blob: a00cd484e2db00723b2dfdc958da3594e9d88bb5 [file] [log] [blame]
/* Copyright 2016 The TensorFlow Authors. All Rights Reserved.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
==============================================================================*/
#include "tensorflow/core/distributed_runtime/session_mgr.h"
#include <algorithm>
#include <string>
#include <utility>
#include "tensorflow/core/common_runtime/device_mgr.h"
#include "tensorflow/core/common_runtime/renamed_device.h"
#include "tensorflow/core/distributed_runtime/coordination/coordination_service.h"
#include "tensorflow/core/distributed_runtime/coordination/coordination_service_agent.h"
#include "tensorflow/core/distributed_runtime/error_payloads.h"
#include "tensorflow/core/distributed_runtime/graph_mgr.h"
#include "tensorflow/core/distributed_runtime/remote_device.h"
#include "tensorflow/core/distributed_runtime/worker_cache_wrapper.h"
#include "tensorflow/core/lib/strings/strcat.h"
#include "tensorflow/core/protobuf/cluster.pb.h"
#include "tensorflow/core/protobuf/coordination_config.pb.h"
#include "tensorflow/core/protobuf/distributed_runtime_payloads.pb.h"
#include "tensorflow/core/protobuf/tensorflow_server.pb.h"
#include "tensorflow/core/util/ptr_util.h"
namespace tensorflow {
SessionMgr::SessionMgr(
WorkerEnv* worker_env, const std::string& default_worker_name,
std::unique_ptr<WorkerCacheInterface> default_worker_cache,
WorkerCacheFactory worker_cache_factory)
: worker_env_(worker_env),
default_worker_cache_(std::move(default_worker_cache)),
legacy_session_(WorkerSession::CreateWithBorrowedDeviceMgr(
"", default_worker_name,
std::unique_ptr<WorkerCacheInterface>(
new WorkerCacheWrapper(default_worker_cache_.get())),
worker_env->device_mgr,
std::unique_ptr<GraphMgr>(
new GraphMgr(worker_env, worker_env->device_mgr)),
nullptr)),
worker_cache_factory_(std::move(worker_cache_factory)) {}
/* static */
std::string SessionMgr::WorkerNameFromServerDef(const ServerDef& server_def) {
return strings::StrCat("/job:", server_def.job_name(),
"/replica:0/task:", server_def.task_index());
}
Status SessionMgr::CreateSession(const std::string& session,
const ServerDef& server_def,
bool isolate_session_state) {
return CreateSession(session, server_def, {}, isolate_session_state);
}
Status SessionMgr::CreateSession(
const std::string& session, const ServerDef& server_def,
const protobuf::RepeatedPtrField<DeviceAttributes>&
cluster_device_attributes,
bool isolate_session_state) {
return CreateSession(
session, server_def, cluster_device_attributes, isolate_session_state,
/*master_task=*/"",
/*master_incarnation=*/0, /*coordination_service_config=*/{});
}
Status SessionMgr::CreateSession(
const std::string& session, const ServerDef& server_def,
const protobuf::RepeatedPtrField<DeviceAttributes>&
cluster_device_attributes,
bool isolate_session_state, std::string master_task,
int64_t master_incarnation,
const CoordinationServiceConfig& coordination_service_config) {
mutex_lock l(mu_);
if (session.empty()) {
return errors::InvalidArgument("Session must be non-empty.");
}
// For given master task name, check if one or more `WorkerSession`s have been
// created previously on this worker, and if so garbage collect the expired
// `WorkerSession`s. This happens when the master fails before sending
// `DeleteSession` requests, which can cause `WorkerSession`s to be leaked.
if (!master_task.empty()) {
auto it_range = master_to_associated_sessions_.equal_range(master_task);
if (it_range.first != it_range.second &&
it_range.first->second.master_incarnation != master_incarnation) {
LOG(INFO) << "When creating WorkerSession for master task " << master_task
<< ", found old WorkerSessions created by the same master task "
<< "with a different incarnation. These sessions will "
<< "be garbage collected. Current WorkerSession count: "
<< sessions_.size();
auto it = it_range.first;
while (it != it_range.second) {
auto session_it = sessions_.find(it->second.session_handle);
if (session_it != sessions_.end()) {
sessions_.erase(session_it);
}
it = master_to_associated_sessions_.erase(it);
}
}
}
WorkerCacheInterface* worker_cache = nullptr;
std::string worker_name;
if (server_def.cluster().job().empty()) {
worker_cache = new WorkerCacheWrapper(default_worker_cache_.get());
worker_name = legacy_session_->worker_name();
} else {
TF_RETURN_IF_ERROR(worker_cache_factory_(server_def, &worker_cache));
worker_name = WorkerNameFromServerDef(server_def);
}
if (worker_cache != nullptr && default_worker_cache_ != nullptr) {
worker_cache->SetLogging(this->is_logging_active_);
}
CHECK(!worker_env_->local_devices.empty())
<< "The WorkerEnv must have at least one device in `local_devices`.";
std::shared_ptr<WorkerSession> worker_session;
std::vector<std::unique_ptr<Device>> cluster_devices;
if (isolate_session_state || server_def.cluster().job_size()) {
if (server_def.cluster().job_size()) {
VLOG(1) << "ClusterSpec propagation is enabled.";
}
if (!isolate_session_state) {
VLOG(1) << "Session state isolation is disabled.";
}
// Create a private copy of the DeviceMgr for the WorkerSession.
