mirror of https://github.com/apache/cloudstack.git
4014169ccb
Factor in inaccurancy of System.nanoTime in class ProfilerThe time diff measurement by the class Profiler, is done by System.nanoTime().
System.nanoTime() is not 100% accurate (lots of info on this on the web). The inaccurance seems to vary/depend with different dependencies (multi-core/OS/kernels).
Problem:
In my case the code which measures `Thread.sleep(ONE_SECOND);` ~~takes~~ measures 999 milliseconds, which breaks the test expecting the sleep of 1000 milliseconds to ~~take~~ measure at least 1000 milliseconds.
Solution:
A `MARGIN` (of 100 milliseconds) is already present to factor in some time for overhead/surrounding code which adds time. Another margin could be added to factor in the possible time measurement error.
To limit additional lines of code, the constant `ONE_SECOND`, which is used by the testcase (sleep), is increased from 1000 milliseconds to `1000l + (MARGIN / 10)` (effectively 1010 milliseconds).
Thereby the inaccurancy on my system of ~1% ("faster") is caught as the `sleep(1010)` is measured as '1009' milliseconds.
* pr/1324:
Factor in inaccurancy of System.nanoTime
Signed-off-by: Remi Bergsma <github@remi.nl>
|
||
|---|---|---|
| .. | ||
| bindir | ||
| conf | ||
| src | ||
| pom.xml | ||