[Citizen4our]

Synopsis

Large Timescale event flushes in eva-db-timescale were slow enough to hit sqlx’s hard-coded 2s slow-statement warning, and retention cleanup stayed expensive (DELETE over hypertables).

Solution

• Replace per-row PL/pgSQL insert loops with a set-based UNNEST / INSERT … SELECT body for both insert_state_history_events and try_insert_state_history_events (stable OID lock order; vanished OIDs fail loudly instead of silent drops).
• Prefer Timescale drop_chunks for hypertables; keep DELETE only for plain tables after an EXISTS probe.
• Add optional config slow_statement_threshold (seconds; 0 disables; default remains 2s warn).
• Coordinate OID cleanup with in-flight inserts via a shared/exclusive PostgreSQL advisory transaction lock, so cleanup cannot delete an OID between the OID insert and the events insert of a batch.
• Add unit tests and ignored Timescale integration tests.