How many threads does Pluto use vs `Threads.nthreads()`?

[adomasbaliuka]

If I start Julia with --threads=1 (or --threads=6) , running Threads.nthreads() in the console gives 1 (or 6), as expected. If I then use the Julia instance to start a Pluto notebook, Threads.nthreads() in the notebook always gives 4. Why is this? Does it have to do with the browser? Is it representative of how many threads the notebook can actually use and relevant for performance?

Julia Version 1.7.1
Commit ac5cc99908 (2021-12-22 19:35 UTC)
Platform Info:
OS: Linux (x86_64-pc-linux-gnu)
CPU: Intel(R) Core(TM) i7-8565U CPU @ 1.80GHz
WORD_SIZE: 64
LIBM: libopenlibm
LLVM: libLLVM-12.0.1 (ORCJIT, skylake)
Environment:
JULIA_REVISE_WORKER_ONLY = 1

[rikhuijzer]

It's related to this code:

Pluto.jl/src/Configuration.jl

Lines 153 to 170 in f03f73c

	function roughly_the_number_of_physical_cpu_cores()
	# https://gist.github.com/fonsp/738fe244719cae820245aa479e7b4a8d
	if Sys.CPU_THREADS == 1
	1
	elseif Sys.CPU_THREADS == 2 || Sys.CPU_THREADS == 3
	2
	else
	Sys.CPU_THREADS ÷ 2
	end
	end
	function default_number_of_threads()
	env_value = get(ENV, "JULIA_NUM_THREADS", "")
	all(isspace, env_value) ?
	roughly_the_number_of_physical_cpu_cores() :
	env_value
	end

Why this is the case only Fons knows.

Can you try overriding it by setting the environment variable JULIA_NUM_THREADS instead of the startup flag? Does that change anything?

[fonsp]

The flags that you give to julia only apply to Pluto's server, not to the notebook processes that it spawns. So if you run Threads.nthreads() in a notebook, it is giving you the number of threads of the notebook process, not the server process.

You can set parameters for notebook processes with keyword arguments to Pluto.run, e.g.

Pluto.run(threads="16", compile="0")