A docker-compose setup to quickly provide MlFlow service with database backend and reverse proxy frontend which can optionally allow for basic authentication.
Originally based on [Guillaume Androz's 10-Jan-2020 Toward-Data-Science post, "Deploy MLflow with docker compose"] (https://towardsdatascience.com/deploy-mlflow-with-docker-compose-8059f16b6039), with some changes to:
- replace AWS usage with local mapping for artifact store
- replace mysql with postgresql and other options.
- optionally apply htpasswd access control to mlflow website via nginx frontend
and overall allowing me to quickly clone to to wherever I'm working whereas the original was just a web article.
There are several docker-compose.yaml files in the compose_variations subdirectory, any of which can be in lieu of the docker-compose.yaml in the root directory to use the desired variation. The docker-compose.yaml file is a copy of compose_variations/docker-compose.mlflow_postgres_nginx.yaml.
Future To-dos:
- Add check for whether the env vars are set in shell (or .env file) before kicking off container - this ia a mistake I comment make myself.
- Add dockerfile ARGS to pass MLFLOW_PORT as well as boolean option to use an htpasswd file into the nginx container. The former really is only useful for the case without nginx reverse proxy. The latter implements steps I've been setting up manually meanwhile, but really only limits scope within an already-firewalled company (it's not secure), mainly to prevent inadvertent deletes by curious browsing colleagues. Finding better way to limit such actions by user (especially deletes/changes) would be ideal.
- As a first step on that last sentence, let's put the postgres image into a new container defined in a subdir after all, and add some lines in its Dockerfile that run PG cmdline routines to add/config an mlflow user so we're not using postgres admin/owner account for mlflow. Then we'd have DBADMIN_USER and DBADMIN_PW as well as MLFLOW_USER and MLFLOW_PW.
First set the following env vars in shell (these are listed in comments at top of the docker-compose.yaml files); set as desired for your own system. These (and the example output below) correspond to the docker-compose.mlflow_postgres_nginx.yaml file in compose_variations; that's what the root dir's docker-compose.yaml file is by default, running mlflow with its backend store in postgresql and its artifact store in a local docker volume. The database is hidden on a backend network, and the mlflow contents are viewable via website or REST API.
export MLFLOW_PORT=5001
export DB_NAME=mlflowdb
export DB_PORT=5432
export DB_USER=postgres
export DB_PW=<somepassword> # (choose an actual pw)(or you can put these into an .env file without the exports.) Warning:
there's a security issue there with putting password in environment variable,
as one can interrogate the Linux process list and/or the Docker inspect
output and see it. But typical use-case here is individual or small-group
usage contained inside a company's internal network behind a firewall, so
not at the top of my concern list. Please beware of use-cases beyond that.
Anyhow, start the containers with:
docker-compose up -d --build (-d for detached mode, --build to build the underlying containers if needed)
The first time will download/build the containers, but after that it will
generate output similar to the below, which can be seen via
docker-compose logs -fwhere the -f acts like in tail -f, allowing open-ended streaming of the
new additions to the logs. Docker-compose logs is like docker logs but
puts the logs from the different containers started by docker-compose all
together. The logs output looks like this:
Creating network "docker_mlflow_db_mydefault" with driver "bridge"
Creating volume "docker_mlflow_db_db_datapg" with default driver
Creating volume "docker_mlflow_db_mlrun_data" with default driver
Building app
Step 1/2 : FROM python:3.7-slim-buster
---> d3fbf7fff365
Step 2/2 : RUN pip install mlflow psycopg2-binary pymysql # boto3
---> Using cache
---> 82425881b36b
Successfully built 82425881b36b
Successfully tagged mlflow_server:latest
Creating mlflow_db ... done
Creating mlflow_server ... done
Attaching to mlflow_db, mlflow_server
mlflow_db | The files belonging to this database system will be owned by user "postgres".
mlflow_db | This user must also own the server process.
mlflow_db |
mlflow_db | The database cluster will be initialized with locale "en_US.utf8".
mlflow_db | The default database encoding has accordingly been set to "UTF8".
mlflow_db | The default text search configuration will be set to "english".
mlflow_db |
mlflow_db | Data page checksums are disabled.
