Docker Engine 29.10.0, published on 16 November 2026, comes with a warning at the top of its release notes: a potential breaking change for local registry connections. If you run a local registry on localhost:5000 for development or CI, and you have any insecure-registries setting in your daemon config, your docker push and docker pull commands may start failing after the upgrade.

This tutorial explains exactly when you’re affected, how to reproduce the problem, and three ways to fix it, plus the other changes in 29.10 that developers should know about.

What changed

Until now, Docker Engine always treated loopback addresses (127.0.0.0/8 and ::1/128) as insecure registries. That meant you could push to a plain-HTTP registry on localhost:5000 without any configuration.

From the 29.10.0 release notes:

If you explicitly configure insecure-registries in daemon.json or use --insecure-registry, Docker Engine no longer automatically treats loopback addresses (127.0.0.0/8 and ::1/128) as insecure registries.

Two related rules:

  • When insecure-registries is omitted, loopback registries remain insecure by default. Nothing changes for you.
  • Setting "insecure-registries": [] in daemon.json now disables all default insecure registries, including loopback.

The change comes from moby/moby#53868.

Are you affected?

Your daemon.json Local HTTP registry on localhost:5000 after 29.10
No insecure-registries key Still works
"insecure-registries": ["registry.internal:5000"] Breaks unless you add localhost:5000
"insecure-registries": [] Breaks (all defaults disabled)
"insecure-registries": ["localhost:5000", ...] Works

The middle two rows are the trap. Many teams added one internal registry to insecure-registries years ago and forgot about it. That single line now switches off the loopback default.

Step 1: Check your Docker version and settings

docker version --format '{{.Server.Version}}'
docker info | sed -n '/Insecure Registries/,/^ [A-Z]/p'

The second command prints the “Insecure Registries” section of docker info. Then look at the config file itself. On Linux it’s usually /etc/docker/daemon.json:

sudo cat /etc/docker/daemon.json

On Docker Desktop, open Settings → Docker Engine to see and edit the same JSON.

Step 2: Reproduce the failure

Start a throwaway local registry, tag an image and push it:

docker run -d -p 5000:5000 --name registry registry:3
docker pull alpine:latest
docker tag alpine:latest localhost:5000/alpine:test
docker push localhost:5000/alpine:test

If you’re affected, the push fails with an error about the registry not using HTTPS or about an untrusted certificate, because the engine now tries to talk TLS to a plain-HTTP registry. If it succeeds, your config doesn’t hit the new rule.

Fix 1: Add the local registry explicitly (quickest)

The release notes recommend adding only the local registry addresses you need to your existing list:

{
  "insecure-registries": [
    "registry.internal:5000",
    "localhost:5000",
    "127.0.0.1:5000"
  ]
}

Include both localhost:5000 and 127.0.0.1:5000 if your scripts use both forms. Then restart the daemon:

sudo systemctl restart docker
docker info | sed -n '/Insecure Registries/,/^ [A-Z]/p'
docker push localhost:5000/alpine:test

On Docker Desktop, click Apply & restart after editing the JSON.

Always validate the JSON before restarting. A syntax error stops the daemon from starting. Run python3 -m json.tool /etc/docker/daemon.json and fix any errors first.

Fix 2: Remove insecure-registries if you don’t need it

If the only reason you had the setting was an old registry you no longer use, delete the key. With insecure-registries omitted, loopback registries go back to being insecure by default, and you’ve removed an exception from your config at the same time.

{
  "log-driver": "json-file",
  "log-opts": { "max-size": "10m" }
}

Fix 3: Use TLS with a trusted certificate (best)

The release notes’ first recommendation is to “configure TLS with a trusted certificate”. For a local registry, you can generate a certificate and tell Docker to trust it. Docker reads registry CA certificates from /etc/docker/certs.d/<host:port>/ca.crt:

mkdir -p certs
openssl req -newkey rsa:4096 -nodes -sha256 \
  -keyout certs/domain.key -x509 -days 365 \
  -subj "/CN=localhost" \
  -addext "subjectAltName=DNS:localhost,IP:127.0.0.1" \
  -out certs/domain.crt

docker rm -f registry
docker run -d -p 5000:5000 --name registry \
  -v "$PWD/certs:/certs" \
  -e REGISTRY_HTTP_TLS_CERTIFICATE=/certs/domain.crt \
  -e REGISTRY_HTTP_TLS_KEY=/certs/domain.key \
  registry:3

sudo mkdir -p /etc/docker/certs.d/localhost:5000
sudo cp certs/domain.crt /etc/docker/certs.d/localhost:5000/ca.crt
docker push localhost:5000/alpine:test

This takes a few more minutes to set up, but it works the same way whatever your insecure-registries setting is, and it’s closer to how production registries behave.

CI pipelines: check them too

The most likely place to hit this is CI, where a job starts a registry service container and pushes to localhost:5000. Check:

  • Self-hosted runners with a custom daemon.json. Upgrading Docker on the runner image can break jobs overnight.
  • Docker-in-Docker setups that pass --insecure-registry to dockerd. Any explicit flag now disables the loopback default, so add the local address to the same flag list:
# Example (GitLab CI style): a dind service with an explicit flag list
services:
  docker:
    image: docker:29-dind
    command: ["--insecure-registry=registry.internal:5000", "--insecure-registry=localhost:5000"]

A quick smoke test at the start of the pipeline saves a lot of confusing errors later:

docker push localhost:5000/smoke:latest || {
  echo "Local registry push failed: check insecure-registries (Docker 29.10 change)"
  exit 1
}

Other changes in Docker 29.10

The release has more than the registry change:

  • Engine API spec moves to OpenAPI 3.2. The source is now api/openapi.yaml. A generated Swagger 2.0 spec is still published in api/swagger.yaml, and a Swagger 2.0 copy pinned to API v1.57 is included for the transition. Future API versions won’t get per-version Swagger 2.0 copies, so update code generators that depend on them.
  • API version 1.57.
  • Experimental jobs. One-shot and cron-scheduled workloads managed by the daemon, as an opt-in extension enabled with --enable-extension org.mobyproject.jobs.v0.
  • Embedded containerd configuration through files in /etc/docker/containerd.d/.
  • A restarting attribute on die events, so event consumers can tell when a container is about to restart.
  • Packages for AlmaLinux, Oracle Linux and Rocky Linux.
  • Security hardening: obsolete socket families are blocked in the default AppArmor and seccomp profiles.
  • Updates: BuildKit v0.35.0, Go 1.27.4, and runc v1.5.3 in static binaries.
  • Fixes for docker cp printing “Successfully copied” when the copy failed, and docker --version printing the CLI version instead of an “unknown command” error.

Takeaway

If your Docker daemon has any insecure-registries entry, Docker 29.10 stops treating localhost registries as insecure automatically. Before upgrading laptops, runners and build servers, search your configs for insecure-registries and --insecure-registry, then either add localhost:5000 explicitly, remove the setting, or, best of all, give your local registry a trusted TLS certificate.