Skip to content

Fix data race in distributed ring on pending send/recv lists #4737

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
zcbenz merged 2 commits into from
Oct 23, 2026
Merged

Fix data race in distributed ring on pending send/recv lists #4737

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
a3649bc
[ring] Only touch the socket task queues while holding the lock
Shakhtar-Sankur Oct 21, 2026
e534838
nit
zcbenz Oct 23, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
23 changes: 18 additions & 5 deletions mlx/distributed/ring/ring.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -80,7 +80,10 @@ class SocketThread {
fcntl(fd, F_SETFL, flags | O_NONBLOCK);
}
~SocketThread() {
stop_ = true;
{
std::unique_lock lock(queue_mutex_);
stop_ = true;
}
condition_.notify_all();
worker_.join();
int flags = fcntl(fd_, F_GETFL, 0);
Expand Down Expand Up @@ -151,6 +154,8 @@ class SocketThread {
bool delete_recv = false;
bool delete_send = false;
while (true) {
SocketTask* recv_task = nullptr;
SocketTask* send_task = nullptr;
{
std::unique_lock lock(queue_mutex_);

Expand All @@ -175,10 +180,17 @@ class SocketThread {
return;
}
}

if (!recvs_.empty()) {
recv_task = &recvs_.front();
}
if (!sends_.empty()) {
send_task = &sends_.front();
}
}

if (!recvs_.empty()) {
auto& task = recvs_.front();
if (recv_task) {
auto& task = *recv_task;
ssize_t r =
::recv(fd_, task.buffer, std::min(task.size, MAX_IO_BYTES), 0);
if (r > 0) {
Expand All @@ -195,8 +207,8 @@ class SocketThread {
true, "Receiving from socket", fd_, "failed with errno", errno);
}
}
if (!sends_.empty()) {
auto& task = sends_.front();
if (send_task) {
auto& task = *send_task;
ssize_t r =
::send(fd_, task.buffer, std::min(task.size, MAX_IO_BYTES), 0);
if (r > 0) {
Expand All @@ -218,6 +230,7 @@ class SocketThread {
// Throw exception in invoker's thread.
auto error = std::make_exception_ptr(
std::runtime_error("[ring] connection to a peer was lost"));
std::unique_lock lock(queue_mutex_);
for (auto& task : recvs_) {
task.promise.set_exception(error);
}
Expand Down
Loading