Skip to main content

Higher Level Zookeeper Client

Project description

Travis Build Latest Version Downloads

kazoo implements a higher level API to Apache Zookeeper for Python clients.

See the full docs for more information.


kazoo is offered under the Apache License 2.0.


kazoo started under the Nimbus Project and through collaboration with the open-source community has been merged with code from Mozilla and the Zope Corporation. It has since gathered an active community of over fifty contributors.


2.X.Y (TBD)


Bug Handling

  • #372: fully resolve multiple records for hosts in the zookeeper connection string


2.2.1 (2015-06-17)


Bug Handling

  • handle NameError with basestring on py3k.


2.2 (2015-06-15)



  • Issue #234: Add support for reconfig cluster membership operation

Bug Handling

  • #315: multiple acquires of a kazoo lock using the lock recipe would block when using acquire even when non-blocking is specified (only when the lock was/has been already acquired).

  • #318: At exit register takes *args and **kwargs not args and kargs


2.1 (2015-05-11)


  • Start running tests against Zookeeper 3.5.0 alpha and explicitly configure the admin.serverPort in tests to avoid port conflicts. The Zookeeper alpha version is not yet officially supported.

  • Integrate eventlet handler support into kazoo so that along with [gevent, threading] handlers there can now be a dedicated eventlet handler for projects that need to (or want to) use eventlet (such as those working in the openstack community). The requirements_eventlet.txt file lists the optional eventlet requirement(s) that needs to be satisfied when this new handler is used.

  • Use six to nicely handle the cross compatibility of kazoo with python 2.x and 3.x (reducing/removing the need to have custom compatibility code that replicates what six already provides).

  • Add state_change_event to kazoo.recipe.partitioner.SetPartitioner which is set on every state change.

  • Add a NonBlockingLease recipe. The recipe allows e.g. cron jobs scheduled on multiple machines to ensure that at most N instances will run a particular job, with lease timeout for graceful handover in case of node failures.

Bug Handling

  • #291: Kazoo lock recipe was only partially re-entrant in that multiple calls to acquire would obtain the the lock but the first call to release would remove the underlying lock. This would leave the X - 1 other acquire statements unprotected (and no longer holding there expected lock). To fix this the comment about that lock recipe being re-entrant has been removed and multiple acquires will now block when attempted.

  • #78: Kazoo now uses socketpairs instead of pipes making it compatible with Windows.

  • #144, #221: Let client.command work with IPv6 addresses.

  • #216: Fixed timeout for ConnectionHandler._invoke.

  • #261: Creating a sequential znode under / doesn’t work.

  • #274: Add server_version() retries (by default 4 attempts will be made) to better handle flakey responses.

  • #271: Fixed handling of KazooState.SUSPENDED in SetPartitioner.

  • #283: Fixed a race condition in SetPartitioner when party changes during handling of lock acquisition.

  • #303: don’t crash on random input as the hosts string.


  • #222: Document that committed on the transaction is used to ensure only one commit and is not an indicator of whether operations in the transaction returned desired results.

2.0 (2014-06-19)


  • Extend support to Python 3.4, deprecating Python 3.2.

  • Issue #198: Mention Zake as a sophisticated kazoo mock testing library.

  • Issue #181: Add documentation on basic logging setup.

2.0b1 (2014-04-24)

API Changes

  • Null or None data is no longer treated as “”. Pull req #165, patch by Raul Gutierrez S. This will affect how you should treat null data in a znode vs. an empty string.

  • Passing acl=[] to create() now works properly instead of an InvalidACLError as it returned before. Patch by Raul Gutierrez S in PR #164.

  • Removed the dependency on zope.interface. The classes in the interfaces module are left for documentation purposes only (issue #131).


  • Logging levels have been reduced.

    • Logging previously at the logging.DEBUG level is now logged at the kazoo.loggingsupport.BLATHER level (5).

