paddle_pserver2 err

[adrianhust]

Hi,
log:

Fri Mar  9 08:34:38 2018[1,4]<stderr>:./start_server.sh: line 33: 42473 Aborted                 GLOG_logtostderr=0 GLOG_log_dir="./log" ./paddle_pserver2 --num_gradient_servers=${OMPI_COMM_WORLD_SIZE} --nics=${nics} ${server_arg} --rdma_tcp=${rdma_tcp} --comment=$comment
Fri Mar  9 08:34:38 2018[1,8]<stderr>:F0309 08:34:38.208967 43912 ParameterServer2.cpp:158] Check failed: !configMap_.count(config.para_id()) Duplicated parameter name: ___fc_layer_1__.wbias
Fri Mar  9 08:34:38 2018[1,4]<stderr>:+ check_return 'paddle_pserver2 failed'
Fri Mar  9 08:34:38 2018[1,4]<stderr>:+ '[' 134 -ne 0 ']'
Fri Mar  9 08:34:38 2018[1,4]<stderr>:+ echo '[./start_server.sh : 34] [main]'
Fri Mar  9 08:34:38 2018[1,7]<stderr>:F0309 08:34:38.206758 20904 ParameterServer2.cpp:158] Check failed: !configMap_.count(config.para_id()) Duplicated parameter name: ___fc_layer_1__.wbias
Fri Mar  9 08:34:38 2018[1,4]<stderr>:[./start_server.sh : 34] [main]
Fri Mar  9 08:34:38 2018[1,4]<stderr>:+ echo '[FATAL]: paddle_pserver2 failed'
Fri Mar  9 08:34:38 2018[1,4]<stderr>:[FATAL]: paddle_pserver2 failed
Fri Mar  9 08:34:38 2018[1,4]<stderr>:+ get_stack

[Yancey1989]

Could you submit the job by receiver service? It's a stable version for PaddlePaddle on internal MPI cluster and it works well for most users.

[wangkuiyi]

@adrianhust and @Yancey1989 : is this question about the job submitter we use inside Baidu? I ask because I cannot find the start_server.sh script in the Paddle repo nor the cloud repo.

[Yancey1989]

@wangkuiyi
Yes, this job was running on the internal MPI cluster, and I think this question has been fixed by submitting through receiver, which is a Python Service to make it easy to submit a job to MPI cluster for users.

@adrianhust , I will close this issue and feel free to reopen it.

[pkuyym]

Close it due to inactivity.