Formatting and filtering

Updated Jul 05, 2026

Logger-level formatting is fixed ("{logger} [LEVEL] message"). To customize output wrap handlers with formatter callables using the builder API:

def json_formatter(record: dict[str, object]) -> str:
    import json

    return json.dumps(
        {
            "logger": record["logger"],
            "level": record["level"],
            "message": record["message"],
        },
        separators=(",", ":"),
    )


stream = StreamHandlerBuilder.stdout().with_formatter(json_formatter).build()

Formatter identifiers are limited to "default" for now. Future releases will connect FormatterBuilder and handler formatter IDs, but this is not yet implemented.
Filters are available through both ConfigBuilder and dictConfig. fileConfig does not yet support filter declarations. Handler-level filters and level attributes remain unsupported, as these require Rust infrastructure changes outside the current scope.
Python callback filters receive a mutable logging.LogRecord view on the producer thread. When they accept the record, any new supported attributes they add are copied into metadata.key_values before the record enters the async queue, making them visible to formatter callables and StdlibHandlerAdapter.

import contextvars
import logging

from femtologging import PythonCallbackFilterBuilder

request_id = contextvars.ContextVar("request_id", default="")


def enrich_request(record: logging.LogRecord) -> bool:
    record.request_id = request_id.get()
    return record.levelname == "INFO"


callback_filter = PythonCallbackFilterBuilder(enrich_request)