    • Some low-level logging previously at the logging.INFO level is now logged at the logging.DEBUG level.

  • Issue #133: Introduce a new environment variable ZOOKEEPER_PORT_OFFSET for the testing support, to run the testing cluster on a different range.

Bug Handling

  • When authenticating via add_auth() the auth data will be saved to ensure that the authentication happens on reconnect (as is the case when feeding auth data via KazooClient’s constructor). PR #172, patch by Raul Gutierrez S.

  • Change gevent import to remove deprecation warning when newer gevent is used. PR #191, patch by Hiroaki Kawai.

  • Lock recipe was failing to use the client’s sleep_func causing issues with gevent. Issue #150.

  • Calling a DataWatch or ChildrenWatch instance twice (decorator) now throws an exception as only a single function can be associated with a single watcher. Issue #154.

  • Another fix for atexit handling so that when disposing of connections the atexit handler is removed. PR #190, patch by Devaev Maxim.

  • Fix atexit handling for kazoo threading handler, PR #183. Patch by Brian Wickman.

  • Partitioner should handle a suspended connection properly and restore an allocated state if it was allocated previously. Patch by Manish Tomar.

  • Issue #167: Closing a client that was never started throws a type error. Patch by Joshua Harlow.

  • Passing dictionaries to KazooClient.__init__() wasn’t actually working properly. Patch by Ryan Uber.

  • Issue #119: Handler timeout takes the max of the random interval or the read timeout to ensure a negative number isn’t used for the read timeout.

  • Fix ordering of exception catches in lock.acquire as it was capturing a parent exception before the child. Patch by ReneSac.

  • Fix issue with client.stop() not always setting the client state to KeeperState.CLOSED. Patch by Jyrki Pulliainen in PR #174.

  • Issue #169: Fixed pipes leaking into child processes.


  • Add section on contributing recipes, add maintainer/status information for existing recipes.

  • Add note about alternate use of DataWatch.

1.3.1 (2013-09-25)

Bug Handling

  • #118, #125, #128: Fix unknown variable in KazooClient command_retry argument handling.

  • #126: Fix KazooRetry.copy to correctly copy sleep function.

  • #118: Correct session/socket timeout conversion (int vs. float).


  • #121: Add a note about kazoo.recipe.queue.LockingQueue requiring a Zookeeper 3.4+ server.

1.3 (2013-09-05)


  • #115: Limit the backends we use for SLF4J during tests.

  • #112: Add IPv6 support. Patch by Dan Kruchinin.

1.2.1 (2013-08-01)

Bug Handling

  • Issue #108: Circular import fail when importing kazoo.recipe.watchers directly has now been resolved. Watchers and partitioner properly import the KazooState from kazoo.protocol.states rather than kazoo.client.

  • Issue #109: Partials not usable properly as a datawatch call can now be used. All funcs will be called with 3 args and fall back to 2 args if there’s an argument error.

  • Issue #106, #107: client.create_async didn’t strip change root from the returned path.

1.2 (2013-07-24)


  • KazooClient can now be stopped more reliably even if its in the middle of a long retry sleep. This utilizes the new interrupt feature of KazooRetry which lets the sleep be broken down into chunks and an interrupt function called to determine if the retry should fail early.

  • Issue #62, #92, #89, #101, #102: Allow KazooRetry to have a max deadline, transition properly when connection fails to LOST, and setup separate connection retry behavior from client command retry behavior. Patches by Mike Lundy.

  • Issue #100: Make it easier to see exception context in threading and connection modules.

  • Issue #85: Increase information density of logs and don’t prevent dynamic reconfiguration of log levels at runtime.

  • Data-watchers for the same node are no longer ‘stacked’. That is, if a get and an exists call occur for the same node with the same watch function, then it will be registered only once. This change results in Kazoo behaving per Zookeeper client spec regarding repeat watch use.