std::vector<std::unique_ptr<Device>> renamed_devices;
for (Device* d : worker_env_->local_devices) {
renamed_devices.push_back(RenamedDevice::NewRenamedDevice(
worker_name, d, false, isolate_session_state));
}
auto device_mgr = MakeUnique<StaticDeviceMgr>(std::move(renamed_devices));
LookupLocalDevice cb = [&device_mgr](StringPiece name, Device** device) {
return device_mgr->LookupDevice(name, device);
};
AsRemoteDevices(worker_env_->env, cluster_device_attributes, cb,
&cluster_devices);
std::unique_ptr<DynamicDeviceMgr> remote_devices;
if (!cluster_device_attributes.empty()) {
remote_devices = MakeUnique<DynamicDeviceMgr>();
TF_RETURN_IF_ERROR(
remote_devices->AddDevices(std::move(cluster_devices)));
}
auto graph_mgr = MakeUnique<GraphMgr>(worker_env_, device_mgr.get());
worker_session.reset(
new WorkerSession(session, worker_name,
std::unique_ptr<WorkerCacheInterface>(worker_cache),
std::move(device_mgr), std::move(graph_mgr),
std::move(remote_devices)));
} else {
AsRemoteDevices(worker_env_->env, cluster_device_attributes, nullptr,
&cluster_devices);
std::unique_ptr<DynamicDeviceMgr> remote_devices;
if (!cluster_device_attributes.empty()) {
remote_devices = MakeUnique<DynamicDeviceMgr>();
TF_RETURN_IF_ERROR(
remote_devices->AddDevices(std::move(cluster_devices)));
}
// Borrow the WorkerEnv's DeviceMgr for the WorkerSession, so
// that resources using it can use its devices after the
// WorkerSession has been deleted.
auto graph_mgr = MakeUnique<GraphMgr>(worker_env_, worker_env_->device_mgr);
worker_session = WorkerSession::CreateWithBorrowedDeviceMgr(
session, worker_name,
std::unique_ptr<WorkerCacheInterface>(worker_cache),
worker_env_->device_mgr, std::move(graph_mgr),
std::move(remote_devices));
}
sessions_.insert(std::make_pair(session, std::move(worker_session)));
if (!master_task.empty()) {
MasterAssociatedSession s{master_incarnation, session};
master_to_associated_sessions_.emplace(master_task, s);
}
// If configured, enable coordination service and agent in the first worker
// session.
if (!coordination_service_config.service_type().empty() &&
coordination_service_agent_ == nullptr) {
std::unique_ptr<CoordinationClientCache> client_cache;
TF_RETURN_IF_ERROR(worker_cache->GetCoordinationClientCache(&client_cache));
// Note: If this worker is not the leader, no service instance will be
// returned. Hence, only the worker leader in the cluster would hold the
// coordination service instance.
coordination_service_ =
CoordinationServiceInterface::EnableCoordinationService(
coordination_service_config.service_type(), worker_env_->env,
server_def, std::move(client_cache));
std::unique_ptr<CoordinationClientCache> agent_cache;
TF_RETURN_IF_ERROR(worker_cache->GetCoordinationClientCache(&agent_cache));
coordination_service_agent_ = CreateCoordinationServiceAgent();
TF_RETURN_IF_ERROR(coordination_service_agent_->Initialize(
worker_env_->env, worker_env_->device_mgr, server_def,
std::move(agent_cache),
/*error_fn=*/[](Status s) {
LOG(ERROR) << "Coordination agent is set to error: " << s;
}));
}
return Status::OK();
}
void SessionMgr::ResetDefaultWorkerCache(WorkerCacheInterface* worker_cache) {
default_worker_cache_.reset(worker_cache);
}
Status SessionMgr::UpdateSession(
const std::string& session, const ServerDef& server_def,
const protobuf::RepeatedPtrField<DeviceAttributes>&
cluster_device_attributes) {
mutex_lock l(mu_);
if (session.empty()) {
return errors::InvalidArgument("Session must be non-empty.");
}
auto it = sessions_.find(session);
if (it == sessions_.end()) {
return errors::InvalidArgument("Cannot update session ", session,
" because it does not exist.");
}
std::shared_ptr<WorkerSession> worker_session = it->second;
WorkerCacheInterface* worker_cache = nullptr;
if (server_def.cluster().job().empty()) {
worker_cache = new WorkerCacheWrapper(default_worker_cache_.get());
} else {
TF_RETURN_IF_ERROR(worker_cache_factory_(server_def, &worker_cache));
}
std::vector<std::string> updated_remote_workers;
worker_cache->ListWorkers(&updated_remote_workers);
std::vector<std::unique_ptr<Device>> cluster_devices;
const DeviceMgr* local_device_mgr = worker_session->device_mgr();
DeviceMgr* remote_device_mgr = worker_session->remote_device_mgr();
std::vector<Device*> curr_remote_devices = remote_device_mgr->ListDevices();
