From 986a664109ead1f48b1ea542f175ae2ee68b3b51 Mon Sep 17 00:00:00 2001
From: JamesWrigley <james@puiterwijk.org>
Date: Tue, 7 Apr 2026 23:51:52 +0200
Subject: [PATCH] Cleanly shutdown the GC messages task

`start_gc_msgs_task()` previously launched a lingering task that could cause
hangs when Julia is started with multiple threads.
---
 src/cluster.jl    | 17 ++++++++++++++++-
 src/remotecall.jl |  6 ++++--
 2 files changed, 20 insertions(+), 3 deletions(-)

diff --git a/src/cluster.jl b/src/cluster.jl
index 848b908..1bfab27 100644
--- a/src/cluster.jl
+++ b/src/cluster.jl
@@ -1337,12 +1337,26 @@ end
 
 using Random: randstring
 
+# Exit handler state
+const shutting_down = Threads.Atomic{Bool}(false)
+
+function atexit_handler()
+    if inited[]
+        terminate_all_workers()
+    end
+
+    shutting_down[] = true
+    @lock any_gc_flag notify(any_gc_flag)
+    if !isnothing(gc_msgs_task)
+        wait(gc_msgs_task::Task)
+    end
+end
+
 # do initialization that's only needed when there is more than 1 processor
 const inited = Threads.Atomic{Bool}(false)
 function init_multi()
     if !Threads.atomic_cas!(inited, false, true)
         push!(Base.package_callbacks, _require_callback)
-        atexit(terminate_all_workers)
         init_bind_addr()
         cluster_cookie(randstring(HDR_COOKIE_LEN))
     end
@@ -1350,6 +1364,7 @@ function init_multi()
 end
 
 function init_parallel()
+    atexit(atexit_handler)
     start_gc_msgs_task()
 
     # start in "head node" mode, if worker, will override later.
diff --git a/src/remotecall.jl b/src/remotecall.jl
index 19ec2f7..e4dea65 100644
--- a/src/remotecall.jl
+++ b/src/remotecall.jl
@@ -272,10 +272,12 @@ end
 # XXX: Is this worth the additional complexity?
 #      `flush_gc_msgs` has to iterate over all connected workers.
 const any_gc_flag = Threads.Condition()
+gc_msgs_task::Union{Task, Nothing} = nothing
+
 function start_gc_msgs_task()
-    errormonitor(
+    global gc_msgs_task = errormonitor(
         @async begin
-            while true
+            while !shutting_down[]
                 lock(any_gc_flag) do
                     # this might miss events
                     wait(any_gc_flag)