update image generation to use uv

Dockerfile_k8s

@@ -34,11 +34,12 @@ ENV HOME /home/sky
 WORKDIR /home/sky
 
 # Install skypilot dependencies
-RUN conda init && export PIP_DISABLE_PIP_VERSION_CHECK=1 && \
-    python3 -m venv ~/skypilot-runtime && \
-    PYTHON_EXEC=$(echo ~/skypilot-runtime)/bin/python && \
-    $PYTHON_EXEC -m pip install 'skypilot-nightly[remote,kubernetes]' 'ray[default]==2.9.3' 'pycryptodome==3.12.0' && \
-    $PYTHON_EXEC -m pip uninstall skypilot-nightly -y && \
+RUN curl -LsSf https://astral.sh/uv/install.sh | UV_INSTALL_DIR=$HOME/.local/bin sh && \
+    export PATH=$PATH:$HOME/.local/bin && \
+    conda init && uv venv --seed ~/skypilot-runtime && \
+    UV_PIP_EXEC="env VIRTUAL_ENV=$(echo ~/skypilot-runtime) uv pip" && \
+    $UV_PIP_EXEC install 'skypilot-nightly[remote,kubernetes]' 'ray[default]==2.9.3' 'pycryptodome==3.12.0' && \
+    $UV_PIP_EXEC uninstall skypilot-nightly && \
     curl -LO "https://dl.k8s.io/release/v1.28.11/bin/linux/amd64/kubectl" && \
     sudo install -o root -g root -m 0755 kubectl /usr/local/bin/kubectl && \
     echo 'export PATH="$PATH:$HOME/.local/bin"' >> ~/.bashrc

Dockerfile_k8s_gpu

@@ -42,11 +42,12 @@ RUN curl https://repo.anaconda.com/miniconda/Miniconda3-py310_23.11.0-2-Linux-x8
     eval "$(~/miniconda3/bin/conda shell.bash hook)" && conda init && conda config --set auto_activate_base true && conda activate base && \
     grep "# >>> conda initialize >>>" ~/.bashrc || { conda init && source ~/.bashrc; } && \
     rm Miniconda3-Linux-x86_64.sh && \
-    export PIP_DISABLE_PIP_VERSION_CHECK=1 && \
-    python3 -m venv ~/skypilot-runtime && \
-    PYTHON_EXEC=$(echo ~/skypilot-runtime)/bin/python && \
-    $PYTHON_EXEC -m pip install 'skypilot-nightly[remote,kubernetes]' 'ray[default]==2.9.3' 'pycryptodome==3.12.0' && \
-    $PYTHON_EXEC -m pip uninstall skypilot-nightly -y && \
+    curl -LsSf https://astral.sh/uv/install.sh | UV_INSTALL_DIR=$HOME/.local/bin sh && \
+    export PATH=$PATH:$HOME/.local/bin && \
+    uv venv --seed ~/skypilot-runtime && \
+    UV_PIP_EXEC="env VIRTUAL_ENV=$(echo ~/skypilot-runtime) uv pip" && \
+    $UV_PIP_EXEC install 'skypilot-nightly[remote,kubernetes]' 'ray[default]==2.9.3' 'pycryptodome==3.12.0' && \
+    $UV_PIP_EXEC uninstall skypilot-nightly && \
     curl -LO "https://dl.k8s.io/release/v1.28.11/bin/linux/amd64/kubectl" && \
     sudo install -o root -g root -m 0755 kubectl /usr/local/bin/kubectl && \
     echo 'export PATH="$PATH:$HOME/.local/bin"' >> ~/.bashrc

docs/source/reference/kubernetes/kubernetes-getting-started.rst

@@ -270,7 +270,7 @@ FAQs
 * **Are autoscaling Kubernetes clusters supported?**
 
   To run on autoscaling clusters, set the :code:`provision_timeout` key in :code:`~/.sky/config.yaml` to a large value to give enough time for the cluster autoscaler to provision new nodes.
-  This will direct SkyPilot to wait for the cluster to scale up before failing over to the next candidate resource (e.g., next cloud). 
+  This will direct SkyPilot to wait for the cluster to scale up before failing over to the next candidate resource (e.g., next cloud).
 
