Performance

How can I profile a SQLAlchemy powered application?

Looking for performance issues typically involves two strategies. One is query profiling, and the other is code profiling.

Query Profiling

Sometimes just plain SQL logging (enabled via python’s logging module or via the echo=True argument on create_engine()) can give an idea how long things are taking. For example, if you log something right after a SQL operation, you’d see something like this in your log:

  1. 17:37:48,325 INFO [sqlalchemy.engine.base.Engine.0x...048c] SELECT ...
  2. 17:37:48,326 INFO [sqlalchemy.engine.base.Engine.0x...048c] {<params>}
  3. 17:37:48,660 DEBUG [myapp.somemessage]

if you logged myapp.somemessage right after the operation, you know it took 334ms to complete the SQL part of things.

Logging SQL will also illustrate if dozens/hundreds of queries are being issued which could be better organized into much fewer queries. When using the SQLAlchemy ORM, the “eager loading” feature is provided to partially (contains_eager()) or fully (joinedload(), subqueryload()) automate this activity, but without the ORM “eager loading” typically means to use joins so that results across multiple tables can be loaded in one result set instead of multiplying numbers of queries as more depth is added (i.e. r + r*r2 + r*r2*r3 …)

For more long-term profiling of queries, or to implement an application-side “slow query” monitor, events can be used to intercept cursor executions, using a recipe like the following:

  1. from sqlalchemy import event
  2. from sqlalchemy.engine import Engine
  3. import time
  4. import logging
  5. logging.basicConfig()
  6. logger = logging.getLogger("myapp.sqltime")
  7. logger.setLevel(logging.DEBUG)
  8. @event.listens_for(Engine, "before_cursor_execute")
  9. def before_cursor_execute(conn, cursor, statement,
  10. parameters, context, executemany):
  11. conn.info.setdefault('query_start_time', []).append(time.time())
  12. logger.debug("Start Query: %s", statement)
  13. @event.listens_for(Engine, "after_cursor_execute")
  14. def after_cursor_execute(conn, cursor, statement,
  15. parameters, context, executemany):
  16. total = time.time() - conn.info['query_start_time'].pop(-1)
  17. logger.debug("Query Complete!")
  18. logger.debug("Total Time: %f", total)

Above, we use the ConnectionEvents.before_cursor_execute() and ConnectionEvents.after_cursor_execute() events to establish an interception point around when a statement is executed. We attach a timer onto the connection using the info dictionary; we use a stack here for the occasional case where the cursor execute events may be nested.

Code Profiling

If logging reveals that individual queries are taking too long, you’d need a breakdown of how much time was spent within the database processing the query, sending results over the network, being handled by the DBAPI, and finally being received by SQLAlchemy’s result set and/or ORM layer. Each of these stages can present their own individual bottlenecks, depending on specifics.

For that you need to use the Python Profiling Module. Below is a simple recipe which works profiling into a context manager:

  1. import cProfile
  2. import io
  3. import pstats
  4. import contextlib
  5. @contextlib.contextmanager
  6. def profiled():
  7. pr = cProfile.Profile()
  8. pr.enable()
  9. yield
  10. pr.disable()
  11. s = io.StringIO()
  12. ps = pstats.Stats(pr, stream=s).sort_stats('cumulative')
  13. ps.print_stats()
  14. # uncomment this to see who's calling what
  15. # ps.print_callers()
  16. print(s.getvalue())

To profile a section of code:

  1. with profiled():
  2. Session.query(FooClass).filter(FooClass.somevalue==8).all()

The output of profiling can be used to give an idea where time is being spent. A section of profiling output looks like this:

  1. 13726 function calls (13042 primitive calls) in 0.014 seconds
  2. Ordered by: cumulative time
  3. ncalls tottime percall cumtime percall filename:lineno(function)
  4. 222/21 0.001 0.000 0.011 0.001 lib/sqlalchemy/orm/loading.py:26(instances)
  5. 220/20 0.002 0.000 0.010 0.001 lib/sqlalchemy/orm/loading.py:327(_instance)
  6. 220/20 0.000 0.000 0.010 0.000 lib/sqlalchemy/orm/loading.py:284(populate_state)
  7. 20 0.000 0.000 0.010 0.000 lib/sqlalchemy/orm/strategies.py:987(load_collection_from_subq)
  8. 20 0.000 0.000 0.009 0.000 lib/sqlalchemy/orm/strategies.py:935(get)
  9. 1 0.000 0.000 0.009 0.009 lib/sqlalchemy/orm/strategies.py:940(_load)
  10. 21 0.000 0.000 0.008 0.000 lib/sqlalchemy/orm/strategies.py:942(<genexpr>)
  11. 2 0.000 0.000 0.004 0.002 lib/sqlalchemy/orm/query.py:2400(__iter__)
  12. 2 0.000 0.000 0.002 0.001 lib/sqlalchemy/orm/query.py:2414(_execute_and_instances)
  13. 2 0.000 0.000 0.002 0.001 lib/sqlalchemy/engine/base.py:659(execute)
  14. 2 0.000 0.000 0.002 0.001 lib/sqlalchemy/sql/elements.py:321(_execute_on_connection)
  15. 2 0.000 0.000 0.002 0.001 lib/sqlalchemy/engine/base.py:788(_execute_clauseelement)
  16. ...

