County health runs Structured Streaming for clinic event feeds and must keep replays from duplicating ops metrics. What operational need should they recognize?

County health running Structured Streaming for clinic event feeds must keep replays from duplicating ops metrics—that operational need is checkpointing. Checkpoints store offsets and stream progress so restarts resume consistently instead of reprocessing the entire clinic event history into waiting-room or throughput counters. Exactly-once-oriented streaming operations for civic clinic KPIs depend on durable checkpoints plus appropriate sinks that do not silently double-write metrics. Using trigger intervals so Structured Streaming clinic event replays stay consistent for ops metrics controls how often micro-batches run, but it does not persist offsets or stream state, so it cannot resume progress consistently after a failure or prevent duplicated ops metrics on replay. Tuning watermark delays so Structured Streaming clinic event replays stay consistent for ops metrics manages event-time lateness, but it does not persist offsets or stream state, so it cannot resume processing consistently after a failure. Using complete mode so Structured Streaming clinic event replays stay consistent for ops metrics rewrites the entire aggregate table but does not checkpoint offsets or stream progress, so replayed ops metrics can still double after a restart. Exam caveat: checkpointing is necessary but not sufficient alone—sink design such as Delta appends or idempotent foreachBatch, plus not deleting checkpoint paths casually, also matter for metric correctness after restarts. Operational check: configure a checkpoint location for the clinic Structured Streaming job, induce a controlled restart, confirm offsets resume and ops metrics do not double, restrict access to the checkpoint path, and alert if the stream falls back to a full reprocess.