Inventory - pyinfra

Overview

The Inventory is pyinfra’s system for managing target hosts, organizing them into groups, and associating configuration data. It’s the foundation of multi-host deployments and enables flexible, data-driven infrastructure automation.

The Inventory class (defined in src/pyinfra/api/inventory.py) represents a collection of hosts with their associated data and group memberships.

Inventory Structure

The Inventory consists of three key components:

Hosts: Individual target machines
Groups: Collections of hosts sharing properties
Data: Configuration values associated with hosts and groups

# From src/pyinfra/api/inventory.py:25-58
class Inventory:
    """
    Represents a collection of target hosts. Stores and provides access to group data,
    host data and default data for these hosts.

    Args:
        names_data: tuple of (names, data)
        override_data: dictionary of data overrides
        **groups: map of group name -> (names, data)
    """

    state: State
    groups: dict[str, list[Host]]
    hosts: dict[str, Host]
    host_data: dict[str, dict]
    group_data: dict[str, dict]
    data: dict  # Global/default data
    override_data: dict  # Override data

Creating Inventories

Simple Inventory

Create an inventory with a list of hostnames:

from pyinfra.api import Inventory

# List of hosts
hosts = ["web1.example.com", "web2.example.com", "db1.example.com"]

# Create inventory
inventory = Inventory(
    (hosts, {}),  # (names, data)
)

Inventory with Host Data

Associate data with specific hosts:

from pyinfra.api import Inventory

# Hosts with individual data
hosts = [
    ("web1.example.com", {"app_port": 8001}),
    ("web2.example.com", {"app_port": 8002}),
    "db1.example.com",  # No individual data
]

inventory = Inventory(
    (hosts, {}),
)

Inventory with Groups

Organize hosts into groups:

from pyinfra.api import Inventory

inventory = Inventory(
    (["web1.example.com", "web2.example.com", "db1.example.com"], {}),
    webservers=(
        ["web1.example.com", "web2.example.com"],
        {"app_env": "production"},
    ),
    databases=(
        ["db1.example.com"],
        {"db_type": "postgresql"},
    ),
)

Inventory with Global Data

Set default values for all hosts:

from pyinfra.api import Inventory

inventory = Inventory(
    (
        ["web1.example.com", "web2.example.com"],
        {"ssh_user": "deploy", "ssh_port": 22},  # Global data
    ),
)

Host Creation

When the inventory is created, it generates Host objects:

# From src/pyinfra/api/inventory.py:60-146
def make_hosts_and_groups(self, names, groups) -> None:
    all_connectors = get_all_connectors()
    execution_connectors = get_execution_connectors()

    # Map name -> data
    name_to_data: dict[str, dict] = defaultdict(dict)
    # Map name -> group names
    name_to_group_names = defaultdict(list)

    # Process groups first
    for group_name, (group_names, group_data) in groups.items():
        self.group_data[group_name] = group_data

        for name, data in extract_name_data(group_names):
            name_to_data[name].update(data)
            name_to_group_names[name].append(group_name)

    # Process top-level hosts
    for name, data in extract_name_data(names):
        name_to_data[name].update(data)

    # Create Host instances
    hosts: dict[str, Host] = {}

    for name, connector_cls in names_connectors:
        host_groups = name_to_group_names[name]

        host = Host(
            name,
            inventory=self,
            groups=host_groups,
            connector_cls=connector_cls,
        )
        hosts[name] = host

        # Add to groups
        for group_name in host_groups:
            if host not in self.groups[group_name]:
                self.groups[group_name].append(host)

    self.hosts = hosts

Host Object

Each host in the inventory is represented by a Host object (from src/pyinfra/api/host.py):

# From src/pyinfra/api/host.py:103-178
class Host:
    """
    Represents a target host. Thin class that links up to facts and host/group data.
    """

    state: State
    connector_cls: type[BaseConnector]
    connector: BaseConnector
    connected: bool = False

    name: str
    inventory: Inventory
    groups: list[str]
    data: HostData  # Waterfall data access

    def __init__(
        self,
        name: str,
        inventory: Inventory,
        groups: list[str],
        connector_cls=None,
    ):
        self.inventory = inventory
        self.groups = groups
        self.connector_cls = connector_cls or get_execution_connector("ssh")
        self.name = name

