Skip to content
体验新版
项目
组织
正在加载...
登录
切换导航
打开侧边栏
PaddlePaddle
Paddle
提交
b089b809
P
Paddle
项目概览
PaddlePaddle
/
Paddle
1 年多 前同步成功
通知
2302
Star
20931
Fork
5422
代码
文件
提交
分支
Tags
贡献者
分支图
Diff
Issue
1423
列表
看板
标记
里程碑
合并请求
543
Wiki
0
Wiki
分析
仓库
DevOps
项目成员
Pages
P
Paddle
项目概览
项目概览
详情
发布
仓库
仓库
文件
提交
分支
标签
贡献者
分支图
比较
Issue
1,423
Issue
1,423
列表
看板
标记
里程碑
合并请求
543
合并请求
543
Pages
分析
分析
仓库分析
DevOps
Wiki
0
Wiki
成员
成员
收起侧边栏
关闭侧边栏
动态
分支图
创建新Issue
提交
Issue看板
提交
b089b809
编写于
6月 14, 2018
作者:
T
tangwei12
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
update rpc to add checkpoint notify
上级
12de20f5
变更
7
显示空白变更内容
内联
并排
Showing
7 changed file
with
64 addition
and
0 deletion
+64
-0
paddle/fluid/operators/detail/grpc_client.cc
paddle/fluid/operators/detail/grpc_client.cc
+16
-0
paddle/fluid/operators/detail/grpc_client.h
paddle/fluid/operators/detail/grpc_client.h
+18
-0
paddle/fluid/operators/detail/grpc_service.h
paddle/fluid/operators/detail/grpc_service.h
+3
-0
paddle/fluid/operators/detail/request_handler.h
paddle/fluid/operators/detail/request_handler.h
+10
-0
paddle/fluid/operators/detail/request_handler_impl.cc
paddle/fluid/operators/detail/request_handler_impl.cc
+6
-0
paddle/fluid/operators/detail/rpc_client.h
paddle/fluid/operators/detail/rpc_client.h
+4
-0
paddle/fluid/operators/detail/send_recv.proto
paddle/fluid/operators/detail/send_recv.proto
+7
-0
未找到文件。
paddle/fluid/operators/detail/grpc_client.cc
浏览文件 @
b089b809
...
...
@@ -229,6 +229,22 @@ void GRPCClient::AsyncSendComplete(const std::string& ep, int64_t time_out) {
req_count_
++
;
}
void
GRPCClient
::
AsyncCheckpointNotify
(
const
std
::
string
&
ep
,
const
std
::
string
&
dir
,
int64_t
time_out
)
{
const
auto
ch
=
GetChannel
(
ep
);
CheckpointNotifyProcessor
*
s
=
new
CheckpointNotifyProcessor
(
ch
);
s
.
prepare
(
time_out
);
sendrecv
::
CheckpointMessage
req
;
req
.
set_notify_type
(
CHECKPOINT_SAVE_MESSAGE
);
req
.
set_checkpoint_dir
(
dir
);
auto
rpc
=
s
->
stub_
->
AsyncCheckpointNotify
(
s
->
context_
.
get
(),
req
,
&
cq
);
rpc
->
Finish
(
&
s
->
reply_
,
&
s
->
status_
,
reinterpret_cast
<
void
*>
(
s
));
req_count_
++
;
}
void
GRPCClient
::
Wait
()
{
std
::
unique_lock
<
std
::
mutex
>
lk
(
sync_mutex_
);
sync_cond_
.
wait
(
lk
,
[
this
]
{
return
req_count_
==
0
;
});
...
...
paddle/fluid/operators/detail/grpc_client.h
浏览文件 @
b089b809
...
...
@@ -165,6 +165,20 @@ class FetchBarrierProcessor : public BaseProcessor {
std
::
unique_ptr
<
sendrecv
::
SendRecvService
::
Stub
>
stub_
;
};
class
CheckpointNotifyProcessor
:
public
BaseProcessor
{
public:
explicit
CheckpointNotifyProcessor
(
std
::
shared_ptr
<
grpc
::
Channel
>
ch
)
:
BaseProcessor
(
ch
)
{
stub_
=
sendrecv
::
SendRecvService
::
NewStub
(
ch
);
}
virtual
~
CheckpointNotifyProcessor
()
{}
virtual
void
Process
()
{}
sendrecv
::
VoidMessage
reply_
;
std
::
unique_ptr
<
sendrecv
::
SendRecvService
::
Stub
>
stub_
;
}
class
GRPCClient
:
public
RPCClient
{
public:
GRPCClient
()
{}
...
...
@@ -193,6 +207,10 @@ class GRPCClient : public RPCClient {
const
std
::
string
&
ep
,
int64_t
time_out
=
RPCClient
::
rpc_time_out
)
override
;
void
AsyncCheckpointNotify
(
const
std
::
string
&
ep
,
const
std
::
string
&
dir
,
int64_t
time_out
=
RPCClient
::
rpc_time_out
)
override
;
void
Wait
()
override
;
void
SendComplete
()
override
;
...
