Python Server SDK

Set up the SDK

1. Install the SDK

   Installation

   shell

   Copy

   pip install statsig-python-core
   

2. Initialize the SDK

   After installation, initialize the SDK using a Server Secret Key from the Statsig console.

   Keep Server Secret Keys private. If you expose one, you can disable and recreate it in the Statsig console.

   An optional options parameter accepts a StatsigOptions object to customize the SDK.

   python

   Copy

   from statsig_python_core import Statsig, StatsigOptions # note, import statement has underscores while install has dashes
   
   options = StatsigOptions()
   options.environment = "development"
   
   statsig = Statsig("secret-key", options)
   statsig.initialize().wait()
   
   # If you're running this in a script, be sure to wait for shutdown at the end to flush event logs to statsig
   statsig.shutdown().wait()
   

   initialize performs a network request. After initialize completes, virtually all SDK operations are synchronous (refer to Evaluating Feature Gates in the Statsig SDK). The SDK fetches updates from Statsig in the background independently of API calls.

   Python Core SDK and prefork workers

   Python-specific prefork support

   The Python Core SDK provides the prefork lifecycle integration described here. Other Server Core language SDKs share underlying Rust recovery logic, but don't provide the same lifecycle integration or establish support for initializing before a fork. For other language bindings, initialize the SDK separately in each child process after forking unless that SDK explicitly documents prefork support.

   Python Core SDK 0.24.0 and later includes recovery for initialized SDK instances when a Unix process forks. The fork callback marks recovery as pending; background work resumes later at a safe child lifecycle point.

   For earlier Python Core versions, initialize the SDK separately in each child process after forking.

   Celery prefork on Linux

   For Linux Celery workers using --pool=prefork, the Python Core SDK registers a handler for Celery's worker_process_init signal. Import Celery before importing the SDK in the parent process so the SDK can register the handler. Initialize the SDK in the parent before Celery creates its worker pool. You don't need to register another signal handler. Recovery also runs when Celery recycles workers.

   Install statsig-python-core 0.24.0 or later and celery[redis]. Run a Redis broker accessible to the worker. Set CELERY_BROKER_URL to its URL and STATSIG_SERVER_SECRET_KEY to your project's server secret key. Then create worker.py:

   python

   Copy

   import os
   from celery import Celery
   from statsig_python_core import Statsig, StatsigUser
   
   app = Celery("worker", broker=os.environ["CELERY_BROKER_URL"])
   statsig = Statsig(os.environ["STATSIG_SERVER_SECRET_KEY"])
   statsig.initialize().wait()
   
   @app.task
   def check_example_gate(user_id):
       return statsig.check_gate(StatsigUser(user_id), "example_gate")
   

   Replace example_gate with a gate in your project. Start the worker with celery -A worker worker --pool=prefork --concurrency=2 --loglevel=INFO. In another terminal with the same environment, submit a task:

   bash

   Copy

   celery -A worker call worker.check_example_gate --args '["example-user"]'
   

   Check the worker log for the task result. The celery -A worker command also imports worker.py and initializes a separate SDK instance in the submitting process. A False result can be a valid gate evaluation. It doesn't by itself indicate a recovery failure.

   Recovery schedules background work. It doesn't guarantee a network refresh has completed before the first task runs. The child doesn't resend events queued in the parent before the fork.

   Other Unix prefork hosts

   For another Unix prefork host, call recover_after_fork once per child from the framework's child-ready callback. Call it after child initialization and before processing work:

   python

   Copy

   from statsig_python_core import recover_after_fork
   
   def on_worker_child_ready():
       if not recover_after_fork():
           # Report the failed recovery through your worker startup diagnostics.
           # Later SDK operations can retry recovery.
           pass
   

   The function covers all live ordinary and shared SDK instances. Don't call it from a raw os.register_at_fork(after_in_child=...) callback. A True return means the SDK scheduled or re-armed recovery within its shared callback budget. It doesn't mean fresh configuration has arrived. A False return means recovery didn't finish scheduling for every instance. Record the failure and allow later SDK operations to retry.

