Minos is a distributed deployment and monitoring system. It was initially developed and used at Xiaomi to deploy and manage the Hadoop, HBase and ZooKeeper clusters used in the company. Minos can be easily extended to support other systems, among which HDFS, YARN and Impala have been supported in the current release.
The Minos system contains the following four components:
This is the command line client tool used to deploy and manage processes of various systems. You can use this client to perform various deployment tasks, e.g. installing, (re)starting, stopping a service. Currently, this client supports ZooKeeper, HDFS, HBase, YARN and Impala. It can be extended to support other systems. You can refer to the following Using Client to learn how to use it.
This is the dashboard system to display the status of all processes, where users can take a overview of the whole clusters managed by Minos. It collects data from servers through JMX interface. And it organizes pages in cluster, job and task corresponding to the definition in cluster configuration. It also provides some utils like health alerter, HDFS quota updater and quota reportor. You can refer to Installing Owl to learn how to install and use it.
This is the process management and monitoring system. Supervisor is an open source project, a client/server system that allows its users to monitor and control a number of processes on a UNIX-like operating system.
Based on the version of supervisor-3.0b1, we extended Supervisor to support Minos. We implemented an RPC interface under the deployment
directory, so that our deploy client can invoke the services supplied by supervisord.
When deploying a Hadoop cluster for the first time, you need to set up supervisord
on every production machine. This only needs to be done once. You can refer to Installing Supervisor to learn how to install and use it.
This is a simple package management Django app server for our deployment tool. When setting up a cluster for the first time, you should set up a tank server first. This also needs to be done only once. You can refer to Installing Tank to learn how to install and use it.
Make sure install Python 2.7 or later from http://www.python.org.
Make sure that the Oracle Java Development Kit 6 is installed (not OpenJDK) from http://www.oracle.com/technetwork/java/javase/downloads/index.html, and that JAVA_HOME
is set in your environment.
To Using Minos, just check out the code on your production machine:
git clone https://github.com/XiaoMi/minos.git
All the Components of Minos run with its own virtual environment. So, before using Minos, building the virtual environment firstly.
cd minos
./build.sh build
Note: If you only use the Client component on your current machine, this operation is enough, then you can refer to Using Client to learn how to deploy and manage a cluster. If you want to use the current machine as a Tank server, you can refer to Installing Tank to learn how to do that. Similarly, if you want to use the current machine as a Owl server or a Supervisor server, you can refer to Installing Owl and Installing Supervisor respectively.
cd minos
./build.sh start tank --tank_ip ${your_local_ip} --tank_port ${port_tank_will_listen}
Note: If you do not specify the
tank_ip
andtank_port
, it will start tank server using0.0.0.0
on8000
port.
./build.sh stop tank
Make sure you have intstalled Tank on one of the production machines.
cd minos
./build.sh start supervisor --tank_ip ${tank_server_ip} --tank_port ${tank_server_port}
When starting supervisor for the first time, the tank_ip
and tank_port
must be specified.
After starting supervisor on the destination machine, you can access the web interface of the supervisord. For example, if supervisord listens on port 9001, and the serving machine's IP address is 192.168.1.11, you can access the following URL to view the processes managed by supervisord:
http://192.168.1.11:9001/
./build.sh stop supervisor
We use Superlance to monitor processes. Superlance is a package of plug-in utilities for monitoring and controlling processes that run under supervisor.
We integrate superlance-0.7
to our supervisor system, and use the crashmail tool to monitor all processes. When a process exits unexpectedly, crashmail will send an alert email to a mailing list that is configurable.
We configure crashmail as an auto-started process. It will start working automatically when the supervisor is started. Following is a config example, taken from minos/build/template/supervisord.conf.tmpl
, that shows how to configure crashmail:
[eventlistener:crashmailbatch-monitor]
command=python superlance/crashmailbatch.py \
--toEmail="[email protected]" \
--fromEmail="[email protected]" \
--password="123456" \
--smtpHost="mail.example.com" \
--tickEvent=TICK_5 \
--interval=0.5
events=PROCESS_STATE,TICK_5
buffer_size=100
stdout_logfile=crashmailbatch.stdout
stderr_logfile=crashmailbatch.stderr
autostart=true
Note: The related configuration information such as the server
port
orusername
is set inminos/build/template/supervisord.conf.tmpl
, if you don't want to use the default value, change it.
Make sure you have intstalled Tank and Supervisor on your production machines.
Here we would like to show you how to use the client in a simple tutorial. In this tutorial we will use Minos to deploy an HDFS service, which itself requires the deployment of a ZooKeeper service.
The following are some conventions we will use in this tutorial:
- Cluster type: we define three types of clusters:
tst
for testing,prc
for offline processing, andsrv
for online serving. - ZooKeeper cluster name: we define the ZooKeeper cluster name using the IDC short name and the cluster type. For example,
dptst
is used to name a testing cluster at IDCdp
. - Other service cluster names: we define other service cluster names using the corresponding ZooKeeper cluster name and the name of the business for which the service is intended to serve. For example, the
dptst-example
is the name of a testing cluster used to do example tests. - Configuration file names: all the services will have a corresponding configuration file, which will be named as
${service}-${cluster}.cfg
. For example, thedptst
ZooKeeper service's configuration file is named aszookeeper-dptst.cfg
, and thedptst
example HDFS service's configuration file is named ashdfs-dptst-example.cfg
.