Bug Handling

  • Issue #53: Throw a warning upon starting if the chroot path doesn’t exist so that it’s more obvious when the chroot should be created before performing more operations.

  • Kazoo previously would let the same function be registered as a data-watch or child-watch multiple times, and then call it multiple times upon being triggered. This was non-compliant Zookeeper client behavior, the same watch can now only be registered once for the same znode path per Zookeeper client documentation.

  • Issue #105: Avoid rare import lock problems by moving module imports in to the module scope.

  • Issue #103: Allow prefix-less sequential znodes.

  • Issue #98: Extend testing ZK harness to work with different file locations on some versions of Debian/Ubuntu.

  • Issue #97: Update some docstrings to reflect current state of handlers.

  • Issue #62, #92, #89, #101, #102: Allow KazooRetry to have a max deadline, transition properly when connection fails to LOST, and setup separate connection retry behavior from client command retry behavior. Patches by Mike Lundy.

API Changes

  • The kazoo.testing.harness.KazooTestHarness class directly inherits from unittest.TestCase and you need to ensure to call its __init__ method.

  • DataWatch no longer takes any parameters besides for the optional function during instantiation. The additional options are now implicitly True, with the user being left to ignore events as they choose. See the DataWatch API docs for more information.

  • Issue #99: Better exception raised when the writer fails to close. A WriterNotClosedException that inherits from KazooException is now raised when the writer fails to close in time.

1.1 (2013-06-08)


  • Issue #93: Add timeout option to lock/semaphore acquire methods.

  • Issue #79 / #90: Add ability to pass the WatchedEvent to DataWatch and ChildWatch functions.

  • Respect large client timeout values when closing the connection.

  • Add a max_leases consistency check to the semaphore recipe.

  • Issue #76: Extend testing helpers to allow customization of the Java classpath by specifying the new ZOOKEEPER_CLASSPATH environment variable.

  • Issue #65: Allow non-blocking semaphore acquisition.

Bug Handling

  • Issue #96: Provide Windows compatibility in testing harness.

  • Issue #95: Handle errors deserializing connection response.

  • Issue #94: Clean up stray bytes in connection pipe.

  • Issue #87 / #88: Allow re-acquiring lock after cancel.

  • Issue #77: Use timeout in initial socket connection.

  • Issue #69: Only ensure path once in lock and semaphore recipes.

  • Issue #68: Closing the connection causes exceptions to be raised by watchers which assume the connection won’t be closed when running commands.

  • Issue #66: Require ping reply before sending another ping, otherwise the connection will be considered dead and a ConnectionDropped will be raised to trigger a reconnect.

  • Issue #63: Watchers weren’t reset on lost connection.

  • Issue #58: DataWatcher failed to re-register for changes after non-existent node was created then deleted.

API Changes

  • KazooClient.create_async now supports the makepath argument.

  • KazooClient.ensure_path now has an async version, ensure_path_async.

1.0 (2013-03-26)


  • Added a LockingQueue recipe. The queue first locks an item and removes it from the queue only after the consume() method is called. This enables other nodes to retake the item if an error occurs on the first node.

Bug Handling

  • Issue #50: Avoid problems with sleep function in mixed gevent/threading setup.

  • Issue #56: Avoid issues with watch callbacks evaluating to false.

1.0b1 (2013-02-24)


  • Refactored the internal connection handler to use a single thread. It now uses a deque and pipe to signal the ZK thread that there’s a new command to send, so that the ZK thread can send it, or retrieve a response. Processing ZK requests and responses serially in a single thread eliminates the need for a bunch of the locking, the peekable queue and two threads working on the same underlying socket.

  • Issue #48: Added documentation for the retry helper module.

  • Issue #55: Fix os.pipe file descriptor leak and introduce a KazooClient.close method. The method is particular useful in tests, where multiple KazooClients are created and closed in the same process.

Bug Handling

  • Issue #46: Avoid TypeError in GeneratorContextManager on process shutdown.

