python-observability

name	python-observability
description	Python observability patterns including structured logging, metrics, and distributed tracing.
when_to_use	Adding logging, implementing metrics collection, setting up tracing, or debugging production systems.

Python Observability

Instrument Python applications with structured logs, metrics, and traces. When something breaks in production, you need to answer "what, where, and why" without deploying new code.

Core Concepts

1. Structured Logging

Emit logs as JSON with consistent fields for production environments. Machine-readable logs enable powerful queries and alerts. For local development, consider human-readable formats.

2. The Four Golden Signals

Track latency, traffic, errors, and saturation for every service boundary.

3. Correlation IDs

Thread a unique ID through all logs and spans for a single request, enabling end-to-end tracing.

4. Bounded Cardinality

Keep metric label values bounded. Unbounded labels (like user IDs) explode storage costs.

Quick Start

import structlog

structlog.configure(
    processors=[
        structlog.processors.TimeStamper(fmt="iso"),
        structlog.processors.JSONRenderer(),
    ],
)

logger = structlog.get_logger()
logger.info("Request processed", user_id="123", duration_ms=45)

Fundamental Patterns

Pattern 1: Structured Logging with Structlog

Configure structlog for JSON output with consistent fields.

import logging
import structlog

def configure_logging(log_level: str = "INFO") -> None:
    """Configure structured logging for the application."""
    structlog.configure(
        processors=[
            structlog.contextvars.merge_contextvars,
            structlog.processors.add_log_level,
            structlog.processors.TimeStamper(fmt="iso"),
            structlog.processors.StackInfoRenderer(),
            structlog.processors.format_exc_info,
            structlog.processors.JSONRenderer(),
        ],
        wrapper_class=structlog.make_filtering_bound_logger(
            getattr(logging, log_level.upper())
        ),
        context_class=dict,
        logger_factory=structlog.PrintLoggerFactory(),
        cache_logger_on_first_use=True,
    )

# Initialize at application startup
configure_logging("INFO")
logger = structlog.get_logger()

Pattern 2: Consistent Log Fields

Every log entry should include standard fields for filtering and correlation.

import structlog
from contextvars import ContextVar

# Store correlation ID in context
correlation_id: ContextVar[str] = ContextVar("correlation_id", default="")

logger = structlog.get_logger()

def process_request(request: Request) -> Response:
    """Process request with structured logging."""
    logger.info(
        "Request received",
        correlation_id=correlation_id.get(),
        method=request.method,
        path=request.path,
        user_id=request.user_id,
    )

    try:
        result = handle_request(request)
        logger.info(
            "Request completed",
            correlation_id=correlation_id.get(),
            status_code=200,
            duration_ms=elapsed,
        )
        return result
    except Exception as e:
        logger.error(
            "Request failed",
            correlation_id=correlation_id.get(),
            error_type=type(e).__name__,
            error_message=str(e),
        )
        raise

Pattern 3: Semantic Log Levels

Use log levels consistently across the application.

Level	Purpose	Examples
DEBUG	Development diagnostics	Variable values, internal state
INFO	Request lifecycle, operations	Request start/end, job completion
WARNING	Recoverable anomalies	Retry attempts, fallback used
ERROR	Failures needing attention	Exceptions, service unavailable

# DEBUG: Detailed internal information
logger.debug("Cache lookup", key=cache_key, hit=cache_hit)

# INFO: Normal operational events
logger.info("Order created", order_id=order.id, total=order.total)

# WARNING: Abnormal but handled situations
logger.warning(
    "Rate limit approaching",
    current_rate=950,
    limit=1000,
    reset_seconds=30,
)

# ERROR: Failures requiring investigation
logger.error(
    "Payment processing failed",
    order_id=order.id,
    error=str(e),
    payment_provider="stripe",
)

Never log expected behavior at ERROR. A user entering a wrong password is INFO, not ERROR.

Pattern 4: Correlation ID Propagation

Generate a unique ID at ingress and thread it through all operations.

from contextvars import ContextVar
import uuid
import structlog

correlation_id: ContextVar[str] = ContextVar("correlation_id", default="")

def set_correlation_id(cid: str | None = None) -> str:
    """Set correlation ID for current context."""
    cid = cid or str(uuid.uuid4())
    correlation_id.set(cid)
    structlog.contextvars.bind_contextvars(correlation_id=cid)
    return cid

# FastAPI middleware example
from fastapi import Request

async def correlation_middleware(request: Request, call_next):
    """Middleware to set and propagate correlation ID."""
    # Use incoming header or generate new
    cid = request.headers.get("X-Correlation-ID") or str(uuid.uuid4())
    set_correlation_id(cid)

    response = await call_next(request)
    response.headers["X-Correlation-ID"] = cid
    return response

Propagate to outbound requests:

import httpx

async def call_downstream_service(endpoint: str, data: dict) -> dict:
    """Call downstream service with correlation ID."""
    async with httpx.AsyncClient() as client:
        response = await client.post(
            endpoint,
            json=data,
            headers={"X-Correlation-ID": correlation_id.get()},
        )
        return response.json()

Pattern 5: Deferred Log Message Formatting

Never f-string a log message. Pass the template and its values as separate args so formatting is deferred -- it runs only when the record's level is enabled (Logger.isEnabledFor short-circuits before interpolation; an f-string, or %/str.format/+ pre-formatting, pays that cost even for filtered-out records). Two mechanisms:

stdlib_logger.info("user=%s orders=%s", user_id, order_count)              # stdlib: logger interpolates %s args
logger.info("orders processed", user_id=user_id, order_count=order_count)  # structlog: constant event + structured kwargs
stdlib_logger.info(f"user={user_id} orders={order_count}")                 # BAD: f-string renders eagerly, every call

%s args are not the structlog form: this skill's processor chain (Pattern 1) omits PositionalArgumentsFormatter, so logger.info("user=%s", user_id) stores the literal user=%s under positional_args instead of interpolating. Use kwargs with structlog, %s args with stdlib.

Ruff flake8-logging-format (G) enforces this: G004 flags f-strings, G001/G002/G003 flag str.format/%/+ pre-formatting.

For exceptions use logger.exception (= error(..., exc_info=True)) -- it attaches the traceback; keep deferred %s args:

except OSError:
    stdlib_logger.exception("save failed, record_id=%s", record.id)
    raise

Detailed worked examples and patterns

Detailed sections (starting with ## Advanced Patterns) live in references/details.md. Read that file when the navigation summary above is insufficient.

Best Practices Summary

1. Use structured logging - JSON logs with consistent fields
2. Propagate correlation IDs - Thread through all requests and logs
3. Track the four golden signals - Latency, traffic, errors, saturation
4. Bound label cardinality - Never use unbounded values as metric labels
5. Log at appropriate levels - Don't cry wolf with ERROR
6. Include context - User ID, request ID, operation name in logs
7. Use context managers - Consistent timing and error handling
8. Separate concerns - Observability code shouldn't pollute business logic
9. Test your observability - Verify logs and metrics in integration tests
10. Set up alerts - Metrics are useless without alerting
11. Defer log formatting - Pass %s args (stdlib) or kwargs (structlog); never f-string a log message (ruff G004)