zc.zk 2.0.0a4

Installation

You can install this as you would any other distribution. It requires the kazoo Python ZooKeeper interface.

Instantiating a ZooKeeper helper

To use the helper API, create a ZooKeeper instance:

>>> import zc.zk
>>> zk = zc.zk.ZooKeeper('zookeeper.example.com:2181')

The ZooKeeper constructor takes a ZooKeeper connection string, which is a comma-separated list of addresses of the form HOST:PORT. It defaults to '127.0.0.1:2181', which is convenient during development.

You can also pass a kazoo client object, instead of a connection string.

Register a server providing a service

To register a server, use the register method, which takes a service path and the address a server is listing on:

>>> zk.register('/fooservice/providers', ('192.168.0.42', 8080))

register creates a read-only ephemeral ZooKeeper node as a child of the given service path. The name of the new node is (a string representation of) the given address. This allows clients to get the list of addresses by just getting the list of the names of children of the service path.

Ephemeral nodes have the useful property that they’re automatically removed when a ZooKeeper session is closed or when the process containing it dies. De-registration is automatic.

When registering a server, you can optionally provide server (node) data as additional keyword arguments to register. By default, the process id is set as the pid property. This is useful to tracking down the server process. In addition, an event is generated, providing subscribers to add properties as a server is being registered. (See Server-registration events.)

Get the addresses of service providers

Getting the addresses providing a service is accomplished by getting the children of a service node:

>>> addresses = zk.children('/fooservice/providers')
>>> sorted(addresses)
['192.168.0.42:8080']

The children method returns an iterable of names of child nodes of the node specified by the given path. The iterable is automatically updated when new providers are registered:

>>> zk.register('/fooservice/providers', ('192.168.0.42', 8081))
>>> sorted(addresses)
['192.168.0.42:8080', '192.168.0.42:8081']

You can also get the number of children with len:

>>> len(addresses)
2

You can call the iterable with a callback function that is called whenever the list of children changes:

>>> @zk.children('/fooservice/providers')
... def addresses_updated(addresses):
...     print 'addresses changed'
...     print sorted(addresses)
addresses changed
['192.168.0.42:8080', '192.168.0.42:8081']

The callback is called immediately with the children. When we add another child, it’ll be called again:

>>> zk.register('/fooservice/providers', ('192.168.0.42', 8082))
addresses changed
['192.168.0.42:8080', '192.168.0.42:8081', '192.168.0.42:8082']

Get service configuration data

You get service configuration data by getting properties associated with a ZooKeeper node. The interface for getting properties is similar to the interface for getting children:

>>> data = zk.properties('/fooservice')
>>> data['database']
u'/databases/foomain'
>>> data['threads']
1

The properties method returns a mapping object that provides access to node data. (ZooKeeper only stores string data for nodes. zc.zk provides a higher-level data interface by storing JSON strings.)

The properties objects can be called with callback functions and used as function decorators to get update notification:

>>> @zk.properties('/fooservice')
... def data_updated(data):
...     print 'data updated'
...     for item in sorted(data.items()):
...         print '%s: %r' % item
data updated
database: u'/databases/foomain'
favorite_color: u'red'
threads: 1

The callback is called immediately. It’ll also be called when data are updated.

Updating node properties

You can update properties by calling the update method:

>>> thread_info = {'threads': 2}
>>> data.update(thread_info, secret='123')
data updated
database: u'/databases/foomain'
favorite_color: u'red'
secret: u'123'
threads: 2

You can also set individual properties:

>>> data['threads'] = 1
data updated
database: u'/databases/foomain'
favorite_color: u'red'
secret: u'123'
threads: 1

If you call the set method, keys not listed are removed:

>>> data.set(threads= 3, secret='1234')
data updated
secret: u'1234'
threads: 3

Both update and set can take data from a positional data argument, or from keyword parameters. Keyword parameters take precedent over the positional data argument.

Getting property data without tracking changes

Sometimes, you want to get service data, but don’t want to watch for changes. If you pass watch=False to properties, Then properties won’t track changes. In this case, you can’t set callback functions, but you can still update data.

Tree-definition format, import, and export

You can describe a ZooKeeper tree using a textual tree representation. You can then populate the tree by importing the representation. Heres an example:

/lb : ipvs
  /pools
    /cms
      # The address is fixed because it's
      # exposed externally
      address = '1.2.3.4:80'
      providers -> /cms/providers
    /retail
      address = '1.2.3.5:80'
      providers -> /cms/providers

/cms : z4m cms
  threads = 3
  /providers
  /databases
    /main
      /providers

/retail : z4m retail
  threads = 1
  /providers
  /databases
    main -> /cms/databases/main
    /ugc
      /providers

This example defines a tree with 3 top nodes, lb and cms, and retail. The retail node has two sub-nodes, providers and databases and a property threads.

