Configure Sampling

Learn how to configure sampling in your app.

If you find that Sentry's tracing functionality is generating too much data, for example, if you notice your spans quota is quickly being exhausted, you can choose to sample your traces.

Effective sampling is key to getting the most value from Sentry's performance monitoring while minimizing overhead. The Python SDK provides two ways to control the sampling rate. You can review the options and examples below.

traces_sample_rate is a floating-point value between 0.0 and 1.0, inclusive, which controls the probability with which each transaction will be sampled:

Copied

sentry_sdk.init(
    # ...

    # Set traces_sample_rate to 1.0 to capture 100%
    # of transactions for tracing.
    # We recommend adjusting this value in production,
    traces_sample_rate=1.0,
)

With traces_sample_rate set to 0.25, each transaction in your application is randomly sampled with a probability of 0.25, so you can expect that one in every four transactions will be sent to Sentry.

For more granular control, you can provide a traces_sampler function. This approach allows you to:

Apply different sampling rates to different types of transactions
Filter out specific transactions entirely
Make sampling decisions based on transaction data
Control the inheritance of sampling decisions in distributed traces
Use custom attributes to modify sampling

It is strongly recommended when using a custom traces_sampler that you respect the parent sampling decision. This ensures your traces will be complete.

In distributed systems, implementing inheritance logic when trace information is propagated between services will ensure consistent sampling decisions across your entire distributed trace.

Copied

import sentry_sdk
from sentry_sdk.types import SamplingContext

def traces_sampler(sampling_context: SamplingContext) -> float:
    # Use the parent sampling decision if we have an incoming trace.
    # Note: we strongly recommend respecting the parent sampling decision,
    # as this ensures your traces will be complete!
    parent_sampling_decision = sampling_context.get("parent_sampled")
    if parent_sampling_decision is not None:
        return float(parent_sampling_decision)

    # Examine provided sampling context along with anything in the
    # global namespace to compute the sample rate for this transaction
    if "...":
        # These are important - take a big sample
        return 0.5
    elif "...":
        # These are less important - only take 1%
        return 0.01
    elif "...":
        # These aren't worth tracking - drop these transactions
        return 0

    # Default sample rate
    return 0.1

sentry_sdk.init(
    # ...
    traces_sampler=traces_sampler,
)

Traces Sampler Examples

Prioritizing Critical User Flows

Copied

import sentry_sdk
from sentry_sdk.types import SamplingContext

def traces_sampler(sampling_context: SamplingContext) -> float:
    # Use the parent sampling decision if we have an incoming trace.
    # Note: we strongly recommend respecting the parent sampling decision,
    # as this ensures your traces will be complete!
    parent_sampling_decision = sampling_context.get("parent_sampled")
    if parent_sampling_decision is not None:
        return float(parent_sampling_decision)

    ctx = sampling_context.get("transaction_context", {})
    name = ctx.get("name")

    # Sample all checkout transactions
    if name and ('/checkout' in name or
                ctx.get("op") == 'checkout'):
        return 1.0

    # Sample 50% of login transactions
    if name and ('/login' in name or
                ctx.get("op") == 'login'):
        return 0.5

    # Sample 10% of everything else
    return 0.1

sentry_sdk.init(
    dsn="your-dsn",
    traces_sampler=traces_sampler,
)

Handling Different Environments and Error Rates

Copied

import sentry_sdk
from sentry_sdk.types import SamplingContext

def traces_sampler(sampling_context: SamplingContext) -> float:
    # Use the parent sampling decision if we have an incoming trace.
    # Note: we strongly recommend respecting the parent sampling decision,
    # as this ensures your traces will be complete!
    parent_sampling_decision = sampling_context.get("parent_sampled")
    if parent_sampling_decision is not None:
        return float(parent_sampling_decision)

    ctx = sampling_context.get("transaction_context", {})
    environment = os.environ.get("ENVIRONMENT", "development")

    # Sample all transactions in development
    if environment == "development":
        return 1.0

    # Sample more transactions if there are recent errors
    # Note: hasRecentErrors is a custom attribute that needs to be set
    if ctx.get("data", {}).get("hasRecentErrors"):
        return 0.8

    # Sample based on environment
    if environment == "production":
        return 0.05  # 5% in production
    elif environment == "staging":
        return 0.2   # 20% in staging

    # Default sampling rate
    return 0.1

# Initialize the SDK with the sampling function
sentry_sdk.init(
    dsn="your-dsn",
    traces_sampler=traces_sampler,
)

