Skip to content

PLUGINS/UCX: Added option to forcibly close UCP endpoints. (#1846) #1950

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

Open
ovidiusm wants to merge 1 commit into
base: release/1.4.0
Choose a base branch
from
Open

PLUGINS/UCX: Added option to forcibly close UCP endpoints. (#1846) #1950

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
2 changes: 1 addition & 1 deletion .gitlab/test_ep.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -64,7 +64,7 @@ run_elastic_test() {
unset UCX_NET_DEVICES # let UCX auto-select GPU-capable transport
# Force NVLink-only transports.
if [[ "$extra_flags" != *--disable-ll-nvlink* ]]; then
export UCX_TLS=cuda_copy,cuda_ipc,sm,self
export UCX_TLS=^rc_gda
fi
PYTHONPATH="${NIXL_BUILD_DIR}/${EP_SRC_DIR}:${EP_SRC_DIR}/tests:${EP_SRC_DIR}/tests/elastic${PYTHONPATH:+:$PYTHONPATH}" \
timeout 300 python3 ${EP_SRC_DIR}/tests/elastic/elastic.py \
Expand Down
3 changes: 2 additions & 1 deletion examples/device/ep/csrc/nixl_ep.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1407,7 +1407,8 @@ void Buffer::_nixl_agent_init() {

const char* num_channels_env = std::getenv("NIXL_EP_NUM_CHANNELS");
init_params["ucx_num_device_channels"] = num_channels_env ? num_channels_env : "4";
init_params["ucx_error_handling_mode"] = "none";
init_params["ucx_error_handling_mode"] = "peer";
init_params["ucx_ep_close_force"] = "yes";
init_params["num_workers"] = std::to_string(1);

nixlBackendH* ucx_backend = nullptr;
Expand Down
13 changes: 12 additions & 1 deletion src/plugins/ucx/ucx_backend.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -32,6 +32,15 @@
#include "absl/strings/str_split.h"
#include <asio.hpp>

namespace {
[[nodiscard]] uint32_t
epCloseFlags(const nixl_b_params_t *custom_params) {
return nixl::getBackendParamDefaulted(custom_params, "ucx_ep_close_force", false) ?
UCP_EP_CLOSE_FLAG_FORCE :
0;
}
} // namespace

/****************************************
* Backend request management
*****************************************/
Expand Down Expand Up @@ -823,6 +832,8 @@ nixlUcxEngine::nixlUcxEngine(const nixlBackendInitParams &init_params)
err_handling_mode = ucx_err_mode_from_string(*opt);
}

const uint32_t ep_close_flags = epCloseFlags(custom_params);

const auto engine_config =
nixl::getBackendParamDefaulted(custom_params, "engine_config", std::string());

Expand All @@ -836,7 +847,7 @@ nixlUcxEngine::nixlUcxEngine(const nixlBackendInitParams &init_params)
uc->warnAboutHardwareSupportMismatch();

for (size_t i = 0; i < num_workers; i++) {
uws.emplace_back(std::make_unique<nixlUcxWorker>(*uc, err_handling_mode));
uws.emplace_back(std::make_unique<nixlUcxWorker>(*uc, err_handling_mode, ep_close_flags));
}

auto &uw = uws.front();
Expand Down
22 changes: 15 additions & 7 deletions src/plugins/ucx/ucx_utils.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -121,9 +121,10 @@ nixlUcxEp::setState(nixl::ucx::ep_state_t new_state) {
}

nixl_status_t
nixlUcxEp::closeImpl(ucp_ep_close_flags_t flags) {
nixlUcxEp::closeImpl() {
ucs_status_ptr_t request = nullptr;
ucp_request_param_t req_param = {.op_attr_mask = UCP_OP_ATTR_FIELD_FLAGS, .flags = flags};
const ucp_request_param_t req_param = {.op_attr_mask = UCP_OP_ATTR_FIELD_FLAGS,
.flags = closeFlags_};

switch (state) {
case nixl::ucx::ep_state_t::UNINITIALIZED:
Expand Down Expand Up @@ -156,7 +157,11 @@ nixlUcxEp::closeImpl(ucp_ep_close_flags_t flags) {
std::terminate();
}

nixlUcxEp::nixlUcxEp(ucp_worker_h worker, void *addr, ucp_err_handling_mode_t err_handling_mode) {
nixlUcxEp::nixlUcxEp(ucp_worker_h worker,
void *addr,
ucp_err_handling_mode_t err_handling_mode,
uint32_t close_flags)
: closeFlags_{close_flags} {
ucp_ep_params_t ep_params;
nixl_status_t status;

