delete_records leads up to a crash when using Python 3.14.6

[xdgrulez]

Hi,

I have an issue with confluent_kafka (latest version 2.14.2) on Python 3.14 (tested with 3.14.6): When I call "delete_records" and then "list_topics", I get a segfault. This is a minimal example to test it:

from confluent_kafka.admin import AdminClient

admin_client = AdminClient({"bootstrap.servers": "localhost:9092"})

from confluent_kafka import TopicPartition
tp = TopicPartition(topic="my_topic", partition=0, offset=100)

print("Deleting records...")
delete_futures = admin_client.delete_records([tp], operation_timeout=10)

for tp, future in delete_futures.items():
    try:
        future.result()
        print(f"Successfully deleted records for {tp.topic}-{tp.partition} up to offset {tp.offset}")
    except Exception as e:
        print(f"Failed to delete records for {tp.topic}: {e}")

print("\nListing topics...")
cluster_metadata = admin_client.list_topics(timeout=10)

for topic_name in cluster_metadata.topics:
    print(f"- {topic_name}")

Crashes at the "list_topics" call:
Segmentation fault (core dumped)

This is on Linux, don't know what happens on Mac/Windows.

Works fine with Python 3.12 and 3.13, just 3.14 seems to be problematic.

Best, Ralph

[jmathai-confluent]

Thanks for the report. Confirmed on macOS (Apple Silicon, Darwin 24.6.0).

	Python 3.13.12	Python 3.14.6
confluent-kafka	2.14.2	2.14.2
Result	Pass	Segfault (exit 139)

Tested with a minimal repro: delete_records followed by list_topics on the same AdminClient. Works fine on 3.13, segfaults on 3.14.

Stack trace with python -X faulthandler:

Fatal Python error: Segmentation fault

Current thread 0x00000001efa0e2c0 (most recent call first):
  File ".../confluent_kafka/admin/__init__.py", line 661 in list_topics
  File "test_segfault_minimal.py", line 22 in <module>

Current thread's C stack trace (most recent call first):

  PyList_New+0x28
  cfl_int32_array_to_py_list+0x1c
  c_partitions_to_py+0x39c
  c_topics_to_py+0x29c
  c_metadata_to_py+0x2b0
  list_topics+0x1f4

The crash is in the C extension during list_topics → c_metadata_to_py → c_topics_to_py → c_partitions_to_py → cfl_int32_array_to_py_list → PyList_New. Likely a Python 3.14 C
API incompatibility (e.g., changes to Py_ssize_t handling or the stable ABI) causing an invalid argument to PyList_New after delete_records corrupts some state.