--- title: OpenAI hits fourth outage in four days, MCP goes... description: OpenAI suffered its fourth service disruption in four days, with 503 errors hitting APIs, ChatGPT, and Codex simultaneously — a pattern that matters more now... canonical: https://daily.dev/posts/openai-hits-fourth-outage-in-four-days-mcp-goes-stateless-in-kubernetes-ttj5nvdvf twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: OpenAI hits fourth outage in four days, MCP goes stateless in Kubernetes | daily.dev og:description: OpenAI suffered its fourth service disruption in four days, with 503 errors hitting APIs, ChatGPT, and Codex simultaneously — a pattern that matters more now... og:url: https://daily.dev/posts/openai-hits-fourth-outage-in-four-days-mcp-goes-stateless-in-kubernetes-ttj5nvdvf og:image: https://api.daily.dev/og/posts/TTJ5nVDVF.png og:image:alt: OpenAI hits fourth outage in four days, MCP goes stateless in Kubernetes og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # OpenAI hits fourth outage in four days, MCP goes stateless in Kubernetes **[Backend Digest](https://daily.dev/sources/backend_digest)** · 5 min read · 1 upvotes · 0 comments ## Summary OpenAI suffered its fourth service disruption in four days, with 503 errors hitting APIs, ChatGPT, and Codex simultaneously — a pattern that matters more now that the platform has 10 million weekly agentic users. MCP's stateless HTTP mode solves the sticky-session problem that made horizontal scaling painful, and a real Kubernetes demo shows it working cleanly across replicas. Istio is retiring two container registries by late 2026, with 1.30 users most exposed. Zalando published how they cut load balancer costs from $450 to $110 per day by moving routing in-process at one million requests per second. ## Content **TLDR:** OpenAI suffered its fourth service disruption in four days, with 503 errors hitting APIs, ChatGPT, and Codex simultaneously — a pattern that matters more now that the platform has 10 million weekly agentic users. MCP's stateless HTTP mode solves the sticky-session problem that made horizontal scaling painful, and a real Kubernetes demo shows it working cleanly across replicas. Istio is retiring two container registries by late 2026, with 1.30 users most exposed. Zalando published how they cut load balancer costs from $450 to $110 per day by moving routing in-process at one million requests per second. --- ## OpenAI's fourth outage in four days hits APIs, ChatGPT, and Codex together Saturday's disruption was OpenAI's fourth in four days — 503 errors labeled `biscuit_baker_service_me_circuit_open` blocked requests from reaching servers entirely, with mitigation applied in about an hour. The reliability picture is harder to ignore now: OpenAI's status page has logged roughly 166 incidents over the past nine months, around 18 per month. The stakes have shifted too. When an outage hits a user mid-chat it's annoying; when it hits one of the platform's 10 million weekly agentic users mid-task, it cascades into failed workflows and broken pipelines. The pattern raises a real question about whether AI infrastructure is ready for the enterprise-grade reliability teams are starting to depend on. [Read more](https://daily.dev/posts/vWSR66D95) ## Stateless MCP enables true horizontal scaling across Kubernetes replicas The new stateless HTTP transport for MCP (streamable-http with stateless_http=True) fixes the fundamental scaling problem with the old SSE-based approach. A concrete Kubernetes test makes the difference clear: old MCP with multiple non-sticky replicas succeeded on only 2 of 6 sessions due to session_id mismatches; sticky sessions fixed correctness but pinned all traffic to one pod. Stateless mode hit 6/6 with load distributed evenly across both pods, no special routing config required. Any replica can serve any request, which is what you actually need for horizontal scaling to mean anything. [Read more](https://daily.dev/posts/7fJMNZm12) ## Zalando's in-process load balancer: $450 to $110/day at one million requests per second Zalando's Product Read API fans a single request into up to 100 parallel calls to individual product pods. Moving routing in-process rather than through their shared Skipper edge proxy let them shrink the Skipper fleet from 50+ pods to 8 and cut daily infrastructure costs from $450 to $110. Key engineering decisions: reimplementing Skipper's consistent hashing (xxHash64, 100 virtual nodes) to avoid cache splits during migration, a watch-based Kubernetes informer instead of polling, and a 30-second N-ring fade-in for new pod warm-up. AZ-aware routing was tried and abandoned — it fragmented caches and caused DynamoDB read explosions. They're explicit that this is only justified at genuine extremes. [Read more](https://daily.dev/posts/gWPrc272c) ## Propeller rebuilt survey processing on hybrid Kubernetes spanning on-prem GPUs and AWS EKS Propeller replaced a decade-old homegrown scheduler with a hybrid Kubernetes cluster connecting on-premise GPU servers in Sydney and Denver to AWS EKS overflow. A custom operator handles cross-zone scheduling