Above, we can see that the instances() SQLAlchemy function was called 222 times (recursively, and 21 times from the outside), taking a total of .011 seconds for all calls combined.

Execution Slowness

The specifics of these calls can tell us where the time is being spent. If for example, you see time being spent within cursor.execute(), e.g. against the DBAPI:

  1. 2 0.102 0.102 0.204 0.102 {method 'execute' of 'sqlite3.Cursor' objects}

this would indicate that the database is taking a long time to start returning results, and it means your query should be optimized, either by adding indexes or restructuring the query and/or underlying schema. For that task, analysis of the query plan is warranted, using a system such as EXPLAIN, SHOW PLAN, etc. as is provided by the database backend.

Result Fetching Slowness - Core

If on the other hand you see many thousands of calls related to fetching rows, or very long calls to fetchall(), it may mean your query is returning more rows than expected, or that the fetching of rows itself is slow. The ORM itself typically uses fetchall() to fetch rows (or fetchmany() if the Query.yield_per() option is used).

An inordinately large number of rows would be indicated by a very slow call to fetchall() at the DBAPI level:

  1. 2 0.300 0.600 0.300 0.600 {method 'fetchall' of 'sqlite3.Cursor' objects}

An unexpectedly large number of rows, even if the ultimate result doesn’t seem to have many rows, can be the result of a cartesian product - when multiple sets of rows are combined together without appropriately joining the tables together. It’s often easy to produce this behavior with SQLAlchemy Core or ORM query if the wrong Column objects are used in a complex query, pulling in additional FROM clauses that are unexpected.

On the other hand, a fast call to fetchall() at the DBAPI level, but then slowness when SQLAlchemy’s CursorResult is asked to do a fetchall(), may indicate slowness in processing of datatypes, such as unicode conversions and similar:

  1. # the DBAPI cursor is fast...
  2. 2 0.020 0.040 0.020 0.040 {method 'fetchall' of 'sqlite3.Cursor' objects}
  3. ...
  4. # but SQLAlchemy's result proxy is slow, this is type-level processing
  5. 2 0.100 0.200 0.100 0.200 lib/sqlalchemy/engine/result.py:778(fetchall)

In some cases, a backend might be doing type-level processing that isn’t needed. More specifically, seeing calls within the type API that are slow are better indicators - below is what it looks like when we use a type like this:

  1. from sqlalchemy import TypeDecorator
  2. import time
  3. class Foo(TypeDecorator):
  4. impl = String
  5. def process_result_value(self, value, thing):
  6. # intentionally add slowness for illustration purposes
  7. time.sleep(.001)
  8. return value

the profiling output of this intentionally slow operation can be seen like this:

  1. 200 0.001 0.000 0.237 0.001 lib/sqlalchemy/sql/type_api.py:911(process)
  2. 200 0.001 0.000 0.236 0.001 test.py:28(process_result_value)
  3. 200 0.235 0.001 0.235 0.001 {time.sleep}

that is, we see many expensive calls within the type_api system, and the actual time consuming thing is the time.sleep() call.

Make sure to check the Dialect documentation for notes on known performance tuning suggestions at this level, especially for databases like Oracle. There may be systems related to ensuring numeric accuracy or string processing that may not be needed in all cases.

There also may be even more low-level points at which row-fetching performance is suffering; for example, if time spent seems to focus on a call like socket.receive(), that could indicate that everything is fast except for the actual network connection, and too much time is spent with data moving over the network.

