sunchao opened a new issue, #5502:
URL: https://github.com/apache/datafusion-comet/issues/5502

   ### Describe the bug
   
   The process-wide object-store cache uses the normalized scheme/authority and 
configuration hash, but does not distinguish Hadoop from native backends. With 
`fs.comet.libhdfs.schemes=s3`, a Hadoop `s3://bucket/...` URL and a native 
`s3a://bucket/...` URL normalize to the same cache key under the same 
configuration. The second lookup can therefore reuse the first backend.
   
   The DataFusion registry has a related, separate limitation: its key contains 
only scheme/authority. Registering another configuration or backend at the same 
key replaces the previous mapping, even though the process cache has two 
distinct stores.
   
   ### Steps to reproduce
   
   Use one configuration with `fs.comet.libhdfs.schemes=s3`, then prepare both 
`s3://bucket/first.parquet` and `s3a://bucket/second.parquet`. Repeat in 
reverse order. The first cache insertion determines which backend the second 
URL receives. For the registry collision, prepare one bucket with two distinct 
object-store configurations in one `RuntimeEnv`, then retrieve both returned 
registration URLs.
   
   ### Expected behavior
   
   Cache and registration identities include the selected backend and 
configuration. Native `s3`/`s3a` aliases share one canonical identity, while 
explicitly Hadoop-routed schemes retain their physical spelling. Registration 
identity is independent of planning order. Synthetic registry identifiers must 
not reach physical encryption-key lookup.
   
   ### Additional context
   
   Extracted from the review of #5453 so this correctness change can be 
evaluated independently of I/O metric definitions and producer shutdown. The 
targeted regression uses distinct in-memory stores for deterministic 
cache/registry/read checks without requiring Hadoop or cloud credentials. This 
issue concerns object-store identity; file-metadata cache namespacing is not 
claimed to be addressed.
   


-- 
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]


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to