25). Not yet
built.
Networking¶
Governed by ENG01-21 and STD01-03 (production/non-production as
fully separate infrastructure boundaries at the networking layer). Not
yet built.
Deployment Patterns¶
Governed by ENG01-22 (secrets via the current-lean HashiCorp Vault,
DR-011, open — injected at pod startup rather than fetched by
application code) and ENG01-10 (CI/CD-driven promotion). Not yet built.
Resource Management¶
Not yet defined. ENG01-15 calls for Azure budget/alert thresholds on
Terraform-provisioned resources for cost visibility, but per-workload
resource requests/limits and quota conventions for the AKS cluster
itself haven't been established yet — that's part of ENG01-21's still-
open scope.
Scaling¶
Not yet defined. STD01-13 sets the general principle that DragonPay doesn't target one global TPS/latency/availability number — SLOs are defined per customer-facing API and per external execution pattern, sized from actual expected CU/product volume, not assumed ahead of it. No concrete AKS autoscaling configuration exists yet.