Pthread_setname_np(3)

Community
1

Let me preface this with the fact that I’m not an expert in pthreads by any measure. Looking around I’ve found that Ruby’s second thread is named:

$ cat /proc/30193/task/*/comm
ruby
ruby-timer-thr

Maybe it would make sense for Crystal to do the same?

Another suggestion would be to expose a way to name the main thread (and hopefully any thread past 1.0) other than writing to /proc.

2

I am not an expert on pthreads but why would we want to match that pattern? Are their tools or something that expect that pattern?

3

It would make sense to name runtime threads in order to be able to identify them.

1 Like
4

But we only have one thread at the moment.

1 Like
5

So, if I’m mistaken here please set me straight.

Underneath each of the /proc/[pid] directories, a task subdirectory contains subdirectories of the form task/[tid], which contain corresponding information about each of the threads in the process, where tid is the kernel thread ID of the thread.

$ grep -c processor /proc/cpuinfo
2
$ cat ./test.cr
#!/usr/bin/env crystal
sleep 1
system  "ls -lA /proc/#{Process.pid}/task"
sleep 1
spawn do
        sleep
end
system  "ls -lA /proc/#{Process.pid}/task"
sleep 1
$ ./test.cr
total 0
dr-xr-xr-x 7 x x 0 May 21 19:28 29680
dr-xr-xr-x 7 x x 0 May 21 19:28 29681
total 0
dr-xr-xr-x 7 x x 0 May 21 19:28 29680
dr-xr-xr-x 7 x x 0 May 21 19:28 29681

The real program has 12 tasks, running on a 12-vcore CPU, so I guess this is a thread pool?

I guess my first hunch that the second thread was the GC/scheduler was plain wrong.

6

This could be part of Crystal’s concurrency goals.

7

The non-main (tid != pid) threads are the GC, but we don’t have control over those threads to name them.

8

I’m curious as to why there are ncpu - 1 of them though.