Project

judo

0.01
No commit activity in last 3 years
No release in over 3 years
The gentle way to manage and control ec2 instances
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
 Dependencies

Runtime

>= 2.3.8
>= 0
 Project Readme

Judo

Judo is a tool for managing a cloud of ec2 servers. It aims to be both simple to get going and powerful.

CONCEPTS

Servers and Groups. Servers are identified by a naked string. Groups always have a colon prefix. The special group :all refers to all groups. A name prepended by a carrot will exclude that selection.

$ judo restart myserver1                ## this restarts myserver1
$ judo restart myserver1 myserver2      ## this restarts myserver1 and myserver2
$ judo restart :default                 ## this restarts all servers in the :default group
$ judo restart myserver1 :default :db   ## this restarts all servers in the :default group, the :db group, and a server named myserver1
$ judo restart :all                     ## this restarts all servers in all groups
$ judo restart :default ^myserver3      ## this restarts all servers in the default group except myserver3
$ judo restart :all ^:default           ## this restarts all servers except those in the default group

Server: Judo does not track EC2 instances, but Servers, which are a collection of state, such as EBS volumes, elastic IP addresses and a current EC2 instance ID. This allows you to abstract all the elastic EC2 concepts into a more traditional concept of a static Server.

STARTING

You will need an AWS account with EC2, S3 and SDB all enabled.

Setting up a new judo repo named "my_cloud" would look like this:

$ export AWS_SECRET_ACCESS_KEY="..."
$ export AWS_ACCESS_KEY_ID="..."
$ mkdir my_cloud
$ cd my_cloud
$ judo setup --bucket BUCKET

The 'judo setup' command will make a .judo folder to store your EC2 keys and S3 bucket. It will also make a folder named "default" to hold the default server config. Feel free to examine the default folder. It consists of some example files for a simple server.

To launch a default server you create it

$ judo create :default            ## make one :default server - have judo pick the name
---> Creating server default1...     done (0.6s)
$ judo create 2:default           ## make two :default servers - have judo pick the name
---> Creating server default2...     done (0.5s)
---> Creating server default3...     done (0.7s)
$ judo create myserver1:default   ## make one :default server named myserver1
---> Creating server myserver1...    done (0.6s)
$ judo list
  SERVERS
--------------------------------------------------------------------------------
  default1             default      v1             m1.small               ebs:0
  default2             default      v1             m1.small               ebs:0
  default3             default      v1             m1.small               ebs:0
  myserver1            default      v1             m1.small               ebs:0
$ judo start myserver1
---> Starting server myserver1...    done (2.3s)
$ judo launch myserver2:default   ## launch does a create and a start in 1 step
---> Creating server myserver2...    done (0.6s)
---> Starting server myserver2...    done (2.9s)
$ judo list
  SERVERS
--------------------------------------------------------------------------------
  default1             default      v1             m1.small               ebs:0
  default2             default      v1             m1.small               ebs:0
  default3             default      v1             m1.small               ebs:0
  myserver1            default      v1 i-6fdf8d09  m1.small    running    ebs:0
  myserver2            default      v1 i-49cef122  m1.small    running    ebs:0

You can examine a groups config by looking in the group folder in the repo. The default group will look something like this.

$ cat default/config.json
{
    "ami32":"ami-2d4aa444",       // public ubuntu 10.04 ami - 32 bit
    "ami64":"ami-fd4aa494",       // public ubuntu 10.04 ami - 64 bit
    "user":"ubuntu",              // this is for ssh acccess - defaults to ubuntu not root
    "kuzushi_version": "0.0.54",  // this will pin the version of kuzushi the server will boot with
    "security_group":"judo",
    "example_config": "example_mode",
    "availability_zone":"us-east-1d"
}

Any changes you make to these files do not stick until you've committed them. To commit a group do the following.

$ judo commit :default
Compiling version 2... done (1.2s)

This will create and start two servers. One named 'myserver1' and one named 'myserver2'. You can ssh into 'myserver1' you can type:

$ judo ssh myserver1

You can stop all the servers in the :default group with:

$ judo stop :default

COMMANDS

   $ judo --help

   judo launch [options] SERVER ...
   judo create [options] SERVER ...
   judo destroy [options] SERVER ...

   # SERVER can be formatted as NAME or NAME:GROUP or N:GROUP
   # where N is the number of servers to create or launch
   # 'launch' only differs from 'create' in that it immediately starts the server

   judo start [options] [SERVER ...]
   judo stop [options] [SERVER ...]
   judo restart [options] [SERVER ...]

   judo commit [options] GROUP

   judo snapshot [options] SERVER SNAPSHOT ## take an ebs snapshot of a server
   judo snapshots [options] [SERVER ...]   ## show current snapshots on servers
   judo animate [options] SNAPSHOT SERVER    ## create a new server from a snapshot
   judo erase [options] SNAPSHOT           ## erase an old snapshot

   judo swap [options] SERVER SERVER     ## swap elastic IP's and names on the two servers

   judo watch [options] SERVER         ## watch the server's boot process
   judo info [options] [SERVER ...]
   judo console [options] [SERVER ...] ## shows AWS console output
   judo ssh [options] [SERVER ...]     ## ssh's into the server

   # SERVER can be formatted as NAME or NAME:GROUP
   # or :GROUP to indicate the whole group.
   # If no servers are listed all servers are assumed.

   judo list [options]    ## lists all servers
   judo groups [options]  ## lists all groups

   judo volumes [options] ## shows all EBS volumes and what they are attached to
   judo ips [options]     ## shows all elastic ips and what they are attached to

EXAMPLES

An example is worth a thousand words.

