Expand description
The worker pool of a command (“Kizz never panics”).
A command that has parallel work builds one WorkerPool before that work, and
passes it to the code that needs it: a rayon pool of as many threads as
std::thread::available_parallelism gives, at most MAX_WORKERS, and 1 when
it gives an error. Each worker starts through kizz_threads, which waits and
retries while the system cannot start a thread for the moment, and never panics;
a worker that still cannot start makes WorkerPool::new return the error, and
rayon stops the workers that started.
The cache check, the recording of observations, cache cleanup, BLAKE3’s parallel
hashing and the discovery walk run inside the pool, through
WorkerPool::install, so BLAKE3’s rayon::join runs on it too, and rayon’s
global pool, which panics when it cannot start a thread, never starts.
Structs§
- Pool
Error - A worker pool that could not be built because a worker could not start.
- Worker
Pool - The pool of threads that runs a command’s parallel work.
Constants§
- MAX_
WORKERS - The most threads a worker pool has. The measurement report times the check of 10,000 cached steps at 0.73 seconds with 4 threads, 0.83 seconds with 1, and 1.53 seconds with the default pool of 32 threads.