-- migrate:up
--
-- One-time cleanup + guard for state.position_history duplicates.
--
-- A parked device re-reports the same gpsTime on every 60s poll, so before the
-- projector's write-time NOT EXISTS guard (shipped alongside this migration)
-- the history hypertable accumulated identical (imei, occurred_at) rows —
-- bloating storage and inflating the "fix count" shown in the trip dock.
--
--   Step 1  delete duplicates, keeping the lowest history_id per
--           (imei, occurred_at).
--   Step 2  add a unique index so duplicates can never reappear. It includes
--           occurred_at (the hypertable partition column), which TimescaleDB
--           requires for a unique index on a hypertable.
--
-- NOTE: the DELETE scans the whole hypertable; run it in a quiet window. The
-- migration is independent of the projector deploy order — the code's
-- NOT EXISTS guard needs no constraint, so applying this before/after a
-- redeploy is both safe.

DELETE FROM state.position_history a
      USING state.position_history b
      WHERE a.imei = b.imei
        AND a.occurred_at = b.occurred_at
        AND a.history_id > b.history_id;

CREATE UNIQUE INDEX IF NOT EXISTS position_history_imei_occurred_uq
    ON state.position_history (imei, occurred_at);

-- migrate:down

DROP INDEX IF EXISTS state.position_history_imei_occurred_uq;