Intro

We will deploy MLflow in Mail.ru Cloud Solutions using S3 as artifact store, DBaaS Postgres as backend entity storage and Tracking Server on separate host.
This is the most production ready scenario of deployment
https://mlflow.org/docs/latest/tracking.html#scenario-4-mlflow-with-remote-tracking-server-backend-and-artifact-stores

1. Create VM for mlflow

https://mcs.mail.ru/help/ru_RU/create-vm/vm-quick-create
Tested with OS - Ubuntu 18.04
Record VM white ip and internal ip

2. Install Conda on VM created on step 1

You can access VM with command

ssh -i REPLACE_WITH_YOUR_KEY ubuntu@REPLACE_WITH_YOUR_VM_IP

Download and install Conda

curl -O https://repo.anaconda.com/archive/Anaconda3-2020.11-Linux-x86_64.sh
bash Anaconda3-2020.11-Linux-x86_64.sh
exec bash

3. Install MLflow

We will install MLflow in separate environment

conda create -n mlflow_env
conda activate mlflow_env
conda install python

pip install mlflow
pip install boto3

sudo apt install gcc
pip install psycopg2-binary

4. Create Postgres as a backend store

https://mcs.mail.ru/help/ru_RU/dbaas-start/db-postgres
Save credentials, DB name

5. Create S3 bucket and directory

The next step is creating a directory for our Tracking Server to log the Machine Learning models and other artifacts. Remember that the Postgres database is only used for storing metadata regarding those models. We will use S3 as artifact storage.

Create bucket and directory inside it
https://mcs.mail.ru/help/ru_RU/s3-start/create-bucket

Create account, access key id and secret key
https://mcs.mail.ru/help/ru_RU/s3-start/s3-account

6. Launch MLflow

Login to VM that was created on Step 1
You can access VM with command

ssh -i REPLACE_WITH_YOUR_KEY ubuntu@REPLACE_WITH_YOUR_VM_IP

Set env variables

sudo nano /etc/environment

#Copy and paste this, replace with your values
MLFLOW_S3_ENDPOINT_URL=https://hb.bizmrg.com
MLFLOW_TRACKING_URI=http://REPLACE_WITH_INTERNAL_IP_MLFLOW_VM:8000

Log out, login to apply changes

Create file with S3 credentials

mkdir .aws

nano ~/.aws/credentials

Copy and paste this in ~/.aws/credentials

[default]
aws_access_key_id = REPLACE_WITH_YOUR_KEY
aws_secret_access_key = REPLACE_WITH_YOUR_SECRET_KEY

This is test run, if you leave terminal MLflow will be unaccessible
Permanent serving of MLflow on the next step

conda activate mlflow_env
mlflow server --backend-store-uri postgresql://pg_user:pg_password@REPLACE_WITH_INTERNAL_IP_POSTGRESQL/db_name --default-artifact-root s3://REPLACE_WITH_YOUR_BUCKET/REPLACE_WITH_YOUR_DIRECTORY/ -h 0.0.0.0 -p 8000

7. Enable MLflow as a systemd service

Create dirs for logs and errors

mkdir ~/mlflow_logs/
mkdir ~/mlflow_errors/

Create file with systemd service

sudo nano /etc/systemd/system/mlflow-tracking.service

Copy and paste in /etc/systemd/system/mlflow-tracking.service

[Unit]
Description=MLflow Tracking Server
After=network.target
[Service]
Environment=MLFLOW_S3_ENDPOINT_URL=https://hb.bizmrg.com
Restart=on-failure
RestartSec=30
StandardOutput=file:/home/ubuntu/mlflow_logs/stdout.log
StandardError=file:/home/ubuntu/mlflow_errors/stderr.log
User=ubuntu
ExecStart=/bin/bash -c 'PATH=/home/ubuntu/anaconda3/envs/mlflow_env/bin/:$PATH exec mlflow server --backend-store-uri postgresql://PG_USER:PG_PASSWORD@REPLACE_WITH_INTERNAL_IP_POSTGRESQL/DB_NAME --default-artifact-root s3://REPLACE_WITH_YOUR_BUCKET/REPLACE_WITH_YOUR_DIRECTORY/ -h 0.0.0.0 -p 8000' 
[Install]
WantedBy=multi-user.target

Enable MLflow service

sudo systemctl daemon-reload
sudo systemctl enable mlflow-tracking
sudo systemctl start mlflow-tracking
sudo systemctl status mlflow-tracking

Check that evrything is ok in logs

head -n 95 ~/mlflow_logs/stdout.log 

8. Create JupyterHub host

https://mcs.mail.ru/help/ru_RU/ml-start/ml-info

Login to VM with JupyterHub

ssh -i REPLACE_WITH_YOUR_KEY ubuntu@REPLACE_WITH_YOUR_VM_IP

Open tmux and launch JupyterHub

tmux

jupyter-notebook --ip '*'

Copy string that looks like that
http://name_of_host:8888/?token=5d3d6b7a0551asdffds4190e8sdffsd329bee345esdfmkdfs2c042c0b7a5

deattach from tmux session
ctrl + b d

9. Config JupyterHub host

Set env variables

sudo nano /etc/environment

MLFLOW_TRACKING_URI=http://REPLACE_WITH_INTERNAL_IP_MLFLOW_VM:8000
MLFLOW_S3_ENDPOINT_URL=https://hb.bizmrg.com

Log out, login to apply changes

Create file with S3 credentials

mkdir .aws

nano ~/.aws/credentials

Copy and paste this in ~/.aws/credentials

[default]
aws_access_key_id = REPLACE_WITH_YOUR_KEY
aws_secret_access_key = REPLACE_WITH_YOUR_SECRET_KEY

