From: naruse@...
Date: 2016-06-01T16:54:17+00:00
Subject: [ruby-dev:49640] [Ruby trunk Feature#8214] デッドロックチェックに全スレッドのバックトレースダンプの追加
Issue #8214 has been updated by Yui NARUSE.
バックトレースも欲しいらしいので
```
% ./miniruby -ve'Thread.current.name="MainThread!";z=Thread.new{Thread.stop};a,b=Thread.new{1until b;b.join},Thread.new{1until a;a.join};a.name="aaaaa";b.name="bbbbb";z.name="zzzz";a.join'
--- thread.o ---
compiling ../../ruby/thread.c
--- miniruby ---
linking miniruby
ruby 2.4.0dev (2016-06-01 trunk 55248) [x86_64-freebsd10.3]
-e:1:in `join': No live threads left. Deadlock? (fatal)
4 threads, 4 sleeps current:0x00000802c1b200 main thread:0x00000802c17300
* #<Thread:0x00000802d8ce58@MainThread! sleep_forever>
rb_thread_t:0x00000802c17300 native:0x00000802c06400 int:1
-e:1:in `join'
-e:1:in `<main>'
* #<Thread:0x00000802d7f780@zzzz@-e:1 sleep_forever>
rb_thread_t:0x00000802c1ac00 native:0x00000802c07800 int:0
-e:1:in `stop'
-e:1:in `block in <main>'
* #<Thread:0x00000802d7f690@aaaaa@-e:1 sleep_forever>
rb_thread_t:0x00000802c1af00 native:0x00000802c07c00 int:0
depended by: tb_thread_id:0x00000802c1b200
depended by: tb_thread_id:0x00000802c17300
-e:1:in `join'
-e:1:in `block in <main>'
* #<Thread:0x00000802d7f5a0@bbbbb@-e:1 sleep_forever>
rb_thread_t:0x00000802c1b200 native:0x00000802c08000 int:0
depended by: tb_thread_id:0x00000802c1af00
-e:1:in `join'
-e:1:in `block in <main>'
from -e:1:in `<main>'
```
```diff
diff --git a/thread.c b/thread.c
index d17b663..84f76d1 100644
--- a/thread.c
+++ b/thread.c
@@ -2618,7 +2618,7 @@ rb_thread_group(VALUE thread)
}
static const char *
-thread_status_name(rb_thread_t *th)
+thread_status_name(rb_thread_t *th, int detail)
{
switch (th->status) {
case THREAD_RUNNABLE:
@@ -2626,8 +2626,9 @@ thread_status_name(rb_thread_t *th)
return "aborting";
else
return "run";
- case THREAD_STOPPED:
case THREAD_STOPPED_FOREVER:
+ if (detail) return "sleep_forever";
+ case THREAD_STOPPED:
return "sleep";
case THREAD_KILLED:
return "dead";
@@ -2687,7 +2688,7 @@ rb_thread_status(VALUE thread)
}
return Qfalse;
}
- return rb_str_new2(thread_status_name(th));
+ return rb_str_new2(thread_status_name(th, FALSE));
}
@@ -2833,7 +2834,7 @@ rb_thread_inspect(VALUE thread)
VALUE str;
GetThreadPtr(thread, th);
- status = thread_status_name(th);
+ status = thread_status_name(th, TRUE);
str = rb_sprintf("#<%"PRIsVALUE":%p", cname, (void *)thread);
if (!NIL_P(th->name)) {
rb_str_catf(str, "@%"PRIsVALUE, th->name);
@@ -4727,27 +4728,46 @@ ruby_native_thread_p(void)
return th != 0;
}
+VALUE rb_vm_backtrace_str_ary(rb_thread_t *th, long lev, long n);
static void
-debug_deadlock_check(rb_vm_t *vm)
+debug_deadlock_check(rb_vm_t *vm, VALUE msg)
{
-#ifdef DEBUG_DEADLOCK_CHECK
rb_thread_t *th = 0;
+ VALUE sep = rb_str_new_cstr("\n ");
- printf("%d %d %p %p\n", vm_living_thread_num(vm), vm->sleeper, GET_THREAD(), vm->main_thread);
+ rb_str_catf(msg, "\n%d threads, %d sleeps current:%p main thread:%p\n",
+ vm_living_thread_num(vm), vm->sleeper, GET_THREAD(), vm->main_thread);
list_for_each(&vm->living_threads, th, vmlt_node) {
- printf("th:%p %d %d", th, th->status, th->interrupt_flag);
if (th->locking_mutex) {
rb_mutex_t *mutex;
+ struct rb_thread_struct volatile *mth;
+ int waiting;
GetMutexPtr(th->locking_mutex, mutex);
native_mutex_lock(&mutex->lock);
- printf(" %p %d\n", mutex->th, mutex->cond_waiting);
+ mth = mutex->th;
+ waiting = mutex->cond_waiting;
native_mutex_unlock(&mutex->lock);
+ rb_str_catf(msg, "* %+"PRIsVALUE"\n rb_thread_t:%p native:%p int:%u mutex:%p cond:%d\n",
+ th->self, th, th->thread_id,
+ th->interrupt_flag, mth, waiting);
}
- else
- puts("");
+ else {
+ rb_str_catf(msg, "* %+"PRIsVALUE"\n rb_thread_t:%p native:%p int:%u\n",
+ th->self, th, th->thread_id,
+ th->interrupt_flag);
+ }
+ {
+ rb_thread_list_t *list = th->join_list;
+ while (list) {
+ rb_str_catf(msg, " depended by: tb_thread_id:%p\n", list->th);
+ list = list->next;
+ }
+ }
+ rb_str_catf(msg, " ");
+ rb_str_concat(msg, rb_ary_join(rb_vm_backtrace_str_ary(th, 0, 0), sep));
+ rb_str_catf(msg, "\n");
}
-#endif
}
static void
@@ -4782,7 +4802,7 @@ rb_check_deadlock(rb_vm_t *vm)
VALUE argv[2];
argv[0] = rb_eFatal;
argv[1] = rb_str_new2("No live threads left. Deadlock?");
- debug_deadlock_check(vm);
+ debug_deadlock_check(vm, argv[1]);
vm->sleeper--;
rb_threadptr_raise(vm->main_thread, 2, argv);
}
```
----------------------------------------
Feature #8214: デッドロックチェックに全スレッドのバックトレースダンプの追加
https://bugs.ruby-lang.org/issues/8214#change-59006
* Author: Motohiro KOSAKI
* Status: Open
* Priority: Normal
* Assignee:
----------------------------------------
さいきん、くくぱど方面のデッドロックバグの調査を手伝っていて気づいたのですが、
現在のRubyのdeadlock detectorは開発者に優しくないです。
例外があがるとスタックが巻戻ってしまって証拠が消えてしまうので、例外をあげるまえに
全スレッドのバックトレースをダンプするオプションがあるべきではないでしょうか。
デッドロックが起きたときは９９％、直さないといけない状況になるはずで、証拠を残す機能は
有意義なはずです。
TBD: ruby level だけで十分か、Cレベルも含めるか
--
https://bugs.ruby-lang.org/