A newer version of this software is available

You are viewing the documentation for an older version of this software. To find the documentation for the current version, visit the Couchbase documentation home page.

This guide provides information for developers who want to use the Couchbase Python SDK to build applications that use Couchbase Server.

Getting Started

The following sections demonstrate how to get started using Couchbase with the Python SDK. We’ll first show how to install the SDK and then demonstrate how it can be used to perform some simple operations.

Download and Installation

Install the following packages to get started using the Python SDK:

  1. Download, install, and start Couchbase server. Come back here when you are done.

  2. Download and install the C library. If you are using Microsoft Windows, you can skip this step because starting with version 1.0 Beta, the C library is bundled with the Python SDK.

    If you already have the C SDK installed, ensure the installed version is at least 2.1.0. To determine the version, either query your package version, or run the cbc command:

    shell> cbc version
    cbc built from: libcouchbase 2.1.3_14_g7684ab7 (rev. 7684ab7832628d520cd2803f3a3d0822e993c3f3)
    using libcouchbase: 2.1.3 (libevent)
    Note that depending on how your libcouchbase was installed, the cbc command may not be available, and is not required for the Python SDK.
  3. Check your Python version. It must be at least version 2.6 (Python version 3.x is supported as well). To check your python version:

    shell> python -V python 2.6.6

  4. Install the Python development files.

    • For EL5-based RPM distributions (e.g. CentOS 5, RHEL 5), do yum install python26-devel

    • For EL6-based RPM distributions (e.g. CentOS 6, RHEL 6), do yum install python-devel

    • For Debian-based systems (including Ubuntu), do apt-get install python-dev

    • For OS X and Windows systems, Python development files are already included in the Python distribution.

  5. Install the Python SDK. The easiest way to do this is via the pip tool:

    shell> pip install couchbase --quiet If all went well, you should not see any errors printed to the screen.

    Alternatively, you can manually download one of the packages at PyPi.

  6. Verify that your Python SDK is available and working by entering the following command:

    shell> python -c 'import couchbase' If this does not print any errors or exceptions, your Python SDK is properly installed!

Hello Couchbase

To follow the tradition of programming tutorials, we’ll start with “Hello Couchbase.” This example works with the “beer-sample” bucket that is provided with the default install.

from couchbase import Couchbase
from couchbase.exceptions import CouchbaseError

c = Couchbase.connect(bucket='beer-sample', host='localhost')

    beer = c.get("aass_brewery-juleol")

except CouchbaseError as e:
    print "Couldn't retrieve value for key", e
    # Rethrow the exception, making the application exit

doc = beer.value

# Because Python 2.x will complain if an ASCII format string is used
# with Unicode format values, we make the format string unicode as well.

print unicode("{name}, ABV: {abv}").format(name=doc['name'], abv=doc['abv'])

doc['comment'] = "Random beer from Norway"

    result = c.replace("aass_brewery-juleol", doc)
    print result

except CouchbaseError as e:
    print "Couldn't replace key"

The following points explain each step in the example:

  • Connecting

    The Couchbase.connect class method constructs a new couchbase.connection.Connection object. This object represents a connection to a single bucket within the cluster. Arguments passed to connect are passed to the constructor (see API documentation on the Connection object for more details and options).

    A bucket represents a logical namespace for a key. All keys must be unique within a single bucket, but multiple buckets can have keys with the same names and they will not conflict. A new connection object must be created for each bucket that you want to interact with in your application. This example creates one connection to the beer-sample bucket.

    The constructor is passed the bucket name, which is beer-sample, and a node on the cluster to connect to. You can pass any node that is a member of the cluster. This example uses the local cluster instance.

  • Retrieving Data

    The get method retrieves the value for the key requested. If the key exists, a Result object that contains the value of the key as well as additional metadata is returned. To get the actual value of the object, you can access the Result object’s value property.

    If the key does not exist on the server, an exception of type CouchbaseError is thrown. This exception object can be caught and examined or printed to see more details about why the operation failed. See the API documentation for more details.

    We treat the value as a dict object. As a documented oriented database, values stored to the server are considered to be JSON by default, and when retrieved from the server are interpreted to be JSON (and unserialized into a Python dict ). It is possible to use other formats than the default JSON, however. The set methods accept a format keyword argument that indicates the conversion type to be used. The default is couchbase.FMT_JSON, but you can also use couchbase.FMT_BYTES, couchbase.FMT_UTF8, or couchbase.FMT_PICKLE instead. If none of these are sufficient, you can write your own custom Transcoder object to handle conversion on your own.

