CrossMedia的预测模式

CrossMedia不支持单机模式，在我部署的分布式环境下做predict遇到如下问题，求大神帮忙看看问题：
1、ps报错，提示Not found variable[user_graphfc1/bias] part[5] in variable_infos when load variable.
2、worker报错，提示：
execute fail retry cnt[1]
I
('An error was raised. This may be due to a preemption in a connected worker or parameter server. The current session will be closed and a new session will be created. Error: %s', 'ErrorCode[4], ErrorMsg[RunUdfChain Version Mismatch\nCheck Status [PS2XDL::ConvertStatus(st)] at [/home/yue.song/x-deeplearning/xdl/xdl/core/ops/ps_ops/ps_pull_op.cc]operator()@41\nCheck Status [PS2XDL::ConvertStatus(st)] at [/home/yue.song/x-deeplearning/xdl/xdl/core/ops/ps_ops/model_server/model_server_client_forward.cc]operator()@44]')
I
('An error was raised. This may be due to a preemption in a connected worker or parameter server. The current session will be closed and a new session will be created. Error: %s', 'ErrorCode[4], ErrorMsg[RunUdfChain Version Mismatch\nCheck Status [PS2XDL::ConvertStatus(st)] at [/home/yue.song/x-deeplearning/xdl/xdl/core/ops/ps_ops/ps_pull_op.cc]operator()@41\nCheck Status [PS2XDL::ConvertStatus(st)] at [/home/yue.song/x-deeplearning/xdl/xdl/core/ops/ps_ops/model_server/model_server_client_forward.cc]operator()@44]')
3、schedule报错提示：
I0308 09:41:26.938618 88 scheduler_impl.cc:268] Restore checkpoint :, Get Status: ErrorCode [11]: network error
I0308 09:41:26.938792 88 scheduler_impl.cc:215] All servers are ready
I0308 09:41:26.938827 88 scheduler_impl.cc:221] Try Restoring to latest checkpoint due to error
I0308 09:41:26.938848 88 scheduler_impl.cc:266] Restore checkpoint:
I0308 09:41:26.955829 88 scheduler_impl.cc:442] Real Checkpoints[ckpt-...............10017] with info_size[23]
report error in process, seq=79, serverid=18
I0308 09:41:28.348634 88 scheduler_impl.cc:268] Restore checkpoint :, Get Status: ErrorCode [11]: network error
I0308 09:41:28.348829 88 scheduler_impl.cc:215] All servers are ready

Start by reproducing CrossMedia prediction in the reported distributed deployment and collect the parameter-server, worker, and scheduler logs shown in the issue. Trace the variable-loading, RunUdfChain version, and checkpoint-restore failures to their relevant xdl components. Done means the root cause is identified and distributed prediction completes with checkpoint restoration.

Written by the indexing model from the issue text.