   Statsig has integration coverage for Linux Celery prefork. This guidance doesn't establish support for macOS child forks, Windows, spawn, forkserver, or other process managers.

   Initializing with WSGI servers

   For production deployments using WSGI servers like uWSGI or Gunicorn, initialize Statsig after the worker processes are forked, not in the main process.

   uWSGIGunicornFastAPI

   ✅ Correct: uWSGI example

   python

   Copy

   # app.py
   from statsig_python_core import Statsig, StatsigOptions
   from flask import Flask
   
   app = Flask(__name__)
   statsig = None
   
   def init_statsig():
       global statsig
       if statsig is None:
           options = StatsigOptions()
           options.environment = "production"
           statsig = Statsig("your-server-secret-key", options)
           statsig.initialize().wait()
   
   # Initialize in each worker process
   @app.before_first_request
   def before_first_request():
       init_statsig()
   
   @app.route('/')
   def index():
       # Use statsig here
       return "Hello World"
   

   ini

   Copy

   # uwsgi.ini
   [uwsgi]
   module = app:app
   master = true
   processes = 4
   # Statsig will be initialized in each worker process
   

   initialize performs a network request. After initialize completes, virtually all SDK operations are synchronous (refer to Evaluating Feature Gates in the Statsig SDK). The SDK fetches updates from Statsig in the background independently of API calls.

Working with the SDK

Checking a feature flag/gate

After you initialize the SDK, you can fetch a Feature Gate. Feature Gates create logic branches in code that you can roll out to different users from the Statsig Console. Gates are always CLOSED or OFF (equivalent to return false;) by default.

All APIs require you to specify the user (refer to Statsig user) associated with the request. For example:

user = StatsigUser("a-user")

if statsig.check_gate(user, "a_gate"):
    # Gate is on, enable new feature
else:
    # Gate is off

Reading a dynamic config

Feature Gates are useful for simple on/off switches with optional user targeting. To send a different set of values (strings, numbers, and similar types) to clients based on user attributes such as country, use Dynamic Configs. The API is similar to Feature Gates, but returns a full JSON object configurable on the server from which you can fetch typed parameters.

# Get a dynamic config for a specific user
config = statsig.get_dynamic_config(StatsigUser("my_user"), "a_config")

# Access config values with type-safe getters and fallback values
product_name = config.get_string("product_name", "Awesome Product v1")  # returns String
price = config.get_float("price", 10.0)  # returns float
should_discount = config.get_bool("discount", False)  # returns bool
quantity = config.get_integer("quantity", 1)  # returns int64

# Advanced Usage:
# You can disable exposure logging for this specific check
options = DynamicConfigEvaluationOptions(disable_exposure_logging=True)
config = statsig.get_dynamic_config(user, "a_config", options)

# The config object also provides metadata about the evaluation
print(config.rule_id)  # The ID of the rule that served this config
print(config.id_type)  # The type of the evaluation (experiment, config, etc)

The get_dynamic_config() method returns a DynamicConfig object that allows you to:

• Fetch typed values with fallback defaults using get_string(), get_float(), get_boolean(), and get_integer()
• Access evaluation metadata through properties like rule_id and id_type
• Configure evaluation behavior using DynamicConfigEvaluationOptions

By default, Statsig automatically logs exposures when it evaluates configs. You can disable exposure logging for specific checks using the evaluation options.

Getting a layer/experiment

Use Layers/Experiments to run A/B/n experiments. Two APIs are available. Statsig recommends Layers because they make parameters reusable and support mutually exclusive experiments.

