Make profiling more robust with many tasks (JuliaLang#42978)

This patch includes two sets of changes.

(1) `jl_thread_suspend_and_get_state` uses `pthread_cond_timedwait` to
recover from the case where the request is not received by the signal
handler. This is required because `usr2_handler` contains some paths
for the case where it is not possible to obtain `ptls`.

(2) `ctx_switch` now makes sure to null out `ptls` of the last task
(`lastt->ptls = NULL`) after changing the current task by updating
pgcstack (`jl_set_pgcstack(&t->gcstack)`).  This closes the gap in
which `usr2_handler` can observe the null `ptls`.

Co-authored-by: Jameson Nash <[email protected]>
(cherry picked from commit 8131580)

Author: 2 people

src/signals-unix.c

@@ -7,6 +7,7 @@
 #include <sys/stat.h>
 #include <sys/mman.h>
 #include <pthread.h>
+#include <time.h>
 #include <errno.h>
 #if defined(_OS_DARWIN_) && !defined(MAP_ANONYMOUS)
 #define MAP_ANONYMOUS MAP_ANON
@@ -364,11 +365,25 @@ static pthread_cond_t signal_caught_cond;
 
 static void jl_thread_suspend_and_get_state(int tid, unw_context_t **ctx)
 {
+    struct timespec ts;
+    clock_gettime(CLOCK_REALTIME, &ts);
+    ts.tv_sec += 1;
     pthread_mutex_lock(&in_signal_lock);
     jl_ptls_t ptls2 = jl_all_tls_states[tid];
     jl_atomic_store_release(&ptls2->signal_request, 1);
     pthread_kill(ptls2->system_id, SIGUSR2);
-    pthread_cond_wait(&signal_caught_cond, &in_signal_lock);  // wait for thread to acknowledge
+    // wait for thread to acknowledge
+    int err = pthread_cond_timedwait(&signal_caught_cond, &in_signal_lock, &ts);
+    if (err == ETIMEDOUT) {
+        sig_atomic_t request = 1;
+        if (jl_atomic_cmpswap(&ptls2->signal_request, &request, 0)) {
+            *ctx = NULL;
+            pthread_mutex_unlock(&in_signal_lock);
+            return;
+        }
+        err = pthread_cond_wait(&signal_caught_cond, &in_signal_lock);
+    }
+    assert(!err);
     assert(jl_atomic_load_acquire(&ptls2->signal_request) == 0);
     *ctx = signal_context;
 }
@@ -750,6 +765,8 @@ static void *signal_listener(void *arg)
             for (int i = jl_n_threads; i-- > 0; ) {
                 // notify thread to stop
                 jl_thread_suspend_and_get_state(i, &signal_context);
+                if (signal_context == NULL)
+                    continue;
 
                 // do backtrace on thread contexts for critical signals
                 // this part must be signal-handler safe

src/task.c

@@ -388,18 +388,19 @@ static void ctx_switch(jl_task_t *lastt)
         else
 #endif
         *pt = NULL; // can't fail after here: clear the gc-root for the target task now
-        lastt->ptls = NULL;
     }
 
     // set up global state for new task and clear global state for old task
     t->ptls = ptls;
     ptls->current_task = t;
     JL_GC_PROMISE_ROOTED(t);
+    jl_signal_fence();
+    jl_set_pgcstack(&t->gcstack);
+    jl_signal_fence();
     lastt->ptls = NULL;
 #ifdef MIGRATE_TASKS
     ptls->previous_task = lastt;
 #endif
-    jl_set_pgcstack(&t->gcstack);
 
 #if defined(JL_TSAN_ENABLED)
     tsan_switch_to_ctx(&t->tsan_state);

test/profile_spawnmany_exec.jl

@@ -0,0 +1,14 @@
+# This file is a part of Julia. License is MIT: https://julialang.org/license
+
+using Profile
+
+function spawnmany(n)
+    if n > 2
+        m = n ÷ 2
+        t = Threads.@spawn spawnmany(m)
+        spawnmany(m)
+        wait(t)
+    end
+end
+
+@profile spawnmany(parse(Int, get(ENV, "NTASKS", "2000000")))

test/threads.jl

@@ -147,3 +147,39 @@ end
 
 # We don't need the watchdog anymore
 close(proc.in)
+
+# https://github.com/JuliaLang/julia/pull/42973
+Sys.islinux() && @testset "spawn and wait *a lot* of tasks in @profile" begin
+    # Not using threads_exec.jl for better isolation, reproducibility, and a
+    # tighter timeout.
+    script = "profile_spawnmany_exec.jl"
+    cmd = `$(Base.julia_cmd()) --depwarn=error --rr-detach --startup-file=no $script`
+    @testset for n in [20000, 200000, 2000000]
+        proc = run(ignorestatus(setenv(cmd, "NTASKS" => n; dir = @__DIR__)); wait = false)
+        done = Threads.Atomic{Bool}(false)
+        timeout = false
+        timer = Timer(100) do _
+            timeout = true
+            for sig in [Base.SIGTERM, Base.SIGHUP, Base.SIGKILL]
+                for _ in 1:1000
+                    kill(proc, sig)
+                    if done[]
+                        if sig != Base.SIGTERM
+                            @warn "Terminating `$script` required signal $sig"
+                        end
+                        return
+                    end
+                    sleep(0.001)
+                end
+            end
+        end
+        try
+            wait(proc)
+        finally
+            done[] = true
+            close(timer)
+        end
+        @test success(proc)
+        @test !timeout
+    end
+end