1 From f12c93efd04991bc982a27e2fa6142538c33ca82 Mon Sep 17 00:00:00 2001
2 From: Rui Ueyama <ruiu@cs.stanford.edu>
3 Date: Sat, 7 May 2022 19:55:24 +0800
4 Subject: [PATCH] Retry if pthread_create fails with EAGAIN
6 On many Unix-like systems, pthread_create can fail spuriously even if
7 the running machine has enough resources to spawn a new thread.
8 Therefore, if EAGAIN is returned from pthread_create, we actually have
11 I observed this issue when running the mold linker
12 (https://github.com/rui314/mold) under a heavy load. mold uses OneTBB
15 As another data point, Go has the same logic to retry on EAGAIN:
16 https://go-review.googlesource.com/c/go/+/33894/
18 nanosleep is defined in POSIX 2001, so I believe that all Unix-like
21 Signed-off-by: Rui Ueyama <ruiu@cs.stanford.edu>
23 src/tbb/rml_thread_monitor.h | 19 ++++++++++++++++++-
24 1 file changed, 18 insertions(+), 1 deletion(-)
26 diff --git a/src/tbb/rml_thread_monitor.h b/src/tbb/rml_thread_monitor.h
27 index 13b556380..5b844b232 100644
28 --- a/src/tbb/rml_thread_monitor.h
29 +++ b/src/tbb/rml_thread_monitor.h
36 #error Unsupported platform
38 @@ -191,8 +192,24 @@ inline thread_monitor::handle_type thread_monitor::launch( void* (*thread_routin
39 check(pthread_attr_init( &s ), "pthread_attr_init has failed");
41 check(pthread_attr_setstacksize( &s, stack_size ), "pthread_attr_setstack_size has failed" );
44 - check( pthread_create( &handle, &s, thread_routine, arg ), "pthread_create has failed" );
47 + int error_code = pthread_create(&handle, &s, thread_routine, arg);
50 + if (error_code != EAGAIN || tries++ > 20) {
51 + handle_perror(error_code, "pthread_create has failed");
55 + // pthreaed_create can spuriously fail on many Unix-like systems.
56 + // Retry after tries * 1 millisecond.
57 + struct timespec ts = {0, tries * 1000 * 1000};
58 + nanosleep(&ts, NULL);
61 check( pthread_attr_destroy( &s ), "pthread_attr_destroy has failed" );