Result Fetching Slowness - ORM

To detect slowness in ORM fetching of rows (which is the most common area of performance concern), calls like populate_state() and _instance() will illustrate individual ORM object populations:

  1. # the ORM calls _instance for each ORM-loaded row it sees, and
  2. # populate_state for each ORM-loaded row that results in the population
  3. # of an object's attributes
  4. 220/20 0.001 0.000 0.010 0.000 lib/sqlalchemy/orm/loading.py:327(_instance)
  5. 220/20 0.000 0.000 0.009 0.000 lib/sqlalchemy/orm/loading.py:284(populate_state)

The ORM’s slowness in turning rows into ORM-mapped objects is a product of the complexity of this operation combined with the overhead of cPython. Common strategies to mitigate this include:

  • fetch individual columns instead of full entities, that is:

    1. session.query(User.id, User.name)

    instead of:

    1. session.query(User)
  • Use Bundle objects to organize column-based results:

    1. u_b = Bundle('user', User.id, User.name)
    2. a_b = Bundle('address', Address.id, Address.email)
    3. for user, address in session.query(u_b, a_b).join(User.addresses):
    4. # ...
  • Use result caching - see Dogpile Caching for an in-depth example of this.

  • Consider a faster interpreter like that of PyPy.

The output of a profile can be a little daunting but after some practice they are very easy to read.

See also

Performance - a suite of performance demonstrations with bundled profiling capabilities.

I’m inserting 400,000 rows with the ORM and it’s really slow!

The SQLAlchemy ORM uses the unit of work pattern when synchronizing changes to the database. This pattern goes far beyond simple “inserts” of data. It includes that attributes which are assigned on objects are received using an attribute instrumentation system which tracks changes on objects as they are made, includes that all rows inserted are tracked in an identity map which has the effect that for each row SQLAlchemy must retrieve its “last inserted id” if not already given, and also involves that rows to be inserted are scanned and sorted for dependencies as needed. Objects are also subject to a fair degree of bookkeeping in order to keep all of this running, which for a very large number of rows at once can create an inordinate amount of time spent with large data structures, hence it’s best to chunk these.

Basically, unit of work is a large degree of automation in order to automate the task of persisting a complex object graph into a relational database with no explicit persistence code, and this automation has a price.

ORMs are basically not intended for high-performance bulk inserts - this is the whole reason SQLAlchemy offers the Core in addition to the ORM as a first-class component.

For the use case of fast bulk inserts, the SQL generation and execution system that the ORM builds on top of is part of the Core. Using this system directly, we can produce an INSERT that is competitive with using the raw database API directly.

Note

When using the psycopg2 dialect, consider making use of the batch execution helpers feature of psycopg2, now supported directly by the SQLAlchemy psycopg2 dialect.

Alternatively, the SQLAlchemy ORM offers the Bulk Operations suite of methods, which provide hooks into subsections of the unit of work process in order to emit Core-level INSERT and UPDATE constructs with a small degree of ORM-based automation.

The example below illustrates time-based tests for several different methods of inserting rows, going from the most automated to the least. With cPython 2.7, runtimes observed:

  1. SQLAlchemy ORM: Total time for 100000 records 6.89754080772 secs
  2. SQLAlchemy ORM pk given: Total time for 100000 records 4.09481811523 secs
  3. SQLAlchemy ORM bulk_save_objects(): Total time for 100000 records 1.65821218491 secs
  4. SQLAlchemy ORM bulk_insert_mappings(): Total time for 100000 records 0.466513156891 secs
  5. SQLAlchemy Core: Total time for 100000 records 0.21024107933 secs
  6. sqlite3: Total time for 100000 records 0.137335062027 sec

We can reduce the time by a factor of nearly three using recent versions of PyPy:

  1. SQLAlchemy ORM: Total time for 100000 records 2.39429616928 secs
  2. SQLAlchemy ORM pk given: Total time for 100000 records 1.51412987709 secs
  3. SQLAlchemy ORM bulk_save_objects(): Total time for 100000 records 0.568987131119 secs
  4. SQLAlchemy ORM bulk_insert_mappings(): Total time for 100000 records 0.320806980133 secs
  5. SQLAlchemy Core: Total time for 100000 records 0.206904888153 secs
  6. sqlite3: Total time for 100000 records 0.165791988373 sec

Script:

