quantlab.factor.kunquant.shared_executor#

quantlab.factor.kunquant.shared_executor(num_threads)[source]#

Return the process-wide KunQuant multi-thread executor for num_threads.

Every KunQuant run in quantlab takes its executor from here instead of calling KunRunner.createMultiThreadExecutor. The first call for a thread count creates the executor; later calls return the same object, which lives until the interpreter exits.

KunQuant’s executor destructor (kunquant 0.1.11) can miss its wake-up of a worker thread that is starting up or settling after a run; its join then never returns, and since it holds the GIL the whole process freezes. So no executor is destroyed while the program runs: one per thread count is reused, and _release_executors destroys them at exit after the workers have settled.

Reuse is safe because runGraph and StreamContext.run wait until all their work is done before returning, so no work is left on the executor between runs. The idle worker threads of a cached executor wait on a condition variable and use no CPU.

Parameters:

num_threadsint: Number of worker threads, as FactorConfig.njobs.

Returns:

KunRunner.Executor: The cached executor for num_threads.

Examples

>>> shared_executor(4) is shared_executor(4)
True
>>> shared_executor(4) is shared_executor(2)
False