kobotoolbox / kpi / 22676123444 / 2

81%

master: 76%

push

github

fix(longRunningMigration): fix 0005 pagination timeout and improve duplicate cleanup DEV-1698 (#6777)

### 💭 Notes

- **LRM 0005 pagination fix**: removed cursor-based pagination (`ORDER
BY pk` + `pk__gt`) that forced full table scans and caused timeouts.
Instances are now fetched without ordering (order is irrelevant — we
just need to exhaust all `root_uuid=NULL` rows for a given xform).

- **`clean_duplicated_submissions` improvements**: switched from
`transaction.atomic()` + raw MongoDB deletion to
`kc_transaction_atomic()` + `delete_instances()` for proper cross-DB
support. `root_uuid` is now backfilled on the surviving reference
*after* duplicates are deleted (to avoid a unique-constraint violation
when a duplicate already held the same value). Added XForm parser
caching (`use_cached_parser=True`) to avoid redundant DB queries per
batch. Comprehensive test suite added.

7325 of 11481 branches covered (63.8%)

27960 of 34423 relevant lines covered (81.22%)

0.81 hits per line