   If you are using GPUs in a scale-to-zero setting, you should also set the :code:`autoscaler` key to the autoscaler type of your cluster. More details in :ref:`config-yaml`.
 
@@ -344,12 +344,12 @@ FAQs
         eval "$(~/miniconda3/bin/conda shell.bash hook)" && conda init && conda config --set auto_activate_base true && conda activate base && \
         grep "# >>> conda initialize >>>" ~/.bashrc || { conda init && source ~/.bashrc; } && \
         rm Miniconda3-Linux-x86_64.sh && \
-        export PIP_DISABLE_PIP_VERSION_CHECK=1 && \
-        python3 -m venv ~/skypilot-runtime && \
-        PYTHON_EXEC=$(echo ~/skypilot-runtime)/bin/python && \
-        $PYTHON_EXEC -m pip install 'skypilot-nightly[remote,kubernetes]' 'ray[default]==2.9.3' 'pycryptodome==3.12.0' && \
-        $PYTHON_EXEC -m pip uninstall skypilot-nightly -y && \
+        curl -LsSf https://astral.sh/uv/install.sh | UV_INSTALL_DIR=$HOME/.local/bin sh && \
+        export PATH=$PATH:$HOME/.local/bin && \
+        uv venv --seed ~/skypilot-runtime && \
+        UV_PIP_EXEC="env VIRTUAL_ENV=$(echo ~/skypilot-runtime) uv pip" && \
+        $UV_PIP_EXEC install 'skypilot-nightly[remote,kubernetes]' 'ray[default]==2.9.3' 'pycryptodome==3.12.0' && \
+        $UV_PIP_EXEC uninstall skypilot-nightly && \
         curl -LO "https://dl.k8s.io/release/v1.28.11/bin/linux/amd64/kubectl" && \
         sudo install -o root -g root -m 0755 kubectl /usr/local/bin/kubectl && \
         echo 'export PATH="$PATH:$HOME/.local/bin"' >> ~/.bashrc
-

sky/clouds/service_catalog/images/README.md

@@ -7,8 +7,9 @@ You only need to do this once.
 ```bash
 packer init plugins.pkr.hcl
 ```
-3. Setup cloud credentials
-4. `cd sky/clouds/service_catalog/images`
+3. Check that docker buildx is installed: `docker buildx` should show a help page. buildx is installed with Docker Desktop but you may have to install separately on Linux.
+4. Setup cloud credentials
+5. `cd sky/clouds/service_catalog/images`
 
 ## Generate Images
 FYI time to packer build images:

sky/clouds/service_catalog/images/plugins.pkr.hcl

@@ -15,3 +15,12 @@ packer {
     }
   }
 }
+
+packer {
+  required_plugins {
+    azure = {
+      source  = "github.com/hashicorp/azure"
+      version = "~> 2"
+    }
+  }
+}

sky/clouds/service_catalog/images/provisioners/skypilot.sh

@@ -32,36 +32,39 @@ conda init
 conda config --set auto_activate_base true
 conda activate base
 
-# Conda, Python
+# Install uv
+curl -LsSf https://astral.sh/uv/install.sh | UV_INSTALL_DIR=$HOME/.local/bin sh
+export PATH=$PATH:$HOME/.local/bin
+
+# Set up Conda, uv, Python
 echo "Creating conda env with Python 3.10"
 conda create -y -n skypilot-runtime python=3.10
 conda activate skypilot-runtime
-export PIP_DISABLE_PIP_VERSION_CHECK=1
 echo PATH=$PATH
-python3 -m venv ~/skypilot-runtime
+uv venv --seed ~/skypilot-runtime
 PYTHON_EXEC=$(echo ~/skypilot-runtime)/bin/python
+UV_EXEC_PIP="env VIRTUAL_ENV=$(echo ~/skypilot-runtime) uv pip"
 
 # Install SkyPilot
-$PYTHON_EXEC -m pip install "skypilot-nightly[remote]"
+$UV_EXEC_PIP install "skypilot-nightly[remote]"
 
