betshare-infra-logging 0.1.0

Shared logging + Slack-alerting for BetShareMarket Python services

betshare-infra-logging

Shared logging + Slack-alerting for BetShareMarket Python services. Extracted from the near-identical per-service conf_logger modules (consumer, joiner-events, betfair-exchange-scraper).

Import name: betshare_infra_logging. Distribution name: betshare-infra-logging.

Why

The per-service copies drifted in accidental ways that were bugs:

• Slack POST hang. joiner-events and betfair-exchange-scraper called requests.post(webhook, ...) with no timeout, so a hung Slack call could freeze their single-threaded loops. This package bakes in a finite timeout (SLACK_TIMEOUT = 10) with no public parameter to override it to None.
• Level env var schism (LOGGER_LEVEL vs LOG_LEVEL).
• Slack channel drift (#error vs #errors).
• Idempotency, noise suppression, and correlation IDs were each reinvented or missing.

Install

Production (private registry, version-pinned per service):

# in the service's pyproject.toml
betshare-infra-logging = "^0.1.0"

Local development (PATH dependency, used by the pilot migration):

betshare-infra-logging = { path = "../../packages/py-infra-logging", develop = true }

Or with pip: pip install -e packages/py-infra-logging.

Usage

One-liner at service startup. A service's thin conf_logger.py shim collapses to:

from betshare_infra_logging import configure_logging

logger = configure_logging("joiner")  # Slack timeout now bounded => hang bug fixed

configure_logging configures the ROOT logger, so existing logging.getLogger(__name__) call sites keep working unchanged via propagation.

Correlation-id / context binding:

log = logger.child(correlation_id=cid, bookmaker="bet365")
log.info("processing message")
# -> "... processing message [correlation_id=... bookmaker=bet365]"
# context shows in console output AND in the Slack payload.

get_logger(name) returns a Logger wrapping logging.getLogger(name).

Notes

• No import-time side effects. Importing the package does NOT configure logging (unlike the old modules which called configure_logger() at import). Call configure_logging(...) explicitly. The package does NOT call load_dotenv() — that stays in the service entrypoint.
• Level resolution priority (Python majority order): level arg ?? LOGGER_LEVEL env ?? LOG_LEVEL env ?? "INFO". Cutover does not change any service's effective level and no .env is edited.
• Slack channel defaults to #errors (unified from the Python services' #error). The channel field is cosmetic for modern Slack incoming webhooks — the destination is fixed by the webhook URL itself, so this change is safe and does not reroute alerts.
• Idempotent. Calling configure_logging twice does not stack duplicate console/Slack handlers; the package tracks and clears the handlers it owns.
• Slack failures are swallowed. A failing Slack POST never propagates into the caller's loop.

Public interface

DEFAULT_QUIET_LOGGERS = (
    "azure", "azure.servicebus", "azure.core",
    "motor", "pymongo", "crawl4ai", "playwright",
)
SLACK_TIMEOUT = 10  # seconds, baked in

def configure_logging(service, *, level=None, slack_webhook=None,
                      quiet_loggers=None) -> Logger: ...
def get_logger(name=None) -> Logger: ...

class Logger:
    def debug(self, message, **context) -> None
    def info(self, message, **context) -> None
    def warning(self, message, **context) -> None
    def error(self, message, exc=None, **context) -> None
    def child(self, **bindings) -> "Logger"