There is a configuration file named deploy.cfg
under the root directory of minos. You should first edit this file to set up the deployment environment. Make sure that all service packages are prepared and configured in deploy.cfg
.
As mentioned in the cluster naming conventions, we will set up a testing ZooKeeper cluster at the dp
IDC, and the corresponding configuration file for the cluster will be named as zookeeper-dptst.cfg
.
You can edit zookeeper-dptst.cfg
under the config/conf/zookeeper
directory to configure the cluster. The zookeeper-dptst.cfg
is well commented and self explained, so we will not explain more here.
To set up a ZooKeeper cluster, just do the following two steps:
-
Install a ZooKeeper package to the tank server:
cd minos/client ./deploy install zookeeper dptst
-
Bootstrap the cluster, this is only needed once when the cluster is setup for the first time:
./deploy bootstrap zookeeper dptst
Here are some handy ways to manage the cluster:
-
Show the status of the ZooKeeper service:
./deploy show zookeeper dptst
-
Start/Stop/Restart the ZooKeeper cluster:
./deploy stop zookeeper dptst ./deploy start zookeeper dptst ./deploy restart zookeeper dptst
-
Clean up the ZooKeeper cluster:
./deploy cleanup zookeeper dptst
-
Rolling update the ZooKeeper cluster:
./deploy rolling_update zookeeper dptst
Now it is time to configure the HDFS system. Here we set up a testing HDFS cluster named dptst-example
, whose configuration file will be named as hdfs-dptst-example.cfg
, as explained in the naming conventions.
You can edit hdfs-dptst-example.cfg
under the config/conf/hdfs
directory to configure the cluster. The hdfs-dptst-example.cfg
is well commented and self explained, so we will not explain more here.
Setting up and managing an HDFS cluster is similar to setting up and managing a ZooKeeper cluster. The only difference is the cluster name, dptst-example
, which implies that the corresponding ZooKeeper cluster is dptst
:
./deploy install hdfs dptst-example
./deploy bootstrap hdfs dptst-example
./deploy show hdfs dptst-example
./deploy stop hdfs dptst-example
./deploy start hdfs dptst-example
./deploy restart hdfs dptst-example
./deploy rolling_update hdfs dptst-example --job=datanode
./deploy cleanup hdfs dptst-example
The client tool also supports a very handy command named shell
. You can use this command to manage the files on HDFS, tables on HBase, jobs on YARN, etc. Here are some examples about how to use the shell
command to perform several different HDFS operations:
./deploy shell hdfs dptst-example dfs -ls /
./deploy shell hdfs dptst-example dfs -mkdir /test
./deploy shell hdfs dptst-example dfs -rm -R /test
You can run ./deploy --help
to see the detailed help messages.
Owl must be installed on the machine that you also use the Client component, they both use the same set of cluster configuration files.
Gnuplot is required for opentsdb, you can install it with the following command.
Centos: sudo yum install gnuplot
Ubuntu: sudo apt-get install gnuplot
Ubuntu:
sudo apt-get install mysql-server
sudo apt-get install mysql-client
Centos:
yum install mysql-server mysql mysql-devel
Configure the clusters you want to monitor with owl in minos/config/owl/collector.cfg
. Following is an example that shows how to modify the configuration.
[collector]
# service name(space seperated)
service = hdfs hbase
[hdfs]
# cluster name(space seperated)
clusters=dptst-example
# job name(space seperated)
jobs=journalnode namenode datanode
# url for collecotr, usually JMX url
metric_url=/jmx?qry=Hadoop:*
Note: Some other configurations such as and
opentsdb port
is set inminos/build/minos_config.py
. You can change the default port for avoiding port conflicts.
cd minos
./build.sh start owl --owl_ip ${your_local_ip} --owl_port ${port_owl_monitor_will_listen}
After starting Owl, you can access the web interface of the Owl. For example, if Owl listens on port 8088, and the machine's IP address is 192.168.1.11, you can access the following URL to view the Owl web interface:
http://192.168.1.11:8088/
./build.sh stop owl
-
When installing Mysql-python, you may get an error of
_mysql.c:44:23: error: my_config.h: No such file or directory (centos)
orEnvironmentError: mysql_config not found (ubuntu)
. As mysql_config is part of mysql-devel, installing mysql-devel allows the installation of Mysql-python. So you may need to install it.ubuntu: sudo apt-get install libmysqlclient-dev centos: sudo yum install mysql-devel
-
When installing twisted, you may get an error of
CompressionError: bz2 module is not available
and compile appears:Python build finished, but the necessary bits to build these modules were not found: _sqlite3 _tkinter bsddb185 bz2 dbm dl
Then, you may need to install bz2 and sqlite3 such as
sudo apt-get install libbz2-dev
sudo apt-get install libsqlite3-dev
-
When setting up the stand-alone hbase on Ubuntu, you may fail to start it because of the
/etc/hosts
file. You can refer to http://hbase.apache.org/book/quickstart.html#ftn.d2907e114 to fix the problem. -
When using the Minos client to install a service package, if you get an error of
socket.error: [Errno 101] Network is unreachable
, please check your tank server configuration indeploy.cfg
file, you might miss it.
Note: See Minos Wiki for more advanced features.