# Values via get_layer
layer = statsig.get_layer(StatsigUser("my_user"), "user_promo_experiments")
title = layer.get_string("title", "Welcome to Statsig!")
discount = layer.get_float("discount", 0.1)

# Via get_experiment
title_exp = statsig.get_experiment(StatsigUser("my_user"), "new_user_promo_title")
price_exp = statsig.get_experiment(StatsigUser("my_user"), "new_user_promo_price")

title = title_exp.get_string("title", "Welcome to Statsig!")
discount = price_exp.get_float("discount", 0.1)

Retrieving feature gate metadata

In certain scenarios, you may need more information about a gate evaluation than a boolean value. For additional metadata about the evaluation, use the Get Feature Gate API, which returns a FeatureGate object:

gate = statsig.get_feature_gate(user, "example_gate")
print(gate.rule_id)
print(gate.value)

Parameter stores

Use Parameter Stores when you want to define a parameter without deciding whether it should be a Feature Gate, Experiment, or Dynamic Config. Parameter Stores let you change the parameter type at any point in the Statsig console without a new deployment. Parameter Stores are optional, but parameterizing your application provides future flexibility and allows non-technical Statsig users to turn parameters into experiments.

# Get a Parameter Store by name
param_store = statsig.get_parameter_store(user, "my_parameter_store")

Retrieving parameter values

Parameter Store provides methods for retrieving values of different types with fallback defaults.

# String parameters
string_value = param_store.get_string("string_param", "default_value")

# Boolean parameters
bool_value = param_store.get_bool("bool_param", False)

# Numeric parameters
float_value = param_store.get_float("float_param", 0.0)
integer_value = param_store.get_integer("integer_param", 0)

# Complex parameters
default_array = ["item1", "item2"]
array_value = param_store.get_array("array_param", default_array)

default_map = {"key": "value"}
map_value = param_store.get_map("map_param", default_map)

Evaluation options

You can disable exposure logging when retrieving a parameter store:

from statsig_python_core import ParameterStoreEvaluationOptions

options = ParameterStoreEvaluationOptions(disable_exposure_logging=True)
param_store = statsig.get_parameter_store(user, "my_parameter_store", options)

Logging an event

To track custom events, call the Log Event API. Specify the user, event name, and an optional value or metadata object:

statsig.log_event(
    user=StatsigUser("user_id"),  # Replace with your user object
    event_name="add_to_cart",
    value="SKU_12345",
    metadata={
        "price": "9.99",
        "item_name": "diet_coke_48_pack"
    }
)

Sending events to Log Explorer

You can forward logs to Logs Explorer for convenient analysis using the Forward Log Line Event API. This lets you include custom metadata and event values with each log.

user = StatsigUser(
    user_id="a-user",
    custom={
        "service": "my-service",
        "pod": "my-pod",
        "namespace": "my-namespace",
        "container": "my-container",
        # ...include any service-specific metadata
    }
)

# levels: trace, debug, info, log, warn, error
statsig.forward_log_line_event(user, "warn", "script failed to load", {
    "custom_metadata": "script_name:my-script"
    # ... include any event-specific metadata
})

Using shared instance

To create a single Statsig instance accessible globally throughout your codebase, use the shared instance functionality, which provides a singleton pattern:

# Create a shared instance that can be accessed globally
statsig = Statsig.new_shared("secret-key", options)
statsig.initialize().wait()

# Access the shared instance from anywhere in your code
shared_statsig = Statsig.shared()
is_feature_enabled = shared_statsig.check_gate(StatsigUser("user_id"), "feature_name")

# Check if a shared instance exists
if Statsig.has_shared_instance():
    # Use the shared instance
    pass

# Remove the shared instance when no longer needed
Statsig.remove_shared()

The shared instance helps when multiple parts of your codebase need Statsig without passing an instance between them.

• Statsig.new_shared(sdk_key, options): Creates a new shared instance of Statsig that you can access globally
• Statsig.shared(): Returns the shared instance
• Statsig.has_shared_instance(): Checks if a shared instance exists (useful when the shared instance may not be ready yet)
• Statsig.remove_shared(): Removes the shared instance (useful when you want to switch to a new shared instance)

Manual exposures

By default, the SDK automatically logs an exposure event when you check a gate, get a config, get an experiment, or call get() on a parameter in a layer. To delay exposure logging (for example, to log only after the user actually uses the feature), use manual exposures.