10. Install MLflow on JupyterHub host

We will create separate environment, install MLflow and create kernel for JupyterHub with MLflow

conda create -n mlflow_env

conda activate mlflow_env

conda install python

pip install mlflow

pip install matplotlib
pip install sklearn
pip install boto3

conda install -c anaconda ipykernel

python -m ipykernel install --user --name ex --display-name "Python (mlflow)"

11. Launch JupyterHub and test MLflow

Your JupyterHub available on url like that
http://name_of_host:8888/?token=5d3d6b7a0551asdffds41sdvlgfd8sdffsd329bee345esdfmkdfs2c042c0b7dffb

You should get this url on step 8

Launch a terminal in Jupyter and clone the mlflow repo

git clone https://github.com/stockblog/webinar_mlflow/ webinar_mlflow

Open mlflow_demo.ipynb and launch cells

12. Serve model from artifact store

Find URI of model in MLFlow UI
Connect to MLflow host created on step 1

#EXAMPLE, REPLACE S3 path with your own path to model
mlflow models serve -m s3://mlflow_webinar/artifacts/5/a7ee769713974c118d0eff20226eb474/artifacts/model -h 0.0.0.0 -p 8001

mlflow models serve -m s3://BUCKET/FOLDER/EXPERIMENT_NUMBER/INTERNAL_MLFLOW_ID/artifacts/model -h 0.0.0.0 -p 8001

13. Serve model from the model registry

Register model in MLflow Models UI. Copy model name and paste it to example string

#EXAMPLE
mlflow models serve -m "models:/diabet_test/Staging"

mlflow models serve -m "models:/YOUR_MODEL_NAME/STAGE"

14. Test model

You may need to replace port from 8001 to default port 5001 if you did not set -p parameter in prev step

#EXAMPLE, replace ip adress with internal ip of your mlflow host. You could also use white ip, set firewall settings 
curl -X POST -H "Content-Type:application/json; format=pandas-split" --data '{"columns":["age", "sex", "bmi", "bp", "s1", "s2", "s3", "s4", "s5", "s6"], "data":[[0.0453409833354632, 0.0506801187398187, 0.0606183944448076, 0.0310533436263482, 0.0287020030602135, 0.0473467013092799, 0.0544457590642881, 0.0712099797536354, 0.133598980013008, 0.135611830689079]]}' http://0.0.0.0:8001/invocations

curl -X POST -H "Content-Type:application/json; format=pandas-split" --data '{"columns":["age", "sex", "bmi", "bp", "s1", "s2", "s3", "s4", "s5", "s6"], "data":[[0.0453409833354632, 0.0506801187398187, 0.0606183944448076, 0.0310533436263482, 0.0287020030602135, 0.0473467013092799, 0.0544457590642881, 0.0712099797536354, 0.133598980013008, 0.135611830689079]]}' http://0.0.0.0:8001/invocations

15. Permanent serving of model

Connect to MLflow host created on step 1
you can find MLFLOW_ENV_OF_MODEL when you launch model on step 11 or 12

sudo nano /etc/systemd/system/mlflow-model.service

[Unit]
Description=MLFlow Model Serving
After=network.target

[Service]
Restart=on-failure
RestartSec=30
StandardOutput=file:/home/ubuntu/mlflow_logs/stdout.log
StandardError=file:/home/ubuntu/mlflow_errors/stderr.log
Environment=MLFLOW_TRACKING_URI=http://REPLACE_WITH_INTERNAL_IP_MLFLOW_VM:8000
Environment=MLFLOW_CONDA_HOME=/home/ubuntu/anaconda3/
Environment=MLFLOW_S3_ENDPOINT_URL=https://hb.bizmrg.com
ExecStart=/bin/bash -c 'PATH=/home/ubuntu/anaconda3/envs/REPLACE_WITH_MLFLOW_ENV_OF_MODEL/bin/:$PATH exec mlflow models serve -m "models:/YOUR_MODEL_NAME/STAGE" -h 0.0.0.0 -p 8001'

[Install]
WantedBy=multi-user.target

Then enable new service

sudo systemctl daemon-reload
sudo systemctl enable mlflow-model
sudo systemctl start mlflow-model
sudo systemctl status mlflow-model

16. Build docker image with model

Prerequisites: install Docker
https://docs.docker.com/engine/install/ubuntu/
https://docs.docker.com/engine/install/linux-postinstall/

You may need to edit conda.yaml and add dependencies
You can find this file in artifact storage in folder artifacts/model
s3://BUCKET/FOLDER/EXPERIMENT_NUMBER/INTERNAL_MLFLOW_ID/artifacts/model

Example conda.yaml where we added scipy and boto3 libs to standart dependencies

channels:
- defaults
- conda-forge
dependencies:
- python=3.6.6
- pip
- pip:
  - mlflow
  - scikit-learn==0.19.2
  - cloudpickle==0.5.5
  - scipy
  - boto3
name: mlflow-env

Build docker image with model

#Example
mlflow models build-docker -m "models:/diabet_test/Staging" -n "mlflow-diabetes-model"

mlflow models build-docker -m "models:/YOUR_MODEL_NAME/STAGE" -n "DOCKER_IMAGE_NAME"

Launch container with model

docker run -p 5001:8080 "mlflow-diabetes-model"

Test model in docker

curl -X POST -H "Content-Type:application/json; format=pandas-split" --data '{"columns":["age", "sex", "bmi", "bp", "s1", "s2", "s3", "s4", "s5", "s6"], "data":[[0.0453409833354632, 0.0506801187398187, 0.0606183944448076, 0.0310533436263482, 0.0287020030602135, 0.0473467013092799, 0.0544457590642881, 0.0712099797536354, 0.133598980013008, 0.135611830689079]]}' http://127.0.0.1:5001/invocations