Get RUN_AS_DEVIN working with app sandbox (#1426)

* get RUN_AS_DEVIN and network=host working with app sandbox

* attempt to fix the workspace base permission

* sandbox might failed in chown due to mounting, but it won't be fatal

* update sshbox instruction

* remove default user id since it will be passed in the instruction

* revert permission fix since it should be resolved by correct SANDBOX_USER_ID

* the permission issue can be fixed by simply provide correct env var

* remove log

* set sandbox user id to getuid by default

* move logging to initializer

* make the uid consistent across host, app container, and sandbox

* remove hostname as it causes sudo issue

* fix permission of entrypoint script

* make the uvicron app run as host user uid for jupyter plugin

* revert use host network

* get docker socket gid and usermod instead of chmod 777

* try to fix app build disk space issue

.github/workflows/ghcr.yml

@@ -42,8 +42,21 @@ jobs:
           username: ${{ github.repository_owner }}
           password: ${{ secrets.GITHUB_TOKEN }}
 
-      - name: Delete huge unnecessary tools folder
+      - name: Free Disk Space (Ubuntu)
-        run: rm -rf /opt/hostedtoolcache
+        uses: jlumbroso/free-disk-space@main
+        with:
+          # this might remove tools that are actually needed,
+          # if set to "true" but frees about 6 GB
+          tool-cache: true
+
+          # all of these default to true, but feel free to set to
+          # "false" if necessary for your workflow
+          android: true
+          dotnet: true
+          haskell: true
+          large-packages: true
+          docker-images: false
+          swap-storage: true
 
       - name: Build and push ${{ matrix.image }}
         if: github.event.pull_request.head.repo.full_name == github.repository

containers/app/Dockerfile

@@ -32,7 +32,8 @@ FROM python:3.12-slim as runtime
 
 WORKDIR /app
 
-ENV RUN_AS_DEVIN=false
+ENV RUN_AS_DEVIN=true
+ENV SANDBOX_USER_ID=1000
 ENV USE_HOST_NETWORK=false
 ENV SSH_HOSTNAME=host.docker.internal
 ENV WORKSPACE_BASE=/opt/workspace_base
@@ -40,13 +41,23 @@ ENV OPEN_DEVIN_BUILD_VERSION=$OPEN_DEVIN_BUILD_VERSION
 RUN mkdir -p $WORKSPACE_BASE
 
 RUN apt-get update -y \
-    && apt-get install -y curl ssh
+    && apt-get install -y curl ssh sudo
+
+RUN useradd -m -u $SANDBOX_USER_ID -s /bin/bash opendevin && \
+    usermod -aG sudo opendevin && \
+    echo '%sudo ALL=(ALL) NOPASSWD:ALL' >> /etc/sudoers
+RUN chown -R opendevin:opendevin /app
+USER opendevin
 
 ENV VIRTUAL_ENV=/app/.venv \
     PATH="/app/.venv/bin:$PATH" \
     PYTHONPATH='/app'
 
 COPY --from=backend-builder ${VIRTUAL_ENV} ${VIRTUAL_ENV}
+# change ownership of the virtual environment to the sandbox user
+USER root
+RUN chown -R opendevin:opendevin ${VIRTUAL_ENV}
+USER opendevin
 
 COPY ./opendevin ./opendevin
 COPY ./agenthub ./agenthub
@@ -55,4 +66,17 @@ RUN playwright install --with-deps chromium
 
 COPY --from=frontend-builder /app/dist ./frontend/dist
 
-CMD ["uvicorn", "opendevin.server.listen:app", "--host", "0.0.0.0", "--port", "3000"]
+USER root
+RUN chown -R opendevin:opendevin /app
+# make group permissions the same as user permissions
+RUN chmod -R g=u /app
+USER opendevin
+
+# change ownership of the app directory to the sandbox user
+COPY ./containers/app/entrypoint.sh /app/entrypoint.sh
+
+# run the script as root
+USER root
+RUN chown opendevin:opendevin /app/entrypoint.sh
+RUN chmod 777 /app/entrypoint.sh
+CMD ["/app/entrypoint.sh"]

containers/app/entrypoint.sh

@@ -0,0 +1,23 @@
+#!/bin/bash
+# check user is root
+if [ "$(id -u)" -ne 0 ]; then
+  echo "Please run as root"
+  exit 1
+fi
+
+if [ -z "$SANDBOX_USER_ID" ]; then
+  echo "SANDBOX_USER_ID is not set"
+  exit 1
+fi
+
+# change uid of opendevin user to match the host user
+# but the group id is not changed, so the user can still access everything under /app
+usermod -u $SANDBOX_USER_ID opendevin
+
+# get the user group of /var/run/docker.sock and set opendevin to that group
+DOCKER_SOCKET_GID=$(stat -c '%g' /var/run/docker.sock)
+echo "Docker socket group id: $DOCKER_SOCKET_GID"
+usermod -aG $DOCKER_SOCKET_GID opendevin
+
+# switch to the user and start the server
+su opendevin -c "cd /app && uvicorn opendevin.server.listen:app --host 0.0.0.0 --port 3000"

opendevin/config.py

@@ -52,6 +52,7 @@ DEFAULT_CONFIG: dict = {
     ConfigType.USE_HOST_NETWORK: 'false',
     ConfigType.SSH_HOSTNAME: 'localhost',
     ConfigType.DISABLE_COLOR: 'false',
+    ConfigType.SANDBOX_USER_ID: os.getuid() if hasattr(os, 'getuid') else None,
     ConfigType.SANDBOX_TIMEOUT: 120,
     ConfigType.GITHUB_TOKEN: None
 }

opendevin/sandbox/docker/ssh_box.py

@@ -41,7 +41,6 @@ if SANDBOX_USER_ID := config.get(ConfigType.SANDBOX_USER_ID):
 elif hasattr(os, 'getuid'):
     USER_ID = os.getuid()
 
-
 class DockerSSHBox(Sandbox):
     instance_id: str
     container_image: str
@@ -62,6 +61,7 @@ class DockerSSHBox(Sandbox):
         timeout: int = 120,
         sid: str | None = None,
     ):
+        logger.info(f'SSHBox is running as {"opendevin" if RUN_AS_DEVIN else "root"} user with USER_ID={USER_ID} in the sandbox')
         # Initialize docker client. Throws an exception if Docker is not reachable.
         try:
             self.docker_client = docker.from_env()
@@ -150,8 +150,10 @@ class DockerSSHBox(Sandbox):
                 workdir=SANDBOX_WORKSPACE_DIR,
             )
             if exit_code != 0:
-                raise Exception(
+                # This is not a fatal error, just a warning
-                    f'Failed to chown workspace directory for opendevin in sandbox: {logs}')
+                logger.warning(
+                    f'Failed to chown workspace directory for opendevin in sandbox: {logs}. But this should be fine if the {SANDBOX_WORKSPACE_DIR=} is mounted by the app docker container.'
+                )
         else:
             exit_code, logs = self.container.exec_run(
                 # change password for root
@@ -351,7 +353,6 @@ class DockerSSHBox(Sandbox):
                 **network_kwargs,
                 working_dir=SANDBOX_WORKSPACE_DIR,
                 name=self.container_name,
-                hostname='opendevin_sandbox',
                 detach=True,
                 volumes={
                     mount_dir: {