First, we have to have KUDU installation package
Here is not to provide a direct download (because there 5G, my server NIC only 4M, under your very slow)
CDH version used here is
Official Download http://archive.cloudera.com/cdh5/
I've got three nodes node01, node02, node03. Here are uploaded to node03
Here imapa do with integration, green is an integrated part you can not see
First upload / export / data / under (the students did not create mkdir / export / data)
This directory
cd /export/data
Decompression in the current directory
tar zxvf cdh5. 14.0 centos6. tar gz
To make a local yum source (duku easy to find their own)
yum -y install httpd
The installation is complete
start up
service httpd start
enter
cd /etc/yum.repos.d
In this installation directory, and opens localimp.repo
vim localimp.repo
change into
Create a connection to read the apache httpd
ln -s /export/data/cdh/5.14.0 /var/www/html/5.14fuckKudu (access path disposed outside the network)
Go var / www / html view
cd / var / www / html ll
Visit http: //node03/5.14fuckKudu
Explained: ln -s establish a flexible connection, fuckKudu want to associate with the back of the 5.14.0
localimp the url will find if yo fuckKudu under / var / www / html, you can find and access , can not configure access 5.14fuckKudu, this right-off and impala
If var / www / html not write path, the default fetch 5.14.0 (disk has been mapped file name)
impala distributed to other nodes node02, node03
cd /etc/yum.repos.d/
scp localimp.repo node02:$PWD
scp localimp.repo node01:$PWD
3 nodes installed kudu
yum -y install must have a master-must-be-client0 eServer must-client-devel
success
Configuration (Modification 3 units)
Enter / etc / kudu / conf
cd / etc / should / conf
master log
vim /etc/default/kudu-master
Modified their mapping or IP node (node map here)
Modify master
vim /etc/kudu/conf/master.gflagfile
Add to
--fromenv=rpc_bind_addresses
--fromenv=log_dir
--fs_wal_dir=/export/servers/kudu/master
--fs_data_dirs=/export/servers/kudu/master
--master_addresses=node:7051,node02:7051,node03:7051
Each must change (change of the current node)
tserver log
Modify rpc own node address
vim /etc/default/kudu-tserver
修改tserver
vim /etc/kudu/conf/tserver.gflagfile
去掉默认的
添加
--fromenv=rpc_bind_addresses
--fromenv=log_dir
--fs_wal_dir=/export/servers/kudu/tserver
--fs_data_dirs=/export/servers/kudu/tserver
--tserver_master_addrs=node01:7051,node02:7051,node03:7051
因为没有这个dir文件夹,所以需要创建
在三台节点上创建
mkdir -p /export/servers/kudu/tserver
mkdir -p /export/servers/kudu/master
kudu的默认权限是kudu用户组。所以需要改变权限为kudu
chown -R kudu:kudu kudu/
kudu非常依赖时间同步，为了确保万一，进行时间同步
/etc/init.d/ntpd restart
启动测试
sudo service kudu-master start
失败，去日志查看
vim /var/log/kudu/此日志文件名
发现master配置疑似多个了空格
vim /etc/kudu/conf/master.gflagfile
删除空格
重新测试sudo service kudu-master start
还是FAILED
查看日志
发现是更改kudu后，但是没有加-R,所以文件夹的内部文件夹还是root
重写chown -R kudu:kudu kudu/
重新测试
发现还是失败，查看日志，不一样了，说明之前的问题解决了，这是个新问题
妈的，端口无效，我居然写了个::端口
修改vim /etc/default/kudu-master
重新测试sudo service kudu-master start
查看日志
原因是我的node01时间不一致，时间修改不了，我去问问那些10年工作经验的大神们.
十年开发经验的大神改了允许默认修改，并且映射了其它机器的时间，现在在来重试一下
启动
service kudu-master start
失败，所有节点全部先启动service ntpd start
然后在启动kudu-master start 成功
启动sudo service kudu-tserver start
需要启动3台节点
访问页面
kudu很不稳定，容易掉（依赖ntpd），开发建议不用，优点高可用,OPAL计算,天生和imapa结合。