  1. import time
  2. import sqlite3
  3. from sqlalchemy.ext.declarative import declarative_base
  4. from sqlalchemy import Column, Integer, String, create_engine
  5. from sqlalchemy.orm import scoped_session, sessionmaker
  6. Base = declarative_base()
  7. DBSession = scoped_session(sessionmaker())
  8. engine = None
  9. class Customer(Base):
  10. __tablename__ = "customer"
  11. id = Column(Integer, primary_key=True)
  12. name = Column(String(255))
  13. def init_sqlalchemy(dbname='sqlite:///sqlalchemy.db'):
  14. global engine
  15. engine = create_engine(dbname, echo=False)
  16. DBSession.remove()
  17. DBSession.configure(bind=engine, autoflush=False, expire_on_commit=False)
  18. Base.metadata.drop_all(engine)
  19. Base.metadata.create_all(engine)
  20. def test_sqlalchemy_orm(n=100000):
  21. init_sqlalchemy()
  22. t0 = time.time()
  23. for i in xrange(n):
  24. customer = Customer()
  25. customer.name = 'NAME ' + str(i)
  26. DBSession.add(customer)
  27. if i % 1000 == 0:
  28. DBSession.flush()
  29. DBSession.commit()
  30. print(
  31. "SQLAlchemy ORM: Total time for " + str(n) +
  32. " records " + str(time.time() - t0) + " secs")
  33. def test_sqlalchemy_orm_pk_given(n=100000):
  34. init_sqlalchemy()
  35. t0 = time.time()
  36. for i in xrange(n):
  37. customer = Customer(id=i + 1, name="NAME " + str(i))
  38. DBSession.add(customer)
  39. if i % 1000 == 0:
  40. DBSession.flush()
  41. DBSession.commit()
  42. print(
  43. "SQLAlchemy ORM pk given: Total time for " + str(n) +
  44. " records " + str(time.time() - t0) + " secs")
  45. def test_sqlalchemy_orm_bulk_save_objects(n=100000):
  46. init_sqlalchemy()
  47. t0 = time.time()
  48. for chunk in range(0, n, 10000):
  49. DBSession.bulk_save_objects(
  50. [
  51. Customer(name="NAME " + str(i))
  52. for i in xrange(chunk, min(chunk + 10000, n))
  53. ]
  54. )
  55. DBSession.commit()
  56. print(
  57. "SQLAlchemy ORM bulk_save_objects(): Total time for " + str(n) +
  58. " records " + str(time.time() - t0) + " secs")
  59. def test_sqlalchemy_orm_bulk_insert(n=100000):
  60. init_sqlalchemy()
  61. t0 = time.time()
  62. for chunk in range(0, n, 10000):
  63. DBSession.bulk_insert_mappings(
  64. Customer,
  65. [
  66. dict(name="NAME " + str(i))
  67. for i in xrange(chunk, min(chunk + 10000, n))
  68. ]
  69. )
  70. DBSession.commit()
  71. print(
  72. "SQLAlchemy ORM bulk_insert_mappings(): Total time for " + str(n) +
  73. " records " + str(time.time() - t0) + " secs")
  74. def test_sqlalchemy_core(n=100000):
  75. init_sqlalchemy()
  76. t0 = time.time()
  77. engine.execute(
  78. Customer.__table__.insert(),
  79. [{"name": 'NAME ' + str(i)} for i in xrange(n)]
  80. )
  81. print(
  82. "SQLAlchemy Core: Total time for " + str(n) +
  83. " records " + str(time.time() - t0) + " secs")
  84. def init_sqlite3(dbname):
  85. conn = sqlite3.connect(dbname)
  86. c = conn.cursor()
  87. c.execute("DROP TABLE IF EXISTS customer")
  88. c.execute(
  89. "CREATE TABLE customer (id INTEGER NOT NULL, "
  90. "name VARCHAR(255), PRIMARY KEY(id))")
  91. conn.commit()
  92. return conn
  93. def test_sqlite3(n=100000, dbname='sqlite3.db'):
  94. conn = init_sqlite3(dbname)
  95. c = conn.cursor()
  96. t0 = time.time()
  97. for i in xrange(n):
  98. row = ('NAME ' + str(i),)
  99. c.execute("INSERT INTO customer (name) VALUES (?)", row)
  100. conn.commit()
  101. print(
  102. "sqlite3: Total time for " + str(n) +
  103. " records " + str(time.time() - t0) + " sec")
  104. if __name__ == '__main__':
  105. test_sqlalchemy_orm(100000)
  106. test_sqlalchemy_orm_pk_given(100000)
  107. test_sqlalchemy_orm_bulk_save_objects(100000)
  108. test_sqlalchemy_orm_bulk_insert(100000)
  109. test_sqlalchemy_core(100000)
  110. test_sqlite3(100000)