QuakeWang opened a new pull request, #8071:
URL: https://github.com/apache/paimon/pull/8071

   ### Purpose
   
   Ray writes create one independent Paimon writer per write task. For 
primary-key tables, this is unsafe unless all rows for each `(partition, 
bucket)` are routed to one writer.
   
   The previous Ray path only guarded HASH_FIXED primary-key tables in 
default/off mode. Default primary-key tables use dynamic buckets (`bucket = 
-1`), so they were written as-is. Multiple Ray tasks could then assign buckets 
and sequence numbers from independent local state, producing duplicate keys 
with overlapping `_SEQUENCE_NUMBER`.
   
   This PR makes Ray primary-key writes fail fast unless the table is 
HASH_FIXED and `hash_fixed_precluster="map_groups"` is selected. Append-only 
non-HASH_FIXED tables still pass through unchanged. The PyPaimon Ray docs are 
updated to match the new behavior.
   
   ### Tests
   
   - `python3 -m py_compile paimon-python/pypaimon/ray/shuffle.py 
paimon-python/pypaimon/ray/ray_paimon.py 
paimon-python/pypaimon/write/table_write.py 
paimon-python/pypaimon/tests/test_ray_shuffle_helper.py 
paimon-python/pypaimon/tests/ray_repartition_test.py`
   - `python3 -m pytest paimon-python/pypaimon/tests/test_ray_shuffle_helper.py 
-q`


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to