The purpose of this package is to provide both reasonable statistical and deterministic profiling to be used in running (multi-threading) production environment. Up to now, I'm just trying to prove the concept and this package is just in the pre-alpha stage.
The deterministic profiler is mainly an container of row profiling
informations. Actual profiling is performed by any object implementing the same
_lsprof.Profiler interface. Trough the provided decorators and context manager,
developers can control what code they want to profile: information are stored into the
container at the end of each single profile. Profiling and information storage
must be thread and sub-call/recursion wise.
Statistical profiling is performed using sys._current_frames. Both signal
and thread based statistical profiling are provided for flexibility. Each
profile is collected and serialized as soon as it's produced.
Both deterministic and statistical profiling can be enabled/disabled trough a class variable, and initially setup with environment variables. Whenever instructed so, a collector class can collect profiling information and serialize for later use. This way, developers can decide when/how to profile.