 # Install Ray
 RAY_ADDRESS=127.0.0.1:6380
-$PYTHON_EXEC -m pip install --exists-action w -U "ray[default]==2.9.3"
-export PATH=$PATH:$HOME/.local/bin
+$UV_EXEC_PIP install -U "ray[default]==2.9.3"
 source ~/skypilot-runtime/bin/activate
 which ray > ~/.sky/ray_path || exit 1
-$PYTHON_EXEC -m pip list | grep "ray " | grep 2.9.3 2>&1 > /dev/null && {
+$UV_EXEC_PIP list | grep "ray " | grep 2.9.3 2>&1 > /dev/null && {
   $PYTHON_EXEC -c "from sky.skylet.ray_patches import patch; patch()" || exit 1
 }
 
 # Install cloud dependencies
 if [ "$CLOUD" = "azure" ]; then
-    $PYTHON_EXEC -m pip install "skypilot-nightly[azure]"
+    $UV_EXEC_PIP install "skypilot-nightly[azure]"
 elif [ "$CLOUD" = "gcp" ]; then
     # We don't have to install the google-cloud-sdk since it is installed by default in GCP machines.
-    $PYTHON_EXEC -m pip install "skypilot-nightly[gcp]"
+    $UV_EXEC_PIP install "skypilot-nightly[gcp]"
 elif [ "$CLOUD" = "aws" ]; then
-    $PYTHON_EXEC -m pip install "skypilot-nightly[aws]"
+    $UV_EXEC_PIP install "skypilot-nightly[aws]"
 else
     echo "Error: Unknown cloud $CLOUD so not installing any cloud dependencies."
 fi
@@ -81,5 +84,5 @@ sudo systemctl disable jupyter > /dev/null 2>&1 || true
 
 # Cleanup
 # Remove SkyPilot in OS image because when user sky launch we will install whatever version of SkyPilot user has on their local machine.
-$PYTHON_EXEC -m pip uninstall "skypilot-nightly" -y
+$UV_EXEC_PIP uninstall "skypilot-nightly"
 rm -rf ~/.sky

sky/clouds/service_catalog/images/provisioners/user-toolkit.sh

@@ -3,10 +3,12 @@
 
 eval "$(~/miniconda3/bin/conda shell.bash hook)"
 conda activate base
-pip install numpy
-pip install pandas
+
+export PATH=$PATH:$HOME/.local/bin
+uv pip install numpy
+uv pip install pandas
 
 if [ "$AZURE_GRID_DRIVER" = 1 ]; then
     # Need PyTorch X.X.X+cu121 version to be compatible with older NVIDIA driver (535.161.08 or lower)
-    pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121
+    uv pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121
 fi

sky/skylet/constants.py

@@ -154,15 +154,15 @@
     f'conda create -y -n {SKY_REMOTE_PYTHON_ENV_NAME} python=3.10 && '
     f'conda activate {SKY_REMOTE_PYTHON_ENV_NAME};'
     # Install uv for venv management and pip installation.
-    'which uv >/dev/null 2>&1 || '
+    f'{SKY_UV_INSTALL_DIR}/uv --version >/dev/null 2>&1 || '
     'curl -LsSf https://astral.sh/uv/install.sh '
     f'| UV_INSTALL_DIR={SKY_UV_INSTALL_DIR} sh;'
     # Create a separate conda environment for SkyPilot dependencies.
     f'[ -d {SKY_REMOTE_PYTHON_ENV} ] || '
     # Do NOT use --system-site-packages here, because if users upgrade any
     # packages in the base env, they interfere with skypilot dependencies.
     # Reference: https://github.com/skypilot-org/skypilot/issues/4097
-    f'{SKY_UV_CMD} venv {SKY_REMOTE_PYTHON_ENV};'
+    f'{SKY_UV_CMD} venv --seed {SKY_REMOTE_PYTHON_ENV};'
     f'echo "$(echo {SKY_REMOTE_PYTHON_ENV})/bin/python" > {SKY_PYTHON_PATH_FILE};'
 )