Prisma Accelerate retires 1 December 2026: four developer cutoffs to fix by 8 December

Swap Prisma Accelerate URLs and Railway config files before 1 December, then move Composer users to VS Code and ARM32 jobs off hosted runners by 8 December.

Read time
17 min
Word count
2.6K
Sections
10
FAQs
8
Share
Engineer adding notes to a planning wall by a laptop, with migration planning text and Prisma, Railway, AWS and GitHub logos
On this page · 10 sections
  1. The four cutoffs at a glance
  2. Prisma Accelerate retires on 1 December
  3. Railway Config as Code: hard cutoff on 1 December
  4. AWS Infrastructure Composer console closes on 7 December
  5. GitHub Actions ends ARM32 on hosted runners by 8 December
  6. Worked example: a 25-engineer team hitting all four
  7. What to do if you see this
  8. Other developer deadlines before February
  9. How eCorpIT can help
  10. References

Summary. Four developer services change between 1 and 8 December 2026, and none of the vendor pages we checked on 21 November offers an extension. On 1 December Prisma retires Accelerate: apps that reach Prisma Postgres through a prisma+postgres://accelerate.prisma-data.net URL must switch to a pooled TCP connection or the Prisma Postgres serverless driver, and every cacheStrategy call has to go, because Accelerate's query cache has no replacement. The same day, Railway stops reading railway.json and railway.toml Config as Code files, a date its docs call a hard cutoff. The replacement is Infrastructure as Code in .railway/railway.ts, applied with railway config plan and railway config apply. On 7 December AWS shuts the standalone Infrastructure Composer console, which stopped taking new customers on 29 October; the same visual editor stays in the AWS Toolkit for Visual Studio Code. By 8 December GitHub-hosted runners end support for ARM32 jobs on Linux and Windows, because Node 24 does not support ARM32. Our verdict: Prisma needs the most testing if you relied on the cache, Railway needs a careful cutover in every environment, AWS is mostly a bookmark and training change, and GitHub can mean rebuilding an armv7 release pipeline.

About this article. For engineering managers, platform and DevOps engineers and startup CTOs who use any of these four services. On 21 November 2026 we checked every date, command, package name and limit against Prisma's changelog, migration guide and pricing page, the npm registry, Railway's documentation and changelog, AWS's Infrastructure Composer developer guide, and the GitHub Changelog and GitHub Docs. The effort figures in the worked example are planning assumptions, not measurements. Vendors sometimes move deadlines, so re-check each page before your change freeze; no vendor paid for inclusion. Researched with AI assistance and checked against the primary sources listed in the references.

GitHub has a fifth change in the same window: from 2 December, pull_request_target workflows are blocked by default in public repositories on GitHub Free, Team and Enterprise Cloud, which our pull_request_target guide covers.

The four cutoffs at a glance

Counted from 21 November, the first two dates are 10 days away, the AWS date 16 days and the GitHub date 17 days.

Date Service What stops Replacement Who must act
1 December 2026 Prisma Accelerate Hosted Accelerate connections, including Prisma Postgres URLs on accelerate.prisma-data.net, and Accelerate in front of your own database Pooled TCP with @prisma/adapter-pg, or the serverless driver with @prisma/adapter-ppg Any app using withAccelerate, @prisma/extension-accelerate or cacheStrategy
1 December 2026 Railway Config as Code Railway stops reading railway.json and railway.toml Infrastructure as Code in .railway/railway.ts (generally available), or .py and .go files (beta) Any service with a Config File path, or a config file at the service or repository root
7 December 2026 AWS Infrastructure Composer console The standalone console, plus the export options in the Lambda and Step Functions consoles Infrastructure Composer in the AWS Toolkit for Visual Studio Code Anyone who designs CloudFormation or SAM templates in the browser
By 8 December 2026 GitHub Actions ARM32 Support for ARM32 jobs on hosted Linux and Windows runners, including native AArch32 user-space execution Build for arm64, or produce armv7 artefacts another way Workflows that run 32-bit ARM containers or binaries on hosted runners

Prisma Accelerate retires on 1 December

What changes. Prisma's 2 October 2026 changelog says Accelerate is retired on 1 December 2026, including Prisma Postgres connections that go through it. The Console no longer offers Accelerate for new setups, and existing users get a Transition tab on the Accelerate page. Prisma's migration guide says only the connection changes: the database, data, workspace and plan stay as they are, and local Postgres URLs such as prisma+postgres://localhost:51213 are out of scope. Prisma repeated the date in its 4 November changelog, the latest entry when we checked.