        # Create waterfall data: override -> host -> group -> global -> deploy
        self.data = HostData(
            self,
            lambda: inventory.get_override_data(),
            lambda: inventory.get_host_data(name),
            lambda: inventory.get_groups_data(groups),
            lambda: inventory.get_data(),
            self.get_deploy_data,  # @deploy function data
        )

Data Hierarchy

pyinfra uses a waterfall data system where more specific data overrides more general data:

Override Data (highest priority)
    ↓
Host Data
    ↓
Group Data
    ↓
Global Data
    ↓
Deploy Data (lowest priority)

Accessing Host Data

from pyinfra import host

# Access data attributes
app_port = host.data.app_port
ssh_user = host.data.ssh_user

# With defaults
app_port = host.data.get("app_port", 8000)

# Check if data exists
if hasattr(host.data, "app_port"):
    print(f"App port: {host.data.app_port}")

Data Example

from pyinfra.api import Inventory

inventory = Inventory(
    # Global data (applies to all hosts)
    (
        [
            ("web1.example.com", {"app_port": 8001}),  # Host data
            ("web2.example.com", {"app_port": 8002}),
            "db1.example.com",
        ],
        {"ssh_user": "deploy", "ssh_port": 22},  # Global data
    ),
    # Group data
    webservers=(
        ["web1.example.com", "web2.example.com"],
        {"app_env": "production", "app_workers": 4},  # Group data
    ),
)

# For web1.example.com:
# - host.data.app_port = 8001 (host data)
# - host.data.app_env = "production" (group data)
# - host.data.ssh_user = "deploy" (global data)

Inventory Methods

Getting Hosts

from pyinfra.api import Inventory

inventory = Inventory(
    (["web1", "web2", "db1"], {}),
    webservers=(["web1", "web2"], {}),
)

# Get specific host
web1 = inventory.get_host("web1")

# Get all hosts
for host in inventory:
    print(host.name)

# Get host count
print(f"Total hosts: {len(inventory)}")

Getting Groups

# Get hosts in a group
webservers = inventory.get_group("webservers")
for host in webservers:
    print(f"Webserver: {host.name}")

# Check if group exists
if "databases" in inventory.groups:
    db_hosts = inventory.get_group("databases")

Getting Data

# Get global data
global_data = inventory.get_data()

# Get host data
host_data = inventory.get_host_data("web1")

# Get group data
group_data = inventory.get_group_data("webservers")

# Get combined group data
groups_data = inventory.get_groups_data(["webservers", "production"])

Inventory Connectors

pyinfra supports connector-based inventory sources:

SSH Connector (Default)

# Explicit SSH hosts
hosts = [
    "@ssh/web1.example.com",
    "@ssh/web2.example.com",
]

inventory = Inventory((hosts, {}))

Local Connector

# Execute on local machine
hosts = ["@local"]

inventory = Inventory((hosts, {}))

Docker Connector

# Target Docker containers
hosts = [
    "@docker/web_container",
    "@docker/db_container",
]

inventory = Inventory((hosts, {}))

Dynamic Connectors

Connectors can generate multiple hosts:

# From src/pyinfra/api/inventory.py:92-113
if name[0] == "@":
    connector_name = name[1:]
    arg_string = None

    if "/" in connector_name:
        connector_name, arg_string = connector_name.split("/", 1)

    if connector_name not in get_all_connectors():
        raise NoConnectorError(f"Invalid connector: {connector_name}")

    # Connector can expand to multiple hosts
    names_data = all_connectors[connector_name].make_names_data(arg_string)
    
    # Each connector-generated host gets its own Host instance
    for sub_name, sub_data, sub_groups in names_data:
        # Create host with connector data
        ...

