AWS Lambda — Node.js 18 Runtime
What to do: Migrate functions to Node.js 22 — this one is already past, so patch and update support has stopped.
The end-of-life and deprecation dates that break cloud-native stacks — curated, prioritized, with what to do and why it matters for your SOC 2 / ISO 27001. The ones that hit AWS & serverless teams first.
Last updated: October 13, 2026 · Verified against vendor sources.
What to do: Migrate functions to Node.js 22 — this one is already past, so patch and update support has stopped.
What to do: Already past — kernel patches have stopped. Move EKS node groups to AL2023 / Bottlerocket and move Lambda to AL2023-based runtimes now.
What to do: The AL2-based branches (Go, Java Corretto 8/11/17, Tomcat, Docker/ECS, .NET Core on AL2) are already retired — no more managed platform updates or security patches. Rebuild affected environments on the AL2023 equivalent now.
What to do: Already past — creating new Python UDFs is blocked and existing ones stopped executing. If any of your ETL still calls them, migrate to Lambda UDFs now; a broken pipeline step is worse for an audit trail than the migration itself.
What to do: Community EOL passed Jul 6, 2026 — plan the upgrade to MariaDB 10.11 (LTS) now. RDS end of standard support for 10.6 is Dec 31, 2026, and RDS for MariaDB has no Extended Support to buy time: after the date it's a hard cutoff, no more patches.
What to do: Maintenance mode has been in effect since Jul 15, 2026 — AWS vendors botocore/s3transfer directly into CLI v1, so pip install -U botocore no longer changes what CLI v1 actually uses; only a new CLI v1 release does. If your CVE-patching routine relies on that trick, it stops working. Still worth planning the real move to CLI v2 before v1's end-of-support Jul 15, 2027 (critical fixes only after that, no new services/regions).
What to do: Standard support ended Jul 29 — clusters still on 1.33 are now auto-enrolled in Extended Support at 6x the per-cluster-hour cost ($0.10 → $0.60), accruing now. Upgrade the control plane and node groups to 1.34+ to stop the charge.
What to do: Since Jul 30, these are closed to new customers — Bedrock Agents Classic, Amazon Kendra, Amazon Q Business, 11 SageMaker AI features (Clarify, Debugger, Ground Truth, Model Monitor and more), Cognito Sync, Simple AD. Existing usage keeps working; just don't build anything new on them, and plan a migration path if one is core to your product.
What to do: These branches (all on Amazon Linux 2023) retired Jul 31 — you're now getting no managed platform updates or security patches. Check the EB console → Configuration → Platform and rebuild any affected environment on a currently supported branch. Beanstalk is exactly where old runtimes get forgotten.
What to do: Standard support has ended — any 8.0 instance still running was auto-enrolled into paid Extended Support on Aug 1, 2026, so per-vCPU-hour charges are on your bill right now. Plan and test the upgrade to MySQL 8.4. Extended Support steps up to higher year-3 pricing Aug 1, 2028 and ends entirely Jul 31, 2029.
What to do: The OS-only provided.al2 runtime (Go, Rust, any custom runtime on Amazon Linux 2) is deprecated — the AL2 blast radius reaching the runtimes people forget are AL2 underneath. Rebuild on provided.al2023. Create is blocked from Jul 29, 2027 and update from Aug 31, 2027. Security patches have stopped, and container-image functions get no deprecation email at all.
What to do: New PCoIP WorkSpaces signups already closed (Jul 31, 2026); full end of support follows Oct 31, 2027 (HP is retiring the underlying PCoIP/HP Anyware tech). Migrate to the DCV protocol — no extra cost, and existing OS/apps/data carry over via Modify Protocol.
What to do: New WorkSpaces Pools signups already closed (Jul 31, 2026); full end of support follows Dec 31, 2027. Migrate to WorkSpaces Applications — it can import your existing pool image directly.
What to do: Standard support has ended — any 10.5 instance is now unpatched. Upgrade to 10.11 (LTS) now. Unlike MySQL/PostgreSQL, RDS for MariaDB has no Extended Support — there's no paid safety net: EOL here is a hard cutoff, not a billing event.
What to do: Nothing urgent if you already use it — Detect closed to new customers Aug 31, but existing users keep it (no end-of-support date). The signal: don't architect a new IoT fleet's "continuous monitoring" story around a feature you can no longer switch on.
What to do: End of support has passed — MSK can now auto-upgrade your 3.7.x clusters to the recommended version at any time, with no notification. Upgrade to 3.9.x on your own terms immediately and own the timing before AWS does; an unplanned broker bump is a change you can't tie to a change ticket.