mlflow_db |
mlflow_db | fixing permissions on existing directory /var/lib/postgresql/data ... ok
mlflow_db | creating subdirectories ... ok
mlflow_db | selecting dynamic shared memory implementation ... posix
mlflow_db | selecting default max_connections ... 100
mlflow_db | selecting default shared_buffers ... 128MB
mlflow_db | selecting default time zone ... Etc/UTC
mlflow_db | creating configuration files ... ok
mlflow_db | running bootstrap script ... ok
mlflow_db | performing post-bootstrap initialization ... ok
mlflow_db | syncing data to disk ... ok
mlflow_db |
mlflow_db |
mlflow_db | Success. You can now start the database server using:
mlflow_db |
mlflow_db | pg_ctl -D /var/lib/postgresql/data -l logfile start
mlflow_db |
mlflow_db | initdb: warning: enabling "trust" authentication for local connections
mlflow_db | You can change this by editing pg_hba.conf or using the option -A, or
mlflow_db | --auth-local and --auth-host, the next time you run initdb.
mlflow_db | waiting for server to start....2020-09-03 07:29:43.392 UTC [46] LOG: starting PostgreSQL 12.4 (Debian 12.4-1.pgdg100+1) on x86_64-pc-linux-gnu, compiled by gcc (Debian 8.3.0-6) 8.3.0, 64-bit
mlflow_db | 2020-09-03 07:29:43.394 UTC [46] LOG: listening on Unix socket "/var/run/postgresql/.s.PGSQL.5432"
mlflow_db | 2020-09-03 07:29:43.411 UTC [47] LOG: database system was shut down at 2020-09-03 07:29:43 UTC
mlflow_db | 2020-09-03 07:29:43.421 UTC [46] LOG: database system is ready to accept connections
mlflow_db | done
mlflow_db | server started
mlflow_db | CREATE DATABASE
mlflow_db |
mlflow_db |
mlflow_db | /usr/local/bin/docker-entrypoint.sh: ignoring /docker-entrypoint-initdb.d/*
mlflow_db |
mlflow_db | waiting for server to shut down....2020-09-03 07:29:43.665 UTC [46] LOG: received fast shutdown request
mlflow_db | 2020-09-03 07:29:43.667 UTC [46] LOG: aborting any active transactions
mlflow_db | 2020-09-03 07:29:43.671 UTC [46] LOG: background worker "logical replication launcher" (PID 53) exited with exit code 1
mlflow_db | 2020-09-03 07:29:43.675 UTC [48] LOG: shutting down
mlflow_db | 2020-09-03 07:29:43.690 UTC [46] LOG: database system is shut down
mlflow_db | done
mlflow_db | server stopped
mlflow_db |
mlflow_db | PostgreSQL init process complete; ready for start up.
mlflow_db |
mlflow_db | 2020-09-03 07:29:43.779 UTC [1] LOG: starting PostgreSQL 12.4 (Debian 12.4-1.pgdg100+1) on x86_64-pc-linux-gnu, compiled by gcc (Debian 8.3.0-6) 8.3.0, 64-bit
mlflow_db | 2020-09-03 07:29:43.780 UTC [1] LOG: listening on IPv4 address "0.0.0.0", port 5432
mlflow_db | 2020-09-03 07:29:43.780 UTC [1] LOG: listening on IPv6 address "::", port 5432
mlflow_db | 2020-09-03 07:29:43.783 UTC [1] LOG: listening on Unix socket "/var/run/postgresql/.s.PGSQL.5432"
mlflow_db | 2020-09-03 07:29:43.798 UTC [64] LOG: database system was shut down at 2020-09-03 07:29:43 UTC
mlflow_db | 2020-09-03 07:29:43.807 UTC [1] LOG: database system is ready to accept connections
mlflow_server | 2020/09/03 07:29:44 INFO mlflow.store.db.utils: Creating initial MLflow database tables...
mlflow_server | 2020/09/03 07:29:44 INFO mlflow.store.db.utils: Updating database tables
mlflow_server | INFO [alembic.runtime.migration] Context impl PostgresqlImpl.
mlflow_server | INFO [alembic.runtime.migration] Will assume transactional DDL.
