This message was deleted.

Hi. This https://github.com/bentoml/yatai-image-builder could be what you are looking for?

Thanks @Jiang for responding I am not using yatai as I am trying to use bentoml in vertex ai end point, so building container image using

bentoml containerize

in order to use that to deploy a custom container. yatai-image-builder would require Kubernetes environment based on my understanding

I see. This error is common when you don't have docker installed or the permission to call docker build.

Do you have Docker setup?

hmm still no luck, I would agree it would be great if bentoml support kaniko for containerize, but I would like to find a solution that is doable as of now Looking at this I would assume that docker is supported, but still didn't figure our the right way to successfully run bentoml containerize using cloud build https://cloud.google.com/build/docs/interacting-with-dockerhub-images

Yeah, Ic. It seems that the key is to make docker available in the cloud builder.

I tried to run the command using this as a base image but it failed as bentoml is not recognised and pip is not available to install it for trial sake, I'll try creating another image based on this and install all the required libraries and try with that if it works

Yeah. I went to a wrong direction. It is just because you are trying to run bentoml containerize in the image

xxx_bentoml

Yes please I would appreciate it

I'm trying to draw a graph for the recommended workflow. It may take a while

Makes sense that's exactly what I was trying to do in last hour but getting errors when trying to install pip

getting errors when trying to install pip

Any details? It might due to Google is using distroless in their cloud builder

This is the dockerfile content

# Create image that is used to run bentoml build
FROM <http://gcr.io/cloud-builders/docker|gcr.io/cloud-builders/docker>

ARG HOME=/home

RUN apt-get update 
RUN apt-get install -y python3-dev
RUN apt-get install -y python3
RUN apt-get install -y wget git
RUN python3 --version
RUN apt-get install -y python3-pip
RUN apt-get clean

RUN pip install bentoml

# Set home directory
WORKDIR ${HOME}

But getting an error at

python3-pip

on executing docker build

Oh it looks like this is an issue related to Ubuntu. For this problem, we are on the same page and we need to do some searching.

Ok good to know this the approach you suggested an we are on the same page with the direction

I noticed a hash mismatch, and it is possible that there has been a man-in-the-middle attack.

hmm strangely with retrying the hash mismatch didn't come up, but the error keep changing I attached the full output of two

docker build

command runs... In the meantime I am googling to know how to modify the source list in the apt configuration, if you have an easily shareable example that would handy

Personally I still suggest to change to https first. I saw it is still using http:

#0 1.772 Get:1 <http://archive.ubuntu.com/ubuntu> focal/main amd64 liblocale-gettext-perl amd64 1.07-4 [17.1 kB]
#0 2.556 Get:2 <http://archive.ubuntu.com/ubuntu> focal-updates/main amd64 binutils-common amd64 2.34-6ubuntu1.4 [207 kB]
#0 4.122 Get:3 <http://archive.ubuntu.com/ubuntu> focal-updates/main amd64 libbinutils amd64 2.34-6ubuntu1.4 [474 kB]

An answer from chatgpt:

FROM <http://gcr.io/cloud-builders/docker|gcr.io/cloud-builders/docker>

RUN sed -i 's/http:/https:/g' /etc/apt/sources.list

oh I totally forgot about consulting chatgpt! trying this now

I'm not familiar with GCP's tools, but I know in AWS's CodeBuild, you have to run your build job in

Privileged

mode if you're going to do Docker-in-Docker (as in, building a Docker image inside CodeBuild's Docker container). That way it has access to the Docker runtime inside the build container.

Finally managed to make it work 🎉. Below are the steps in case someone else needed to do the same: 1- Create following

Dockerfile

# Create image that is used to run bentoml containerize
FROM docker:20.10.24 as static-docker-source

FROM python:3.9
COPY --from=static-docker-source /usr/local/bin/docker /usr/local/bin/docker
ARG HOME=/home

RUN sed -i 's/http:/https:/g' /etc/apt/sources.list

RUN pip install bentoml

# Set home directory
WORKDIR ${HOME}

2- run

docker build -t <http://gcr.io/project-id/bentoml_builder|gcr.io/project-id/bentoml_builder> -f bentobuilder.Dockerfile .

&

docker push <http://gcr.io/autograb-dev/bentoml_builder|gcr.io/autograb-dev/bentoml_builder>

to build the image and push it to google container registry 3- Create following

cloudbuild.yaml

steps:
# Build bentoml
- name: '<http://gcr.io/project-id/poc_bentoml:latest|gcr.io/project-id/poc_bentoml:latest>'
  script: |
    python import_model.py --mlflow_run_id=${_MLFLOW_RUN_ID} --mlflow_model_uri=${_MLFLOW_MODEL_URI}
    echo running bentoml build
    bentoml build
    bentoml list
  env:
  - 'MLFLOW_TRACKING_URI=${_MLFLOW_TRACKING_URI}'
  - '_MLFLOW_RUN_ID=${_MLFLOW_RUN_ID}'
  - '_MLFLOW_MODEL_URI=${_MLFLOW_MODEL_URI}'
# Containerize bentoml
- name: '<http://gcr.io/project-id/bentoml_builder|gcr.io/project-id/bentoml_builder>'
  script: |
    echo running bentoml containerize ${_BENTO_SERVICE} -t <http://gcr.io/$PROJECT_ID/${_BENTO_SERVICE}|gcr.io/$PROJECT_ID/${_BENTO_SERVICE}>
    bentoml containerize ${_BENTO_SERVICE} -t <http://gcr.io/$PROJECT_ID/${_BENTO_SERVICE}|gcr.io/$PROJECT_ID/${_BENTO_SERVICE}>
  env:
  - 'PROJECT_ID=$PROJECT_ID'
  - '_BENTO_SERVICE=${_BENTO_SERVICE}'
# Push image to container registry
- name: '<http://gcr.io/cloud-builders/docker|gcr.io/cloud-builders/docker>'
  args: ['push', '<http://gcr.io/$PROJECT_ID/${_BENTO_SERVICE}|gcr.io/$PROJECT_ID/${_BENTO_SERVICE}>']
substitutions:  # default values
    _BENTO_SERVICE: "poc_service:latest"
    _MLFLOW_TRACKING_URI: "xxx"
    _MLFLOW_RUN_ID: "xxx"
    _MLFLOW_MODEL_URI: "xxx"
images:
- <http://gcr.io/$PROJECT_ID/${_BENTO_SERVICE}|gcr.io/$PROJECT_ID/${_BENTO_SERVICE}>
options:
    dynamic_substitutions: true
    logging: CLOUD_LOGGING_ONLY

4- Now it is possible to trigger the build by running

gcloud beta builds submit --project=project-id --config=cloudbuild.yaml --substitutions=_MLFLOW_TRACKING_URI="xxx",_BENTO_SERVICE="xxx",_MLFLOW_RUN_ID="xxx",_MLFLOW_MODEL_URI="xxx"

Note: The first step in the cloudbuild script is using an imaged built to conclude the required files to run

bentoml build

command such as

bentofile.yaml

,

service.py

, and other code required to create the bento service. Second and third steps would be common for any bento deployment. It would be great if there is a better way to do step 1 without having to bake a new image with the source code for every service.