12. History and archive

Stores and lifetimes

Store	Contents	Service	Lifetime
Live state	What a run needs to continue: task statuses, timers, awaited events	Durable Task Scheduler	Until the run ends, then purged after 7 days
History	One row per run and per task; logs; event executions; poll data	Azure SQL	13 months by default, settable per workflow name
Payloads	Inputs and outputs above the size threshold	Blob	Same as history
Archive	Finished runs past retention, one JSON file each	Blob, cool tier, archive tier after 12 months	Indefinite

Every read by the API or the UI goes to Azure SQL or the archive. Nothing reads live state directly.

History tables

workflow_run    (workflow_id, name, version, status, correlation_id, priority, input, output,
                 variables_json, started_at, ended_at, reason, parent_workflow_id, parent_task_id,
                 owner, instruction_versions_json, idempotency_key, retry_of, restart_of, rerun_of)
task_run        (task_id, workflow_id, task_ref, task_type, task_def_name, status, attempt, worker_id,
                 domain, scheduled_at, started_at, ended_at, input, output, reason,
                 callback_after_s, poll_count, queue_wait_ms)
task_log        (id, task_id, logged_at, line)
poll_data       (task_type, domain, worker_id, last_poll_at)
event_execution (id, handler_name, source, message_id, status, action, result_json, at)
workflow_message(workflow_id, seq, payload_json, pushed_at, pulled_at)
task_cache      (key, output_json, expires_at)

The engine writes workflow_run at start and end, and task_run at every status change. Task logs are capped at 100 lines per task.

Search

Indexes: (name, status, started_at), (correlation_id), (status, ended_at), and a full-text index on reason. GET /api/workflow/search and GET /api/tasks/search use them. Analytics over history runs on the Fabric mirror of these tables, never on the API.

Redaction

• maskedFields in a definition lists input and output paths to redact before writing.
• Values from ${workflow.secrets.*} are resolved at the moment of use and never written to history or logs.

Archival

A daily function selects runs with a terminal status older than the retention period, writes each run with its tasks and logs to archive/{yyyy}/{mm}/{workflow_id}.json, and deletes the rows. GET /api/workflow/{id} reads Azure SQL first and the archive second, so an archived run is still readable.