via two custom resources, routing jobs on-prem first and failing over to AWS after time-based thresholds. Results are concrete: 100+ GPU nodes in production, 10,000+ concurrent pods at peak, 67,000+ daily processing steps, roughly 50% reduction in AlignSourcePhotos runtime, and AWS overflow costs of only ~$250/week versus tens of thousands for equivalent cloud GPUs. A second Denver region was stood up on the same platform with no from-scratch build. [Read more](https://daily.dev/posts/4inmn4EtO) --- ## Also notable - **Istio retiring gcr.io and registry.istio.io registries by late 2026, migrate to docker.io/istio:** Istio is shutting down both the gcr.io/istio-release and registry.istio.io/release registries by late 2026 — the proxy is being killed because Cloudflare egress IPs trigger rate-limiting on free container hosting platforms — and Istio 1.30 users are most exposed since that version defaults to registry.istio.io/release. [Read more](https://daily.dev/posts/SDQwaH5rH) - **Go HTTP client idle connection timeout longer than load balancer cutoff caused silent production timeouts:** A production incident traced intermittent pricing service timeouts to the Go HTTP transport's IdleConnTimeout exceeding the upstream load balancer's own cutoff — pooled connections were silently dropped then handed back as healthy — fixed by tightening IdleConnTimeout below the load balancer's threshold. [Read more](https://daily.dev/posts/q21dknunO) - **Kubernetes schema catalog covers 115 projects and ~9,000 schemas, cuts agent token usage 57%:** A public MCP server at schemas.fluxoperator.dev/mcp serves current CRD schemas for 115 Kubernetes ecosystem projects, reducing token usage by 57% and tool calls by 80% compared to web search when agents generate or validate manifests. [Read more](https://daily.dev/posts/FZ3QYl5qR) - **AutoMQ drops Kafka broker local disk dependency by offloading storage to S3:** AutoMQ is a drop-in Kafka replacement that offloads broker storage to object storage (S3 or S3-compatible), eliminating stateful brokers and cross-AZ replication costs while keeping the full Kafka protocol and client compatibility intact. [Read more](https://daily.dev/posts/zF0H70fEr) - **Podman migration on a six-service Compose stack: five services zero-touch, daemonless saves 180MB RSS on Linux:** A four-week evaluation found five of six services migrated to Podman without changes, rootless containers saved ~180MB idle RSS on Linux but added ~120ms latency to podman ps at scale, and the macOS laptop stayed mixed because daemonless benefits don't apply under a VM abstraction. [Read more](https://daily.dev/posts/da7QpUl9i) ## Similar posts on daily.dev - [MCP Is Going Stateless. Here’s How This Changes Things](https://daily.dev/posts/mcp-is-going-stateless-here-s-how-this-changes-things-7fjmnzm12) · Medium · 0 upvotes · 0 comments - [AAIF's MCP Dev Summit: Gateways, gRPC, and Observability Signal Protocol Hardening](https://daily.dev/posts/aaif-s-mcp-dev-summit-gateways-grpc-and-observability-signal-protocol-hardening-jnnoh4b7a) · InfoQ · 1 upvotes · 0 comments --- Tags: [#kubernetes](https://daily.dev/tags/kubernetes), [#openai](https://daily.dev/tags/openai), [#mcp](https://daily.dev/tags/mcp) [View this post on daily.dev](https://daily.dev/posts/openai-hits-fourth-outage-in-four-days-mcp-goes-stateless-in-kubernetes-ttj5nvdvf) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"DiscussionForumPosting","mainEntityOfPage":"https://daily.dev/posts/openai-hits-fourth-outage-in-four-days-mcp-goes-stateless-in-kubernetes-ttj5nvdvf","headline":"OpenAI hits fourth outage in four days, MCP goes stateless in Kubernetes","text":"OpenAI suffered its fourth service disruption in four days, with 503 errors hitting APIs, ChatGPT, and Codex simultaneously — a pattern that matters more now that the platform has 10 million weekly agentic users. MCP's stateless HTTP mode solves the sticky-session problem that made horizontal scaling painful, and a real Kubernetes demo shows it working cleanly across replicas. Istio is retiring two container registries by late 2026, with 1.30 users most exposed. Zalando published how they cut load balancer costs from $450 to $110 per day by moving routing in-process at one million requests per second.","url":"https://daily.dev/posts/openai-hits-fourth-outage-in-four-days-mcp-goes-stateless-in-kubernetes-ttj5nvdvf","datePublished":"2026-07-26T04:19:15.651Z","dateModified":"2026-09-13T21:29:58.551Z","author":{"@type":"Organization","name":"Backend Digest","logo":"https://media.daily.dev/image/upload/s--9uCSrsAb--/f_auto,q_auto/v1773839398/logos/backend_digest?_a=BAMAMiiu0","url":"https://daily.dev/sources/backend_digest"},"interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"isPartOf":{"@type":"WebPage","url":"https://daily.dev/sources/backend_digest","name":"Backend Digest"}} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Backend Digest","item":"https://daily.dev/sources/backend_digest"},{"@type":"ListItem","position":3,"name":"OpenAI hits fourth outage in four days, MCP goes stateless in Kubernetes"}]} ```