Backend.AI Manager
Project description
Package Structure
ai.backend
manager: Abstraction of agents and computation kernels
gateway: RESTful API gateway based on aiohttp
Installation
Backend.AI Agent requires Python 3.6 or higher. We highly recommend to use pyenv for an isolated setup of custom Python versions that might be different from default installations managed by your OS or Linux distros.
pip install backend.ai-managerTo use optional monitoring service (Datadog and Sentry) supports, add monitor extras tag to the pip command:
pip install 'backend.ai-manager[monitor]'For development
We recommend to use virtual environments in Python. You may share a virtual environment with other Backend.AI projects.
git clone https://github.com/lablup/backend.ai-manager.git
python -m venv /home/user/venv
source /home/user/venv/bin/activate
pip install -U pip setuptools # ensure latest versions
pip install -U -r requirements-dev.txtThe above example shows a standalone installation process for the manager, but normally you would want to install all other depedencies like agents and databases for integration tests.
Running and Deployment
Prepare databases
An RDBMS (PostgreSQL)
An etcd (v3) server
A Redis server
The manager uses the following database IDs
0 (default): to track realtime performance metrics and statistics of computing sessions
1: to track realtime request rate-limits of each API access key
Check out README on the meta-repo for the docker-compose example to run above databases with a single command.
Configuration
You need to specify configuration parameters using either CLI arguments or environment variables. The default values are for development settings so you should set most of them explicitly in production. For details about arguments and their equivalent environment variable names, run the server module with --help.
$ cp alembic.ini.sample alembic.ini
$ edit alembic.ini
$ python -m ai.backend.manager.cli schema oneshot head
Creating tables...
Stamping alembic version to ...Optionally you can populate pre-defined fixtures. You may add your own ones in fixtures directory for deployment. example_keypair fixture is required to run the test suite.
$ python -m ai.backend.manager.cli fixture populate example_keypair
populating fixture 'example_keypair'Running the API gateway server
$ python -m ai.backend.gateway.server \
--etcd-addr localhost:2379 \
--namespace my-cluster \
--redis-addr localhost:6379 \
--db-addr localhost:5432 \
--db-name my-cluster \
--db-user dbuser \
--db-password dbpass \
--docker-registry docker.example.com:5000 \
--service-ip 127.0.0.1 \
--service-port 8080 \
--events-port 5002The gateway server can directly serve the public traffic, either via plain HTTP or HTTPS (with --ssl-cert and --ssl-key options), but we recommend to use a dedicated reverse-proxy such as nginx for advanced HTTPS handling (e.g., SNI). Note that the gateway itself can fully utilize all the CPU cores in the system without limits from GIL (global interpreter lock).
Please check out --help to see more options and their defaults.
Example configs
/etc/supervisor/conf.d/manager.conf:
[program:backend.ai-manager]
user = user
stopsignal = TERM
stopasgroup = true
command = /home/user/run-manager.sh/home/user/run-manager.sh:
#!/bin/sh
source /home/user/venv/bin/activate
# AWS API keypair for S3 file uploads (optional)
export AWS_ACCESS_KEY_ID="..."
export AWS_SECRET_ACCESS_KEY="..."
# Datadog monitoring (optional)
export DATADOG_API_KEY="..."
export DATADOG_APP_KEY="..."
# Sentry monitoring (optional)
export RAVEN_URI="..."
# the main command
exec python -m ai.backend.gateway.server \
--etcd-addr localhost:2379 \
--namespace my-cluster \
--redis-addr localhost:6379 \
# ... other options ...
--service-ip 127.0.0.1 \
--service-port 8080/etc/nginx/sites-enabled/gateway:
ssl_session_cache shared:SSL:10m;
ssl_session_timeout 10m;
ssl_protocols TLSv1 TLSv1.1 TLSv1.2;
ssl_prefer_server_ciphers on;
ssl_ciphers EECDH+CHACHA20:EECDH+AES128:RSA+AES128:EECDH+AES256:RSA+AES256:EECDH+3DES:RSA+3DES:!MD5;
map $http_connection $connection_upgrade {
default upgrade;
'' close;
}
server {
listen 443 ssl;
server_name my-cluster.example.com;
charset utf-8;
client_max_body_size 32M;
ssl_certificate /path/to/ssl.crt
ssl_certificate_key /path/to/ssl.key
add_header Strict-Transport-Security "max-age=31536000; includeSubdomains";
location / {
proxy_pass http://127.0.0.1:8080;
proxy_pass_request_headers on;
proxy_set_header Host "my-cluster.example.com";
proxy_redirect off;
proxy_buffering off;
proxy_read_timeout 600s;
}
location ~ ^/v\d+/stream/ {
proxy_pass http://127.0.0.1:8080;
proxy_pass_request_headers on;
proxy_set_header Host "my-cluster.example.com";
proxy_redirect off;
proxy_buffering off;
proxy_read_timeout 60s;
proxy_http_version 1.1;
proxy_set_header Upgrade $http_upgrade;
proxy_set_header Connection $connection_upgrade;
}
}Networking
The manager and agent should run in the same local network or different networks reachable via VPNs.
You need to check the firewall settings to allow the following access patterns (all ports are TCP):
The manager’s service port: open to the reverse-proxy or the public Internet
The manager’s events port: open to the agents
The etcd’s service port: open to the manager and agents
The redis’ service port: open to the manager and agents
The (optional) private docker registry’s service port: open to the manager and agents
The database’s service port: open to the manager
The agents’ ALL ports: open to the manager
Note that etcd/redis server may run on different physical servers or cloud instances as long as the manager and agents can access them. The PostgreSQL database is only accessed by the manager.
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
File details
Details for the file backend.ai-manager-1.0.2.tar.gz.
File metadata
- Download URL: backend.ai-manager-1.0.2.tar.gz
- Upload date:
- Size: 41.2 kB
- Tags: Source
- Uploaded using Trusted Publishing? No
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 | 7046def9464baa847c12f9a2f2a7db7642315e3c55c79a5489649a3d838ae1a6 |
|
| MD5 | 8f83bba3a728460282e2a438fa6c6928 |
|
| BLAKE2b-256 | cfed679a67e660b5b39a20b12a733e8d02035af23ae16a25cebfdd37b70e8e9a |
Provenance
File details
Details for the file backend.ai_manager-1.0.2-py36-none-any.whl.
File metadata
- Download URL: backend.ai_manager-1.0.2-py36-none-any.whl
- Upload date:
- Size: 53.6 kB
- Tags: Python 3.6
- Uploaded using Trusted Publishing? No
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 | b4f14ed22fb14f436666adf57b2edc98b72f337d2b40cd2d5f4d9468fb7f7135 |
|
| MD5 | 5914b2d5dd2228ee5f60679d7987c0e9 |
|
| BLAKE2b-256 | cafe435334e1577f99551a7a97f0de386bd8392e4733e86309d4e8bf5ff69886 |
