MongoDB Connection Management in Python

Question

Is it better to do

from pymongo import Connection

conn = Connection()
db = conn.db_name

def contrivedExample(firstName)
    global db
    return db.people.find_one({'first-name': firstName})

or

from pymongo import Connection

def contrivedExample(firstName):
    with Connection() as conn:
        return conn.db_name.people.find_one({'first-name': firstName})

Various basic MongoDB tutorials (Python-oriented or not) imply that an app should connect once at startup; is that actually the case? Does the answer change for non-trivial, long-running applications? Does the answer change for web applications specifically? What are the pros/cons of going single-connection vs. connection-per-request?

Assuming "once at startup" is the right answer, would it be appropriate to start that connection in __init__.py?

Answer 1

The pymongo Connection class supports connection pooling and, as of version 2.2, the auto_start_request option ensures that the same socket will be used for a connection activity during the lifetime of the thread (default behavior). Additionally, there is built-in support for reconnecting when necessary, although your application code should handle the immediate exception.

To your question, I believe it'd be preferable to rely on pymongo's own connection pooling and request a new connection per thread. This Stack Overflow thread also discusses some best practices and explains some of the options at play, which you may find helpful. If necessary, you have the option of sharing the same socket between threads.