    Starting from version 1.1.0, you may also use the couchbase.FMT_AUTO flag which will guess the suitable format based on the Python datatype being passed.

  • Storing Data

    To store documents in the server, you can use one of the set family of methods. Here we use replace, which enforces the constraint that a previous value of the document must already exist. This can be thought of as an update operation in terms of CRUD (create, read, update, delete).

    The storage methods also return a Result object that contains metadata about the value stored.

Now we’re ready to run our first Couchbase Program:

The first line outputs the name field of the document, and the second line outputs the Result object of the replace operation.

Working With Documents

A document in Couchbase server consists of a key, value, and metadata.

  • Key

    A key is a unique identifier for your data. Each document must have a unique key. The key can be any valid Unicode string.

  • Value

    The value is your own application data that exists under the key. The format of the value can be anything. By default, only JSON-serializable objects are supported — Python str, unicode, dict, list, tuple, int, long, float, bool, and None types — in short, anything that the standard json.dumps accepts. The reason JSON is the default format is for the ability to later query the database based on value contents, as will be explained later.

    You can also store arbitrary Python objects using the FMT_PICKLE value for the format option.

  • Metadata

    The metadata contains information concerning the format of the value that is, whether it’s JSON, Pickle, or something else. It also contains revision information such as the CAS, which we’ll read about later.

You can store documents by providing the unique key under which the document will be stored, and the value which contains the actual document. You can retrieve documents either by directly specifying the unique key under which the document was stored or by querying views that retrieve documents based on specific criteria.

Storing Documents

This section provides a bit more insight on how to store documents. This is a prerequisite to demonstrate how to retrieve documents because there must be something to retrieve.

There are additional storage methods beyond those described here, which are covered in the Advanced section, see Advanced Usage. These include manipulating numeric counters, setting expiration times for documents, and appending/prepending to existing values.

The Connection object provides the following store operations, which conform to the CRUD model:

  • set(key, value)

    Stores the document value under the key. If the key did not previously exist, it is created. If the key already exists, its existing value is overwritten with the new contents of value.

  • add(key, value)

    Stores the document value under the key, but only if key does not already exist. If key already exists, an exception is thrown.

  • replace(key, value)

    Replace is the inverse of add. It sets the contents of key to value, but only if the key already exists. If the key does not already exist, an exception is thrown.

  • delete(key)

    Deletes the key from the bucket. Future attempts to access this key via get raise an exception until something is stored again for this key using one of the set methods.

CRUD Example

The following code demonstrates the store operations.

from couchbase import Couchbase
from couchbase.exceptions import CouchbaseError

key = "demo_key"
value = "demo_value"

# We use the 'default' bucket.
c = Couchbase.connect(bucket='default', host='localhost')

print "Setting key {0} with value {1}".format(key, value)
result = c.set(key, value)
print "...", result

print ""
print "Getting value for key {0}".format(key)
result = c.get(key)
print "...", result

print ""
print "Creating new key {0} with value 'new_value'".format(key)
print "This will fail as '{0}' already exists".format(key)
    c.add(key, "another value")
except CouchbaseError as e:
    print e

print "Replacing existing key {0} with new value".format(key)
result = c.replace(key, "new value")
print "...", "result"

print ""
print "Getting new value for key {0}".format(key)
result = c.get(key)
print "...", result

print ""
print "Deleting key", key
result = c.delete(key)
print "...", result

print ""
print "Getting value for key {0}. This will fail as it has been deleted".format(key)
except CouchbaseError as e:
    print e

print ""
print "Creating new key {0} with value 'added_value'".format(key)
result = c.add(key, "added_value")
print "...", result

print "Getting the new value"
result = c.get(key)
print "...", result


Setting key demo_key with value demo_value
... OperationResult<RC=0x0, Key=demo_key, CAS=0x2fbbf239b45e0400>

Getting value for key demo_key
... ValueResult<RC=0x0, Key=demo_key, Value=u'demo_value', CAS=0x2fbbf239b45e0400, Flags=0x0>

Creating new key demo_key with value 'new_value'
This will fail as 'demo_key' already exists
<Key=u'demo_key', RC=0xC[Key exists (with a different CAS value)], Operational Error, Results=1, C Source=(src/multiresult.c,148)>
Replacing existing key demo_key with new value
... result

Getting new value for key demo_key
... ValueResult<RC=0x0, Key=demo_key, Value=u'new value', CAS=0xa047003ab45e0400, Flags=0x0>

