Adopt multi‑Region Aurora, migrate Lambda to AL2023, and prepare for Kubernetes v1.37

What Happened

Several major cloud and platform updates were announced that require operational action and planning:

• Amazon Aurora DSQL expanded multi‑Region cluster support to four additional Regions (Europe — Stockholm, Europe — Spain, Asia Pacific — Mumbai, Asia Pacific — Singapore). Multi‑Region active‑active clusters are now available across 16 Regions; single‑Region clusters in additional Regions are also listed. AWS provides an entry path via the Free Tier and documentation for getting started [1].
• Kubernetes v1.37 (sneak peek) is planned for release (details subject to change). Notable changes: deprecation of kubectl run –filename/-f; Kubelet static Pods can no longer reference Secrets/ConfigMaps; ipvs mode in kube‑proxy is deprecated with disable-by-default planned by v1.40 and removal by v1.43; cgroup v1 is being phased out (failCgroupV1 defaulted true since v1.35); SELinux volume relabeling (SELinuxMount) graduates to GA; metrics.k8s.io goes GA; Kubelet UserNS (rootless) graduates to Beta; Volume Health Monitor reset to Alpha with new CSI RPCs [2].
• Amazon RDS for Oracle now offers 1‑year and 3‑year Reserved Instances for R8i and M8i instance classes (up to ~53% savings vs On‑Demand). R8i/M8i provide CPU/memory‑bandwidth improvements vs prior Intel generation; RI benefits apply to Multi‑AZ and Single‑AZ and support size flexibility under BYOL within the instance family [3].
• AWS Lambda now supports Java 8, 11, and 17 on Amazon Linux 2023 (AL2023) as managed runtimes and container base images to enable migration from AL2; AL2 reached end of life and AL2 runtimes will receive critical/security patches only through a one‑year extension period. AWS recommends moving to Java 21 or 25 on AL2023 where practical [4].
• Amazon CloudWatch added fully managed Prometheus collectors so customers no longer need to self‑manage OpenTelemetry Collectors. Supported discovery includes EKS, ECS (via Cloud Map/DNS), EC2 direct scraping, MSK/OpenSearch endpoints; metrics are delivered in OpenTelemetry format and queryable with PromQL. Feature is available in most Regions (not New Zealand) and billed hourly plus standard OTLP ingestion charges [5].

Why It Matters to Businesses

These updates affect availability, cost, compliance and operational complexity across data platforms and runtime fleets:

• Global availability and disaster tolerance: Aurora DSQL multi‑Region adds active‑active writable endpoints, enabling stronger regional resilience for transactional workloads — but with higher cross‑region latency and egress cost considerations [1].
• Cost optimization and procurement: R8i/M8i Reserved Instances provide material discounts for Oracle workloads but require commitment and capacity planning; BYOL size flexibility mitigates some sizing risk [3].
• Runtime security and support: Lambda customers running on Amazon Linux 2 must migrate to AL2023 to maintain SLA and full support after AL2 EOL; the AL2023 Java 8/11/17 runtimes ease migration but upgrading to Java 21/25 is recommended for performance and features [4].
• Monitoring and observability ops overhead: Managed Prometheus collectors reduce agent maintenance and scaling overhead and enable unified PromQL queries across AWS and user metrics — helpful for teams that want faster time to insight, but it centralizes control with CloudWatch [5].
• Cluster compatibility and security posture: Kubernetes v1.37 deprecations and feature changes (cgroups, kube‑proxy ipvs, static pod API references) require planned cluster and node OS changes; ignoring them risks future disruption or insecure configurations [2].

Kimbodo Engineering Perspective

When building production systems we weigh availability, latency, cost, and operational complexity. Key engineering judgments from these updates:

