Activity Tracking

Optional Redis-backed unique-user sets so dashboards can show calendar DAU / MAU / QAU without treating User.last_login as “active today.”

What you get

WAU is not implemented (no weekly keys or queries).

These are calendar periods, not rolling windows: MAU is “active in January,” not “active in the last 30 days.”

last_login alone undercounts: long-lived refresh tokens mean users stay active for days without a fresh login. Activity tracking records presence on authenticated traffic instead.

Requirements

• redis_url — required when tracking is on (Redis Sets)
• enable_activity_tracking=True — default is False

auth = EnterpriseRBAC(  # or SimpleRBAC / OutlabsAuth
    database_url=...,
    secret_key=...,
    redis_url="redis://localhost:6379/0",
    enable_activity_tracking=True,
)

If tracking is enabled without Redis, initialization fails with a clear error. Leave the flag off when you do not need these metrics.

How it works

1. On authenticated requests (and some login paths), the library schedules activity_tracker.track_activity_detached(user_id) — non-blocking.
2. Redis Sets hold the current day / month / quarter memberships (O(1) add).
3. A maintenance sync writes counts into Postgres activity_metrics for history and optionally updates User.last_activity.

Production APIs default to background_job_mode="disabled". One external maintenance owner should call run_maintenance_once() at the host-selected cadence. The 30-minute activity_sync_interval applies only to explicitly enabled embedded development loops.

You rarely talk to Redis yourself for dashboards older than the Redis TTL — query activity_metrics for history; use tracker helpers for “right now.”

Configuration knobs

Querying

Access the tracker after await auth.initialize() via auth.activity_tracker (only when enabled).

Live (Redis)

from datetime import date

dau = await auth.activity_tracker.get_daily_active_users()
mau = await auth.activity_tracker.get_monthly_active_users()
qau = await auth.activity_tracker.get_quarterly_active_users()

# Specific periods
dau_past = await auth.activity_tracker.get_daily_active_users(date(2025, 1, 15))
mau_past = await auth.activity_tracker.get_monthly_active_users(2025, 1)
qau_past = await auth.activity_tracker.get_quarterly_active_users(2025, 1)

Redis answers only while keys remain within TTL (roughly days for DAU, months for MAU, longer for QAU). Older periods → Postgres.

Historical (Postgres)

from datetime import date, timedelta
from sqlmodel import select
from outlabs_auth.models.sql.activity_metric import ActivityMetric

end = date.today()
start = end - timedelta(days=30)

result = await session.execute(
    select(ActivityMetric)
    .where(
        ActivityMetric.metric_type == "dau",
        ActivityMetric.metric_date >= start,
        ActivityMetric.metric_date <= end,
    )
    .order_by(ActivityMetric.metric_date)
)
for row in result.scalars():
    print(row.metric_date, row.count)

Monthly rows are keyed on the 1st of the month.

There is no packaged HTTP router for DAU/MAU — expose your own admin endpoint that calls the tracker or reads activity_metrics.

Ops notes

• Production sync runs through one external maintenance owner; enabling tracking does not start a loop while background_job_mode="disabled".
• Prometheus / structured logs may emit activity-related signals when observability is on — see Observability and Metrics Reference.
• For “last seen” UX, prefer User.last_activity (when sync updates it) over inventing your own counter.

Related

• Configuration
• Background Maintenance
• JWT Tokens — why refresh lifetime skews last_login
• Design decision DD-049 — maintainer depth