Finally figured out a fix, hope this would help somebody.

The problem occurs because get_all_post_dates is called many times in process_ticker_attention (thousands of times per ticker). The fix is to share one mongo_client inside each subprocess (NOT across processes).

def init_process(ticker, finish_num, lock):
    """Initialize a MongoDB connection for each worker process."""
    global mongo_p
    mongo_p = pymongo.MongoClient("mongodb://localhost:27017/", connectTimeoutMS=600000)
    process_ticker_attention(ticker, finish_num, lock)