• Adopt multi‑Region databases selectively. Use Aurora DSQL active‑active where application-level strong consistency across regions is business‑critical (financial ledgers, global order books). For read‑heavy global apps, prefer read‑replica or regional caches to avoid cross‑region write latency and egress costs [1].
• Reserve capacity only after validation. Purchase R8i/M8i RIs after load testing shows steady capacity needs; leverage size flexibility for BYOL Oracle deployments to reduce lock‑in risk, and stagger RI purchases to avoid synchronized renewals [3].
• Migrate Lambda early, but plan compatibility tests. Use AL2023 Java 8/11/17 runtimes for lift‑and‑shift migrations when you cannot immediately upgrade Java versions. Prefer upgrading to Java 21/25 where feasible to reduce future migrations and gain performance/security improvements [4].
• Embrace managed observability sensibly. Managed Prometheus collectors remove agent ops but trade off control. Use them for standard scraping and correlation while retaining a minimal self‑managed collector for special‑case, high‑security, or custom processing needs [5].
• Treat Kubernetes deprecations as a release train. Plan node OS and kube‑proxy migrations (cgroup v2 and iptables alternatives), remove static pod references to Secrets/ConfigMaps, and add test coverage for Kubelet rootless and Volume Health Monitor changes to avoid disruptive surprises during cluster upgrades [2].

How We Would Implement It

Aurora DSQL multi‑Region (when to deploy)

• Evaluate candidate databases by SLA and RPO/RTO; run cross‑region latency and throughput benchmarks using representative workloads.
• Start with a two‑Region active‑active proof of concept in non‑production using the Free Tier where applicable; validate conflict/latency behavior and failover drills [1].
• Architect clients for region affinity (prefer local writes when possible), add regional caches, and measure cross‑region egress to validate cost assumptions.

RDS R8i/M8i Reserved Instances

• Run performance tests on R8i/M8i to confirm the claimed throughput/memory bandwidth gains for your Oracle workload before committing to RIs [3].
• Stagger 1‑ and 3‑year RI purchases, reserve only steady baseline capacity, and use size flexibility under BYOL to handle instance size fluctuations.

Lambda AL2023 Java migration

• Create a migration branch: convert a small subset of functions to AL2023 managed runtimes or container images; run unit and integration tests, then canary traffic using traffic shifting in Lambda.
• If code relies on JVM internals or native libs, container images give more control; plan for eventual move to Java 21/25 and codify runtime expectations in CI pipelines [4].

CloudWatch managed Prometheus collectors

• Define scrape configs centrally; start by importing EKS metrics and a limited set of EC2 targets into the managed collector and verify PromQL queries and dashboards.
• Enforce IAM least privilege for the managed collector to access targets, and use VPC endpoints or private connectivity for scraping sensitive targets; monitor ingestion and hourly collector costs during trial runs [5].

Kubernetes v1.37 upgrades

• Run cluster upgrade rehearsals in staging: migrate nodes to OS images with cgroup v2, switch kube‑proxy to a supported mode if using ipvs, and update manifests to remove static pod references to Secrets/ConfigMaps.
• Add e2e tests for SELinux relabeling behavior, metrics.k8s.io collection, and Volume Health Monitor interactions with CSI drivers before upgrading production clusters [2].

Risks, Costs and Security

• Cross‑region latency and egress costs: Active‑active multi‑Region writes increase latency and inter‑region data transfer costs; budget and SLOs must reflect that [1].
• RI commitment risk: Reserved Instances lower compute cost but commit to capacity; misforecasting can waste budget or require On‑Demand fallback [3].
• Runtime support window: AL2 EOL forces migration planning; relying on the AL2023 Java 8/11/17 interim runtimes prolongs compatibility but plan to upgrade for long‑term security and feature support [4].
• Observability and data exposure: Managed collectors centralize metric ingestion. Ensure collectors use secure endpoints, restrict scrape targets with IAM, encrypt in transit, and audit access to avoid exposing internal telemetry [5].
• Upgrade-induced outages: Kubernetes deprecations (cgroup changes, kube‑proxy ipvs removal, static pod API reference restrictions) can break workloads during upgrades; exhaustive staging tests and phased rollouts are required to limit blast radius [2].
• Compliance and licensing: Moving Oracle under BYOL with RIs retains licensing responsibilities; verify license mobility and compliance before capacity purchases [3].

Sources

1. [1] Amazon Aurora DSQL adds multi-Region cluster support in four more Regions
2. [2] Kubernetes v1.37 Sneak Peek
3. [3] Amazon RDS for Oracle now offers Reserved Instances for R8i and M8i instances
4. [4] AWS Lambda now supports Java 8, 11, and 17 on Amazon Linux 2023
5. [5] Amazon CloudWatch announces managed Prometheus collectors