Skip to content

Docker driver: no sandbox can start on Debian 13 with Debian's docker.io (26.1.5): "getent unable to find entry "1000:1000"" #5314

Closed

Description

@Guilospanck

User Story

I use OpenShell for running coding agents in sandboxes through osh (a terminal dashboard over the OpenShell CLI), on macOS and Linux, with the Docker and Podman drivers.
I directly encountered this when moving our Linux test machine from Ubuntu 24.04 to Debian 13 (trixie): every sandbox on the Docker driver fails to provision.
This affects anyone running the Docker driver on Debian 13 with Debian's own Docker package.

Problem Statement

On a Debian 13 host with Debian's docker.io 26.1.5, openshell sandbox create on a Docker-driver gateway never reaches Ready. Provisioning fails with BoundaryConfigWriteFailed ... getent unable to find entry "1000:1000" in passwd database (status 500 from Docker). It happens for every sandbox, every time, with any image we tried.

Impact / Why This Matters

When this happens, users cannot use the Docker driver at all on Debian 13 with the distribution's Docker package: no sandbox ever starts, and the error does not say that the Docker version is involved.
The workaround is to replace docker.io with Docker's official docker-ce (29.9.3 works), or to use the Podman driver.
This matters because docker.io is what apt install docker.io gives on Debian's current stable release, and what many setup guides use.

Observed on the same host and setup:

  • Debian 13 + docker.io 26.1.5: fails
  • Debian 13 + docker-ce 29.9.3: works
  • Debian 13 + Podman driver: works
  • Ubuntu 24.04 + its docker.io + Docker driver: works

Acceptance Criteria

  • On Debian 13 with docker.io 26.1.5, openshell sandbox create --from <image> --detach on a Docker-driver gateway reaches Ready.
  • Or, if that Docker version is not supported: the gateway or sandbox create fails up front with a clear message naming the Docker server version and the minimum supported one, instead of BoundaryConfigWriteFailed.
  • The supported Docker version range for the Docker driver is documented.

Reproduction Steps

  1. Debian 13 (trixie) host: sudo apt-get install -y docker.io (gives 26.1.5+dfsg1-9+deb13u1), and add the user to the docker group.
  2. Install OpenShell 0.1.7 and start a gateway with the Docker driver.
  3. openshell sandbox create --name repro --from node:24-trixie-slim --detach
  4. Observe that the sandbox enters the error phase while provisioning (log below).
  5. Replace Docker with docker-ce 29.9.3 from download.docker.com and repeat step 3: the sandbox reaches Ready.

Suggested UX (if applicable)

No response

Environment

  • OpenShell: 0.1.7
  • OS: Debian 13 (trixie), arm64, systemd (a privileged test container with systemd as PID 1, also expected on a plain Debian 13 host)
  • Deployment: local gateway (systemd user service), Docker driver
  • Docker: docker.io 26.1.5+dfsg1-9+deb13u1 (fails); docker-ce 5:29.9.3-1debian.13trixie (works)

Logs

$ openshell sandbox create --name repro --from claude-code --policy sandbox-policy.yaml --detach
[0.0s] Sandbox allocated
[0.0s] Pulling image claude-code
[0.0s] Image already present
[2.8s] Starting sandbox Container created
Error:   × sandbox entered error phase while provisioning: BoundaryConfigWriteFailed:
  │ stage Docker sandbox bundle: Docker responded with status code 500: getent
  │ unable to find entry "1000:1000" in passwd database

# The same Docker error without OpenShell, on that host (Docker 26.1.5),
# for a container whose user is uid:gid and a copy that keeps ownership:
$ id=$(docker create --user 1000:1000 claude-code true)
$ docker cp -a /tmp/f "$id":/tmp/f
Error response from daemon: getent unable to find entry "1000:1000" in passwd database
# --user 1000 and --user node work on 26.1.5; --user 1000:1000 works on docker-ce 29.9.3.

Activity

  1. added theissue type on Nov 5, 2026
  2. elezar commented on Nov 6, 2026

    @elezar
    Member

    📋 triage-agent

    Thanks for the detailed report. The User Story, impact, environment, and standalone docker create --user 1000:1000 + docker cp -a repro are all present.

    Assessment: valid, but the failing environment is outside the documented support range.

    • The failure is in Docker 26.1.5, not OpenShell logic. The driver stages its bundle with upload_to_container using copy_uidgid=true (crates/openshell-driver-docker/src/lib.rs, stage_docker_sandbox_bundle). Your non-OpenShell repro shows that 26.1.x resolves a numeric uid:gid through getent in the image and rejects it when the group has no passwd entry. Newer Docker releases do not.
    • Docker 26.1 is EOL. The docs already state Docker Engine 28.0 or later is required (docs/about/installation.mdx, docs/about/support-matrix.mdx). So the second acceptance criterion's documentation item is already met, and the first (make 26.1.5 work) is out of scope for a supported configuration.
    • The real gap is that the Docker driver never checks the daemon version. docker.version() in the driver startup path (lib.rs ~L920) only verifies the call succeeds. Users on an unsupported Docker get an opaque BoundaryConfigWriteFailed instead of a clear message.

    Not reproduced locally: I did not set up Debian 13 + docker.io 26.1.5. The analysis is from the report and a code read, so the exact Docker-side cause is inferred from your repro.

    Suggested scope for a maintainer to decide: a startup check that compares the Docker server version (from the existing version() response) against 28.0 and fails (or warns) with a message naming the detected and minimum versions, plus a pointer to the support matrix. Open questions: hard error vs. warning, since distro packages may backport fixes, and whether to check API version instead of engine version.

    Next human decision: whether to accept the "version preflight check with a clear error" as the work for this issue and retitle accordingly. Making 26.1.x work is not recommended given EOL.

  3. added and removed
    state:triage-neededOpened without agent diagnostics and needs triage
    on Nov 6, 2026
  4. TaylorMutch commented on Nov 10, 2026

    @TaylorMutch
    Collaborator

    Accepting the issue with the narrower scope of warning when using a version older than the documented supported version.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    Type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions