gedih3.doctor.diagnoses.log_state#

log_state diagnosis — stuck recovery flags + log↔disk drift.

Detects:

  • _pending_variable_update left in the log without matching state.

  • Partitions present on disk but not listed in h3_partition_ids.

  • Partitions listed in the log but missing from disk.

  • Granule status drift — granules in granule_info with a non-INDEXED status whose (orbit, granule, track) is present in a finalized partition’s metadata JSON. Caused by the merge-only resume shortcut (gh3_build v0.8.9+), which skips reconcile and leaves stale PENDING statuses behind even though the data is already on disk.

Remedies (safe):

  • Clear the stuck flag when no work is pending.

  • Repopulate h3_partition_ids from disk via set_post_build_info.

  • Disk-missing partitions are reported only (manual intervention).

  • Granule status drift: flip stale entries to INDEXED and save log.

Performance pillar (v0.8.x cross-pipeline lessons):

  • Granule-triple discovery is fanned out via gedih3.doctor.parallel.parallel_map() — one task per partition reads its few meta JSONs and returns the triples found. Without this, the default --check db audit on a continental-scale (10k+ partitions) database stalls for minutes in two driver-side recursive globs over h3_*/.

Functions#

log_state_check(→ gedih3.doctor.report.Report)

log_state_fix(→ gedih3.doctor.report.Report)

Module Contents#

gedih3.doctor.diagnoses.log_state.log_state_check(ctx: gedih3.doctor.report.DoctorContext) → gedih3.doctor.report.Report[source]#
gedih3.doctor.diagnoses.log_state.log_state_fix(ctx: gedih3.doctor.report.DoctorContext, report: gedih3.doctor.report.Report) → gedih3.doctor.report.Report[source]#