The /retail/databases node has symbolic link, main and a ugc sub-node. The symbolic link is implemented as a property named `` We’ll say more about symbolic links in a later section.

The lb, cms and retail nodes have types. A type is indicated by following a node name with a colon and a string value. The string value is used to populate a type property. Types are useful to document the kinds of services provided at a node and can be used by deployment tools to deploy service providers.

You can import a tree definition with the import_tree method:

>>> zk.import_tree(tree_text)

This imports the tree at the top of the ZooKeeper tree.

We can also export a ZooKeeper tree:

>>> print zk.export_tree(),
/cms : z4m cms
  threads = 3
  /databases
    /main
      /providers
  /providers
/fooservice
  secret = u'1234'
  threads = 3
  /providers
/lb : ipvs
  /pools
    /cms
      address = u'1.2.3.4:80'
      providers -> /cms/providers
    /retail
      address = u'1.2.3.5:80'
      providers -> /cms/providers
/retail : z4m retail
  threads = 1
  /databases
    main -> /cms/databases/main
    /ugc
      /providers
  /providers

Note that when we export a tree:

• The special reserved top-level zookeeper node is omitted.

• Ephemeral nodes are omitted.

• Each node’s information is sorted by type (properties, then links,

• then sub-nodes) and then by name,

You can export just a portion of a tree:

>>> print zk.export_tree('/fooservice'),
/fooservice
  secret = u'1234'
  threads = 3
  /providers

You can optionally see ephemeral nodes:

>>> print zk.export_tree('/fooservice', ephemeral=True),
/fooservice
  secret = u'1234'
  threads = 3
  /providers
    /192.168.0.42:8080
      pid = 9999
    /192.168.0.42:8081
      pid = 9999
    /192.168.0.42:8082
      pid = 9999

We can import a tree over an existing tree and changes will be applied. Let’s update our textual description:

/lb : ipvs
  /pools
    /cms
      # The address is fixed because it's
      # exposed externally
      address = '1.2.3.4:80'
      providers -> /cms/providers

/cms : z4m cms
  threads = 4
  /providers
  /databases
    /main
      /providers

and re-import:

>>> zk.import_tree(tree_text)
extra path not trimmed: /lb/pools/retail

We got a warning about nodes left over from the old tree. We can see this if we look at the tree:

>>> print zk.export_tree(),
/cms : z4m cms
  threads = 4
  /databases
    /main
      /providers
  /providers
/fooservice
  secret = u'1234'
  threads = 3
  /providers
/lb : ipvs
  /pools
    /cms
      address = u'1.2.3.4:80'
      providers -> /cms/providers
    /retail
      address = u'1.2.3.5:80'
      providers -> /cms/providers
/retail : z4m retail
  threads = 1
  /databases
    main -> /cms/databases/main
    /ugc
      /providers
  /providers

If we want to trim these, we can add a trim option. This is a little scary, so we’ll use the dry-run option to see what it’s going to do:

>>> zk.import_tree(tree_text, trim=True, dry_run=True)
would delete /lb/pools/retail.

If we know we’re not trimming things and want to avoid a warning, we can use trim=False:

>>> zk.import_tree(tree_text, trim=False)

We can see that this didn’t trim by using dry-run again:

>>> zk.import_tree(tree_text, trim=True, dry_run=True)
would delete /lb/pools/retail.

We do want to trim, so we use trim=True:

>>> zk.import_tree(tree_text, trim=True)
>>> print zk.export_tree(),
/cms : z4m cms
  threads = 4
  /databases
    /main
      /providers
  /providers
/fooservice
  secret = u'1234'
  threads = 3
  /providers
/lb : ipvs
  /pools
    /cms
      address = u'1.2.3.4:80'
      providers -> /cms/providers
/retail : z4m retail
  threads = 1
  /databases
    main -> /cms/databases/main
    /ugc
      /providers
  /providers

Note that nodes containing (directly or recursively) ephemeral nodes will never be trimmed. Also node that top-level nodes are never automatically trimmed. So we weren’t warned about the unreferenced top-level nodes in the import.

Recursive deletion

ZooKeeper only allows deletion of nodes without children. The delete_recursive method automates removing a node and all of it’s children.

If we want to remove the retail top-level node, we can use delete_recursive:

>>> zk.delete_recursive('/retail')
>>> print zk.export_tree(),
/cms : z4m cms
  threads = 4
  /databases
    /main
      /providers
  /providers
/fooservice
  secret = u'1234'
  threads = 3
  /providers
/lb : ipvs
  /pools
    /cms
      address = u'1.2.3.4:80'
      providers -> /cms/providers

Bt default, delete_recursive won’t delete ephemeral nodes, or nodes that contain them:

>>> zk.delete_recursive('/fooservice')
Not deleting /fooservice/providers/192.168.0.42:8080 because it's ephemeral.
Not deleting /fooservice/providers/192.168.0.42:8081 because it's ephemeral.
Not deleting /fooservice/providers/192.168.0.42:8082 because it's ephemeral.
/fooservice/providers not deleted due to ephemeral descendent.
/fooservice not deleted due to ephemeral descendent.

