Project
AWS's managed Kubernetes service, used as a control plane for containerised workloads on AWS infrastructure.
One revoked personal OAuth token stopped runs across 276 Terraform Enterprise workspaces, according to a dev.to write-up. Nobody was paged and the team needed two days to notice, so it moved the workspaces onto a GitHub App owned by the org.
Reality
Kubernetes 1.34 loses upstream patches on October 27, but AKS stops patching it on November 30 and EKS starts billing it at six times the rate on December 2. Teams need an upgrade plan for each provider, because the managed dates decide what a 1.34 cluster costs and whether it still gets fixes.
Reality
AWS now lets SageMaker Studio users launch JupyterLab and Code Editor Spaces on HyperPod EKS clusters from the browser, with no kubectl step. On a shared GPU cluster, who may launch them comes down to three managed IAM policies and a per-user identity setting that older Studio domains must switch on.
Reality
Amazon EKS now lets customers set selected scheduler, controller-manager and API-server parameters, including MostAllocated bin-packing, through its API. Teams that self-host only for such tuning now need a parameter AWS has not exposed.
Reality
One Karpenter operator wrote about 250 lines of plain client-go to flag pods that will not come back when a tainted node goes away. Doing it without controller-runtime or Kubebuilder put every piece of the watch, queue and reconcile loop on the page.
Reality
EKS Pod Identity, launched at re:Invent 2023, replaces IRSA's per-cluster OIDC trust with one pods.eks.amazonaws.com principal on every IAM role. A dev.to migration guide has each role trust both paths during cutover, so teams can move and verify one workload at a time.
Reality
AWS says running DeepEP over its EFA network on EKS gives mixture-of-experts reinforcement learning 40% more throughput. Whether that reaches another cluster depends on how much of each training step goes to expert traffic between nodes.
Reality
Lorin Hochstein rates Buildkite's public write-up above GitHub's on detail, and the first thing his own retelling has to do is teach clusters, nodes and headroom, the layer a platform keeps out of sight.
Reality
The spec's three JavaScript APIs handle codecs, encoding and transport once two peers have found each other. Finding them takes a signaling service, STUN and TURN, and all three are yours to run and budget for.
Reality
Apache Storm ships no maintained Helm chart and no migration guide, so one team wrote its own templates for EKS. The first thing that broke was a StorageClass name, and the only symptom was an init container log line.
Reality
PDFik dropped --no-sandbox from its Playwright workers and kept rendering customer-submitted HTML in non-root, capability-stripped pods. Its own security auditor later wrote that the flag was mandatory.
Reality
Amazon's new EKS managed addon reads Prometheus metrics from every model pod and sends each request to the one with room in its KV cache. The advertised 82% cut in first-token latency rests on a single 4.4-second baseline.
Reality
A dev.to writeup makes the case for trying Bedrock Agents or AgentCore Runtime before containers, and for putting real limits on the gateway, because a looping agent spends money without ever raising an error.
Reality
AWS now says it takes full operational responsibility for Elastic Beanstalk production environments for the life of the application. Underneath, Cluster Mode puts multiple apps on one shared Amazon EKS cluster.
Reality
AWS's NVRx walkthrough puts synchronous checkpointing ahead of GPU faults as the source of idle time on its FSDP jobs, and pairs a background save with a restart that re-enters training without cycling the container.
Reality
A dev.to walkthrough of the provisioning loop shows Karpenter batching pending pods for up to ten seconds, packing them in memory against their own scheduling constraints, then letting EC2 Fleet price the candidates.
Reality
Cilium 1.19 puts routing, mTLS and observability in one eBPF program per node, with a single Envoy per node for Layer 7. The 100 GB of RAM the post says that frees assumes 1,000 pods at the top of the sidecar range.
Reality
The environment and techteam tags that drive cost allocation also drive Prometheus EC2 service discovery, so an add-only tagging run in ap-south-1 enrolled Kubernetes nodes in a scrape job for a host agent they do not run.
Reality
An audit of a post-Series A fintech serving 35 requests a second puts one observability line at $3,800 a month, more than five times the two bare-metal machines that replaced the whole AWS cluster.
Reality
A k3s workload moved to Amazon EKS without an application code change, and the three things that had to be fixed were the control plane endpoint policy, the instance type AWS would allow, and the credential doing the work.
Reality