alibaba / alibaba/euler

zookeeper分布式训练时报错

Open
#217 0 comments 0 reactions 0 assignees View on GitHub
Dominant language
C++
Stars
2.9k
Forks
553
PR merge metrics
No merged PRs in 30d

Description

请教下zookeeper只需要在master上安装就行吗? 我只在hadoop-master节点上装了zookeeper,测试连接正常,用zookeeper自带客户端(zkCli.sh -server localhost:2181)能连接成功并运行create命令。但是运行euler训练的时候报错了,不知道是什么原因,euler_zk_path尝试过zookeeper里头配置的/tmp/zookeeper以及任意的/path/euler的不行,一直报:ZK error when creating meta: no node。麻烦有空指导一下,谢谢。

运行指令如下:
./dist_tf_euler.sh \
--data_dir hdfs://hadoop-master:9000/user/root/ppi \
--euler_zk_addr hadoop-master:2181 --euler_zk_path /path/euler \
--max_id 56944 --feature_idx 1 --feature_dim 50 --label_idx 0 --label_dim 121 \
--model graphsage_supervised --mode train

错误信息work0(work1也一样):
I0104 13:37:16.686715 9816 graph_builder.cc:81] Load Done: hdfs://hadoop-master:9000/user/root/ppi/ppi_data_0.dat
I0104 13:37:16.686715 9818 graph_builder.cc:81] Load Done: hdfs://hadoop-master:9000/user/root/ppi/ppi_data_4.dat
I0104 13:37:16.700754 9817 graph_builder.cc:81] Load Done: hdfs://hadoop-master:9000/user/root/ppi/ppi_data_2.dat
I0104 13:37:16.701020 9656 graph_builder.cc:127] Each Thread Load Finish! Node Count:34166 Edge Count:963932
I0104 13:37:16.701040 9656 graph_builder.cc:135] Graph Loading Finish!
I0104 13:37:16.982077 9656 graph_builder.cc:147] Graph Load Finish! Node Count:34166 Edge Count:963932
I0104 13:37:16.988603 9656 graph_builder.cc:152] Done: build node sampler
I0104 13:37:16.988626 9656 graph_builder.cc:162] Graph build finish
I0104 13:37:16.988660 9656 graph_service.cc:179] service init finish
I0104 13:37:16.989260 9656 graph_service.cc:131] bound port: 172.20.0.2:46415
E0104 13:37:16.993388 9656 zk_server_register.cc:77] ZK error when creating root node: no node.
W0104 13:37:16.993427 9656 graph.h:198] global sampler is not ok
E0104 13:37:16.994678 9656 zk_server_register.cc:99] ZK error when creating meta: no node.
I0104 13:37:16.994695 9656 graph_service.cc:146] service start。

hdfs正常,分片也正常。

不好意思,我发现是我没有手动建立这个node。现在运行正常。

Contributor guide

No contributing guide indexed for this repository

Research direction

Start with zk_server_register.cc and the graph_service.cc startup path, then reproduce the provided dist_tf_euler.sh command against the stated HDFS and ZooKeeper addresses. Check the ZooKeeper setup and the requested euler_zk_path; done is a training run that starts without the reported root-node or meta-node errors.

Written by the indexing model from the issue text.

Assessment

Tech stack
cpp, hadoop
Domain
distributed-systems
Issue type
Bug
Difficulty
3/5
Estimated time
1-2 days
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
15/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.