The part that bites. Accelerate's query cache is not replaced. Both replacement paths keep Prisma Postgres connection pooling, neither keeps the cache, and Prisma Postgres does not currently offer replacement query caching. Every cached read becomes an ordinary database query, which can raise load and latency on hot paths. It can also push you up a plan, because Prisma Postgres is metered in operations: the Free plan includes 200,000 a month, Starter 1 million and Pro 10 million, and one query counts as one operation.

A second trap. Prisma's changelog warns that the latest tag on npm points at Prisma ORM 8, so Prisma ORM 7 users should change any script or CI job that installs prisma@latest to prisma@7. On 21 November the npm registry still listed a release candidate, 8.0.0-rc.34, under latest, with 7.10.1 under prev. Prisma's guide adds that ORM 8 does not use the hosted Accelerate integration, so the steps below apply to Prisma ORM 6.19 and 7.

Prisma checklist

  1. Find every use. Search source, build scripts and deployment config for the strings Prisma lists, and record every match:

            prisma+postgres://accelerate.prisma-data.net
withAccelerate
accelerateUrl
@prisma/extension-accelerate
@prisma/client/edge
cacheStrategy
withAccelerateInfo
$accelerate
--no-engine
--accelerate
--data-proxy
          
  1. Pick the replacement. A Node.js or Bun server with PostgreSQL TCP support uses pooled TCP and @prisma/adapter-pg; an edge or TCP-constrained runtime uses the serverless driver and @prisma/adapter-ppg; migrations, introspection and Prisma Studio use a direct TCP URL.
  1. Generate new credentials in the Prisma Console under Connect to your database. Prisma says not to hand-edit the hostname or credentials of an existing string. The pooled URL uses pooled.db.prisma.io and the direct URL db.prisma.io.
  1. Update the client. On Prisma ORM 7, uninstall @prisma/extension-accelerate, install @prisma/adapter-pg and pg (or @prisma/adapter-ppg and @prisma/ppg), point prisma.config.ts at the direct URL and construct PrismaClient with the adapter. On Prisma 6, first move to 6.19 (Prisma's commands pin 6.19.3). For pooled TCP, return to the engine-backed client by removing the client engine type setting and the --no-engine, --accelerate or --data-proxy flags; for the serverless driver, keep the engine-less client and swap the extension for the driver adapter.
  1. Remove caching calls. Delete cacheStrategy, .withAccelerateInfo(), $accelerate.invalidate() and $accelerate.invalidateAll(). Untyped code or casts can defer the failure to runtime, so repeat the search even if the build passes.
  1. Load-test the hot paths on a preview deployment, watching connections, timeouts, latency and errors. If a busy endpoint depended on the cache, add your own caching, such as Redis or HTTP caching, before production.
  1. Roll secrets everywhere. Update local, preview, production and CI, deploy, and confirm the runtime no longer receives an Accelerate URL.
  1. Keep the old credential briefly as a rollback path, then revoke it once production is stable.

If you use Accelerate in front of your own database, Prisma says to remove it before 1 December: connect to the database directly or, for PostgreSQL, move it to Prisma Postgres. One more sign you are affected: Prisma says P6000 errors reporting that a worker exceeded its resource limits come from inside Accelerate, and switching fixes them.

Railway Config as Code: hard cutoff on 1 December

What changes. Railway's docs mark Config as Code as deprecated. Existing railway.json and railway.toml files keep working for services that already use them until 1 December 2026, which Railway calls a hard cutoff, and new services cannot opt in. The replacement, Infrastructure as Code, defines a project or environment in one file: .railway/railway.ts is generally available, and the Python and Go versions are in beta.

Why it is not a rename. Config as Code was per service and read during each deploy. Infrastructure as Code is evaluated by the Railway CLI, which compares the file with the live environment, shows a plan and applies it only after confirmation. Railway does not read .railway/ during deploys, and railway config plan refuses a service that still has a Config File path set.

The silent failure. If a service has no stored Config File path but a railway.toml or railway.json sits at the service or repository root, the builder still reads it. That file stops being read when it leaves the deployed commit or on the cutoff date, whichever comes first. If its settings (build and start commands, health check, restart policy) are not in your IaC file by then, the next deploy goes out without them.

