gedih3.doctor.diagnoses.orphans#

orphans diagnosis — leftover temp files and empty partition directories.

Detects:

  • *.tmp, *.join.tmp, *.fill.tmp, *.dedup.tmp under db / tmp roots.

  • _s3_download/ and dask-worker-space/ directories left behind by crashed builds.

  • Empty h3_*/ partition directories (no parquet files).

  • Empty year subdirs.

A finding is only flagged when the file/dir’s mtime is older than --orphan-age-hours (default 24h) so an in-progress build isn’t disturbed.

Performance pillar (v0.8.x lessons backport):

  • The temp-file/leftover-dir scan over the database root is pushed to workers via gedih3.doctor.parallel.parallel_map() — one task per h3_* partition. Mirrors the build merge phase’s _list_year_subdirs pattern (gh3builder.py:906-924) which eliminated the same driver-side recursive walk on continental builds.

  • Empty-partition / empty-year-dir checks are O(1) os.scandir instead of glob.glob('**/*.parquet', recursive=True) (matches the build pipeline’s emptiness check at gh3builder.py:1356-1362).

  • The temp directory’s scan stays driver-side because it’s narrow and not nested as deeply as the database tree.

Functions#

orphans_check(→ gedih3.doctor.report.Report)

orphans_fix(→ gedih3.doctor.report.Report)

Module Contents#

gedih3.doctor.diagnoses.orphans.orphans_check(ctx: gedih3.doctor.report.DoctorContext) → gedih3.doctor.report.Report[source]#
gedih3.doctor.diagnoses.orphans.orphans_fix(ctx: gedih3.doctor.report.DoctorContext, report: gedih3.doctor.report.Report) → gedih3.doctor.report.Report[source]#