mutex_rdlm cookbook

Contains some functions that work with an external mutex server.

Also supports managing the mutex server using RDLM

Supported Platforms

For clients: Nothing native, so anything goes I guess.

For mutex server: Tested on CentOS 6.7.

Attributes

<table>
<tr>
<th>Key</th>
<th>Type</th>
<th>Used by</th>
<th>Description</th>
<th>Default</th>
</tr>
<tr>
<td><tt>['mutex_rdlm']['scheme']</tt></td>
<td>String</td>
<td>Clients</td>
<td>Default for mutex scheme</td>
<td><tt>http</tt></td>
</tr>
<tr>
<td><tt>['mutex_rdlm']['hostname']</tt></td>
<td>String</td>
<td>Clients</td>
<td>Default for mutex server name</td>
<td><tt>nil</tt></td>
</tr>
<tr>
<td><tt>['mutex_rdlm']['port']</tt></td>
<td>Integer</td>
<td>Server and Clients</td>
<td>Port for mutex server. Used as default by clients</td>
<td><tt>7305</tt></td>
</tr>
<tr>
<td><tt>['mutex_rdlm']['wait']</tt></td>
<td>Integer</td>
<td>Clients</td>
<td>Default for number of seconds to wait to acquire the lock before raising an error</td>
<td><tt>5</tt></td>
</tr>
<tr>
<td><tt>['mutex_rdlm']['lifetime']</tt></td>
<td>Integer</td>
<td>Clients</td>
<td>Default for how long the lock holds before expiring</td>
<td><tt>300</tt></td>
</tr>
</table>

Recipes

mutex_rdlm::server

Installs and configures the simple RDLM server.

It uses the ['mutex_rdlm']['port'] attribute to determine the port the daemon will be listening on.

I'm using an init file tested on CentOS 6.7.

Library resources

Common parameters

Used to locate the mutex server, appear under additional_config in method descriptions:

• mutex_url: How to reach the mutex server. Defaults to building the url from attributes (if possible)
• mutex_wait: Number of seconds to wait to acquire the lock before raising an error. Defaults to node['mutex_rdlm']['wait']
• mutex_lifetime: Number of seconds before the lock will expire on its own. Defaults to node['mutex_rdlm']['lifetime']

MutexRDLM::with_mutex

Yields (runs a code block) while locking the resource specified on the mutex.

Parameters:

• node: Calling node's object
• mutex_resource: name for mutex
• "additional_config" detailed above

Can be thought of like the synchronized keyword in java.

Example:
```ruby

Add apple to databag if it has none

MutexRDLM::with_mutex(node,'dbvendingjuice') do
db=data_bag_item('vending','juice')
db['kinds']<<'apple' unless db['kinds'].member? 'apple'
db.save
end
``
This will ensure the databag has only oneapple` item, even if several clients run this recipe at the same time.

MutexRDLM::assign_identity

Used to assign a unique identity.

Parameters:

• node: Calling node's object
• assignment_path: Path for attribute to store unique value in
• range: Pool of possibe values
• additional_config: Detailed above

Resulting effects

Upon successful completion of the function, this node will be assigned a unique identity in its node object.

The identity will be stored as a node attribute specified by assignment_path, where [:a,0,'bla'] is used to address the node attribute node[:a][0]['bla'].

The value is selected from range and is guranteed to be different from any value currently present on other nodes in the Chef server.

Selection is done without concurrency, thanks to the mutex.

The resulting value is saved on both the current node object (for use in other parts of the recipe) and in the server's version of the node object immediately, to stop other nodes from choosing the same value. This is important to mention because by default, Chef client only updates the server's version of the node object if the run is successful. In our case, this attribute is updated during resource compilation.

There are several ways this method can fail:

• Can't talk to mutex server
  Fix the server, firewalls etc.
• Can't get mutex lock (other client is busy with mutex).
  • My library might be buggy and not release the mutex for some reason. Create a PR/Issue :)
  • Enumerating all of the nodes takes longer than the mutex time constraints. Allow more wait time.
• Range is exhausted
  • Increase the range
  • Delete dead nodes from the Chef server

Basic walk-through

Assuming a working and reachable mutex server:

1. Function is called with the following:
   • assignment_path = [:slasher,:id]
   • range = (1..5).to_a
   • mutex_url = 'http://mutex:8080/'
   • mutex_wait = 5
   • mutex_lock = 300
2. Current value of attribute is checked (node[:slasher][:id]). If not-empty, function returns. Assuming empty
3. Mutex is locked for a normalized version of the assignment_path (slasherid)
4. If mutex can't be locked for some reason, raise error
5. All node objects on Chef server are enumerated for that value. Resulting collection is filtered out from range.
6. Take a single element from range (first). If empty, raise error (release mutex before). Assuming got 3.
7. Load our node object from Chef server, modify the node attribute and save it.
8. Update the current node's attribute (node.normal[:slasher][:id]=3)
9. Release mutex
10. Return 3

MutexRDLM::find_duplicate_identity

Parameters:

• node: Calling node's object
• assignment_path: Path for attribute unique value is stored in
• only_me: Only check for conflicts involving my node
• additional_config: Detailed above

Used to enforce uniqueness of the identity attribute without modifying anything.

Is useful in monitoring.

Return value

Empty hash ({}) if there are no duplicates.

If there are duplicates, the hash is composed of a key for the duplicate identity and the value is an array of nodes holding said value, like this:
ruby
{
3 => ['node1.backslasher.net', 'node2.backslasher.net']
}


License and Authors

Licensed GPL v2

Author:: Nitzan Raz (backslasher)