From 6a3fd4e40236b37399720d5620afd12ddbe2168c Mon Sep 17 00:00:00 2001 From: drebs Date: Fri, 30 Jun 2017 11:48:35 -0300 Subject: [test] change memory sampling interval to 0.1 on benchmarks --- testing/tests/benchmarks/README.md | 44 ++++++++++++++++++++++++++++++++++++ testing/tests/benchmarks/conftest.py | 2 +- 2 files changed, 45 insertions(+), 1 deletion(-) create mode 100644 testing/tests/benchmarks/README.md diff --git a/testing/tests/benchmarks/README.md b/testing/tests/benchmarks/README.md new file mode 100644 index 00000000..06a75282 --- /dev/null +++ b/testing/tests/benchmarks/README.md @@ -0,0 +1,44 @@ +Benchmark tests +=============== + +This folder contains benchmark tests for Soledad. It aims to provide a fair +account on the time and resources taken to perform some actions. + +These benchmarks are built on top of `pytest-benchmark`, a `pytest` fixture that +provides means for running test functions multiple times and generating +reports. The results are printed to screen and also posted to elasticsearch. + +`pytest-benchmark` runs tests multiple times so it can provide meaningful +statistics for the time taken for a tipical run of a test function. The number +of times that the test is run can be manually or automatically configured. When +automatically configured, the number of runs is decided by taking into account +multiple `pytest-benchmark` configuration parameters. See the following page +for more details on how `pytest-benchmark` works: + + https://pytest-benchmark.readthedocs.io/en/stable/calibration.html + +Some graphs and analysis resulting from these tests can be seen on: + + https://benchmarks.leap.se/ + + +Resource consumption +-------------------- + +For each test, CPU and memory usage statistics are also collected, by querying +`cpu_percent()` and `memory_percent()` from `psutil.Process` for the current +test process. Some notes about the current resource consumption estimation process: + +* Currently, resources are measured for the whole set of rounds that a test + function is run. That means that the CPU and memory percentage include the + `pytest` and `pytest-benchmark` machinery overhead. Anyway, for now this might + provide a fair approximation of per-run test function resource usage. + +* CPU is measured before and after the run of the benchmark function and + returns the percentage that the currnet process occupied of the CPU time + between the two calls. + +* Memory is sampled during the benchmark run by a separate thread. Sampling + interval might have to be configured on a per-test basis, as different tests + take different times to execute (from milliseconds to tens of seconds). For + now, an interval of 0.1s seems to cover all tests. diff --git a/testing/tests/benchmarks/conftest.py b/testing/tests/benchmarks/conftest.py index cfad458a..4dbc4377 100644 --- a/testing/tests/benchmarks/conftest.py +++ b/testing/tests/benchmarks/conftest.py @@ -88,7 +88,7 @@ def txbenchmark_with_setup(monitored_benchmark_with_setup): class ResourceWatcher(threading.Thread): - sampling_interval = 1 + sampling_interval = 0.1 def __init__(self): threading.Thread.__init__(self) -- cgit v1.2.3