All main SDK functions (check_gate, get_dynamic_config, get_experiment, get_layer) accept an optional disable_exposure_logging parameter. When set to True, the SDK doesn't automatically log an exposure. You can then log the exposure manually at a later time:

result = statsig.check_gate(aUser, 'a_gate_name', FeatureGateEvaluationOptions(disable_exposure_logging=True))

statsig.manually_log_gate_exposure(aUser, 'a_gate_name')

Statsig user

The StatsigUser object represents a user in Statsig. You must provide a userID or at least one of the customIDs to identify the user.

When calling APIs that require a user, pass as much information as possible. Complete user information lets you take advantage of advanced gate and config conditions (like country or OS/browser-level checks) and correctly measure the impact of experiments on metrics and events. As explained in why an ID is always required for server SDKs, you must provide at least one identifier (userID or customID) to ensure a consistent experience for a given user.

In addition to userID, the top-level fields on StatsigUser are: email, ip, userAgent, country, locale, and appVersion. You can also pass any key-value pairs in an object or dictionary to the custom field and create targeting based on them.

Private attributes

Private attributes are user attributes that Statsig uses for evaluation but doesn't forward to any integrations. They are useful for PII or sensitive data that you don't want to send to third-party services.

from statsig_python_core import StatsigUser

user = StatsigUser(
    user_id="a-user-id",
    email="user@example.com",
    ip="192.168.1.1",
    user_agent="Mozilla/5.0...",
    country="US",
    locale="en_US",
    app_version="1.0.0",
    custom={
        # Custom fields
        "plan": "premium",
        "age": 25
    },
    custom_ids={
        # Custom ID types
        "stable_id": "stable-id-123"
    },
    private_attributes={
        # Private attributes not forwarded to integrations
        "email": "private@example.com"
    }
)

Statsig options

You can pass an optional options parameter in addition to sdkKey during initialization to customize the Statsig client.

Proxy and custom network routing

Use proxy_config if your service needs a standard outbound HTTP proxy. Use spec_adapter_configs if you need to route spec downloads through Statsig Forward Proxy or another custom spec source.

from statsig_python_core import ProxyConfig, Statsig, StatsigOptions

proxy_config = ProxyConfig(
    proxy_host="proxy.example.com",
    proxy_port=8080,
    proxy_protocol="https",
    ca_cert_path="/etc/ssl/certs/corporate-ca.pem",  # Optional
)

options = StatsigOptions()
options.proxy_config = proxy_config

statsig = Statsig("secret-key", options)
statsig.initialize().wait()

Set ca_cert_path when your environment requires a custom PEM CA bundle for outbound TLS.

Using spec_adapter_configs with multiple sources

from statsig_python_core import StatsigOptions, SpecAdapterConfig

# Configure multiple spec adapters with priority order
# First source: Statsig CDN
primary_adapter = SpecAdapterConfig(
    adapter_type="http",
    specs_url="https://api.statsigcdn.com/v2/download_config_specs",
    init_timeout_ms=3000
)

# Second source: Data Store adapter
# The SDK will try this source if the primary source fails
data_store_adapter = SpecAdapterConfig(
    adapter_type="data_store",
    init_timeout_ms=5000
)

options = StatsigOptions()
options.spec_adapter_configs = [primary_adapter, data_store_adapter]
options.environment = "production"

statsig = Statsig("secret-key", options)
statsig.initialize().wait()

Shutting Statsig down

Statsig batches and periodically flushes events. To flush all logged events before shutdown, call shutdown() before your app or server shuts down:

statsig.shutdown().wait()

Local overrides

Local Overrides let you override the values of gates, configs, experiments, and layers for testing. This is useful for local development or testing when you want to force a specific value without changing the configuration in the Statsig console.

