Configuration

Здесь описаны различные опции для конфигурации MLPanel

Description of environment variables (.env)

Примеры конфигурационных файлов можно найти в разделе Install

General

Required

Default

Description

WORKSPACE

Yes

No

Absolute Path to a workspace folder - any path on you disk (taking into account permission).

The folder contains:

  • projects database;

  • deploy database;

  • folders for each project, each folder is named by the project id and contains:

    • mlflow.db - mlflow database;

    • artifact storage folder - if variable ARTIFACT_STORE points to local path.

Note: if artifacts are stored locally then folder WORKSPACE can have huge size, it depends on number of projects and artifacts (especially ML models) volume.

LOGLEVEL

No

INFO

Logging level in service projects.

Возможные значения соответствуют уровням логирования принятым в Python:

  • DEBUG

  • INFO

  • WARNING

  • WARN

  • ERROR

  • FATAL

  • CRITICAL

HOST_IP

No

0.0.0.0

Defines ip address for tracking servers

Projects

Required

Default

Description

ARTIFACT_STORE

Yes

mlruns

Path to artifact store. It can be:

  • Default:$WORKSPACE/<project_folder>/mlruns

  • Name of store folder in workspace $WORKSPACE/<project_folder>/<store_name>

  • Google Storage bucket in format gs://<bucket_name>; this case requires to define variable GOOGLE_APPLICATION_CREDENTIALS.

TRACKING_SERVER_PORTS

Yes

'‌5000-5020'

Range of ports for tracking servers. It also defines how many projects can be created! One port corresponds to one project (tracking server).

TRACKING_SERVER_WORKERS

No

1

Number of gunicorn workers for MLflow tracking servers. Details on MLflow workers number

Database

Required

Default

Description

PROJECTS_DB_NAME

Yes

No

Database name of service projects

DEPLOY_DB_NAME

Yes

No

Database name of service deploy

POSTGRES_USER

Yes

user

PostgresSQL user name.

POSTGRES_PASSWORD

Yes

passw

PostgresSQL password

DB_HOST

Yes*

db

Database host in mlpanel docker network

*required value = db, by name of database service

DB_PORT

Yes

5432

Database port

Deploy

Now two types of deployments exist:

  • local - deploys model locally (run new MLflow model server process inside a container)

  • gcp - start new GCE instance and run MLflow model server process there

Required

Default

Description

DEPLOY_SERVER_WORKERS

No

1

Number of gunicorn workers for MLflow model servers. Details on serve workers number

GOOGLE_APPLICATION_CREDENTIALS

No*

No

Defines path to your Google credentials JSON inside docker container.

Consists from two parts: fixed (/home/config/ - exact path of folder config inside container) and name of your JSON. The JSON you must put to folder config/.

Note: Required to use Google Cloud Storage (GCP) to store experiment artifacts and/or deploy models on Google Compute Engine (GCE).

GCP_PROJECT

No*

No

ID of your project on GCP.

Note: Required to deploy models on GCE.

GCP_ZONE

No*

No

GCP instance zone type.

Note: Required to deploy models on GCE.

GCP_MACHINE_TYPE

No*

No

GCE machine type. Minimum requirement of machine type is g1-small.

Note: Required to deploy models on GCE.

GCP_OS_IMAGE

No*

No

Your private (custom) OS image. Docker must be installed inside OS image. More on working with OS images

GCP_BUCKET

No*

No

Google Storage bucket name

MODEL_DEPLOY_DOCKER_IMAGE

No

No

Name of docker image which provides running of deployment container on GCE instance. MLflow must be installed in the image

Note: example

MODEL_DEPLOY_DEFAULT_PORT

No*

No

Port on which deployments are available.

Note: Required to deploy models on GCE. All GCP deployment have the same port, but theirs external IP addresses are different.

MODEL_DEPLOY_FIREWALL_RULE

No*

No

Name of firewall rule. Firewall rule is required to open some port(s) for communication with services running on GCE instances. More about firewall

Data Validation

Required

Default

Description

VALIDATE_ON_PREDICT

No

false

Turn on/off data validation on predict. Values true or false.

BIG_DATASET_MIN_SIZE

No

10e7

Minimum dataset rows number when full data validation (using TFDV) is performed. When dataset has less rows then only column names and types are checked.

CHECK_INTERVALS

No

false

If true then numeric data intervals will be checked on predict.

Values: true or false.

Authorization

Required

Default

Description

AUTH_REQUIRED

No

false

Turns on/off authorization: if true authorization is required.

Values: true or false.

AUTH_SERVER_WORKERS

Yes

-

Number of gunicorn workers for authorization service

AUTH0_DOMAIN

Yes

-

Auth0 domain string - tenant name. See details

AUTH0_API_IDENTIFIER

Yes

passw

Auth0 API identifier. See details

SSL settings

When you use self-signed ssl sertificate your browser throws warning. Working in browser you can just agree risk and continue connection. But to avoid problems in Swagger UI it's recommended to allow your browser untrusterd sertificate from localhost.

Google Chrome:

Open in address bar chrome://flags#allow-insecure-localhos and select Enabled in list box.

Mozilla Firefox:

  • in address bar type about:config;

  • search browser.ssl_override_behavior;

  • change value from 2 to 1.

Opera:

Like for Google Chrome: open opera://flags/unsafely-treat-insecure-origin-as-secure and select Enabled

Create account into Auth0

  1. go to Auth0;

  2. create account - then you'll give tenant name - name of third level domain on auth0:

  1. read Get started;

  2. go to Dashboard;

  3. create:

    • new Machine to Machine application;

    • new API;

  4. turn on authorization your API by the application in tab "Machine to Machine Applications " in the API page.

Generate openssl

PreviousConcepts and ArchitectureNextContributing

Last updated

Was this helpful?