mongo-profile

Module mongoprofile contains functions and objects to retreive and parse the output of MongoDB's "db.system.profile.find()"

To get more information about MongoDB profiling, see http://www.mongodb.org/display/DOCS/Database+Profiler

mongoprofile.MongoProfiler

Class MongoProfiler is a "with"-wrapper around any set of MongoDB queries. Typical usecase contains three steps:

Step1. Open connection:

>>> from pymongo import Connection
>>> db = Connection().test

Step 2. Execute and profile queries:

>>> profiler = MongoProfiler(db)
>>> with profiler:
...     db.people.insert(dict(name='John', age=20))
...     db.people.insert(dict(name='Mary', age=30))
...     db.people.update({'name': 'John'}, {'age': 21})
...     db.people.remove({'name': 'Mary'})
...     list(db.people.find({'age': {'$gt': 20.0}}))
...     db.people.find({'age': {'$gt': 20.0}}).count()

Step3. Get profile info

As a result, you will get the more or less comprehensive list of dict subclasses, containing all profile information, including parsed "info". Every subclass has redefined __str__ method returning the convenient presentation of request. See the example below to get the point:

>>> for record in profiler.get_records():
...    print str(record)

test> db.people.insert({...})
test> db.people.insert({...})
test> db.people.update({ name: "John" }, {...})
test> db.people.remove({ name: "Mary" })
test> db.people.find({ $query: { age: { $gt: 20.0 } } })
test> db.runCommand({ count: "people", query: { age: { $gt: 20.0 } }, fields: null })

A few more facts about record objects worth to be known:

• There is a record.short_info() method returning the one-line string with short information about the query.
• Every record class is a subclass of dict, and because of that it's possible to get a bunch of ordered information using calls such as record['millis'], record['ts'], etc.

Markers

The MongoProfiler class has .mark(text) method. When mark is invoked, mongodb client do the fake query to phony collection just to record data in log. After the job has ended, these markers will be available as '==== text ====' records.

Having changed previous example, we get something like this.

Commands:

>>> profiler = MongoProfiler(db)
>>> with profiler:
...     profiler.mark('insert')
...     db.people.insert(dict(name='John', age=20))
...     db.people.insert(dict(name='Mary', age=30))
...     profiler.mark('search')
...     list(db.people.find({'age': {'$gt': 20.0}}))
...     db.people.find({'age': {'$gt': 20.0}}).count()

Will lead to the output:

'==== insert ===='
test> db.people.insert({...})
test> db.people.insert({...})
'==== search ===='
test> db.people.find({ $query: { age: { $gt: 20.0 } } })
test> db.runCommand({ count: "people", query: { age: { $gt: 20.0 } }, fields: null })

DummyMongoProfiler

It is probable that depending on some circumstances, you want or don't want to spend extra resources on your query profiling. Stub DummyMongoProfiler class mocking MongoProfiler interface can be used for that purpose. Below is the usage sample with Django-nonrel in mind:

>>> from django.conf import settings
>>> Profiler = settings.DEBUG and MongoProfiler or DummyMongoProfiler
>>> profiler = Profiler(db)
>>> with profiler:
...     ModelClass.objects.filter(...)
...

Miscellaneous remarks

Collection db.system.profile is capped with a relatively small capacity. If you want to profile large amount of records at once, it is worth to extend its size. The following set of commands creates capped collection of 100Mb:

> db.setProfilingLevel(0)
> db.system.profile.drop()
> db.createCollection("system.profile", {capped:true, size:100*1e6})

Command db.system.profile.stats() shows you the current state of collection.