std::vector<std::unique_ptr<Device>> added_remote_devices;
std::vector<Device*> removed_remote_devices;
std::vector<DeviceAttributes> added_cluster_device_attrs;
for (const auto& da : cluster_device_attributes) {
Device* device;
if (!local_device_mgr->LookupDevice(da.name(), &device).ok() &&
!remote_device_mgr->LookupDevice(da.name(), &device).ok()) {
added_cluster_device_attrs.emplace_back(da);
} else if (device != nullptr &&
device->attributes().incarnation() != da.incarnation()) {
removed_remote_devices.emplace_back(device);
added_cluster_device_attrs.emplace_back(da);
}
}
for (Device* device : curr_remote_devices) {
std::string task_name;
DeviceNameUtils::GetTaskName(device->parsed_name(), &task_name);
if (std::find(updated_remote_workers.begin(), updated_remote_workers.end(),
task_name) == updated_remote_workers.end()) {
removed_remote_devices.emplace_back(device);
}
}
protobuf::RepeatedPtrField<DeviceAttributes> added_cluster_device_attrs_pb(
added_cluster_device_attrs.begin(), added_cluster_device_attrs.end());
AsRemoteDevices(worker_env_->env, added_cluster_device_attrs_pb, nullptr,
&added_remote_devices);
TF_RETURN_IF_ERROR(worker_session->UpdateWorkerCacheAndDevices(
std::unique_ptr<WorkerCacheInterface>(worker_cache),
std::move(added_remote_devices), removed_remote_devices));
return Status::OK();
}
Status SessionMgr::DeleteSession(const std::string& session) {
mutex_lock l(mu_);
auto it = sessions_.find(session);
if (it != sessions_.end()) {
sessions_.erase(it);
}
return Status::OK();
}
Status SessionMgr::WorkerSessionForSessionLocked(
const std::string& session_handle,
std::shared_ptr<WorkerSession>* out_session) {
if (session_handle.empty()) {
*out_session = legacy_session_;
} else {
auto it = sessions_.find(session_handle);
if (it == sessions_.end()) {
return errors::AbortedWithPayloads(
strings::StrCat("Session handle is not found: ", session_handle,
". Possibly this worker (\"",
legacy_session_->worker_name(),
"\") just restarted."),
{{kWorkerPossiblyRestarted,
distributed_runtime::WorkerPossiblyRestarted()
.SerializeAsString()}});
} else {
*out_session = it->second;
}
}
return Status::OK();
}
Status SessionMgr::WorkerSessionForSession(
const std::string& session_handle,
std::shared_ptr<WorkerSession>* out_session) {
mutex_lock l(mu_);
return WorkerSessionForSessionLocked(session_handle, out_session);
}
std::shared_ptr<WorkerSession> SessionMgr::LegacySession() {
return legacy_session_;
}
CoordinationServiceAgent* SessionMgr::GetCoordinationServiceAgent() {
return coordination_service_agent_.get();
}
void SessionMgr::SetLogging(bool active) {
mutex_lock l(mu_);
this->is_logging_active_ = active;
// Legacy Session
if (legacy_session_) {
auto* worker_cache = legacy_session_->worker_cache();
if (worker_cache) {
worker_cache->SetLogging(active);
}
}
for (const auto& session_kv : sessions_) {
auto session = session_kv.second.get();
if (session) {
auto* worker_cache = session->worker_cache();
if (worker_cache) {
worker_cache->SetLogging(active);
}
}
}
}
void SessionMgr::RetrieveLogs(int64_t step_id, LoggingResponse* response) {
mutex_lock l(mu_);
// Legacy Session
if (legacy_session_) {
auto* worker_cache = legacy_session_->worker_cache();
if (worker_cache) {
auto step_stats = StepStats();
if (worker_cache->RetrieveLogs(step_id, &step_stats)) {
auto* labeled_step_stats = response->add_step();
labeled_step_stats->set_step_id(step_id);
labeled_step_stats->mutable_step_stats()->Swap(&step_stats);
}
}
}
for (const auto& session_kv : sessions_) {
auto session = session_kv.second.get();
if (session) {
auto* worker_cache = session->worker_cache();
if (worker_cache) {
auto step_stats = StepStats();
if (worker_cache->RetrieveLogs(step_id, &step_stats)) {
auto* labeled_step_stats = response->add_step();
labeled_step_stats->set_step_id(step_id);
labeled_step_stats->mutable_step_stats()->Swap(&step_stats);
}
}
}
}
}
void SessionMgr::ClearLogs() {
mutex_lock l(mu_);
// Legacy Session
if (legacy_session_) {
auto* worker_cache = legacy_session_->worker_cache();
if (worker_cache) {
worker_cache->ClearLogs();
}
}
for (const auto& session_kv : sessions_) {
auto session = session_kv.second.get();
if (session) {
auto* worker_cache = session->worker_cache();
if (worker_cache) {
worker_cache->ClearLogs();
}
}
}
}
} // namespace tensorflow