You can use the force option to force ephemeral nodes to be deleted.

Symbolic links

ZooKeeper doesn’t have a concept of symbolic links, but zc.zk provides a convention for dealing with symbolic links. When trying to resolve a path, if a node lacks a child, but has a property with a name ending in ' ->', the child will be found by following the path in the property value.

The resolve method is used to resolve a path to a real path:

>>> zk.resolve('/lb/pools/cms/providers')
u'/cms/providers'

In this example, the link was at the endpoint of the virtual path, but it could be anywhere:

>>> zk.register('/cms/providers', '1.2.3.4:5')
>>> zk.resolve('/lb/pools/cms/providers/1.2.3.4:5')
u'/cms/providers/1.2.3.4:5'

Note a limitation of symbolic links is that they can be hidden by children. For example, if we added a real node, at /lb/pools/cms/provioders, it would shadow the link.

children, properties, and register will automatically use resolve to resolve paths.

When the children and properties are used for a node, the paths they use will be adjusted dynamically when paths are removed. To illustrate this, let’s get children of /cms/databases/main:

>>> main_children = zk.children('/cms/databases/main')
>>> main_children.path
'/cms/databases/main'
>>> main_children.real_path
'/cms/databases/main'

Children and Properties objects have a path attribute that has the value passed to the children or properties methods. They have a real_path attribute that contains the path after resolving symbolic links. Let’s suppose we want to move the database node to ‘/databases/cms’. First we’ll export it:

>>> export = zk.export_tree('/cms/databases/main', name='cms')
>>> print export,
/cms
  /providers

Note that we used the export name option to specify a new name for the exported tree.

Now, we’ll create a databases node:

>>> zk.create('/databases')
'/databases'

And import the export:

>>> zk.import_tree(export, '/databases')
>>> print zk.export_tree('/databases'),
/databases
  /cms
    /providers

Next, we’ll create a symbolic link at the old location. We can use the ln convenience method:

>>> zk.ln('/databases/cms', '/cms/databases/main')
>>> zk.get_properties('/cms/databases')
{u'main ->': u'/databases/cms'}

Now, we can remove /cms/databases/main and main_children will be updated:

>>> zk.delete_recursive('/cms/databases/main')
>>> main_children.path
'/cms/databases/main'
>>> main_children.real_path
u'/databases/cms'

If we update /databases/cms, main_children will see the updates:

>>> sorted(main_children)
['providers']
>>> _ = zk.delete('/databases/cms/providers')
>>> sorted(main_children)
[]

Symbolic links can be relative. If a link doesn’t start with a slash, it’s interpreted relative to the node the link occurs in. The special names . and .. have their usual meanings.

So, in:

/a
  /b
    l -> c
    l2 -> ../c
    /c
  /c

The link at /a/b/l resolves to /a/b/c and /a/b/l2 resolves to /a/c.

Property links

In addition to symbolic links between nodes, you can have links between properties. In our earlier example, both the /cms and /fooservice nodes had threads properties:

/cms : z4m cms
  threads = 4
  /databases
    /main
      /providers
  /providers
/fooservice
  secret = u'1234'
  threads = 3
  /providers
/lb : ipvs
  /pools
    /cms
      address = u'1.2.3.4:80'
      providers -> /cms/providers

If we wanted /cms to have the same threads settings, we could use a property link:

/cms : z4m cms
  threads => /fooservice threads
  /databases
    /main
      /providers
  /providers
/fooservice
  secret = u'1234'
  threads = 3
  /providers
/lb : ipvs
  /pools
    /cms
      address = u'1.2.3.4:80'
      providers -> /cms/providers

Property links are indicated with =>. The value is a node path and optional property name, separated by whitespace. If the name is ommitted, then the refering name is used. For example, the name could be left off of the property link above.

Node deletion

If a node is deleted and Children or Properties instances have been created for it, and the paths they were created with can’t be resolved using symbolic links, then the instances’ data will be cleared. Attempts to update properties will fail. If callbacks have been registered, they will be called without arguments, if possible. It would be bad, in practice, to remove a node that processes are watching.

Registering a server with a blank hostname

It’s common to use an empty string for a host name when calling bind to listen on all IPv4 interfaces. If you pass an address with an empty host to register and netifaces is installed, then all of the non-local IPv4 addresses [1] (for the given port) will be registered.