# You can use the sampling function by setting custom attributes:
# Option 1: When creating the transaction
with sentry_sdk.start_transaction(name="GET /api/users", op="http.request") as transaction:
    # Set custom attribute
    transaction.set_data("hasRecentErrors", True)
    # Your code here

# Option 2: During the transaction's lifecycle
with sentry_sdk.start_transaction(name="GET /api/users", op="http.request") as transaction:
    # Your code here
    transaction.set_data("hasRecentErrors", True)  # Set custom attribute

Controlling Sampling Based on User and Transaction Properties

Copied

import sentry_sdk
from sentry_sdk.types import SamplingContext

def traces_sampler(sampling_context: SamplingContext) -> float:
    # Use the parent sampling decision if we have an incoming trace.
    # Note: we strongly recommend respecting the parent sampling decision,
    # as this ensures your traces will be complete!
    parent_sampling_decision = sampling_context.get("parent_sampled")
    if parent_sampling_decision is not None:
        return float(parent_sampling_decision)

    ctx = sampling_context.get("transaction_context", {})
    data = ctx.get("data", {})

    # Always sample for premium users
    # Note: user.tier is a custom attribute that needs to be set
    if data.get("user", {}).get("tier") == "premium":
        return 1.0

    # Sample more transactions for users experiencing errors
    # Note: hasRecentErrors is a custom attribute
    if data.get("hasRecentErrors"):
        return 0.8

    # Sample less for high-volume, low-value paths
    # Note: name is an SDK-provided attribute
    if (ctx.get("name") or "").startswith("/api/metrics"):
        return 0.01

    # Sample more for slow transactions
    # Note: duration_ms is a custom attribute
    if data.get("duration_ms", 0) > 1000:  # Transactions over 1 second
        return 0.5

    # Default sampling rate
    return 0.2

# Initialize the SDK with the sampling function
sentry_sdk.init(
    dsn="your-dsn",
    traces_sampler=traces_sampler,
)

# To set custom attributes for this example:
with sentry_sdk.start_transaction(name="GET /api/users", op="http.request") as transaction:
    # Set custom attributes
    transaction.set_data("user", {"tier": "premium"})  # Custom user data
    transaction.set_data("hasRecentErrors", True)      # Custom error flag
    transaction.set_data("duration_ms", 1500)          # Custom timing data
    # Your code here

Complex Business Logic Sampling

Copied

import sentry_sdk
from sentry_sdk.types import SamplingContext

def traces_sampler(sampling_context: SamplingContext) -> float:
    # Use the parent sampling decision if we have an incoming trace.
    # Note: we strongly recommend respecting the parent sampling decision,
    # as this ensures your traces will be complete!
    parent_sampling_decision = sampling_context.get("parent_sampled")
    if parent_sampling_decision is not None:
        return float(parent_sampling_decision)

    ctx = sampling_context.get("transaction_context", {})
    data = ctx.get("data", {})

    # Always sample critical business operations
    # Note: op is an SDK-provided attribute
    if ctx.get("op") in ["payment.process", "order.create", "user.verify"]:
        return 1.0

    # Sample based on user segment
    # Note: user.segment is a custom attribute
    user_segment = data.get("user", {}).get("segment")
    if user_segment == "enterprise":
        return 0.8
    elif user_segment == "premium":
        return 0.5

    # Sample based on transaction value
    # Note: transaction.value is a custom attribute
    transaction_value = data.get("transaction", {}).get("value", 0)
    if transaction_value > 1000:  # High-value transactions
        return 0.7

    # Sample based on error rate in the service
    # Note: service.error_rate is a custom attribute
    error_rate = data.get("service", {}).get("error_rate", 0)
    if error_rate > 0.05:  # Error rate above 5%
        return 0.9

    # Default sampling rate
    return 0.1

# Initialize the SDK with the sampling function
sentry_sdk.init(
    dsn="your-dsn",
    traces_sampler=traces_sampler,
)

# To set custom attributes for this example:
with sentry_sdk.start_transaction(name="Process Payment", op="payment.process") as transaction:
    # Set custom attributes
    transaction.set_data("user", {"segment": "enterprise"})  # Custom user data
    transaction.set_data("transaction", {"value": 1500})     # Custom transaction data
    transaction.set_data("service", {"error_rate": 0.03})    # Custom service data
    # Your code here