Railway checklist

  1. Upgrade the CLI, then run railway login and railway link. Railway's 16 October changelog says it fixed migrations that could omit build settings or drop services sharing a config file.
  1. Preview. railway config migrate prints the generated file without writing anything. In a monorepo it merges every Config as Code file into one .railway/railway.ts; in a multi-repo project, run it in each repository.
  1. Write the file. railway config migrate --apply writes .railway/railway.ts. It does not clear the Config File path and does not deploy.
  1. Install the SDK with npm install railway so the CLI and CI can evaluate the TypeScript file.
  1. Cut over. railway config migrate cutover clears the Config File path on each declared service, in the linked environment only. It does not change build settings, variables or volumes, and does not deploy. railway config migrate undo restores the old paths from .railway/.cac-migration.json.
  1. Plan, then apply. Read every line of railway config plan before railway config apply. Railway warns that cutover cannot delete a volume but apply can, if the generated file leaves one out.
  1. Only then delete the old file. If the path is still set and the file is gone from the deployed commit, the build reports that the config file was not found.
  1. Repeat per environment. A fork with its own path keeps it, and a new fork still auto-detects a root config file until the file is gone.
  1. Gate CI on drift. railway config plan --detailed-exit-code exits 0 when nothing would change and 2 when changes are pending. Railway's railwayapp/config GitHub Action plans on pull requests and applies the reviewed plan on merge, using a project token scoped to one environment.

            railway config migrate                 # preview only
railway config migrate --apply         # write .railway/railway.ts
railway config migrate cutover         # clear Config File paths in this environment
railway config plan                    # read every line, especially volumes
railway config apply                   # confirm and apply
          

AWS Infrastructure Composer console closes on 7 December

What changes. AWS's developer guide says it is ending support for Infrastructure Composer on 7 December 2026 and stopped accepting new customers from 29 October 2026. Existing customers can use the standalone console until 7 December, after which console.aws.amazon.com/composer will no longer be reachable.

What stays. The visual authoring experience remains in the AWS Toolkit for Visual Studio Code, with no changes planned. Existing CloudFormation stacks, deployments and workflows keep working, as do templates stored in Amazon S3 or on local disk. Composer has no customer-facing APIs, SDKs, CLI commands or service endpoints, so there is nothing to decommission in code.

What else moves. The Export to Infrastructure Composer button is being removed from the Lambda console, and the same option leaves the Actions menu on the Step Functions state machine details page. The Composer view embedded in the CloudFormation console is being replaced by a new visual editor built into that console. AWS says it will not publish a blog post; notices go through the documentation page and AWS Health.

AWS Composer checklist

  1. Find the users. Ask who designs templates in the browser, and check AWS Health in each account for the notice.
  1. Save work in progress. AWS says templates saved locally or in S3 are not affected. Our advice: export anything that exists only in a browser session to your repository now.
  1. Install the AWS Toolkit for VS Code, open a CloudFormation or SAM template and choose the Infrastructure Composer button in the upper-right corner of the editor. You can also right-click the file and choose Open with Infrastructure Composer, or search the Command Palette.
  1. Update runbooks and training. Replace console links and the Lambda and Step Functions export steps with the VS Code path.
  1. Check engineers without VS Code. Decide whether they install it for template work or use the CloudFormation console's new visual editor.

GitHub Actions ends ARM32 on hosted runners by 8 December

What changes. GitHub's 12 November 2026 changelog says its hosted runners will end support for ARM32 job execution on Windows and Linux by 8 December 2026, including native AArch32 user-space execution. GitHub ties this to the end of Node 20, because Node 24 does not support ARM32. On 23 September GitHub said Node 20 was no longer available on Actions runners: JavaScript actions now run on Node 24, which is incompatible with macOS 13.4 and earlier and does not officially support ARM32, and self-hosted runners on those systems are no longer supported.

Who is affected. GitHub's hosted arm64 labels are ubuntu-24.04-arm, ubuntu-22.04-arm, ubuntu-26.04-arm, windows-11-arm and windows-11-vs2026-arm. In our reading, the likely cases are container jobs using arm32v7 or armhf images, test steps that execute armv7 binaries, and Raspberry Pi or embedded Linux pipelines. The notice does not cover arm64 jobs, and it does not say whether ARM32 code emulated on x64 runners is affected.