Deleting key demo_key
... OperationResult<RC=0x0, Key=demo_key, CAS=0xa147003ab45e0400>

Getting value for key demo_key. This will fail as it has been deleted
<Key=u'demo_key', RC=0xD[No such key], Operational Error, Results=1, C Source=(src/multiresult.c,148)>

Creating new key demo_key with value 'added_value'
... OperationResult<RC=0x0, Key=demo_key, CAS=0xe91f0c3ab45e0400>
Getting the new value
... ValueResult<RC=0x0, Key=demo_key, Value=u'added_value', CAS=0xe91f0c3ab45e0400, Flags=0x0>

Getting Documents By Key

Couchbase provides two ways to fetch your documents: you can retrieve a document by its key, or you can retrieve a set of documents that match some constraint by using Views. Because views are more complex, we’ll first demonstrate getting documents by their keys.

To get a single document, simply supply the key as the first argument to the get method. It returns a Result object on success that can then be used to extract the value.

Getting A Single Document"my list", [])
result = client.get("my list")
doc = result.value

To get multiple documents, you can use the more efficient get_multi method. You pass it an iterable sequence of keys, and it returns a MultiResult object with the keys passed to get_multi as keys, and the values being a Result object for the result of each key. MultiResult is a subclass of dict.

Getting Multiple Documents

    'sheep_counting' : ['first sheep', 'second sheep'],
    'famous_sheep' : {'sherry lewis' : 'Lamb Chop'}

keys = ('sheep_counting', 'famous_sheep')
results = client.get_multi(keys)
for key, result in results.items():
    doc = result.value

Error Handling

If a document does not exist, a couchbase.exceptions.NotFoundError (which is a subclass of couchbase.exceptions.CouchbaseError ) is thrown.

You can change this behavior by using the quiet keyword parameter and setting it to true (to suppress exceptions for a specific get call) or by setting the Connection.quiet property on the Connection object (which suppresses exceptions on get for subsequent calls).

When using quiet, you can still determine if a key was retrieved successfully by examining the success property of the value object.

Passing quiet to get

result = client.get("non-exist-key", quiet=True)
if result.success:
    print "Got document OK"
    print ("Couldn't retrieve document. "
           "Result was received with code"), result.rc

Setting quiet in the constructor

client = Couchbase.connect(bucket='default', quiet=True)
result = client.get("non-exist-key")
if result.success:
    print "Got document OK"
    print "Couldn't retrieve document"

The rc property of the Result object contains the error code received on failure (on success, its value is 0 ). You can also obtain the exception class which would have been thrown by using

>>> CouchbaseError.rc_to_exctype(result.rc)
<class 'couchbase.exceptions.NotFoundError'>

This class method is passed an error code and produces the appropriate exception class.

When using get_multi with the quiet option enabled, you can immediately determine whether all the keys were fetched successfully by examining the all_ok property of the returned MultiResult object.

results = client.get_multi(("i exist", "but i don't"), quiet=True)
if not results.all_ok:
    print "Couldn't get all keys"

Getting Documents by Querying Views

In addition to fetching documents by keys, you can also employ Views to retrieve information using secondary indexes. This guide gets you started on how to use them from the Python SDK. If you want to learn more about views, see the chapter in the Couchbase Server 2.0 documentation.

First, create your view definition using by the web UI (though you may also do this directly from the Python SDK, as will be shown later).

You can then query the view results by calling the query method on the Connection object. Simply pass it the design and view name.

view_results = client.query("beer", "brewery_beers")
for result in view_results:
    print "Mapped key: %r" % (result.key,)
    print "Emitted value: %r" % (result.value,)
    print "Document ID: %s" % (result.docid,)

The query method returns a couchbase.views.iterator.View object, which is an iterator. You can simply iterate over it to retrieve the results for the query. Each object yielded is a ViewRow, which is a simple object containing the key, value, document ID, and optionally the document itself for each of the results returned by the view.

In addition to the design and view name, the query method accepts additional keyword arguments that control the behavior of the results returned. You can thus use it as follows:

results = client.query("beer", "brewery_beers", opt1=value1, opt2=value2, ...)
for result in results:
    # do something with result..

Here are some of the available parameters for the query method. A full listing can be found in the API documentation.

  • include_docs

    This boolean parameter indicates whether the corresponding document should be retrieved for each row fetched. If this is true, the doc property of the ViewRow object yielded by the iterator returned by query contains a Result object that contains the document for the key.

