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
0a7cb901
P
Paddle
项目概览
PaddlePaddle
/
Paddle
1 年多 前同步成功
通知
2305
Star
20932
Fork
5423
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
未验证
提交
0a7cb901
编写于
1月 10, 2022
作者:
Y
yaoxuefeng
提交者:
GitHub
1月 10, 2022
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
add retry on pull dense sync (#38793)
上级
1f8fe035
变更
2
隐藏空白更改
内联
并排
Showing
2 changed file
with
53 addition
and
4 deletion
+53
-4
paddle/fluid/framework/fleet/fleet_wrapper.cc
paddle/fluid/framework/fleet/fleet_wrapper.cc
+53
-3
paddle/fluid/framework/fleet/heter_context.h
paddle/fluid/framework/fleet/heter_context.h
+0
-1
未找到文件。
paddle/fluid/framework/fleet/fleet_wrapper.cc
浏览文件 @
0a7cb901
...
...
@@ -632,6 +632,7 @@ void FleetWrapper::PullSparseToTensorSync(const uint64_t table_id, int fea_dim,
if
(
ret
!=
0
)
{
LOG
(
ERROR
)
<<
"fleet pull sparse failed, status["
<<
ret
<<
"]"
;
sleep
(
sleep_seconds_before_fail_exit_
);
exit
(
-
1
);
}
#else
for
(
size_t
index
=
0
;
index
<
inputs
->
size
();
++
index
)
{
...
...
@@ -685,9 +686,36 @@ void FleetWrapper::PullDenseVarsSync(
paddle
::
ps
::
Region
reg
(
w
,
tensor
->
numel
());
regions
.
emplace_back
(
std
::
move
(
reg
));
}
auto
status
=
pslib_ptr_
->
_worker_ptr
->
pull_dense
(
regions
.
data
(),
regions
.
size
(),
tid
);
status
.
wait
();
int32_t
status
=
-
1
;
int32_t
cnt
=
0
;
while
(
true
)
{
auto
tt
=
pslib_ptr_
->
_worker_ptr
->
pull_dense
(
regions
.
data
(),
regions
.
size
(),
tid
);
bool
flag
=
true
;
tt
.
wait
();
try
{
status
=
tt
.
get
();
}
catch
(
const
std
::
future_error
&
e
)
{
VLOG
(
0
)
<<
"Caught a future_error with code"
<<
e
.
code
()
<<
", Message:"
<<
e
.
what
();
}
if
(
status
!=
0
)
{
VLOG
(
0
)
<<
"fleet pull dense sync failed, status["
<<
status
<<
"]"
;
sleep
(
sleep_seconds_before_fail_exit_
);
flag
=
false
;
cnt
++
;
}
if
(
cnt
>
3
)
{
VLOG
(
0
)
<<
"fleet pull dense sync failed, retry 3 times"
;
exit
(
-
1
);
}
if
(
flag
)
{
break
;
}
}
#endif
}
...
...
@@ -1248,6 +1276,7 @@ void FleetWrapper::LoadModelOneTable(const uint64_t table_id,
if
(
ret
.
get
()
!=
0
)
{
LOG
(
ERROR
)
<<
"load model of table id: "
<<
table_id
<<
", from path: "
<<
path
<<
" failed"
;
exit
(
-
1
);
}
#else
VLOG
(
0
)
<<
"FleetWrapper::LoadModel does nothing when no pslib"
;
...
...
@@ -1263,6 +1292,7 @@ void FleetWrapper::LoadWithWhitelist(const uint64_t table_id,
if
(
ret
.
get
()
!=
0
)
{
LOG
(
ERROR
)
<<
"load model of table id: "
<<
table_id
<<
", from path: "
<<
path
<<
" failed"
;
exit
(
-
1
);
}
#else
VLOG
(
0
)
<<
"FleetWrapper::LoadWhitelist does nothing when no pslib"
;
...
...