mlflow_server | INFO [alembic.runtime.migration] Running upgrade -> 451aebb31d03, add metric step
mlflow_server | INFO [alembic.runtime.migration] Running upgrade 451aebb31d03 -> 90e64c465722, migrate user column to tags
mlflow_server | INFO [alembic.runtime.migration] Running upgrade 90e64c465722 -> 181f10493468, allow nulls for metric values
mlflow_server | INFO [alembic.runtime.migration] Running upgrade 181f10493468 -> df50e92ffc5e, Add Experiment Tags Table
mlflow_server | INFO [alembic.runtime.migration] Running upgrade df50e92ffc5e -> 7ac759974ad8, Update run tags with larger limit
mlflow_server | INFO [alembic.runtime.migration] Running upgrade 7ac759974ad8 -> 89d4b8295536, create latest metrics table
mlflow_server | INFO [89d4b8295536_create_latest_metrics_table_py] Migration complete!
mlflow_server | INFO [alembic.runtime.migration] Running upgrade 89d4b8295536 -> 2b4d017a5e9b, add model registry tables to db
mlflow_server | INFO [2b4d017a5e9b_add_model_registry_tables_to_db_py] Adding registered_models and model_versions tables to database.
mlflow_server | INFO [2b4d017a5e9b_add_model_registry_tables_to_db_py] Migration complete!
mlflow_server | INFO [alembic.runtime.migration] Running upgrade 2b4d017a5e9b -> cfd24bdc0731, Update run status constraint with killed
mlflow_server | INFO [alembic.runtime.migration] Running upgrade cfd24bdc0731 -> 0a8213491aaa, drop_duplicate_killed_constraint
mlflow_server | INFO [alembic.runtime.migration] Running upgrade 0a8213491aaa -> 728d730b5ebd, add registered model tags table
mlflow_server | INFO [alembic.runtime.migration] Running upgrade 728d730b5ebd -> 27a6a02d2cf1, add model version tags table
mlflow_server | INFO [alembic.runtime.migration] Running upgrade 27a6a02d2cf1 -> 84291f40a231, add run_link to model_version
mlflow_server | INFO [alembic.runtime.migration] Context impl PostgresqlImpl.
mlflow_server | INFO [alembic.runtime.migration] Will assume transactional DDL.
mlflow_server | [2020-09-03 07:29:44 +0000] [8] [INFO] Starting gunicorn 20.0.4
mlflow_server | [2020-09-03 07:29:44 +0000] [8] [INFO] Listening at: http://0.0.0.0:5001 (8)
mlflow_server | [2020-09-03 07:29:44 +0000] [8] [INFO] Using worker: sync
mlflow_server | [2020-09-03 07:29:44 +0000] [11] [INFO] Booting worker with pid: 11
mlflow_server | [2020-09-03 07:29:44 +0000] [13] [INFO] Booting worker with pid: 13
mlflow_server | [2020-09-03 07:29:44 +0000] [14] [INFO] Booting worker with pid: 14
mlflow_server | [2020-09-03 07:29:44 +0000] [15] [INFO] Booting worker with pid: 15We can verify it's all up and ready via:
docker_mlflow_db 00:10:10> docker ps
CONTAINER ID IMAGE COMMAND CREATED STATUS PORTS NAMES
22900b897469 mlflow_server "sh -c 'mlflow serve…" 2 minutes ago Up 2 minutes 0.0.0.0:5001->5001/tcp mlflow_server
8d452ae5dcb5 postgres:latest "docker-entrypoint.s…" 2 minutes ago Up 2 minutes 0.0.0.0:5432->5432/tcp mlflow_dbWhile it's up we can access the MLFlow website via http://localhost:5001 and its database directly via:
make psqld # which simply calls psql within the containerThe makefile also contains macros to start various containers individually, which I mainly use for debugging.
Shut the docker-compose all down via:
docker_mlflow_db 00:33:26> docker-compose down --volumes
Stopping mlflow_server ... done
Stopping mlflow_db ... done
Removing mlflow_server ... done
Removing mlflow_db ... done
Removing network docker_mlflow_db_mydefault
Removing volume docker_mlflow_db_db_datapg
Removing volume docker_mlflow_db_mlrun_datahttps://github.com/ymym3412/mlflow-docker-compose
https://medium.com/vantageai/keeping-your-ml-model-in-shape-with-kafka-airflow-and-mlflow-143d20024ba6
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent