Refresh writer identity when an Iceberg table name is reused
After recreating a table under the same name, check its UUID and actual data and manifest write targets. Matching names and schema IDs cannot approve an old factory location for the new table.
A reused name does not preserve identity
A recreated table has a new identity even when its name and schema ID repeat. Truthound Core can compare prepared writer locations with the authoritative current table location before approving the next checkpoint. [1, 2, 3]
A stale factory location report
Iceberg issue #18623 reports DynamicIcebergSink 1.12.0 caching factories by name-related keys across drop/recreate. Writers and manifest factories retain the old Table location, producing an S3 Tables access-point 403 at the next checkpoint in the report. [1]
Join UUID and both write targets
Connect old and current UUIDs, current location, actual data and manifest targets, and checkpoint ID. A new table starting at schema ID 0 can still hit an old cache entry. Record identity rather than treating equal names as continuity. [1, 3]
The new-location reference
Recreate illustrative db.t with a new UUID at new-location. Expected writer target=new-location. That passes; old-location fails even if the old storage is still accessible. Reusing db.t and schema ID 0 does not change the independent target. [1, 3]
The SDK does not clear caches
ColumnPairValidator in Truthound Core Python SDK 3.3.2 compares prepared non-null location tokens with ==. The operator resolves identities and captures actual factory targets. The SDK does not invalidate caches or restart the job. [1, 2, 3]
Separate restart and recreation
Verify a supported table-identity or cache-invalidation path, separating recreation tests from restart tests. A restart clearing the cache explains a reported control but is not proof that every reuse path is safe. [1, 3]
Processor refresh is insufficient
Keep data-file and manifest targets together. Updating only the record processor’s metadata view cannot establish that factories created earlier now write for the new table. [1, 3]
Application conditions
Prepare old/current UUIDs, current location, and data/manifest factory targets.
Alternatives
Test supported cache invalidation across restart and recreation separately.
Limitations
The same name and schema ID 0 do not establish recreated-table identity.
Evidence
- Event checked
- Subject checked
- Modified
