1/* Copyright 2016 The TensorFlow Authors. All Rights Reserved.
2
3Licensed under the Apache License, Version 2.0 (the "License");
4you may not use this file except in compliance with the License.
5You may obtain a copy of the License at
6
7 http://www.apache.org/licenses/LICENSE-2.0
8
9Unless required by applicable law or agreed to in writing, software
10distributed under the License is distributed on an "AS IS" BASIS,
11WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12See the License for the specific language governing permissions and
13limitations under the License.
14==============================================================================*/
15
16#include "tensorflow/core/distributed_runtime/partial_run_mgr.h"
17
18#include "tensorflow/core/util/ptr_util.h"
19
20namespace tensorflow {
21
22bool PartialRunMgr::FindOrCreate(int step_id,
23 CancellationManager** cancellation_manager) {
24 mutex_lock l(mu_);
25 auto it = step_id_to_partial_run_.find(step_id);
26 if (it != step_id_to_partial_run_.end()) {
27 *cancellation_manager = it->second->cancellation_manager.get();
28 return false;
29 }
30
31 std::unique_ptr<PartialRunState> partial_run =
32 tensorflow::MakeUnique<PartialRunState>();
33 partial_run->cancellation_manager =
34 tensorflow::MakeUnique<CancellationManager>();
35 *cancellation_manager = partial_run->cancellation_manager.get();
36 step_id_to_partial_run_[step_id] = std::move(partial_run);
37 return true;
38}
39
40void PartialRunMgr::ExecutorDone(int step_id, const Status& executor_status) {
41 StatusCallback done;
42 Status callback_status;
43 {
44 mutex_lock l(mu_);
45 auto run_it = step_id_to_partial_run_.find(step_id);
46 if (run_it == step_id_to_partial_run_.end()) {
47 return;
48 }
49 // If we found the partial_run, we call the final callback, if it
50 // exists.
51 // It is guaranteed that run_it->second->final_callback is left empty
52 // after the std::move call.
53 done = std::move(run_it->second->final_callback);
54 if (!executor_status.ok()) {
55 run_it->second->final_status = executor_status;
56 }
57 callback_status = run_it->second->final_status;
58 run_it->second->executor_done = true;
59 }
60 if (done != nullptr) {
61 done(callback_status);
62 mutex_lock l(mu_);
63 step_id_to_partial_run_.erase(step_id);
64 }
65}
66
67void PartialRunMgr::PartialRunDone(int step_id, StatusCallback done,
68 const Status& status) {
69 Status callback_status;
70 {
71 mutex_lock l(mu_);
72 auto run_it = step_id_to_partial_run_.find(step_id);
73 if (run_it == step_id_to_partial_run_.end()) {
74 return;
75 }
76 run_it->second->final_status.Update(status);
77 if (!run_it->second->executor_done) {
78 // If we found the partial_run, we set the final callback to call only
79 // when the executor is completely done.
80 run_it->second->final_callback = std::move(done);
81 return;
82 }
83 callback_status = run_it->second->final_status;
84 }
85 // Otherwise we call the callback immediately.
86 done(callback_status);
87 mutex_lock l(mu_);
88 step_id_to_partial_run_.erase(step_id);
89}
90
91} // namespace tensorflow
92