From 5d25487b8786419b6dd4fa36fa87014c20591cbc Mon Sep 17 00:00:00 2001 From: Wataru Ishida Date: Tue, 12 Mar 2024 11:36:11 +0000 Subject: [PATCH] chore(test): more complex setting for ring test Signed-off-by: Wataru Ishida --- Dockerfile | 1 + reduction_server/src/ring.rs | 6 ++++-- 2 files changed, 5 insertions(+), 2 deletions(-) diff --git a/Dockerfile b/Dockerfile index 22bcbaf..473282d 100644 --- a/Dockerfile +++ b/Dockerfile @@ -47,6 +47,7 @@ RUN cd reduction_server && cargo build -r FROM optcast AS unittest ENV RUST_LOG=info +ENV NCCL_SOCKET_IFNAME=lo RUN cd reduction_server && cargo test --all -- --nocapture FROM nvcr.io/nvidia/cuda:12.3.1-devel-ubuntu22.04 AS final diff --git a/reduction_server/src/ring.rs b/reduction_server/src/ring.rs index ce18ca9..a839c5b 100644 --- a/reduction_server/src/ring.rs +++ b/reduction_server/src/ring.rs @@ -603,15 +603,17 @@ mod tests { .map(|i| { std::thread::spawn(move || { let ring_rank = format!("{}", i + 1); - let address = format!("127.0.0.1:{}", 9090 + i); + let address = format!("127.0.0.1:{},127.0.0.1:{}", 9090 + i, 9100 + i); let args = Args::parse_from([ "--bench", "--nrank", "4", "--reduce-threads", - "1", + "2", "--address", &address, + "--nreq", + "1", // when using socket plugin, concurrent recv/send requests doesn't work "--ring-rank", &ring_rank, ]);