Bug Description

The HotpotQA evaluation benchmark has two file handle resource leaks that can exhaust file descriptors:

Issue 1: Download function (line 37)

save_file = open(os.path.join(dataset_full_path, 'dev_distractor.json'), 'wb')
response = requests.get(url, stream=True)
for chunk in tqdm.tqdm(response.iter_content(chunk_size=chunk_size)):
    if chunk:
        save_file.write(chunk)
# File never closed; if exception occurs, handle leaks

Issue 2: Run function (line 74)

f = open(dataset_path)
query_objects = json.loads(f.read())
# File never closed

Impact

• File descriptor exhaustion if evaluation is run repeatedly
• Memory leaks in long-running processes
• Violates context manager best practices

Proposed Fix

• Wrap file open calls in with statements
• Issue 1: Add save_file.close() in finally block or use context manager
• Issue 2: Replace with with open(dataset_path) as f:

File: llama-index-core/llama_index/core/evaluation/benchmarks/hotpotqa.py Lines: 37-50, 74-75