HASH partition creation reports “already exists”, then immediate open reports “was not found”

[hsballoon]

Data writing to partitions: 8、60、121、 94.....（as far as we know only 34 is normal）consistently fails.

This is reproducible by calling dldb directly with a single process, single thread, and single session:

from dldb.utils import stable_hash

job_id = "codex-bucket94-probe-1787650993-232"
assert stable_hash(job_id) % 128 == 94

session.add("landing_test", dataframe)

Actual result:

ValueError: Table
'landing_test_type_HASH_column_job_id_partitions_128_partition_94'
already exists

During exception handling, dldb calls open_table(), which then fails with:

ValueError: Table
'landing_test_type_HASH_column_job_id_partitions_128_partition_94'
was not found

Other observations:

• 483 records have been written successfully to other partitions.
• Restarting the worker and creating a fresh dldb session does not resolve it.
• The partition is not visible through table listing/opening.
• No corresponding partition 94 objects are visible under the S3 prefix.
• The issue reproduces without WTGatewayClient.

Expected behavior:

dldb should create and open partition 94 successfully, or return a consistent and actionable error explaining why the partition cannot be created.

Environment:

dldb==1.0.0
lancedb==0.34.0
pylance==9.0.0
pyarrow==23.0.1
pandas==2.3.3