Example: Terraform connector generates hosts from Terraform state:

hosts = ["@terraform"]
inventory = Inventory((hosts, {}))
# Automatically discovers hosts from terraform.tfstate

CLI Inventory

When using pyinfra from the command line, inventory is specified differently:

Inventory File

# inventory.py
hosts = ["web1.example.com", "web2.example.com"]

webservers = ["web1.example.com", "web2.example.com"]
databases = ["db1.example.com"]

Usage:

pyinfra inventory.py deploy.py

Inline Hosts

# Single host
pyinfra web1.example.com deploy.py

# Multiple hosts
pyinfra web1.example.com,web2.example.com deploy.py

# With connector
pyinfra @docker/my_container deploy.py

Inventory + Data Files

# inventory.py
hosts = ["web1", "web2", "db1"]

webservers = ["web1", "web2"]

# group_data/webservers.py
app_env = "production"
app_workers = 4

# host_data/web1.py
app_port = 8001

Limiting Hosts

Run operations on a subset of inventory:

# Limit to specific hosts
pyinfra inventory.py deploy.py --limit web1.example.com

# Limit to a group
pyinfra inventory.py deploy.py --limit webservers

# Multiple limits
pyinfra inventory.py deploy.py --limit web1,web2

In code:

from pyinfra.api import State, Inventory

inventory = Inventory((["web1", "web2", "db1"], {}))
state = State()
state.init(inventory, config)

# Limit to specific hosts
web1 = inventory.get_host("web1")
state.limit_hosts = [web1]

# Operations will only run on web1

Active vs Activated Hosts

# From src/pyinfra/api/state.py:254-280
class State:
    # Hosts we've activated at any time
    activated_hosts: set[Host]
    
    # Active hosts that *haven't* failed yet
    active_hosts: set[Host]
    
    # Hosts that have failed
    failed_hosts: set[Host]

# Get active hosts (not failed)
active = inventory.get_active_hosts()

# Get all activated hosts (including failed)
activated = list(inventory.iter_activated_hosts())

Inventory Patterns

Environment-Based Inventory

# inventory/production.py
hosts = [
    ("web1.prod.example.com", {"env": "production"}),
    ("web2.prod.example.com", {"env": "production"}),
]

# inventory/staging.py  
hosts = [
    ("web1.staging.example.com", {"env": "staging"}),
]

pyinfra inventory/production.py deploy.py
pyinfra inventory/staging.py deploy.py

Role-Based Groups

inventory = Inventory(
    (["host1", "host2", "host3", "host4"], {}),
    webservers=(["host1", "host2"], {}),
    appservers=(["host2", "host3"], {}),  # host2 in multiple groups
    databases=(["host4"], {}),
)

Dynamic Inventory

import boto3
from pyinfra.api import Inventory

# Fetch EC2 instances
ec2 = boto3.client('ec2')
response = ec2.describe_instances(Filters=[{'Name': 'tag:App', 'Values': ['web']}])

hosts = []
for reservation in response['Reservations']:
    for instance in reservation['Instances']:
        hosts.append((
            instance['PublicDnsName'],
            {
                'instance_id': instance['InstanceId'],
                'instance_type': instance['InstanceType'],
            },
        ))

inventory = Inventory((hosts, {'ssh_user': 'ec2-user'}))

Conditional Deployment

from pyinfra import host
from pyinfra.operations import apt

# Only run on hosts in production group
if "production" in host.groups:
    apt.packages(
        name="Install production packages",
        packages=["nginx", "postgresql"],
    )

# Only run on specific hosts
if host.name == "web1.example.com":
    apt.packages(
        name="Install monitoring agent",
        packages=["datadog-agent"],
    )

Host Data Patterns

Port Mapping

inventory = Inventory(
    (
        [
            ("web1", {"app_port": 8001}),
            ("web2", {"app_port": 8002}),
            ("web3", {"app_port": 8003}),
        ],
        {},
    ),
)

# In deploy
from pyinfra import host
from pyinfra.operations import server

server.shell(
    name="Start application",
    commands=[f"./start.sh --port {host.data.app_port}"],
)

Connection Configuration

inventory = Inventory(
    (
        [
            ("bastion", {"ssh_port": 22, "ssh_user": "admin"}),
            ("internal1", {"ssh_port": 2222, "ssh_user": "deploy"}),
        ],
        {"ssh_key": "~/.ssh/id_rsa"},  # Global default
    ),
)

