• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /* Copyright 2016 The TensorFlow Authors. All Rights Reserved.
2 
3 Licensed under the Apache License, Version 2.0 (the "License");
4 you may not use this file except in compliance with the License.
5 You may obtain a copy of the License at
6 
7     http://www.apache.org/licenses/LICENSE-2.0
8 
9 Unless required by applicable law or agreed to in writing, software
10 distributed under the License is distributed on an "AS IS" BASIS,
11 WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 See the License for the specific language governing permissions and
13 limitations under the License.
14 ==============================================================================*/
15 
16 #ifndef TENSORFLOW_CORE_DISTRIBUTED_RUNTIME_WORKER_SESSION_H_
17 #define TENSORFLOW_CORE_DISTRIBUTED_RUNTIME_WORKER_SESSION_H_
18 
19 #include <string>
20 
21 #include "tensorflow/core/common_runtime/device_mgr.h"
22 #include "tensorflow/core/distributed_runtime/cluster_function_library_runtime.h"
23 #include "tensorflow/core/distributed_runtime/graph_mgr.h"
24 #include "tensorflow/core/distributed_runtime/worker_cache.h"
25 
26 namespace tensorflow {
27 
28 class ClusterFunctionLibraryRuntime;
29 class GraphMgr;
30 class WorkerCacheInterface;
31 
32 // WorkerSession encapsulates all of the state relating to a given session.
33 class WorkerSession {
34  public:
35   // Collection of local devices. These devices are typically
36   // RenamedDevices in all except the SessionMgr.legacy_session_ and
37   // sessions created with `isolate_session_state == false`. In the
38   // those cases, this method returns a pointer to a borrowed
39   // DeviceMgr (typically the `worker_env.device_mgr`).
device_mgr()40   const DeviceMgr* device_mgr() {
41     return device_mgr_ ? device_mgr_.get() : borrowed_device_mgr_;
42   }
43 
remote_device_mgr()44   DynamicDeviceMgr* remote_device_mgr() { return remote_device_mgr_.get(); }
45 
session_name()46   const string& session_name() const { return session_name_; }
worker_name()47   const string& worker_name() const { return worker_name_; }
48 
worker_cache()49   WorkerCacheInterface* worker_cache() const {
50     tf_shared_lock l(worker_session_state_mu_);
51     return worker_cache_.get();
52   }
graph_mgr()53   GraphMgr* graph_mgr() const { return graph_mgr_.get(); }
54 
cluster_flr()55   ClusterFunctionLibraryRuntime* cluster_flr() const {
56     return cluster_flr_.get();
57   }
58 
59   WorkerSession(const string& session_name, const string& worker_name,
60                 std::unique_ptr<WorkerCacheInterface> worker_cache,
61                 std::unique_ptr<DeviceMgr> device_mgr,
62                 std::unique_ptr<GraphMgr> graph_mgr,
63                 std::unique_ptr<DynamicDeviceMgr> remote_device_mgr);
64 
65   static std::shared_ptr<WorkerSession> CreateWithBorrowedDeviceMgr(
66       const string& session_name, const string& worker_name,
67       std::unique_ptr<WorkerCacheInterface> worker_cache,
68       const DeviceMgr* borrowed_device_mgr, std::unique_ptr<GraphMgr> graph_mgr,
69       std::unique_ptr<DynamicDeviceMgr> remote_device_mgr);
70 
71   // In the eager runtime we allow WorkerSession to be updated, where the
72   // worker cache will be recreated. If WorkerSession upate is expected and a
73   // worker in the cache is used in RPCs, the caller should hold a shared
74   // pointer to avoid the workers getting deleted.
GetSharedWorkerCache()75   std::shared_ptr<WorkerCacheInterface> GetSharedWorkerCache() {
76     tf_shared_lock l(worker_session_state_mu_);
77     return worker_cache_;
78   }
79 
80   // Update an existing worker session with new set of remote workers and
81   // devices. Added devices will be owned by the worker session, and removed
82   // devices will be freed by their names.
83   Status UpdateWorkerCacheAndDevices(
84       std::unique_ptr<WorkerCacheInterface> new_worker_cache,
85       std::vector<std::unique_ptr<Device>> added_remote_devices,
86       const std::vector<Device*>& removed_remote_devices);
87 
88   ~WorkerSession();
89 
90  private:
91   WorkerSession(const string& session_name, const string& worker_name,
92                 std::unique_ptr<WorkerCacheInterface> worker_cache,
93                 const DeviceMgr* borrowed_device_mgr,
94                 std::unique_ptr<GraphMgr> graph_mgr,
95                 std::unique_ptr<DynamicDeviceMgr> remote_device_mgr);
96 
97   // The name of the session.
98   const string session_name_;
99 
100   // The name of the worker. E.g., /job:mnist/replica:0/task:1.
101   const string worker_name_;
102 
103   mutable mutex worker_session_state_mu_;
104   // Object from which WorkerInterface instances can be obtained.
105   std::shared_ptr<WorkerCacheInterface> worker_cache_
106       TF_GUARDED_BY(worker_session_state_mu_);
107 
108   // graph_mgr keeps track of the registered graphs of this session.
109   //
110   // Note: graph_mgr must be deleted before rendezvous_mgr!
111   // Note: graph_mgr must be deleted before device_mgr!
112   const std::unique_ptr<GraphMgr> graph_mgr_;
113 
114   std::unique_ptr<ClusterFunctionLibraryRuntime> cluster_flr_;
115 
116   const std::unique_ptr<const DeviceMgr> device_mgr_;
117   const DeviceMgr* const borrowed_device_mgr_;  // Not owned.
118   std::unique_ptr<DynamicDeviceMgr> remote_device_mgr_;
119 };
120 
121 }  // namespace tensorflow
122 
123 #endif  // TENSORFLOW_CORE_DISTRIBUTED_RUNTIME_WORKER_SESSION_H_
124