...
paddle/fluid/operators/detail/grpc_service.h
浏览文件 @
b089b809
...
...
@@ -79,6 +79,7 @@ enum class GrpcMethod {
kSendVariable
,
kGetVariable
,
kPrefetchVariable
,
kCheckpointNotify
,
};
static
const
int
kGrpcNumMethods
=
...
...
@@ -92,6 +93,8 @@ inline const char* GrpcMethodName(GrpcMethod id) {
return
"/sendrecv.SendRecvService/GetVariable"
;
case
GrpcMethod
::
kPrefetchVariable
:
return
"/sendrecv.SendRecvService/PrefetchVariable"
;
case
GrpcMethod
::
kCheckpointNotify
:
return
"/sendrecv.SendRecvService/CheckpointNotify"
;
}
// Shouldn't be reached.
...
...
paddle/fluid/operators/detail/request_handler.h
浏览文件 @
b089b809
...
...
@@ -43,6 +43,9 @@ constexpr char kRequestCheckpoint[] = "RequestCheckpoint";
#define FETCH_BARRIER_MESSAGE "FETCH_BARRIER@RECV"
#define COMPLETE_MESSAGE "COMPLETE@RECV"
#define CHECKPOINT_SAVE_MESSAGE "SAVE"
#define CHECKPOINT_LOAD_MESSAGE "LOAD"
class
RPCServer
;
class
RequestHandler
{
...
...
@@ -70,6 +73,11 @@ class RequestHandler {
prefetch_var_name_to_prepared_ctx_
=
g
;
}
void
SetCheckpointNotifyPreparedCtx
(
std
::
shared_ptr
<
framework
::
ExecutorPrepareContext
>
g
)
{
checkpoint_prepared_ctx_
=
g
;
}
// Used for async.
void
SetGradToPreparedCtx
(
std
::
unordered_map
<
...
...
@@ -116,6 +124,8 @@ class RequestHandler {
std
::
unordered_map
<
std
::
string
,
std
::
shared_ptr
<
framework
::
ExecutorPrepareContext
>>*
prefetch_var_name_to_prepared_ctx_
;
// used for checkpoint notify
std
::
shared_ptr
<
framework
::
ExecutorPrepareContext
>
checkpoint_prepared_ctx_
;
// Used for async.
std
::
unordered_map
<
std
::
string
,
...
...
paddle/fluid/operators/detail/request_handler_impl.cc
浏览文件 @
b089b809
...
...
@@ -119,6 +119,12 @@ bool RequestPrefetchHandler::Handle(const std::string& varname,
return
true
;
}
bool
RequestCheckpointHandler
::
Handle
(
const
std
::
string
&
varname
,
framework
::
Scope
*
scope
,
framework
::
Variable
*
invar
,
framework
::
Variable
**
outvar
,
const
std
::
string
&
out_var_name
)
{}
}
// namespace detail
}
// namespace operators
}
// namespace paddle
paddle/fluid/operators/detail/rpc_client.h
浏览文件 @
b089b809
...
...
@@ -53,6 +53,10 @@ class RPCClient {
virtual
void
AsyncSendFetchBarrier
(
const
std
::
string
&
ep
,
int64_t
time_out
=
rpc_time_out
)
=
0
;
virtual
void
AsyncCheckpointNotify
(
const
std
::
string
&
ep
,
const
std
::
string
&
dir
,
int64_t
time_out
=
rpc_time_out
)
=
0
;
// SendComplete tells all the server that current trainer have no more data
// to train, so that the pserver can reduce it's barrier count, and continue
// to train with other trainers.
...
...
paddle/fluid/operators/detail/send_recv.proto
浏览文件 @
b089b809
...
...
@@ -25,6 +25,8 @@ service SendRecvService {
rpc
GetVariable
(
VariableMessage
)
returns
(
VariableMessage
)
{}
// pre-fetch variable by given variable name and Ids
rpc
PrefetchVariable
(
VariableMessage
)
returns
(
VariableMessage
)
{}
rpc
CheckpointNotify
(
CheckpointMessage
)
returns
(
VoidMessage
)
{}
}
// VariableMessage is serialized paddle variable message.
...
...
@@ -79,3 +81,8 @@ message VariableMessage {
}
message
VoidMessage
{}
message
CheckpointMessage
{
string
notify_type
=
1
;
string
checkpoint_dir
=
2
;
}
编辑
预览
Markdown
is supported
0%
请重试
或
添加新附件
.
添加附件
取消
You are about to add
0
people
to the discussion. Proceed with caution.
先完成此消息的编辑!
取消
想要评论请
注册
或
登录