Status field Governance / APIWhat to do: Past — the old Status field has left the Organizations APIs (DescribeAccount, ListAccounts, ListAccountsForParent). If any code, IaC or aws_organizations_* Terraform resource still reads Status, switch it to the new State field now. Silent breaking change: account-vending, offboarding and compliance automation reading Status can already be returning wrong results with no error — audit for it immediately.
What to do: The whole service shuts down Sep 30 — migrate to Amazon VPC Lattice (EKS / cross-VPC), ECS Service Connect (ECS) or direct ALB routing. Not a version bump: after the date the managed Envoy sidecars stop getting updates, leaving an unpatched data plane. Removing sidecars is an architectural migration — plan it.
What to do: The platform retired Sep 30 — move Windows environments to Server 2019 / 2022 / 2025 with IIS 10. The hard wall is next: default AMIs go inaccessible Jan 15, 2027 — after that, any auto-scaling event, instance replacement or deploy that launches a new instance fails. That's an outage, not just missing patches.
What to do: Console and API access end Oct 7 — your deployed infrastructure keeps running, but you lose Proton's templating/management layer. Migrate service templates to plain IaC (CDK or Terraform) before then.
What to do: Console and V1 resources go away Oct 7 — migrate edge fleets to Greengrass V2 before then. V1 devices keep running but get no more security updates; an unplanned edge-fleet migration is painful, so start the rollout early.
What to do: Console and resources go away Oct 7 (new customers already blocked since Oct 7, 2025). Export your datasets and move analytics off FinSpace before then — after the date you lose access to the environment and its managed kdb/Managed kdb Insights workloads.
What to do: Console and resources go away Oct 7 (new customers already blocked since Oct 7, 2025). AWS points predictive-maintenance workloads to AWS IoT SiteWise — export your models/data and re-platform anomaly detection before the date.
What to do: Console and resources (endpoints, segments, campaigns, journeys, analytics) go away Oct 30. SMS/voice/push/OTP APIs aren't affected — those already run on AWS End User Messaging. Migrate campaign/journey logic before then.
What to do: Move python3.10 functions to python3.12 or python3.13 — it's the last Amazon-Linux-2-based Python runtime. Patches stop Oct 31; create is blocked from Jul 29, 2027 and update from Aug 31, 2027.
What to do: Both .NET runtimes deprecate the same day — plan the move to .NET 10 (the LTS successor). For dotnet8 AWS blocks create from Jul 29, 2027 and update from Aug 31, 2027. Note .NET 9 is container-only with no scheduled create/update block: you're fully on your own after Nov 10.
What to do: The natural upgrade target for anyone who just left 1.33 — plan the jump once, not twice. After Dec 2 it enters (paid) Extended Support.
What to do: AWS is retiring Amazon WorkMail — closed to new customers since Apr 30, 2026, full end of support Mar 31, 2027. This is the rare one where the date does delete data: after Mar 31, 2027 the console and all mailbox content (emails, contacts, calendars, attachments) become permanently inaccessible. Export mailboxes and migrate to a third-party mail provider (Zoho Mail, Zoom Mail, Kopano) well before the cutoff.
What to do: In maintenance mode since Feb 25, 2026 (critical security fixes only); end of support Feb 25, 2027 — no more updates for any X-Ray SDK, the daemon or the Lambda X-Ray layers. The X-Ray service itself stays. Find aws-xray-sdk* in your dependencies and move instrumentation to OpenTelemetry (ADOT or upstream OTel) sending to X-Ray / CloudWatch.
What to do: Plan ahead — the current Lambda target itself sunsets here, with Lambda deprecation phases following after.
What to do: AWS is retiring re:Post Private — closed to new customers, full decommission Jun 30, 2027. Another one where the date deletes data: all private spaces (questions, answers, knowledge articles) are permanently deleted, no read-only access after. Export via your TAM (final egress deadline Jun 15, 2027) and move to Stack Overflow for Teams / Stack Internal or AWS Builder Center.
What to do: AMS Advanced ends Jun 30, 2027 — no new landing zones after Jun 30, 2026. Your workloads and AWS accounts are not deleted, but the whole managed layer (RFC/change management, AMS patching, monitoring, incident + access management, landing-zone controls) goes away. If a SOC 2 / ISO 27001 control leans on AMS-run patching or change management, you need to re-home it — plan the move to AMS Accelerate, Unified Operations, a partner, or self-managed (target Mar 31, 2027).
What to do: Migrate these deprecated runtimes to a current version (Node.js 22, Python 3.12/3.13, Ruby 3.4). Corrected again (Oct 2026): AWS pushed the blocks a second time — create blocked from Jul 29, 2027, update blocked from Aug 31, 2027 (the Feb 1 / Mar 3, 2027 dates no longer apply). More runway, but security patches already stopped, so migrate anyway.