GitHub ARM32 checklist

  1. Search your workflows for arm32v7, armhf, armv7, linux/arm/v7 and arm32 in .github/workflows/, container image names and build matrices.
  1. Drop ARM32 where you can. If your users have moved to 64-bit hardware, publish arm64 only.
  1. If you must ship armv7, change how you build it. Cross-compile on an x64 or arm64 runner, or build under emulation on an x64 runner. Because GitHub's notice speaks of ARM32 job execution in general, test the emulated path in a branch before 8 December.
  1. Consider a self-hosted fallback. A 64-bit self-hosted runner can drive armv7 tooling, but the runner itself must be on a supported operating system and architecture.
  1. Check actions on Node 24. The temporary opt-out is gone. GitHub asks action maintainers to set runs.using to node24 and publish a release, and workflow owners to move to action versions that support Node 24.
  1. Update release notes if you drop armv7 artefacts.

Our windows-11-arm runner guide covers the separate Visual Studio 2026 image change on Windows ARM runners.

Worked example: a 25-engineer team hitting all four

Take a 25-engineer SaaS company running a Next.js app on Railway with Prisma ORM 7 against Prisma Postgres through Accelerate, two Railway services configured with railway.toml, one platform engineer who sketches Lambda architectures in the Composer console, and an IoT agent that ships armv7 and arm64 builds from GitHub Actions.

Task Owner Planning estimate Finish by
Prisma: search, switch to the pg adapter, remove 14 cacheStrategy calls, add Redis for two hot endpoints, load-test Backend lead 1.5 days 26 November
Prisma: pin prisma@7 in CI and Dockerfiles Backend lead 1 hour 24 November
Railway: migrate, cut over, plan and apply in staging, then production; delete railway.toml afterwards Platform engineer 4 hours 25 November
Railway: add the railwayapp/config workflow with a project token Platform engineer 2 hours 27 November
AWS: export templates to Git, install the Toolkit, update two runbooks Platform engineer 2 hours 3 December
GitHub: replace the native armv7 test job with a cross-compile and an emulated smoke test; update action pins Firmware lead 1 day 4 December

The day-sized tasks add up to 2.5 days (1.5 plus 1) and the hour-sized ones to 9 hours (1 plus 4 plus 2 plus 2). At 8 working hours a day that is about 3.6 engineer-days, finished before a 5 December freeze. These are planning assumptions, not measurements. The biggest swing is the Prisma cache: a busy endpoint that leaned on cacheStrategy needs its own caching tested under load, which is where most of the 1.5 days goes.

What to do if you see this

The fixes below follow the Prisma and Railway guides cited above.

Symptom Likely cause Fix
Prisma P6000 errors about worker resource limits Limit hit inside Accelerate Switch to pooled TCP or the serverless driver now
CI suddenly installs a Prisma 8 release candidate prisma@latest points at Prisma ORM 8 Pin prisma@7 in scripts and Dockerfiles
railway config plan refuses a service Config File path still set Run railway config migrate cutover, then plan again
Railway build says the config file was not found File deleted before cutover Restore the file, or finish the IaC file and run cutover to clear the path
Railway deploy ignores your health check after 1 December Root railway.toml no longer read Copy the settings into .railway/railway.ts and apply

Other developer deadlines before February

  • January 2027: npm removes direct publishing of new package versions with a granular access token. See our npm token publishing guide.

How eCorpIT can help

eCorpIT is a Gurugram-based technology organisation, founded in 2021, assessed at CMMI Level 5 and MSME certified, with senior-led engineering teams working across AWS, Microsoft and Google platforms. Our application development services team can inventory these dependencies, move Prisma connections and caching, convert Railway services to Infrastructure as Code with CI gating, and rebuild pipelines for new runner architectures. Talk to us at /contact-us/.

Last updated: 21 November 2026.