  • Issue #43: Let DataWatch return node data if allow_missing_node is used.

0.9 (2013-01-07)

API Changes

  • When a retry operation ultimately fails, it now raises a kazoo.retry.RetryFailedError exception, instead of a general Exception instance. RetryFailedError also inherits from the base KazooException.


  • Improvements to Debian packaging rules.

Bug Handling

  • Issue #39 / #41: Handle connection dropped errors during session writes. Ensure client connection is re-established to a new ZK node if available.

  • Issue #38: Set CLOEXEC flag on all sockets when available.

  • Issue #37 / #40: Handle timeout errors during select calls on sockets.

  • Issue #36: Correctly set ConnectionHandler.writer_stopped even if an exception is raised inside the writer, like a retry operation failing.

0.8 (2012-10-26)

API Changes

  • The KazooClient.__init__ took as watcher argument as its second keyword argument. The argument had no effect anymore since version 0.5 and was removed.

Bug Handling

  • Issue #35: KazooClient.__init__ didn’t pass on retry_max_delay to the retry helper.

  • Issue #34: Be more careful while handling socket connection errors.

0.7 (2012-10-15)


  • DataWatch now has a allow_missing_node setting that allows a watch to be set on a node that doesn’t exist when the DataWatch is created.

  • Add new Queue recipe, with optional priority support.

  • Add new Counter recipe.

  • Added debian packaging rules.

Bug Handling

  • Issue #31 fixed: Only catch KazooExceptions in catch-all calls.

  • Issue #15 fixed again: Force sleep delay to be a float to appease gevent.

  • Issue #29 fixed: DataWatch and ChildrenWatch properly re-register their watches on server disconnect.

0.6 (2012-09-27)

API Changes

  • Node paths are assumed to be Unicode objects. Under Python 2 pure-ascii strings will also be accepted. Node values are considered bytes. The byte type is an alias for str under Python 2.

  • New KeeperState.CONNECTED_RO state for Zookeeper servers connected in read-only mode.

  • New NotReadOnlyCallError exception when issuing a write change against a server thats currently read-only.


  • Add support for Python 3.2, 3.3 and PyPy (only for the threading handler).

  • Handles connecting to Zookeeper 3.4+ read-only servers.

  • Automatic background scanning for a Read/Write server when connected to a server in read-only mode.

  • Add new Semaphore recipe.

  • Add a new retry_max_delay argument to the client and by default limit the retry delay to at most an hour regardless of exponential backoff settings.

  • Add new randomize_hosts argument to KazooClient, allowing one to disable host randomization.

Bug Handling

  • Fix bug with locks not handling intermediary lock contenders disappearing.

  • Fix bug with set_data type check failing to catch unicode values.

  • Fix bug with gevent 0.13.x backport of peekable queue.

  • Fix PatientChildrenWatch to use handler specific sleep function.

0.5 (2012-09-06)

Skipping a version to reflect the magnitude of the change. Kazoo is now a pure Python client with no C bindings. This release should run without a problem on alternate Python implementations such as PyPy and Jython. Porting to Python 3 in the future should also be much easier.


  • Docs have been restructured to handle the new classes and locations of the methods from the pure Python refactor.

Bug Handling

This change may introduce new bugs, however there is no longer the possibility of a complete Python segfault due to errors in the C library and/or the C binding.

  • Possible segfaults from the C lib are gone.

  • Password mangling due to the C lib is gone.

  • The party recipes didn’t set their participating flag to False after leaving.


  • New client.command and client.server_version API, exposing Zookeeper’s four letter commands and giving access to structured version information.

  • Added ‘include_data’ option for get_children to include the node’s Stat object.

  • Substantial increase in logging data with debug mode. All correspondence with the Zookeeper server can now be seen to help in debugging.

API Changes

  • The testing helpers have been moved from testing.__init__ into a testing.harness module. The official API’s of KazooTestCase and KazooTestHarness can still be directly imported from testing.

