hannah-logging (Python)

Ships the logs of a Hannah component to the Hannah log collector, so a user can download one archive of all components' logs from the WebUI.

The library adds one logging.Handler. Your existing handlers (stdout, journald, syslog) keep working unchanged.

Installation

pip install hannah-logging

The library relies on the hannah-proto version the component already uses, because every call to Hannah Core has to carry the matching protocol version.

Usage

import logging
import hannah_logging

logging.basicConfig(level=logging.INFO)

# As early as possible: everything logged from here on is buffered.
shipping = hannah_logging.install(
    "telegram",
    version=__version__,
    secrets=[config.bot_token],                         # masked wherever they appear
    logger_categories={"hannah.stt": hannah_logging.TRANSCRIPT},
)

# Once the config is loaded:
shipping.connect(
    hannah_address="localhost:50051",                   # discovery via Hannah Core
    collector_address=None,                             # optional static fallback
)

Setting the collector address directly

A component that knows the collector address itself (Hannah Core, which announces it) passes it in instead of subscribing to discovery:

shipping.set_collector_address("192.168.1.10:50061")   # None withdraws it

It starts shipping if connect hasn't been called yet, takes precedence over the static collector_address, and a new address moves the stream.

Categories

Each line is tagged GENERAL, TRANSCRIPT (speech transcripts, user utterances) or METADATA (room and device names, presence). Exports can leave out the last two, and the WebUI does so by default.

• By logger name: logger_categories={"hannah.stt": TRANSCRIPT} also covers child loggers such as hannah.stt.whisper.
• Per call: log.info("heard: %s", text, extra={hannah_logging.CATEGORY_ATTR: hannah_logging.TRANSCRIPT})

Secrets

Before a line leaves the process, the library masks:

• values of keys such as password, token, api_key, secret, psk
• Bearer tokens, JWTs, credentials in URLs (mqtt://user:pw@host), PEM private keys
• Telegram bot tokens and well-known token prefixes (glpat-, ghp_, …)
• every value passed as secrets=[…] or later through shipping.add_secret(…)

Extra regexes go in secret_patterns=[…]. The component's own handlers still see the unmasked line.

Behaviour

• Buffer: 4 MiB by default (max_buffer_bytes). When it is full, the oldest lines are dropped and reported to the collector as a gap. While no collector is known, the buffer simply keeps running as a ring.
• Timestamps are taken when a line is logged, not when it is sent.
• Discovery: subscribes to Hannah Core's infrastructure announcements and follows the log collector when it moves. If Core is unreachable, the last known collector is kept. The static collector_address is used while none is announced.
• Never blocks the component: sending runs on its own threads. Connection problems are logged once to the hannah_logging logger (not shipped) and retried with backoff.
• Shutdown: at exit, the library tries for up to 2 seconds to send what is still buffered (shipping.close(timeout=…)).