What to do: Runway extended from Feb 2027 to Jun 2029 — more time, but Extended Support still bills per vCPU, so plan the upgrade.
What to do: Find every caller of amazon.nova-canvas-v1:0, amazon.nova-reel-v1:0 and amazon.nova-reel-v1:1 (code, IaC, IAM, CloudTrail InvokeModel / StartAsyncInvoke) and move to a supported image or video model before Sep 30. After EOL the requests fail — an outage for any feature that generates images or video. AWS lists these models in us-east-1, eu-west-1 and ap-northeast-1 only.
What to do: Find every place anthropic.claude-sonnet-4-20250514-v1:0 is pinned (app config, prompts-as-code, Provisioned Throughput) and move to a current Claude model. Re-run your evals — a newer model is a behaviour change, not a drop-in swap. After Oct 14 the requests fail.
What to do: Two dates, not one. Nov 12, 2026 is the community end of life (upstream fixes stop). RDS and Aurora end standard support on Feb 28, 2027; paid Extended Support starts Mar 1, 2027 (per-vCPU-hour surcharge). Plan the major upgrade this quarter, with a test-restore dry run first — major upgrades are never automatic.
What to do: 1.31 is already in the paid Extended Support tier; on Nov 26 even that ends and the control plane has no patch path. Upgrade control plane and node groups now, and jump to a version with real runway rather than stepping through 1.32.
What to do: AI21 Jamba 1.5 Large and Mini end Nov 26; TwelveLabs Marengo Embed 2.7 ends Nov 30. If you use one, plan the swap now — embedding models in particular mean re-indexing. The Bedrock model lifecycle page lists every Legacy model with its EOL date.
What to do: Scripts that download the agent from the public URL break Oct 29 (end of support follows Sep 29, 2027). Agent-format backups can't be restored by the alternatives — restore anything under retention with the agent while it's supported, then move to SAP ASE 16.1 native S3 backup (syb_aws).
What to do: The standalone console and the "Export to Infrastructure Composer" buttons in Lambda and Step Functions go away Dec 7. Stacks and templates are unaffected. Switch to Composer in AWS Toolkit for VS Code; save projects locally or to S3.
What to do: Developer and Business must move to Business Support+ themselves (from $29/month per account) before Jan 1, 2027 — no automatic transition. Enterprise On-Ramp is upgraded to Enterprise automatically during 2026.
What to do: Closed to new customers Oct 29. Remove AWS::DevOpsGuru::ResourceCollection / NotificationChannel from IaC — they'll fail stack create/update once withdrawn. Export insights; move to CloudWatch anomaly detection + DevOps Agent.
What to do: Closed to new customers Oct 29; public blockchain access, Query and Hyperledger Fabric networks end Sep 29, 2027. Plan self-managed nodes or an external RPC/data provider.
What to do: Maintenance mode — no new customers from Oct 29, no end date yet. Existing call flows keep working; don't build new PSTN logic here. AWS points to Amazon Connect.
What to do: Maintenance mode — no new customers from Oct 29, no end date yet. Don't build new access patterns on it; alternatives are WorkSpaces Applications with Chrome or an enterprise browser.
What to do: Support ended Sep 29, 2026. Remove remaining integrations and credentials.
What to do: Applies to Elasticsearch 1.5–2.3, 5.1–5.5, 6.0–6.7, 7.1–7.8 and OpenSearch 1.0–1.2, 2.3–2.9. Their extended support was extended to Nov 7, 2027 — but from Nov 7, 2026 the surcharge equals your instance price, so the instance bill roughly doubles. Check every domain's EngineVersion and upgrade to OpenSearch 2.19 (or at least 1.3 / 2.11+, standard support to Nov 7, 2027).
What to do: From Feb 1, 2027 any Redis OSS v6 cache is enrolled in paid Extended Support automatically (to Jan 31, 2030). Move to Valkey or Redis OSS 7 before then. v4/v5 are already in Extended Support since Feb 1, 2026 — their price tier steps up Feb 1, 2027.
What to do: The next batch of Amazon-Linux-2-based runtimes: java8.al2, java11, java17 and python3.11 deprecate Jun 30, 2027 (create blocked Jul 29, update Aug 31, 2027). Move Java to the .al2023 variants or java21/java25, Python to 3.12+. Also: ruby3.3 deprecates Mar 31, 2027.
No matches. Try a different keyword, or reset the filter to All.
Curated for AWS / serverless / cloud-native teams that must stay compliant. This is a focused radar, not an exhaustive list — dates can change, so always confirm on the vendor source before acting. SunsetSignal is independent and not affiliated with AWS.
The radar is updated weekly. The newsletter brings the new ones straight to you — ranked, with what to do and why it matters for compliance. Free, 5 minutes.
Free forever · 5 min/week · No spam, unsubscribe anytime.