  • The kazoo.handlers.util module was removed.

  • Backwards compatible exception class aliases are provided for now in kazoo exceptions for the prior C exception names.

  • Unicode strings now work fine for node names and are properly converted to and from unicode objects.

  • The data value argument for the create and create_async methods of the client was made optional and defaults to an empty byte string. The data value must be a byte string. Unicode values are no longer allowed and will raise a TypeError.

0.3 (2012-08-23)

API Changes

  • Handler interface now has an rlock_object for use by recipes.

Bug Handling

  • Fixed password bug with updated zc-zookeeper-static release, which retains null bytes in the password properly.

  • Fixed reconnect hammering, so that the reconnection follows retry jitter and retry backoff’s.

  • Fixed possible bug with using a threading.Condition in the set partitioner. Set partitioner uses new rlock_object handler API to get an appropriate RLock for gevent.

  • Issue #17 fixed: Wrap timeout exceptions with staticmethod so they can be used directly as intended. Patch by Bob Van Zant.

  • Fixed bug with client reconnection looping indefinitely using an expired session id.

0.2 (2012-08-12)


  • Fixed doc references to start_async using an AsyncResult object, it uses an Event object.

Bug Handling

  • Issue #16 fixed: gevent zookeeper logging failed to handle a monkey patched logging setup. Logging is now setup such that a greenlet is used for logging messages under gevent, and the thread one is used otherwise.

  • Fixed bug similar to #14 for ChildrenWatch on the session listener.

  • Issue #14 fixed: DataWatch had inconsistent handling of the node it was watching not existing. DataWatch also properly spawns its _get_data function to avoid blocking session events.

  • Issue #15 fixed: sleep_func for SequentialGeventHandler was not set on the class appropriately leading to additional arguments being passed to gevent.sleep.

  • Issue #9 fixed: Threads/greenlets didn’t gracefully shut down. Handler now has a start/stop that is used by the client when calling start and stop that shuts down the handler workers. This addresses errors and warnings that could be emitted upon process shutdown regarding a clean exit of the workers.

  • Issue #12 fixed: gevent 0.13 doesn’t use the same start_new_thread as gevent 1.0 which resulted in a fully monkey-patched environment halting due to the wrong thread. Updated to use the older kazoo method of getting the real thread module object.

API Changes

  • The KazooClient handler is now officially exposed as KazooClient.handler so that the appropriate sync objects can be used by end-users.

  • Refactored ChildrenWatcher used by SetPartitioner into a publicly exposed PatientChildrenWatch under recipe.watchers.


  • connect/connect_async has been renamed to start/start_async to better match the stop to indicate connection handling. The prior names are aliased for the time being.


  • Added Barrier and DoubleBarrier implementation.

0.2b1 (2012-07-27)

Bug Handling

  • ZOOKEEPER-1318: SystemError is caught and rethrown as the proper invalid state exception in older zookeeper python bindings where this issue is still valid.

  • ZOOKEEPER-1431: Install the latest zc-zookeeper-static library or use the packaged ubuntu one for ubuntu 12.04 or later.

  • ZOOKEEPER-553: State handling isn’t checked via this method, we track it in a simpler manner with the watcher to ensure we know the right state.


  • Exponential backoff with jitter for retrying commands.

  • Gevent 0.13 and 1.0b support.

  • Lock, Party, SetPartitioner, and Election recipe implementations.

  • Data and Children watching API’s.

  • State transition handling with listener registering to handle session state changes (choose to fatal the app on session expiration, etc.)

  • Zookeeper logging stream redirected into Python logging channel under the name ‘Zookeeper’.

  • Base client library with handler support for threading and gevent async environments.

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distributions

No source distribution files available for this release.See tutorial on generating distribution archives.

Built Distribution

kedo_kazoo- (122.9 kB view hashes)

Uploaded py2 py3

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page