@@ -1311,6 +1341,7 @@ void FleetWrapper::SaveModelOneTable(const uint64_t table_id,
if
(
ret
.
get
()
!=
0
)
{
LOG
(
ERROR
)
<<
"save model of table id: "
<<
table_id
<<
", to path: "
<<
path
<<
" failed"
;
exit
(
-
1
);
}
#else
VLOG
(
0
)
<<
"FleetWrapper::SaveModelOneTable does nothing when no pslib"
;
...
...
@@ -1328,6 +1359,7 @@ void FleetWrapper::SaveModelOneTablePrefix(const uint64_t table_id,
if
(
ret
.
get
()
!=
0
)
{
LOG
(
ERROR
)
<<
"save model (with prefix) of table id: "
<<
table_id
<<
", to path: "
<<
path
<<
" failed"
;
exit
(
-
1
);
}
#else
VLOG
(
0
)
<<
"FleetWrapper::SaveModelOneTablePrefix does nothing when no pslib"
;
...
...
@@ -1351,6 +1383,7 @@ void FleetWrapper::SetDate(const uint64_t table_id, const std::string& date) {
ret
.
wait
();
if
(
ret
.
get
()
!=
0
)
{
LOG
(
ERROR
)
<<
"setdate : "
<<
date
<<
" failed"
;
exit
(
-
1
);
}
#else
VLOG
(
0
)
<<
"FleetWrapper::SetDate does nothing when no pslib-gpu"
;
...
...
@@ -1463,6 +1496,11 @@ void FleetWrapper::ShrinkSparseTable(int table_id) {
#ifdef PADDLE_WITH_PSLIB
auto
ret
=
pslib_ptr_
->
_worker_ptr
->
shrink
(
table_id
);
ret
.
wait
();
int32_t
err_code
=
ret
.
get
();
if
(
err_code
==
-
1
)
{
LOG
(
ERROR
)
<<
"Shrink Sparse Table failed"
;
exit
(
-
1
);
}
#else
VLOG
(
0
)
<<
"FleetWrapper::ShrinkSparseTable does nothing when no pslib"
;
#endif
...
...
@@ -1472,6 +1510,10 @@ void FleetWrapper::ClearModel() {
#ifdef PADDLE_WITH_PSLIB
auto
ret
=
pslib_ptr_
->
_worker_ptr
->
clear
();
ret
.
wait
();
int32_t
err_code
=
ret
.
get
();
if
(
err_code
==
-
1
)
{
LOG
(
ERROR
)
<<
"Clear Model failed"
;
}
#else
VLOG
(
0
)
<<
"FleetWrapper::ClearModel does nothing when no pslib"
;
#endif
...
...
@@ -1481,6 +1523,10 @@ void FleetWrapper::ClearOneTable(const uint64_t table_id) {
#ifdef PADDLE_WITH_PSLIB
auto
ret
=
pslib_ptr_
->
_worker_ptr
->
clear
(
table_id
);
ret
.
wait
();
int32_t
err_code
=
ret
.
get
();
if
(
err_code
==
-
1
)
{
LOG
(
ERROR
)
<<
"Clear One Table failed table_id: "
<<
table_id
;
}
#else
VLOG
(
0
)
<<
"FleetWrapper::ClearOneTable does nothing when no pslib"
;
#endif
...
...
@@ -1541,6 +1587,10 @@ void FleetWrapper::ClientFlush() {
#ifdef PADDLE_WITH_PSLIB
auto
ret
=
pslib_ptr_
->
_worker_ptr
->
flush
();
ret
.
wait
();
int32_t
err_code
=
ret
.
get
();
if
(
err_code
==
-
1
)
{
LOG
(
ERROR
)
<<
"Client Flush failed"
;
}
#else
VLOG
(
0
)
<<
"FleetWrapper::ServerFlush does nothing when no pslib"
;
#endif
...
...
paddle/fluid/framework/fleet/heter_context.h
浏览文件 @
0a7cb901
...
...
@@ -235,7 +235,6 @@ class HeterContext {
}
VLOG
(
3
)
<<
"heter_context unique keys with dynamic mf dimention"
;
}
for
(
std
::
thread
&
t
:
threads
)
{
t
.
join
();
}
...
...
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