References

  1. Changelog, 2 October 2026: Let your coding agent set up Prisma, and decide what needs your approval — Prisma
  1. Using Config as Code — Railway Docs
  1. Infrastructure as Code, including Migrating from Config as Code — Railway Docs
  1. AWS Infrastructure Composer end of support, Developer Guide — Amazon Web Services
  1. Actions for ARM32 Linux and Windows retirement, 12 November 2026 — GitHub Changelog
  1. pull_request_target blocked by default December 2, 2 November 2026 — GitHub Changelog
  1. Connect to Prisma Postgres without Accelerate — Prisma Documentation
  1. Changelog, 4 November 2026: Build and deploy a Prisma app from one ChatGPT prompt — Prisma
  1. Pricing: usage-based plans for Postgres and Compute — Prisma
  1. prisma package dist-tags, checked 21 November 2026 — npm registry
  1. Changelog #0312, 16 October 2026: Activity feed filters, guided IaC migration, cron jobs in the CLI — Railway
  1. railwayapp/config: GitHub Action for railway config plan and apply — Railway on GitHub
  1. Node 20 is no longer available in GitHub Actions, 23 September 2026 — GitHub Changelog
  1. GitHub-hosted runners reference — GitHub Docs
  1. Metadata syntax reference: runs for JavaScript actions — GitHub Docs
  1. Chroma Cloud wind-down timeline, 23 October 2026 — Chroma
  1. Upcoming deprecation of selected GitHub Copilot models in December, 13 November 2026 — GitHub Changelog
  1. About access tokens — npm Docs

Frequently asked

Quick answers.

01 When exactly does Prisma Accelerate stop working?
Prisma's changelog and migration guide give 1 December 2026 as the retirement date for hosted Accelerate, including Prisma Postgres connections through accelerate.prisma-data.net. Neither page gives a time of day or a grace period, so finish and test the switch before your last November deploy rather than on the morning of 1 December.
02 Is there a replacement for Prisma Accelerate query caching?
No. Prisma says both replacement paths keep Prisma Postgres connection pooling but neither keeps Accelerate's query cache, and Prisma Postgres does not currently offer one. Remove every cacheStrategy call, add your own caching layer, such as Redis or HTTP caching, for endpoints that depended on it, and load-test before production.
03 Do I need to move my data off Prisma Postgres?
No. Prisma's guide says the change affects only how your application connects; your database, data, workspace and plan stay the same. You generate a pooled URL for application traffic and a direct URL for migrations in the Prisma Console, update the client adapter and deploy. Local Postgres development URLs are not affected.
04 What happens to railway.json and railway.toml after 1 December?
Railway's docs call 1 December 2026 a hard cutoff: existing Config as Code files stop being read. Services must then be managed through Infrastructure as Code in a .railway/railway.ts file. If a root config file was supplying build or start settings and you have not copied them into that file, the next deploy goes out without them.
05 Will deleting railway.toml break my Railway deploy?
It can if you delete it too early. Railway says that if a service still has a Config File path set and the file is missing from the deployed commit, the build reports that the config file was not found. Run railway config migrate cutover first, check the plan, and only then push the deletion.
06 Do my CloudFormation stacks break when the Infrastructure Composer console closes?
No. AWS says existing CloudFormation stacks, deployments and workflows keep working, and templates stored locally or in Amazon S3 are unaffected. Only the standalone console at console.aws.amazon.com/composer closes on 7 December 2026. The same visual editor remains in the AWS Toolkit for Visual Studio Code, with no changes planned.
07 Which GitHub Actions jobs does the ARM32 retirement affect?
Jobs that execute 32-bit ARM code on GitHub-hosted Linux or Windows runners, such as arm32v7 container jobs or armv7 test binaries, lose support by 8 December 2026. Arm64 jobs are not part of the notice. Self-hosted ARM32 runners lost support in September, when Actions moved JavaScript actions to Node 24.
08 Can these deadlines move?
Vendors sometimes extend deadlines, but none of the four pages we checked on 21 November 2026 offered an extension or opt-out. Railway calls its date a hard cutoff, Prisma has repeated its date in every changelog entry since October, and AWS says the console URL will stop working. Plan to the published dates and re-check before your freeze.

About the author

Manu Shukla

Founder & Director

Founder of eCorpIT. Hands-on engineer leading senior-only delivery for AI apps, custom software, and cloud systems for global clients.

Subscribe

One engineering note a week. No fluff, no spam.

Senior-architect playbooks on AI agents, mobile apps, cloud, security, data, and marketing — delivered every Wednesday.

Past the reading

Read enough. Let's build something.

A senior architect responds in 24 working hours with scope, indicative cost, and a timeline. NDA before any technical conversation.

Loading search…

Browse all guides