# Overrides the given gate to the specified value
statsig.override_gate("a_gate_name", True)

# Overrides the given dynamic config to the provided value
statsig.override_dynamic_config("a_config_name", {"key": "value"})

# Overrides the given experiment to the provided value
statsig.override_experiment("an_experiment_name", {"key": "value"})

# Overrides the given layer to the provided value
statsig.override_layer("a_layer_name", {"key": "value"})

# Overrides the given experiment to a particular groupname, available for experiments only:
statsig.override_experiment_by_group_name("an_experiment_name", "a_group_name")

Client SDK bootstrapping | SSR

If you use the Statsig client SDK in a browser or mobile app, you can bootstrap the client SDK with values from the server SDK to avoid a network request on the client. This is useful for server-side rendering (SSR) or to reduce network requests on the client.

Client initialize response

The Python Core SDK provides a method to generate a client initialize response that you can use to bootstrap client SDKs without requiring network requests.

import json
from statsig_python_core import Statsig, StatsigUser

# Get client initialize response for a user
response_data = statsig.get_client_initialize_response(user)
response = json.loads(response_data)

# Pass response to a client SDK to initialize without a network request

Persistent storage

The Persistent Storage interface lets you implement custom storage for user-specific configurations. This enables you to persist user assignments across sessions, ensuring consistent experiment groups when a user returns. This is useful for client-side A/B testing where users must always receive the same variant.

class PersistentStorage(PersistentStorageBaseClass):
    def __init__():
        # When you initialize, remember to call super.__init__()
        super().__init__()
        self.load_fn = self.load
        self.save_fn = self.save
        self.delete_fn = self.delete

    def load(self, key: str) -> Optional[UserPersistedValues]:
        """
        Load persisted values for a user from storage

        Args:
            key: A string key that uniquely identifies a user

        Returns:
            Dictionary mapping config names to their persisted values
        """
        pass

    def save(self, key: str, config_name: str, data: StickyValues):
        """
        Save a persistent value for a user

        Args:
            key: A string key that uniquely identifies a user
            config_name: The name of the config/experiment
            data: The values to persist
        """
        pass

    def delete(self, key: str, config_name: str):
        """
        Delete a persistent value for a user

        Args:
            key: A string key that uniquely identifies a user
            config_name: The name of the config/experiment to delete
        """
        pass

Data store

The Data Store interface lets you implement custom storage for Statsig configurations, enabling advanced caching strategies and integration with your preferred storage systems.

class DataStore(DataStoreBase):
    def initialize(self):
        """
        Initialize the data store. Called when the Statsig client initializes.
        """
        pass

    def shutdown(self):
        """
        Clean up resources when the Statsig client shuts down.
        """
        pass

    def get(self, key: str) -> Optional[DataStoreResponse]:
        """
        Retrieve value from the data store.

        Args:
            key: The key to retrieve the value for

        Returns:
            DataStoreResponse containing the result and time
        """
        pass

    def set(self, key: str, value: str, time: Optional[int] = None):
        """
        Store a value in the data store.

        Args:
            key: The key to store the value under
            value: The value to store
            time: Optional timestamp
        """
        pass

    def support_polling_updates_for(self, key: str) -> bool:
        """
        Whether the data store supports polling for updates for the given key.

        Args:
            key: The key to check

        Returns:
            True if polling is supported, False otherwise
        """
        return False

Custom output logger

The Output Logger Provider interface lets you customize how the SDK logs internal messages.

class OutputLoggerProvider(OutputLoggerProviderBase):
    def init(self):
        """
        Initialize the logger. Called when the Statsig client initializes.
        """
        pass

    def debug(self, tag: str, msg: str):
        """
        Log a debug message.

        Args:
            tag: Category/component tag for the message
            msg: The message to log
        """
        pass

    def info(self, tag: str, msg: str):
        """
        Log an info message.

        Args:
            tag: Category/component tag for the message
            msg: The message to log
        """
        pass

    def warn(self, tag: str, msg: str):
        """
        Log a warning message.

        Args:
            tag: Category/component tag for the message
            msg: The message to log
        """
        pass

    def error(self, tag: str, msg: str):
        """
        Log an error message.

