Speed up id generation

py_zipkin/util.py

@@ -1,7 +1,7 @@
 # -*- coding: utf-8 -*-
-import codecs
-import os
+import random
 import struct
+import time
 
 
 def generate_random_64bit_string():
@@ -11,16 +11,22 @@ def generate_random_64bit_string():
 
     :returns: random 16-character string
     """
-    return str(codecs.encode(os.urandom(8), 'hex_codec').decode('utf-8'))
+    return '{:016x}'.format(random.getrandbits(64))
 
 
 def generate_random_128bit_string():
     """Returns a 128 bit UTF-8 encoded string. Follows the same conventions
     as generate_random_64bit_string().
 
-    :returns: random 32-character string
+    The upper 32 bits are the current time in epoch seconds, and the
+    lower 96 bits are random. This allows for AWS X-Ray `interop
+    <https://github.com/openzipkin/zipkin/issues/1754>`_
+
+    :returns: 32-character hex string
     """
-    return str(codecs.encode(os.urandom(16), 'hex_codec').decode('utf-8'))
+    t = int(time.time())
+    lower_96 = random.getrandbits(96)
+    return '{:032x}'.format((t << 96) | lower_96)
 
 
 def unsigned_hex_to_signed_int(hex_string):

requirements-dev.txt

@@ -2,3 +2,4 @@ coverage
 mock
 ordereddict
 pytest
+pytest-benchmark[histogram]

tests/profiling/zipkin_span_benchmark_test.py

@@ -0,0 +1,11 @@
+import pytest
+
+import py_zipkin.zipkin as zipkin
+
+
+@pytest.mark.parametrize('use_128', [False, True])
+def test_create_attrs_for_span(benchmark, use_128):
+    benchmark(
+        zipkin.create_attrs_for_span,
+        use_128bit_trace_id=use_128,
+    )

tests/util_test.py

@@ -5,21 +5,24 @@
 from py_zipkin import util
 
 
-@mock.patch('py_zipkin.util.codecs.encode', autospec=True)
+@mock.patch('py_zipkin.util.random.getrandbits', autospec=True)
 def test_generate_random_64bit_string(rand):
-    rand.return_value = b'17133d482ba4f605'
+    rand.return_value = 0x17133d482ba4f605
     random_string = util.generate_random_64bit_string()
     assert random_string == '17133d482ba4f605'
     # This acts as a contract test of sorts. This should return a str
     # in both py2 and py3. IOW, no unicode objects.
     assert isinstance(random_string, str)
 
 
-@mock.patch('py_zipkin.util.codecs.encode', autospec=True)
-def test_generate_random_128bit_string(rand):
-    rand.return_value = b'17133d482ba4f60517133d482ba4f605'
+@mock.patch('py_zipkin.util.time.time', autospec=True)
+@mock.patch('py_zipkin.util.random.getrandbits', autospec=True)
+def test_generate_random_128bit_string(rand, mock_time):
+    rand.return_value = 0x2ba4f60517133d482ba4f605
+    mock_time.return_value = float(0x17133d48)
     random_string = util.generate_random_128bit_string()
     assert random_string == '17133d482ba4f60517133d482ba4f605'
+    rand.assert_called_once_with(96)  # 96 bits
     # This acts as a contract test of sorts. This should return a str
     # in both py2 and py3. IOW, no unicode objects.
     assert isinstance(random_string, str)

tox.ini

@@ -5,7 +5,7 @@ envlist = pre-commit, py27, py34, py35, py36, flake8
 deps = -rrequirements-dev.txt
 commands =
     coverage erase
-    coverage run --source=py_zipkin/ -m pytest -vv {posargs:tests}
+    coverage run --source=py_zipkin/ -m pytest -vv --benchmark-skip {posargs:tests}
     coverage report -m --show-missing --fail-under 100
 
 [testenv:pre-commit]
@@ -19,6 +19,17 @@ deps = flake8
 commands =
     flake8 py_zipkin tests
 
+[testenv:benchmark]
+basepython = python3.6
+deps =
+    -rrequirements-dev.txt
+commands =
+    python -m pytest -vv --capture=no {posargs:tests/profiling} \
+        --benchmark-only --benchmark-min-rounds=15 \
+        --benchmark-group-by func --benchmark-name short \
+        --benchmark-save=benchmark --benchmark-save-data \
+        --benchmark-histogram=.benchmarks/benchmark
+
 [flake8]
 exclude = .svn,CVS,.bzr,.hg,.git,__pycache__,.tox,docs,python_zipkin/thrift/zipkinCore
 max_line_length = 83