Application Configuration

inventory = Inventory(
    (["web1", "web2"], {}),
    webservers=(
        ["web1", "web2"],
        {
            "app_env": "production",
            "app_workers": 4,
            "app_threads": 2,
            "app_timeout": 30,
            "db_host": "db.internal",
            "db_port": 5432,
        },
    ),
)

# Access in deploy
from pyinfra import host

db_url = f"postgresql://{host.data.db_host}:{host.data.db_port}/app"

Override Data

Override inventory data from the command line:

pyinfra inventory.py deploy.py --data ssh_user=root --data app_port=9000

In code:

inventory = Inventory(
    (["web1", "web2"], {}),
    override_data={"ssh_user": "root", "app_port": 9000},
)

Best Practices

Organize by Role

Group hosts by their function (webservers, databases, etc.) for targeted deployments.

Use Data Hierarchy

Place common config in global data, role-specific in groups, and unique values in host data.

Environment Separation

Use separate inventory files for production, staging, and development environments.

Descriptive Names

Use clear, consistent naming for hosts and groups (e.g., web1.prod not server23).

Document Data Keys

Document expected data keys and their purposes for your deployments.

Version Control

Keep inventory files in version control to track infrastructure changes.

Testing Inventory

import pytest
from pyinfra.api import Inventory

def test_inventory_structure():
    inventory = Inventory(
        (["web1", "web2", "db1"], {"ssh_user": "deploy"}),
        webservers=(["web1", "web2"], {"app_env": "prod"}),
    )
    
    # Test host count
    assert len(inventory) == 3
    
    # Test group membership
    webservers = inventory.get_group("webservers")
    assert len(webservers) == 2
    
    # Test data hierarchy
    web1 = inventory.get_host("web1")
    assert web1.data.ssh_user == "deploy"  # Global data
    assert web1.data.app_env == "prod"  # Group data

Host

Learn about the Host class and host-specific operations

Connectors

Understand how connectors enable different inventory sources

State

See how State manages active and failed hosts

CLI Usage

Learn command-line inventory options

Get Started

Core Concepts

Guides

CLI Reference

Advanced

​Overview

​Inventory Structure

​Creating Inventories

​Simple Inventory

​Inventory with Host Data

​Inventory with Groups

​Inventory with Global Data

​Host Creation

​Host Object

​Data Hierarchy

​Accessing Host Data

​Data Example

​Inventory Methods

​Getting Hosts

​Getting Groups

​Getting Data

​Inventory Connectors

​SSH Connector (Default)

​Local Connector

​Docker Connector

​Dynamic Connectors

​CLI Inventory

​Inventory File

​Inline Hosts

​Inventory + Data Files

​Limiting Hosts

​Active vs Activated Hosts

​Inventory Patterns

​Environment-Based Inventory

​Role-Based Groups

​Dynamic Inventory

​Conditional Deployment

​Host Data Patterns

​Port Mapping

​Connection Configuration

​Application Configuration

​Override Data

​Best Practices

Organize by Role

Use Data Hierarchy

Environment Separation

Descriptive Names

Document Data Keys

Version Control

​Testing Inventory

​Related Concepts

Host

Connectors

State

CLI Usage

Build docs developers (and LLMs) love

Overview

Inventory Structure

Creating Inventories

Simple Inventory

Inventory with Host Data

Inventory with Groups

Inventory with Global Data

Host Creation

Host Object

Data Hierarchy

Accessing Host Data

Data Example

Inventory Methods

Getting Hosts

Getting Groups

Getting Data

Inventory Connectors

SSH Connector (Default)

Local Connector

Docker Connector

Dynamic Connectors

CLI Inventory

Inventory File

Inline Hosts

Inventory + Data Files

Limiting Hosts

Active vs Activated Hosts

Inventory Patterns

Environment-Based Inventory

Role-Based Groups

Dynamic Inventory

Conditional Deployment

Host Data Patterns

Port Mapping

Connection Configuration

Application Configuration

Override Data

Best Practices

Testing Inventory

Related Concepts