Skip to content

Commit 01a0c81

Browse files
XilunWufacebook-github-bot
authored andcommitted
Increase socket buffer size to allow ProcessGroup init up to 12k ranks
Summary: The c10d socket and gloo listener both set their buffer size to 2048 which causes connection issue at 4k scale. This diff sets the buffer size to `-1` which uses `somaxconn` as the actual buffer size, aiming to enable 24k PG init without crash. The experiment shows the ability to successful creation of 12k ranks without crash. split the original diff for OSS vs. internal. Caution: we need the change on both gloo and c10d to enable 12k PG init. Updating only one side may not offer the benefit. Reviewed By: wconstab, bmaurer Differential Revision: D48617912 fbshipit-source-id: 3ba40d1b94c113a268ded0ea8f51a03daa1233d3
1 parent cf1e1ab commit 01a0c81

File tree

1 file changed

+1
-1
lines changed

1 file changed

+1
-1
lines changed

gloo/transport/tcp/listener.h

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -32,7 +32,7 @@ class Listener final : public Handler {
3232
using connect_callback_t =
3333
std::function<void(std::shared_ptr<Socket> socket, Error error)>;
3434

35-
static constexpr int kBacklog = 2048;
35+
static constexpr int kBacklog = -1; // allow somaxconn
3636

3737
Listener(std::shared_ptr<Loop> loop, const attr& attr);
3838

0 commit comments

Comments
 (0)