Skip to content

Latest commit

 

History

History
117 lines (95 loc) · 4.92 KB

2012-05-28-machine_batch.html.markdown

File metadata and controls

117 lines (95 loc) · 4.92 KB
Raw

Chef Metal Parallelization

Provisioning clusters with a lot of machines would be horrifically slow if you had to do it one machine at a time. Fortunately, Chef Metal 0.11 proffers powerful parallelization primitives, and even parallelizes machines by default wherever it can!

Here we'll describe how that works.

Default parallelization

By default, whenever you write machine resources several times in a row, Chef Metal notices and parallelizes them with a machine_batch. So, if you write a recipe like this:

# In recipe
machine 'a'
machine 'b'
machine 'c'

Then the output will look a bit like this:

$ CHEF_DRIVER=fog:AWS chef-apply cluster.rb
...
Converging 1 resources
Recipe: @recipe_files::/Users/jkeiser/oc/environments/metal-test-local/cluster.rb
  * machine_batch[default] action converge
    - [a] creating machine a on fog:AWS:862552916454
    - [a]   key_name: "metal_default"
    - [a]   tags: {"Name"=>"a", "BootstrapId"=>"http://localhost:8889/nodes/a", "BootstrapHost"=>"Johns-MacBook-Pro-2.local", "BootstrapUser"=>"jkeiser"}
    - [a]   name: "a"
    - [b] creating machine b on fog:AWS:862552916454
    - [b]   key_name: "metal_default"
    - [b]   tags: {"Name"=>"b", "BootstrapId"=>"http://localhost:8889/nodes/b", "BootstrapHost"=>"Johns-MacBook-Pro-2.local", "BootstrapUser"=>"jkeiser"}
    - [b]   name: "b"
    - [c] creating machine c on fog:AWS:862552916454
    - [c]   key_name: "metal_default"
    - [c]   tags: {"Name"=>"c", "BootstrapId"=>"http://localhost:8889/nodes/c", "BootstrapHost"=>"Johns-MacBook-Pro-2.local", "BootstrapUser"=>"jkeiser"}
    - [c]   name: "c"
    - [b] machine b created as i-eb778fb9 on fog:AWS:862552916454
    - create node b at http://localhost:8889
    -   add normal.tags = nil
    -   add normal.metal = {"location"=>{"driver_url"=>"fog:AWS:862552916454", "driver_version"=>"0.5.beta.2", "server_id"=>"i-eb778fb9", "creator"=>"user/jkeiser", "allocated_at"=>"2014-05-31 03:40:16 UTC", "key_name"=>"metal_default"}}
    - [a] machine a created as i-e9778fbb on fog:AWS:862552916454
    - create node a at http://localhost:8889
    -   add normal.tags = nil
    -   add normal.metal = {"location"=>{"driver_url"=>"fog:AWS:862552916454", "driver_version"=>"0.5.beta.2", "server_id"=>"i-e9778fbb", "creator"=>"user/jkeiser", "allocated_at"=>"2014-05-31 03:40:16 UTC", "key_name"=>"metal_default"}}
    - [c] machine c created as i-816d95d3 on fog:AWS:862552916454
    - create node c at http://localhost:8889
    -   add normal.tags = nil
    -   add normal.metal = {"location"=>{"driver_url"=>"fog:AWS:862552916454", "driver_version"=>"0.5.beta.2", "server_id"=>"i-816d95d3", "creator"=>"user/jkeiser", "allocated_at"=>"2014-05-31 03:40:17 UTC", "key_name"=>"metal_default"}}
    - [b] waiting for b (i-eb778fb9 on fog:AWS:862552916454) to be ready ...
    - [c] waiting for c (i-816d95d3 on fog:AWS:862552916454) to be ready ...
    - [a] waiting for a (i-e9778fbb on fog:AWS:862552916454) to be ready ...
...
        Running handlers:
        Running handlers complete

        Chef Client finished, 0/0 resources updated in 4.053363945 seconds
    - [c] run 'chef-client -l auto' on c

Running handlers:
Running handlers complete
Chef Client finished, 1/1 resources updated in 59.64014 seconds

You'll notice that, at the very end, it says 1/1 resources. This is because the three machines are replaced with a machine_batch resource that does the parallelization.

Since it's automatic, Chef Metal tries not to be overly aggressive:

  • Complex scripts don't parallelize: If you write machine, then put another resource in between (like remote_file), then another machine, Chef Metal will run the machines sequentially instead of parallelizing.
  • Different actions don't parallelize: machine 'a' followed by machine 'b' do action :delete end will not parallelize.

Default parallelization can be turned off by writing auto_batch_machines = false in a recipe or auto_batch_machines false in your Chef config (knife.rb or client.rb).

machine_batch

machine_batch can also be used explicitly. You can write a machine_batch that does a simple :setup, :converge, :stop or :delete (among others) like this:

# In recipe
machine_batch do
  action :setup
  machines 'a', 'b', 'c', 'd', 'e'
end

You can even mix and match different types of machines:

machine_batch do
  machine 'db' do
    recipe 'mysql'
  end
  1.upto(50) do |i|
    machine "#{web}#{i}" do
      recipe 'apache'
    end
  end
end

Even machines with different drivers and chef_servers can be mixed and parallelized. 50 in AWS and 50 in Azure? No problem! Run them all at once, in the same batch.

max_simultaneous

Specify the maximum number of threads to use with the max_simultaneous attribute on the machine_batch resource:

machine_batch do
  1.upto(50) do |i|
    machine "#{web}#{i}" do
      recipe 'apache'
    end
  end
  max_simultaneous 10
end

This will only create 10 machines at a time.