ci: add docker file

epfl-dlab · Sep 24, 2023 · 2007dc5 · 2007dc5
1 parent aabf630
commit 2007dc5
Show file tree

Hide file tree

Showing 4 changed files with 192 additions and 4 deletions.
diff --git a/.github/workflows/docker-image.yml b/.github/workflows/docker-image.yml
@@ -2,9 +2,9 @@ name: Docker Image CI
 
 on:
   push:
-    branches: [ "main" ]
+    branches: [ "main", "dev" ]
   pull_request:
-    branches: [ "main" ]
+    branches: [ "main"]
 
 jobs:
 

diff --git a/docker/Dockerfile b/docker/Dockerfile
@@ -1,5 +1,144 @@
-# Container image that runs your code
-FROM alpine:3.10
+# Choose a docker template
+# This will set what OS, CUDA, and perhaps even packages / python versions
+# you can preemptly have. You can find more templates in
+ARG BASE_IMAGE=nvcr.io/nvidia/pytorch:22.02-py3
+FROM $BASE_IMAGE
+ARG BASE_IMAGE
+RUN echo "Installing Apex on top of ${BASE_IMAGE}"
+# make sure we don't overwrite some existing directory called "apex"
+WORKDIR /tmp/unique_for_apex
+# uninstall Apex if present, twice to make absolutely sure :)
+RUN pip uninstall -y apex || :
+RUN pip uninstall -y apex || :
+# SHA is something the user can touch to force recreation of this Docker layer,
+# and therefore force cloning of the latest version of Apex
+RUN SHA=ToUcHMe git clone https://github.com/NVIDIA/apex.git
+WORKDIR /tmp/unique_for_apex/apex
+RUN pip install -v --no-cache-dir --global-option="--cpp_ext" --global-option="--cuda_ext" .
+WORKDIR /workspace
+
+#################################################
+#
+# Don't change the above lines, otherwise you may get error with installing apex
+# I'm not sure about the error and don't want to spend time on it, so please don't change it
+# unless you know what you are doing
+#
+#################################################
+
+# use /bin/bash to execute the SHELL command, by default it is /bin/sh
+# -c means run the command and -u means fail if the command fails
+SHELL ["/bin/bash", "-cu"]
+
+# Setup your user profile with the right group permission to access NFS folder
+# For the command that gives the ids and names you would need checkout the .env file
+WORKDIR /
+
+# force recompile
+RUN echo "20230512" >/dev/null
+
+# install sudo
+RUN apt update && \
+    apt -y install sudo
+
+# create a new user and add it to the relevant groups, including sudo group
+RUN --mount=type=secret,id=dot_env source /run/secrets/dot_env && \
+     groupadd -g ${GROUP_ID1} ${GROUP_NAME1} && \
+     groupadd -g ${GROUP_ID2} ${GROUP_NAME2} && \
+     useradd -rm -d /home/${USER_NAME} -s /bin/bash -g ${GROUP_ID1} -G sudo,${GROUP_ID2} -u ${USER_ID} ${USER_NAME} && \
+     chown ${USER_ID} -R /home/${USER_NAME} && \
+     echo -e "${USER_NAME}\n${USER_NAME}" | passwd ${USER_NAME}
+
+# Set some basic ENV vars for readability
+ENV USER_NAME="geng"
+ENV HOME=/home/${USER_NAME}
+ENV CONDA_PREFIX=${HOME}/.conda
+ENV CONDA=${CONDA_PREFIX}/condabin/conda
+
+# the repo name on github
+#ENV REPO=llama
+#ENV REPO_DIR=${HOME}/${REPO}
+
+# WORKDIR instruction sets the directory the following instructions should be run from
+WORKDIR ${HOME}
+
+# [optional and recommended] Install conda
+RUN wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh -O miniconda.sh
+RUN bash miniconda.sh -b -p ${CONDA_PREFIX}
+RUN ${CONDA} config --set auto_activate_base false
+RUN ${CONDA} init bash
+
+# [optional] Github user configuration (necessary for accessing private repos)
+RUN --mount=type=secret,id=dot_env source /run/secrets/dot_env && \
+    git config --global user.name ${GITHUB_NAME}
+RUN --mount=type=secret,id=dot_env source /run/secrets/dot_env && \
+    git config --global user.email ${GITHUB_EMAIL}
+# RUN git config --global pull.rebase false
+
+# Prepare the NFS mount
+RUN mkdir /mnt/dlabdata1
+RUN mkdir /mnt/scratch
+
+# Avoid ascii errors when reading files in Python
+# https://stackoverflow.com/a/60084243/12234753
+RUN apt-get install -y locales && locale-gen en_US.UTF-8
+ENV LANG='en_US.UTF-8' LANGUAGE='en_US:en' LC_ALL='en_US.UTF-8'
+
+
+##############################################
+#
+#        project specific
+#
+##############################################
+RUN echo "20230308" >/dev/null
+
+
+# Create conda environment and install requirements
+COPY requirements.txt .
+ENV ENV_NAME="gcd"
+RUN ${CONDA} create -y -n ${ENV_NAME} python=3.9
+RUN ${CONDA} run --name ${ENV_NAME} pip install -r requirements.txt
+
+# [Optional] Login to wandb
+RUN --mount=type=secret,id=dot_env source /run/secrets/dot_env && \
+    ${CONDA} run -n ${ENV_NAME} wandb login ${WANDB_API_KEY}
+
+# Login to HF
+RUN --mount=type=secret,id=dot_env source /run/secrets/dot_env && \
+     ${CONDA} run -n ${ENV_NAME} python -c "from huggingface_hub.hf_api import HfFolder; HfFolder.save_token('${HF_API_KEY}')"
+#################################################
+#
+#          GF
+#
+##################################################
+
+RUN wget https://github.com/GrammaticalFramework/gf-core/releases/download/3.11/gf-3.11-ubuntu-20.04.deb
+RUN apt-get install ./gf-3.11-ubuntu-20.04.deb
+
+
+# Change ownership of the whole /home/USER forlder, so that the files created by root can be accesible by world user, such as git cloned repo etc.
+# By default, they are owned by root.
+RUN --mount=type=secret,id=dot_env source /run/secrets/dot_env && \
+    chown ${USER_ID} -R /home/${USER_NAME}
+
+
+# fix problem running a Haskell app on an Ubuntu docker image.
+# https://github.com/snoyberg/http-client/issues/292
+RUN apt-get install netbase
+#################################################
+#
+#          Install zsh
+#
+##################################################
+RUN apt install -y zsh
+
+# N.B. the following commands are run as the user, not as root; Don't move them above the USER command
+# Uses "Spaceship" theme with some customization. Uses some bundled plugins and installs some more from github
+RUN sh -c "$(wget -O- https://github.com/deluan/zsh-in-docker/releases/download/v1.1.5/zsh-in-docker.sh)" -- \
+    -t robbyrussell \
+    -p https://github.com/zsh-users/zsh-autosuggestions \
+    -p https://github.com/zsh-users/zsh-completions
+
+USER ${USER_NAME}
 