If there are no non-local interfaces (not connected to network), then the local IPV4 interface will be registered.

If netifaces isn’t installed and you pass an empty host name, then the fully-qualified domain name, as returned by socket.getfqdn() will be used for the host.

Server-registration events

When register is called, a zc.zk.RegisteringServer event is emmitted with a properties attribute that can be updated by subscribers prior to creating the ZooKeeper ephemeral node. This allows third-party code to record extra server information.

Events are emitted by passing them to zc.zk.event.notify. If zope.event is installed, then zc.zk.event.notify is an alias for zope.event.notify, otherwise, zc.zk.event.notify is an empty function that can be replaced by applications.

ZooKeeper Session Management

Kazoo takes care of reestablishing ZooKeeper sessions. Watches created with the children and properties methods are reestablished when new sessions are established. zc.zk also recreates ephemeral nodes created via register.

zookeeper_export script

The zc.zk package provides a utility script for exporting a ZooKeeper tree:

$ zookeeper_export -e zookeeper.example.com:2181 /fooservice
/fooservice
  secret = u'1234'
  threads = 3
  /providers
    /192.168.0.42:8080
      pid = 9999
    /192.168.0.42:8081
      pid = 9999
    /192.168.0.42:8082
      pid = 9999

The export script provides the same features as the export_tree method. Use the --help option to see how to use it.

zookeeper_import script

The zc.zk package provides a utility script for importing a ZooKeeper tree. So, for example, given the tree:

/provision
  /node1
  /node2

In the file mytree.txt, we can import the file like this:

$ zookeeper_import zookeeper.example.com:2181 mytree.txt /fooservice

The import script provides the same features as the import_tree method, with the exception that it provides less flexibility for specifing access control lists. Use the --help option to see how to use it.

Propery-update script

The zc.zk package provides a utility script for updating individual properties:

zookeeper_set_property zookeeper.example.com:2181 /fooservice \
    threads=4 debug=True comment='ok'

The first argument to the script is the path of the node to be updated. Any number of additional arguments of the form: NAME=PYTHONEXPRESSION are provided to supply updates. If setting strings, you may have to quote the argument, as in “comment=’a comment’”.

Iterating over a tree

The walk method can be used to walk over the nodes in a tree:

>>> for path in zk.walk():
...     print path
/
/cms
/cms/databases
/cms/providers
/cms/providers/1.2.3.4:5
/databases
/databases/cms
/fooservice
/fooservice/providers
/fooservice/providers/192.168.0.42:8080
/fooservice/providers/192.168.0.42:8081
/fooservice/providers/192.168.0.42:8082
/fooservice/provision
/fooservice/provision/node1
/fooservice/provision/node2
/lb
/lb/pools
/lb/pools/cms
/zookeeper
/zookeeper/quota

>>> for path in zk.walk('/fooservice'):
...     print path
/fooservice
/fooservice/providers
/fooservice/providers/192.168.0.42:8080
/fooservice/providers/192.168.0.42:8081
/fooservice/providers/192.168.0.42:8082
/fooservice/provision
/fooservice/provision/node1
/fooservice/provision/node2

You can omit ephemeral nodes:

>>> for path in zk.walk('/fooservice', ephemeral=False):
...     print path
/fooservice
/fooservice/providers
/fooservice/providers/192.168.0.42:8080
/fooservice/providers/192.168.0.42:8081
/fooservice/providers/192.168.0.42:8082
/fooservice/provision
/fooservice/provision/node1
/fooservice/provision/node2

You can also get a mutable list of children, which you can mutate:

>>> i = zk.walk('/fooservice', children=True)
>>> path, children = i.next()
>>> path, children
('/fooservice', ['providers', 'provision'])

>>> del children[0]
>>> for path in i:
...     print path
/fooservice/provision
/fooservice/provision/node1
/fooservice/provision/node2

Modifications to nodes are reflected while traversing:

>>> for path in zk.walk('/fooservice'):
...     print path
...     if 'provision' in zk.get_children(path):
...         zk.delete_recursive(path+'/provision')
/fooservice
/fooservice/providers
/fooservice/providers/192.168.0.42:8080
/fooservice/providers/192.168.0.42:8081
/fooservice/providers/192.168.0.42:8082

Graph analysis

The textual tree representation can be used to model and analyze a system architecturte. You can get a parsed representation of a tree using zc.zk.parse_tree to parse a text tree representation generated by hand for import, or using the export_tree method.

>>> tree = zc.zk.parse_tree(tree_text)
>>> sorted(tree.children)
['cms', 'lb']
>>> tree.children['lb'].properties
{'type': 'ipvs'}

The demo module, zc.zk.graphvis shows how you might generate system diagrams from tree models.

Reference

>>> import zc.zk.testing
>>> help(zc.zk.testing)

Change History