Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@rag/utils/azure_sas_conn.py`:
- Around line 80-84: The obj_exist method logs "Fail put ..." on exceptions
which is misleading; change the logging.exception call inside obj_exist to a
existence-specific message like "Fail checking existence of {bucket}/{fnm}"
(include the bucket/fnm) and keep logging.exception so the exception details are
recorded; also ensure obj_exist returns a boolean on error (e.g., return False)
so callers of obj_exist (which calls self.conn.get_blob_client(...).exists())
receive a consistent result.
- Line 65: The delete/get/exists/read methods currently call self.conn using the
new path format f"{bucket}/{fnm}" (e.g., the
self.conn.delete_blob(f"{bucket}/{fnm}") call) but must fall back to the legacy
plain fnm if the new path is not found; update each occurrence (the delete_blob,
get_blob/read operations and existence checks at the same spots) to first try
the f"{bucket}/{fnm}" call and on a not-found/error response attempt the legacy
fnm value before failing, returning the successful result or logging/raising the
original error only if both attempts fail.

In `@rag/utils/azure_spn_conn.py`:
- Around line 86-87: Modify the non-write operations to try the new prefixed
path first and fall back to the legacy plain filename on failure: for the delete
flow using self.conn.delete_file(f"{bucket}/{fnm}") (and similarly for
get/get_presigned_url/obj_exist), catch exceptions or a false/None result and
then attempt the same operation with the legacy path self.conn.delete_file(fnm)
(or self.conn.get_object(fnm), self.conn.get_presigned_url(fnm),
self.conn.object_exists(fnm) as appropriate); ensure both attempts preserve
return values/errors so callers receive the correct success/failure outcome and
avoid raising the first-path exception when the legacy attempt succeeds.
- Around line 102-107: The obj_exist method currently logs a misleading message
"Fail put ..." on exception; update the exception logging in obj_exist to use an
existence-specific message such as "Fail checking existence for {bucket}/{fnm}"
(use the same f-string pattern) so logs accurately reflect the operation and
include the bucket/fnm details; modify the logging.exception call inside
obj_exist to that new message.

In `@test/unit_test/rag/utils/test_azure_blob_bucket_prefix.py`:
- Around line 31-33: The helper _install_stubs currently mutates sys.modules
directly and never restores entries, causing cross-test pollution; change the
stubbing to use pytest's monkeypatch.setitem to insert the fake modules (e.g.,
replace sys.modules["ragflow.runtime"] etc.) so patches are automatically
reverted after each test, and update the three other stub blocks (the ones
creating fake modules around the same area) to use monkeypatch.setitem instead
of direct sys.modules assignment or add explicit teardown that restores
originals if monkeypatch is unavailable; locate uses in function _install_stubs
and the other stub creation sites and replace sys.modules[...] = fake_module
with monkeypatch.setitem(sys.modules, key, fake_module).

---

Outside diff comments:
In `@rag/utils/azure_spn_conn.py`:
- Around line 71-82: The put method's retry loop is broken because the except
block returns immediately, preventing further retries; update put (method put in
class using self.conn.create_file and f.append_data/f.flush_data) to remove the
premature return inside the except so that on exception it logs the error, calls
self.__open__(), sleeps, and then continues the for _ in range(3) loop to retry;
after the loop finishes without success, return None (preserve existing
successful return of f.flush_data when no exception).

---

Nitpick comments:
In `@test/unit_test/rag/utils/test_azure_blob_bucket_prefix.py`:
- Around line 109-186: Add a new test in both TestAzureSpnBucketPrefix and
TestAzureSasBucketPrefix that calls the connector health() (on the
RAGFlowAzureSpnBlob and RAGFlowAzureSasBlob instances) with a bucket like "kb_a"
and asserts the underlying mocked connection was invoked with a path containing
the bucket prefix; for the SPN test call spn.health("kb_a") and assert any
recorded call in spn.conn.call_args_list has a first positional arg that starts
with "kb_a/" (or contains "kb_a"), and for the SAS test call sas.health("kb_a")
and assert any sas.conn.upload/download/delete/get call_args/kwargs include a
"name" or positional arg that starts with "kb_a/". Ensure you reference the
instance names spn and sas, the health() method, and the conn mock when adding
the assertions.