Performance-Based Sampling

Copied

import sentry_sdk
from sentry_sdk.types import SamplingContext

def traces_sampler(sampling_context: SamplingContext) -> float:
    # Use the parent sampling decision if we have an incoming trace.
    # Note: we strongly recommend respecting the parent sampling decision,
    # as this ensures your traces will be complete!
    parent_sampling_decision = sampling_context.get("parent_sampled")
    if parent_sampling_decision is not None:
        return float(parent_sampling_decision)

    ctx = sampling_context.get("transaction_context", {})
    data = ctx.get("data", {})

    # Sample all slow transactions
    # Note: duration_ms is a custom attribute
    if data.get("duration_ms", 0) > 2000:  # Over 2 seconds
        return 1.0

    # Sample more transactions with high memory usage
    # Note: memory_usage_mb is a custom attribute
    if data.get("memory_usage_mb", 0) > 500:  # Over 500MB
        return 0.8

    # Sample more transactions with high CPU usage
    # Note: cpu_percent is a custom attribute
    if data.get("cpu_percent", 0) > 80:  # Over 80% CPU
        return 0.8

    # Sample more transactions with high database load
    # Note: db_connections is a custom attribute
    if data.get("db_connections", 0) > 100:  # Over 100 connections
        return 0.7

    # Default sampling rate
    return 0.1

# Initialize the SDK with the sampling function
sentry_sdk.init(
    dsn="your-dsn",
    traces_sampler=traces_sampler,
)

# To set custom attributes for this example:
with sentry_sdk.start_transaction(name="Process Data", op="data.process") as transaction:
    # Set custom attributes
    transaction.set_data("duration_ms", 2500)      # Custom timing data
    transaction.set_data("memory_usage_mb", 600)   # Custom memory data
    transaction.set_data("cpu_percent", 85)        # Custom CPU data
    transaction.set_data("db_connections", 120)    # Custom database data
    # Your code here

When the traces_sampler function is called, the Sentry SDK passes a sampling_context object with information from the relevant span to help make sampling decisions:

Copied

{
    "transaction_context": {
        "name": str,                    # transaction title at creation time (SDK-provided)
        "op": str,                      # short description of transaction type (SDK-provided)
        "data": Optional[Dict[str, Any]] # custom data you've added to the transaction
    },
    "parent_sampled": Optional[bool],  # whether the parent transaction was sampled (SDK-provided)
    "parent_sample_rate": Optional[float], # the sample rate used by the parent (SDK-provided)
    "custom_sampling_context": Optional[Dict[str, Any]]  # additional custom data for sampling
}

The sampling context contains both SDK-provided attributes and custom attributes:

SDK-Provided Attributes:

transaction_context.name: The name of the transaction
transaction_context.op: The operation type
parent_sampled: Whether the parent transaction was sampled
parent_sample_rate: The sample rate used by the parent

Custom Attributes:

Any data you add to the set_data method on the transaction object. Use this for data that you want to include in the transaction data that gets sent to Sentry.
Any data you add to the custom_sampling_context parameter in start_transaction. Use this for data that you want to use for sampling decisions but don't want to include in the transaction data that gets sent to Sentry. Read more about sampling context here.

When multiple sampling mechanisms could apply, Sentry follows this order of precedence:

If a sampling decision is passed to start_transaction, that decision is used
If traces_sampler is defined, its decision is used. Although the traces_sampler can override the parent sampling decision, most users will want to ensure their traces_sampler respects the parent sampling decision
If no traces_sampler is defined, but there is a parent sampling decision from an incoming distributed trace, we use the parent sampling decision
If neither of the above, traces_sample_rate is used
If none of the above are set, no transactions are sampled. This is equivalent to setting traces_sample_rate=0.0

Sentry uses a "head-based" sampling approach:

A sampling decision is made in the originating service (the "head")
This decision is propagated to all downstream services

The two key headers are:

sentry-trace: Contains trace ID, span ID, and sampling decision
baggage: Contains additional trace metadata including sample rate

The Sentry Python SDK automatically attaches these headers to outgoing HTTP requests when using auto-instrumentation with libraries like requests, urllib3, or httpx. For other communication channels, you can manually propagate trace information. Learn more about customizing tracing in custom trace propagation

Set Up Distributed Tracing

Instrumentation

Was this helpful?

Help improve this content
Our documentation is open source and available on GitHub. Your contributions are welcome, whether fixing a typo (drat!) or suggesting an update ("yeah, this would be better").

How to contribute | Edit this page | Create a docs issue | Get support