A couchdb server:

./couchdb/config.json

{
    "ami32":"ami-2d4aa444",       // public ubuntu 10.04 ami - 32 bit
    "ami64":"ami-fd4aa494",       // public ubuntu 10.04 ami - 64 bit
    "user":"ubuntu",              // this is for ssh acccess - defaults to ubuntu not root
    "security_group":"judo",
    "availability_zone":"us-east-1d"
    "elastic_ip" : true,
    "packages" : "couchdb",
    "volumes" : { "device" : "/dev/sde1",
                  "media"  : "ebs",
                  "size"   : 64,
                  "format" : "ext3",
                  // this is where couchdb looks for its data by default
                  "mount"  : "/var/lib/couchdb/0.10.0",
                  "user"   : "couchdb",
                  "group"  : "couchdb" }
                  // make sure the data is owned by the couchdb user
                  // bounce couch since the data dir changed
}

./couchdb/setup.sh

service couchdb restart

./memcache/config.json

{
    "ami32":"ami-2d4aa444",       // public ubuntu 10.04 ami - 32 bit
    "ami64":"ami-fd4aa494",       // public ubuntu 10.04 ami - 64 bit
    "user":"ubuntu",              // this is for ssh acccess - defaults to ubuntu not root
    "security_group":"judo",
    "availability_zone":"us-east-1d"
    "elastic_ip" : true,
    "instance_type" : "m1.xlarge"
}

./memcache/setup.sh

apt-get install -y memcached
echo 'ENABLE_MEMCACHED=yes' > /etc/default/memcached
kuzushi-erb memcached.conf.erb > /etc/memcached.conf
service memcached start

./memcache/memcached.conf.erb

-d
logfile /var/log/memcached.log
## ohai gives memory in Kb so div by 1024 to get megs
## use 75% of total ram (* 0.75)
-m <%= (@system.memory["total"].to_i / 1024 * 0.75).to_i %>
-u nobody

A redis server with a 2 disk xfs raid 0:

./redis/config.json

{
    "ami32":"ami-2d4aa444",       // public ubuntu 10.04 ami - 32 bit
    "ami64":"ami-fd4aa494",       // public ubuntu 10.04 ami - 64 bit
    "user":"ubuntu",              // this is for ssh acccess - defaults to ubuntu not root
    "security_group":"judo",
    "availability_zone":"us-east-1d"
    "elastic_ip" : true,
    "instance_type" : "m2.xlarge",
    "volumes" : [{ "device" : "/dev/sde1",
                   "media"  : "ebs",
                   "scheduler" : "deadline",
                   "size"   : 16 },
                 { "device" : "/dev/sde2",
                   "media"  : "ebs",
                   "scheduler" : "deadline",
                   "size"   : 16 },
                 { "device"    : "/dev/md0",
                  "media"     : "raid",
                  "mount"     : "/var/lib/redis",
                  "drives"    : [ "/dev/sde1", "/dev/sde2" ],
                  "level"     : 0,
                  "format"    : "xfs" }]
}

./redis/redis-server_1.2.6-1_i686.deb

## the deb package can be included the the folder and pushed to the server

./redis/setup.sh

dpkg -i redis-server_1.2.6-1_i686.deb
chown redis:redis -R /var/lib/redis
service redis restart

CONFIG - LAUNCHING THE SERVER

"instance_type":"m1.small",

Specify the instance type for the server type here. See: http://aws.amazon.com/ec2/instance-types/. If nothing is specified m1.small is used.

"ami32":"ami-2d4aa444", "ami64":"ami-fd4aa494", "user":"ubuntu",

This is where you specify the AMI's to use. The defaults (above) are the ubuntu 10.04 public AMI's. The "user" value is which user has the keypair bound to it for ssh'ing into the server.

"security_group":"judo",

What security group to launch the server in. A judo group is created for you which only has port 22 access. Manually create new security groups as needed and name them here.

"availability_zone":"us-east-1d"

What zone to launch the server in.

"elastic_ip" : true,

If this is true, an elastic IP will be allocated for the server when it is created. This means that if the server is rebooted it will keep the same IP address.

"volumes" : [ { "device" : "/dev/sde1", "media" : "ebs", "size" : 64 } ],

You can specify one or more volumes for the group. If the media is of type "ebs" judo will create an elastic block device with a number of gigabytes specified under size. AWS currently allows values from 1 to 1000. If the media is anything other than "ebs" judo will ignore the entry. The EBS drives are tied to the server and attached as the specified device when started. Only when the server is destroyed are the EBS drives deleted.

Meta

Created by Orion Henry and Adam Wiggins. Forked from the gem 'sumo'.

Patches contributed by Blake Mizerany, Jesse Newland, Gert Goet, and Tim Lossen

Released under the MIT License: http://www.opensource.org/licenses/mit-license.php

http://github.com/orionz/judo