Pull mode

For devices on a LAN that cannot push, the server connects to them (default port 4370) using pyzk.

pip install "django-fingerprint-attendance[pull]"

FINGERPRINT_ATTENDANCE = {"PULL_ENABLED": True}

Create the device with mode="pull", pull_host, and optionally pull_port and pull_comm_key, then approve it.

Command	Purpose
fpa_pull_attendance [--device SN] [--full] [--loop]	import punches newer than the device's pull_last_record_at marker; --full re-imports everything (dedupe prevents duplicates); --loop polls every PULL_POLL_INTERVAL
fpa_live_capture [--device SN]	long-running listener, one thread per device, reconnect with exponential backoff (up to PULL_RECONNECT_BACKOFF_MAX), graceful SIGINT/SIGTERM shutdown
fpa_sync_device [--device SN] [--remove-extra]	write missing users and templates directly (pull devices) or queue commands (push devices)

Gap-fill: every time live capture (re)connects, it first imports from the marker and then resumes live capture, so punches made while disconnected are never missed.

Pull-mode punches go through the same ingestion pipeline as ADMS (source="pull"), with the same dedupe keys, so the same punch arriving by both routes is stored once.

Clock drift is measured on every connection (the device clock is read directly), and log counts and capacity come from read_sizes().

Run fpa_live_capture under a process manager (systemd, supervisord, a container) and restart it on failure.

Custom SDKs

Implement fingerprint_attendance.pull.base.BasePullAdapter (connect, get_info, get_users, get_templates, get_attendance, set_user, set_templates, delete_user, set_time, live_capture) and set PULL_ADAPTER. Tests can use fingerprint_attendance.testing.FakePullAdapter.