segrails.monitoring.tracker

Monitoring tracker — writes inference metrics to Azure Table Storage.

Mode no-op if azure-data-tables is not installed or no connection is configured. Writes are fire-and-forget in a background thread to avoid blocking predict().

Environment variables:
AZURE_TABLES_CONNECTION_STRING: Storage account connection string.

When set, used for direct authentication (local dev, CI).

AZURE_TABLES_ACCOUNT_URL: Storage account table endpoint URL

(e.g. https://deployment4dev.table.core.windows.net). When set without a connection string, authenticates via DefaultAzureCredential (Managed Identity, az login, SP env vars).

AZURE_TABLES_TABLE_NAME: Table name. Default: SegrailsInferences.

Classes

TableTracker

Écrit les métriques d'inférence dans Azure Table Storage (une ligne par appel).

Module Contents

class segrails.monitoring.tracker.TableTracker(table_name: str = 'SegrailsInferences', env: str = 'dev', connection_string: str = '', account_url: str = '')

Écrit les métriques d’inférence dans Azure Table Storage (une ligne par appel).

Auth (par ordre de priorité) : connection string → DefaultAzureCredential (MI/az login).

Parameters:

  • table_name – Nom de la table Azure (créée automatiquement si absente).

  • env – Environnement de déploiement (dev / test / beta / prod).

  • connection_string – Surcharge AZURE_TABLES_CONNECTION_STRING.

  • account_url – Surcharge AZURE_TABLES_ACCOUNT_URL (utilisé avec DefaultAzureCredential).

Example

>>> tracker = TableTracker(table_name="SegrailsInferences", env="dev")
>>> tracker.log_inference(result)
_table_name = 'SegrailsInferences'
_env = 'dev'
_session_id
_client: Any = None
property enabled: bool

Return True if the tracker is connected and ready to write.

log_inference(result: altametris.segrails.result.SegrailsResult, session_id: str | None = None, pipeline_name: str | None = None, env: str | None = None, end_time: str | None = None) → None

Write one inference row to the table (background thread).

No-op if the tracker is disabled. Errors during write are caught and logged as warnings — they never propagate to the caller.

Parameters:

  • result – SegrailsResult returned by SegrailsClient.predict().

  • session_id – Logical session identifier used as PartitionKey. Defaults to the machine hostname (socket.gethostname()).

  • pipeline_name – Calling pipeline identifier.

  • env – Override the tracker’s env tag for this row only.

  • end_time – ISO 8601 timestamp of when predict() returned. Used to compute End_time and Duration columns.

_build_client(connection_string: str, account_url: str, table_name: str) → Any

Build and return a TableClient, or None on failure.

_write(entity: dict[str, Any]) → None

Write entity to the table. Logs warning on error.

_build_entity(result: altametris.segrails.result.SegrailsResult, session_id: str, pipeline_name: str | None = None, env: str | None = None, end_time: str | None = None) → dict[str, Any]

Build the table entity dict from a SegrailsResult.