1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
|
/* 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.
==============================================================================*/
#ifndef TENSORFLOW_CORE_DISTRIBUTED_RUNTIME_WORKER_H_
#define TENSORFLOW_CORE_DISTRIBUTED_RUNTIME_WORKER_H_
#include <unordered_map>
#include "tensorflow/core/distributed_runtime/graph_mgr.h"
#include "tensorflow/core/distributed_runtime/partial_run_mgr.h"
#include "tensorflow/core/distributed_runtime/session_mgr.h"
#include "tensorflow/core/distributed_runtime/worker_interface.h"
namespace tensorflow {
class CancellationManager;
class Device;
struct WorkerEnv;
struct WorkerSession;
// A TensorFlow Worker runs registered graphs and supports worker-to-worker
// Tensor transfer.
//
// See `../protobuf/worker_service.proto` for more details about each method.
//
// This class may be subclassed to provide specialized implementations of
// particular methods for different transport mechanism. For example,
// `GrpcWorker` specializes the `RecvTensorAsync()` method to support a more
// efficient gRPC data structure for handling large binary data.
class Worker : public WorkerInterface {
public:
Worker(WorkerEnv* env);
virtual ~Worker() {}
void GetStatusAsync(const GetStatusRequest* request,
GetStatusResponse* response,
StatusCallback done) override;
void CreateWorkerSessionAsync(const CreateWorkerSessionRequest* request,
CreateWorkerSessionResponse* response,
StatusCallback done) override;
void DeleteWorkerSessionAsync(CallOptions* opts,
const DeleteWorkerSessionRequest* request,
DeleteWorkerSessionResponse* response,
StatusCallback done) override;
void RegisterGraphAsync(const RegisterGraphRequest* request,
RegisterGraphResponse* response,
StatusCallback done) override;
void DeregisterGraphAsync(const DeregisterGraphRequest* request,
DeregisterGraphResponse* response,
StatusCallback done) override;
void RunGraphAsync(CallOptions* opts, RunGraphRequestWrapper* request,
MutableRunGraphResponseWrapper* response,
StatusCallback done) override;
MutableRunGraphRequestWrapper* CreateRunGraphRequest() override;
MutableRunGraphResponseWrapper* CreateRunGraphResponse() override;
void CleanupGraphAsync(const CleanupGraphRequest* request,
CleanupGraphResponse* response,
StatusCallback done) override;
void CleanupAllAsync(const CleanupAllRequest* request,
CleanupAllResponse* response,
StatusCallback done) override;
void RecvTensorAsync(CallOptions* opts, const RecvTensorRequest* request,
TensorResponse* response, StatusCallback done) override;
void LoggingAsync(const LoggingRequest* request, LoggingResponse* response,
StatusCallback done) override;
void TracingAsync(const TracingRequest* request, TracingResponse* response,
StatusCallback done) override;
void RecvBufAsync(CallOptions* opts, const RecvBufRequest* request,
RecvBufResponse* response, StatusCallback done) override;
void CompleteGroupAsync(CallOptions* opts,
const CompleteGroupRequest* request,
CompleteGroupResponse* response,
StatusCallback done) override;
void CompleteInstanceAsync(CallOptions* opts,
const CompleteInstanceRequest* request,
CompleteInstanceResponse* response,
StatusCallback done) override;
void GetStepSequenceAsync(const GetStepSequenceRequest* request,
GetStepSequenceResponse* response,
StatusCallback done) override;
protected:
WorkerEnv* const env_; // Not owned.
Status PrepareRecvTensor(const Rendezvous::ParsedKey& parsed,
Device** src_dev);
void AbortStep(int64);
private:
PartialRunMgr partial_run_mgr_;
CancellationManager cancellation_manager_;
Status PrepareRunGraph(RunGraphRequestWrapper* req,
GraphMgr::NamedTensors* in,
GraphMgr::NamedTensors* out);
void DoRunGraph(CallOptions* opts, RunGraphRequestWrapper* request,
MutableRunGraphResponseWrapper* response,
StatusCallback done);
void DoPartialRunGraph(CallOptions* opts, RunGraphRequestWrapper* request,
MutableRunGraphResponseWrapper* response,
StatusCallback done);
TF_DISALLOW_COPY_AND_ASSIGN(Worker);
};
} // namespace tensorflow
#endif // TENSORFLOW_CORE_DISTRIBUTED_RUNTIME_WORKER_H_
|