 # Copies your code file from your action repository to the filesystem path `/` of the container
 COPY docker/entrypoint.sh /entrypoint.sh

diff --git a/docker/prod.Dockerfile b/docker/prod.Dockerfile
diff --git a/runai/example.yml b/runai/example.yml
@@ -0,0 +1,49 @@
+ApiVersion: run.ai/v1
+kind: RunaiJob
+metadata:
+  name: JOBTEST
+  labels:
+    priorityClassName: "build" # Interactive Job if present, for Train Job REMOVE this line
+    user: GASPAR_NAME
+spec:
+  template:
+    metadata:
+      labels:
+        user: GASPAR:NAME # User e.g. firstname.lastname
+    spec:
+      hostIPC: true
+      schedulerName: runai-scheduler
+      restartPolicy: Never
+      securityContext:
+        runAsUser: UID # insert uid found in people.epfl in admistrative data
+        runAsGroup: GID # insert gid as found in people.epfl in admistrative data
+        fsGroup: GID
+      containers:
+      - name: container-name
+        image: ic-registry.epfl.ch/LAB/FOLDER/IMAGE:VERSION
+        workingDir : /PATH/TO/WORKINGDIR
+        command: ["/bin/bash"]
+        args:
+        - "-c"
+        - "SOME COMMANDS TO EXECUTE && SOME OTHER COMMANDES TO EXECUTE"
+
+        env:
+          - name: HOME
+            value: "/PATH/TO/HOMEDIR"
+          - name: PYTHONPATH
+            value: "/PATH/TO/PYLIB1:/PATH/TO/PYLIB2:/PATH/TO/PYLIB3"
+        resources:
+          limits:
+            nvidia.com/gpu: 1
+        volumeMounts:
+          - mountPath: /MOUNTPATH1
+            name: VOLUME_NAME1
+          - mountPath: /MOUNTPATH2
+            name: VOLUME_NAME2
+      volumes:
+        - name: VOLUME_NAME1
+          persistentVolumeClaim:
+            claimName: runai-pv-VOLUME1
+        - name: VOLUME_NAME2
+          persistentVolumeClaim:
+            claimName: runai-pv-VOLUME2