        Args:
            tag: Category/component tag for the message
            msg: The message to log
        """
        pass

    def shutdown(self):
        """
        Clean up resources when the Statsig client shuts down.
        """
        pass

Observability client

The Observability Client interface lets you monitor the health of the SDK by integrating with your own observability systems. This enables you to track metrics, errors, and performance data. For more information on the metrics emitted by Statsig SDKs, refer to the Monitoring documentation.

class ObservabilityClient(ObservabilityClientBase):
    def init(self):
        """
        Initialize the observability client. Called when the Statsig client initializes.
        """
        pass

    def increment(self, metric_name: str, value: float, tags: Optional[Dict[str, str]] = None):
        """
        Report a counter metric.

        Args:
            metric_name: The name of the metric
            value: The amount to increment by
            tags: Optional tags to associate with the metric
        """
        pass

    def gauge(self, metric_name: str, value: float, tags: Optional[Dict[str, str]] = None):
        """
        Report a gauge metric.

        Args:
            metric_name: The name of the metric
            value: The current value
            tags: Optional tags to associate with the metric
        """
        pass

    def dist(self, metric_name: str, value: float, tags: Optional[Dict[str, str]] = None):
        """
        Report a distribution metric.

        Args:
            metric_name: The name of the metric
            value: The value to record
            tags: Optional tags to associate with the metric
        """
        pass

    def error(self, tag: str, error: str):
        """
        Report an error.

        Args:
            tag: Category/component tag for the error
            error: The error message
        """
        pass

    def should_enable_high_cardinality_for_this_tag(self, tag: str) -> bool:
        """
        Determine if high cardinality should be enabled for a tag.

        Args:
            tag: The tag to check

        Returns:
            True if high cardinality should be enabled, False otherwise
        """
        pass

FAQ

Reference

API methods

• check_gate(user: StatsigUser, gate_name: str, options: Optional[FeatureGateEvaluationOptions] = None) -> bool
• get_dynamic_config(user: StatsigUser, config_name: str, options: Optional[DynamicConfigEvaluationOptions] = None) -> DynamicConfig
• get_experiment(user: StatsigUser, experiment_name: str, options: Optional[ExperimentEvaluationOptions] = None) -> DynamicConfig
• get_layer(user: StatsigUser, layer_name: str, options: Optional[LayerEvaluationOptions] = None) -> Layer
• get_feature_gate(user: StatsigUser, gate_name: str, options: Optional[FeatureGateEvaluationOptions] = None) -> FeatureGate
• get_parameter_store(user: StatsigUser, parameter_store_name: str, options: Optional[ParameterStoreEvaluationOptions] = None) -> ParameterStore
• log_event(user: StatsigUser, event_name: str, value: Optional[Union[str, float]] = None, metadata: Optional[Dict[str, str]] = None) -> None
• manually_log_gate_exposure(user: StatsigUser, gate_name: str) -> None
• manually_log_dynamic_config_exposure(user: StatsigUser, config_name: str) -> None
• manually_log_experiment_exposure(user: StatsigUser, experiment_name: str) -> None
• manually_log_layer_parameter_exposure(user: StatsigUser, layer_name: str, parameter_name: str) -> None
• override_experiment_by_group_name(experiment_name: str, group_name: str, id: Optional[str] = None) -> None
• get_client_initialize_response(user: StatsigUser, options: Optional[ClientInitializeResponseOptions] = None) -> ClientInitializeResponse
• shutdown() -> AsyncResult[None]

Fields needed methods

The following methods return information about which user fields evaluation requires:

• get_gate_fields_needed(gate_name: str) -> List[str]
• get_dynamic_config_fields_needed(config_name: str) -> List[str]
• get_experiment_fields_needed(experiment_name: str) -> List[str]
• get_layer_fields_needed(layer_name: str) -> List[str]

These methods return a list of strings representing the user fields that are required to properly evaluate the specified gate, config, experiment, or layer.