dibbler/sqlalchemy/test/profiling.py
2010-05-07 17:33:49 +00:00

223 lines
6.9 KiB
Python

"""Profiling support for unit and performance tests.
These are special purpose profiling methods which operate
in a more fine-grained way than nose's profiling plugin.
"""
import os, sys
from sqlalchemy.test import config
from sqlalchemy.test.util import function_named, gc_collect
from nose import SkipTest
__all__ = 'profiled', 'function_call_count', 'conditional_call_count'
all_targets = set()
profile_config = { 'targets': set(),
'report': True,
'sort': ('time', 'calls'),
'limit': None }
profiler = None
def profiled(target=None, **target_opts):
"""Optional function profiling.
@profiled('label')
or
@profiled('label', report=True, sort=('calls',), limit=20)
Enables profiling for a function when 'label' is targetted for
profiling. Report options can be supplied, and override the global
configuration and command-line options.
"""
# manual or automatic namespacing by module would remove conflict issues
if target is None:
target = 'anonymous_target'
elif target in all_targets:
print "Warning: redefining profile target '%s'" % target
all_targets.add(target)
filename = "%s.prof" % target
def decorator(fn):
def profiled(*args, **kw):
if (target not in profile_config['targets'] and
not target_opts.get('always', None)):
return fn(*args, **kw)
elapsed, load_stats, result = _profile(
filename, fn, *args, **kw)
report = target_opts.get('report', profile_config['report'])
if report:
sort_ = target_opts.get('sort', profile_config['sort'])
limit = target_opts.get('limit', profile_config['limit'])
print "Profile report for target '%s' (%s)" % (
target, filename)
stats = load_stats()
stats.sort_stats(*sort_)
if limit:
stats.print_stats(limit)
else:
stats.print_stats()
#stats.print_callers()
os.unlink(filename)
return result
return function_named(profiled, fn.__name__)
return decorator
def function_call_count(count=None, versions={}, variance=0.05):
"""Assert a target for a test case's function call count.
count
Optional, general target function call count.
versions
Optional, a dictionary of Python version strings to counts,
for example::
{ '2.5.1': 110,
'2.5': 100,
'2.4': 150 }
The best match for the current running python will be used.
If none match, 'count' will be used as the fallback.
variance
An +/- deviation percentage, defaults to 5%.
"""
# this could easily dump the profile report if --verbose is in effect
version_info = list(sys.version_info)
py_version = '.'.join([str(v) for v in sys.version_info])
try:
from sqlalchemy.cprocessors import to_float
cextension = True
except ImportError:
cextension = False
while version_info:
version = '.'.join([str(v) for v in version_info])
if cextension:
version += "+cextension"
if version in versions:
count = versions[version]
break
version_info.pop()
if count is None:
return lambda fn: fn
def decorator(fn):
def counted(*args, **kw):
try:
filename = "%s.prof" % fn.__name__
elapsed, stat_loader, result = _profile(
filename, fn, *args, **kw)
stats = stat_loader()
calls = stats.total_calls
stats.sort_stats('calls', 'cumulative')
stats.print_stats()
#stats.print_callers()
deviance = int(count * variance)
if (calls < (count - deviance) or
calls > (count + deviance)):
raise AssertionError(
"Function call count %s not within %s%% "
"of expected %s. (Python version %s)" % (
calls, (variance * 100), count, py_version))
return result
finally:
if os.path.exists(filename):
os.unlink(filename)
return function_named(counted, fn.__name__)
return decorator
def conditional_call_count(discriminator, categories):
"""Apply a function call count conditionally at runtime.
Takes two arguments, a callable that returns a key value, and a dict
mapping key values to a tuple of arguments to function_call_count.
The callable is not evaluated until the decorated function is actually
invoked. If the `discriminator` returns a key not present in the
`categories` dictionary, no call count assertion is applied.
Useful for integration tests, where running a named test in isolation may
have a function count penalty not seen in the full suite, due to lazy
initialization in the DB-API, SA, etc.
"""
def decorator(fn):
def at_runtime(*args, **kw):
criteria = categories.get(discriminator(), None)
if criteria is None:
return fn(*args, **kw)
rewrapped = function_call_count(*criteria)(fn)
return rewrapped(*args, **kw)
return function_named(at_runtime, fn.__name__)
return decorator
def _profile(filename, fn, *args, **kw):
global profiler
if not profiler:
if sys.version_info > (2, 5):
try:
import cProfile
profiler = 'cProfile'
except ImportError:
pass
if not profiler:
try:
import hotshot
profiler = 'hotshot'
except ImportError:
profiler = 'skip'
if profiler == 'skip':
raise SkipTest('Profiling not supported on this platform')
elif profiler == 'cProfile':
return _profile_cProfile(filename, fn, *args, **kw)
else:
return _profile_hotshot(filename, fn, *args, **kw)
def _profile_cProfile(filename, fn, *args, **kw):
import cProfile, gc, pstats, time
load_stats = lambda: pstats.Stats(filename)
gc_collect()
began = time.time()
cProfile.runctx('result = fn(*args, **kw)', globals(), locals(),
filename=filename)
ended = time.time()
return ended - began, load_stats, locals()['result']
def _profile_hotshot(filename, fn, *args, **kw):
import gc, hotshot, hotshot.stats, time
load_stats = lambda: hotshot.stats.load(filename)
gc_collect()
prof = hotshot.Profile(filename)
began = time.time()
prof.start()
try:
result = fn(*args, **kw)
finally:
prof.stop()
ended = time.time()
prof.close()
return ended - began, load_stats, result