SSHKit is a toolkit for running commands in a structured way on one or more servers.
How might it work?
The typical use-case looks something like this:
require 'sshkit/dsl'
on %w{1.example.com 2.example.com}, in: :sequence, wait: 5 do
within "/opt/sites/example.com" do
as :deploy do
with rails_env: :production do
rake "assets:precompile"
runner "S3::Sync.notify"
end
end
end
end
One will notice that it's quite low level, but exposes a convenient API, the
as()/within()/with() are nestable in any order, repeatable, and stackable.
When used inside a block in this way, as() and within() will guard
the block they are given with a check.
In the case of within(), an error-raising check will be made that the directory
exists; for as() a simple call to sudo su -<user> whoami wrapped in a check for
success, raising an error if unsuccessful.
The directory check is implemented like this:
if test ! -d <directory>; then echo "Directory doesn't exist" 2>&1; false; fi
And the user switching test implemented like this:
if ! sudo su <user> -c whoami > /dev/null; then echo "Can't switch user" 2>&1; false; fi
According to the defaults, any command that exits with a status other than 0
raises an error (this can be changed). The body of the message is whatever was
written to stdout by the process. The 1>&2 redirects the standard output
of echo to the standard error channel, so that it's available as the body of
the raised error.
Helpers such as runner() and rake() which expand to execute(:rails, "runner", ...) and
execute(:rake, ...) are convenience helpers for Ruby, and Rails based apps.
Parallel
Notice on the on() call the in: :sequence option, the following will do
what you might expect:
on(in: :parallel, limit: 2) { ...}
on(in: :sequence, wait: 5) { ... }
on(in: :groups, limit: 2, wait: 5) { ... }
The default is to run in: :parallel with no limit, if you have 400 servers,
this might be a problem, and you might better look at changing that to run in
groups, or sequence.
Groups were designed in this case to relieve problems (mass Git checkouts) where you rely on a contested resource that you don't want to DDOS by hitting it too hard.
Sequential runs were intended to be used for rolling restarts, amongst other similar use-cases.
Synchronisation
The on() block is the unit of synchronisation, one on() block will wait
for all servers to complete before it returns.
For example:
all_servers = %w{one.example.com two.example.com three.example.com}
site_dir = '/opt/sites/example.com'
# Let's simulate a backup task, assuming that some servers take longer
# then others to complete
on servers do |host|
in site_dir do
execute :tar, '-czf', "backup-#{host.hostname}.tar.gz", 'current'
# Will run: "/usr/bin/env tar -czf backup-one.example.com.tar.gz current"
end
end
# Now we can do something with those backups, safe in the knowledge that
# they will all exist (all tar commands exited with a success status, or
# that we will have raised an exception if one of them failed.
on servers do |host|
in site_dir do
backup_filename = "backup-#{host.hostname}.tar.gz"
target_filename = "backups/#{Time.now.utc.iso8601}/#{host.hostname}.tar.gz"
puts capture(:s3cmd, 'put', backup_filename, target_filename)
end
end
The Command Map
It's often a problem that programatic SSH sessions don't share the same environmental variables as sessions that are started interactively.
This problem often comes when calling out to executables, expected to be on
the $PATH which, under conditions without dotfiles or other environmental
configuration are not where they are expected to be.
To try and solve this there is the with() helper which takes a hash of variables and makes them
available to the environment.
with path: '/usr/local/bin/rbenv/shims:$PATH' do
execute :ruby, '--version'
end
Will execute:
( PATH=/usr/local/bin/rbenv/shims:$PATH /usr/bin/env ruby --version )
Often more preferable is to use the command map.
The command map is used by default when instantiating a Command object
The command map exists on the configuration object, and in principle is quite simple, it's a Hash structure with a default key factory block specified, for example:
puts SSHKit.config.command_map[:ruby]
# => /usr/bin/env ruby
The /usr/bin/env prefix is applied to all commands, to make clear that the
environment is being deferred to to make the decision, this is what happens
anyway when one would simply attempt to execute ruby, however by making it
explicit, it was hoped that it might lead people to explore the documentation.
One can override the hash map for individual commands:
SSHKit.config.command_map[:rake] = "/usr/local/rbenv/shims/rake"
puts SSHKit.config.command_map[:rake]
# => /usr/local/rbenv/shims/rake
One can also override the command map completely, this may not be wise, but it would be possible, for example:
SSHKit.config.command_map = Hash.new do |hash, command|
hash[command] = "/usr/local/rbenv/shims/#{command}"
end
This would effectively make it impossible to call any commands which didn't provide an executable in that directory, but in some cases that might be desirable.
Note: All keys should be symbolised, as the Command object will symbolize it's first argument before attempting to find it in the command map.
Output Handling
The output handling comprises two objects, first is the output itself, by default this is $stdout, but can be any object responding to a StringIO-like interface. The second part is the formatter.
The formatter and output have a strange relationship:
SSHKit.config.output = SSHKit.config.formatter.new($stdout)
The formatter will typically delegate all calls to the output, depending
on it's implementation it will almost certainly override the implementation of
write() (alias <<()) and query the objects it receives to determine what
should be printed.
Known Issues
- No handling of slow / timed out connections
- No handling ot slow / hung remote commands
- No built-in way to background() something (execute and background the process)
- No environment handling
- No arbitrary
Hostproperties