Expand Down Expand Up @@ -185,7 +190,7 @@ nixlUcxEp::~nixlUcxEp() {

nixl_status_t
nixlUcxEp::disconnect_nb() {
nixl_status_t status = closeImpl(ucp_ep_close_flags_t(0));
const nixl_status_t status = closeImpl();

// At step of disconnect we can ignore the remote disconnect error.
return (status == NIXL_ERR_REMOTE_DISCONNECT) ? NIXL_SUCCESS : status;
Expand Down Expand Up @@ -517,9 +522,12 @@ nixlUcxWorker::createUcpWorker(const nixlUcxContext &ctx) {
return worker;
}

nixlUcxWorker::nixlUcxWorker(const nixlUcxContext &ctx, ucp_err_handling_mode_t err_handling_mode)
nixlUcxWorker::nixlUcxWorker(const nixlUcxContext &ctx,
ucp_err_handling_mode_t err_handling_mode,
uint32_t ep_close_flags)
: worker(createUcpWorker(ctx), &ucp_worker_destroy),
err_handling_mode_(err_handling_mode) {}
err_handling_mode_(err_handling_mode),
epCloseFlags_(ep_close_flags) {}

std::string
nixlUcxWorker::epAddr() {
Expand All @@ -540,7 +548,7 @@ nixlUcxWorker::epAddr() {
std::unique_ptr<nixlUcxEp>
nixlUcxWorker::connect(void *addr, std::size_t size) {
try {
return std::make_unique<nixlUcxEp>(worker.get(), addr, err_handling_mode_);
return std::make_unique<nixlUcxEp>(worker.get(), addr, err_handling_mode_, epCloseFlags_);
}
catch (const std::exception &e) {
NIXL_ERROR << "UCX endpoint create failed: " << e.what();
Expand Down
14 changes: 10 additions & 4 deletions src/plugins/ucx/ucx_utils.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -45,11 +45,12 @@ class nixlUcxEp {
private:
ucp_ep_h eph{nullptr};
nixl::ucx::ep_state_t state = nixl::ucx::ep_state_t::UNINITIALIZED;
const uint32_t closeFlags_;

void
setState(nixl::ucx::ep_state_t new_state);
nixl_status_t
closeImpl(ucp_ep_close_flags_t flags);
closeImpl();

/* Connection */
nixl_status_t
Expand All @@ -67,7 +68,10 @@ class nixlUcxEp {
return nixl::ucx::toNixlStatus(state);
}

nixlUcxEp(ucp_worker_h worker, void *addr, ucp_err_handling_mode_t err_handling_mode);
nixlUcxEp(ucp_worker_h worker,
void *addr,
ucp_err_handling_mode_t err_handling_mode,
uint32_t close_flags);
~nixlUcxEp();
nixlUcxEp(const nixlUcxEp &) = delete;
nixlUcxEp &
Expand Down Expand Up @@ -187,7 +191,8 @@ class nixlUcxWorker {
public:
explicit nixlUcxWorker(
const nixlUcxContext &,
ucp_err_handling_mode_t ucp_err_handling_mode = UCP_ERR_HANDLING_MODE_NONE);
ucp_err_handling_mode_t ucp_err_handling_mode = UCP_ERR_HANDLING_MODE_NONE,
uint32_t ep_close_flags = 0);

nixlUcxWorker(nixlUcxWorker &&) = delete;
nixlUcxWorker(const nixlUcxWorker &) = delete;
Expand Down Expand Up @@ -241,7 +246,8 @@ class nixlUcxWorker {
createUcpWorker(const nixlUcxContext &);

const std::unique_ptr<ucp_worker, void (*)(ucp_worker *)> worker;
ucp_err_handling_mode_t err_handling_mode_;
const ucp_err_handling_mode_t err_handling_mode_;
const uint32_t epCloseFlags_;
};

[[nodiscard]] nixl_b_params_t
Expand Down
Loading