gedih3.doctor.fused

Fused per-partition scan for gh3_doctor’s h3db-tree diagnoses.

When the user asks for multiple per-partition diagnoses in one invocation (e.g. --check db or --check parquet_health,metadata), this module turns N separate parallel_map dispatches with N file opens per partition into a single dispatch where each partition’s parquet listing, meta JSON, and parquet footers are opened once and shared across every enabled scan.

The shared state is built by _build_shared_state() at worker entry, passed as a shared dict to each diagnosis’s _scan_partition_*(dir, shared=None) worker, and dropped when the partition’s work is done. Each scan in _SCAN_REGISTRY returns the per-partition payload its diagnosis already produces today (no schema change), so driver-side _finalize_check_from_results aggregation is identical to the single- diagnosis path.

Read-amplification savings on a continental DB scale with the parquet footer reuse — the heaviest reads — between parquet_health, geoparquet_bbox, and backfill.

Functions

register_scan(→ None)	Register a per-partition scan worker under name.
fused_scan_partition(→ Dict[str, Any])	Worker: run every enabled per-partition scan on one partition.
fused_eligible_names(→ set)	Return the set of diagnosis names that participate in fusion.

Module Contents

gedih3.doctor.fused.register_scan(name: str, fn: Callable[Ellipsis, Any]) → None

Register a per-partition scan worker under name.

The worker must be picklable (module-level), accept partition_dir as the first positional arg, and accept shared=None plus any diagnosis-specific keyword args.

gedih3.doctor.fused.fused_scan_partition(partition_dir: str, *, enabled_scans: List[str], scan_kwargs: Dict[str, Dict[str, Any]] | None = None) → Dict[str, Any]

Worker: run every enabled per-partition scan on one partition.

Returns {scan_name: per_partition_result_or_exception}. A per-scan exception is captured in the result dict (mirroring parallel_map’s exception-passthrough) so one scan’s failure on one partition never aborts the others for that same partition.

scan_kwargs carries per-diagnosis broadcast args namespaced by diagnosis name — e.g. {'orphans': {'age_seconds': 86400}, 'backfill': {'products': [...], 'expected_by_product': {...}}}.

gedih3.doctor.fused.fused_eligible_names() → set

Return the set of diagnosis names that participate in fusion.

The runner uses this to decide whether --check X,Y is fusible. Only diagnoses whose worker has been registered via register_scan() are eligible.