  • reduce

    This boolean parameter indicates whether the server should also pass the results to the view’s reduce function. An exception is raised if the view does not have a reduce method defined.

  • limit

    This numeric parameter indicates the maximum amount of results to fetch from the query. This parameter is handy if your query can produce a lot of results.

  • descending

    This boolean parameter indicates that the results should be returned in reverse order.

  • stale

    This boolean parameter controls the tradeoff between performance and freshness of data.

  • debug

    This boolean parameter fetches low-level debugging information from the view engine.

  • streaming

    This boolean parameter indicates whether the view results should be decoded in a streaming manner. When enabled, the iterator internally fetches chunks of the response as required.

    As this is less efficient than fetching all results at once, it is disabled by default, but can be very useful if you have a large dataset because it prevents the entire view from being buffered in memory.

results = client.query("beer", "brewery_beers",
                       include_docs=True, limit=5)

for result in results:
    print "key is %r" % (result.key)
    doc = result.doc.value
    if doc['type'] == "beer":
        print "Got a beer. It's got %0.2f ABV" % (doc['abv'],)

Encoding and Serialization

The default encoding format for the Python SDK is JSON. This means you can pass any valid object that is accepted by the standard json.dumps library function and you will receive it back when you retrieve it.

# -*- coding: utf-8 -*-

import pprint
from couchbase import Couchbase

client = Couchbase.connect(bucket='default', host='localhost')
value = {
    "this is a" : "dictionary",
    "and this is a list" : ["with", "some", "elements"],
    "and this is a tuple" : ("with", "more", "elements"),
    "you can also use floats" : 3.14,
    "integers" : 42,
    "strings" : "hello",
    "unicode" : "שלום!",
    "blobs" : "\x00",
    "or a None" : None

client.set("a_key", value)
result = client.get("a_key")
print result.value['unicode']

Which then prints

{u'and this is a list': [u'with', u'some', u'elements'],
 u'and this is a tuple': [u'with', u'more', u'elements'],
 u'blobs': u'\x00',
 u'integers': 42,
 u'or a None': None,
 u'strings': u'hello',
 u'this is a': u'dictionary',
 u'unicode': u'\u05e9\u05dc\u05d5\u05dd!',
 u'you can also use floats': 3.14}

To view the document you just created, go to localhost:8091 in your browser, type in your administrative credentials, go over to the Data Buckets pane, click on the Documents button for the default bucket, and type in the ID for the document (in this case, it’s a_key )). The document can now be indexed and queried against using views.

Other Formats

While JSON is the default format, you might want to use other formats. For example, if you want to store complex custom Python objects and classes and don’t require that they be indexed with views, you can use the pickle serialization format. It allows you to store types that are not accepted by JSON:

import pprint

from couchbase import Couchbase, FMT_PICKLE

c = Couchbase.connect(bucket='default')
c.set("a_python_object", object(), format=FMT_PICKLE)
c.set("a_python_set", set([1,2,3]), format=FMT_PICKLE)



<object object at 0x7fa7d0ad80e0> set([1, 2, 3]) You can also store arbitrary strings of bytes by using FMT_BYTES.

In Python 2.6 and above bytes and str are the same type; however in Python 3, a str is a string with an encoding (i.e. Python 2’s unicode ) while bytes is a sequence of bytes which must be explicitly converted in order to be used with text operations.

import pprint

from couchbase import Couchbase, FMT_BYTES

c = Couchbase.connect(bucket='default')
c.set("blob", b"\x01\x02\x03\x04", format=FMT_BYTES)


b'\x01\x02\x03\x04' Alternatively, you can use FMT_UTF8 to store a unicode object represented as UTF-8.

While JSON is also capable of storing strings and Unicode, the JSON specification mandates that all strings begin and end with a quote ( " ). This uses additional space and costs extra processing power to decode and encode your JSON string. Therefore,you can save on performance by using FMT_UTF8 for simple strings.

It is possible to encode your data by using encodings other than UTF-8. However, because the view engine operates using UTF-8, we selected this as the default. If you need a different encoding, consider using the Transcoder interface.

from couchbase import Couchbase, FMT_UTF8

c = Couchbase.connect(bucket='default')
c.set("EXCALIBUR", u"\u03EE", format=FMT_UTF8)
print c.get("EXCALIBUR")


ValueResult<RC=0x0, Key=EXCALIBUR, Value=u'\u03ee', CAS=0x6552956a925e0400, Flags=0x4>

Setting The Default Format

You can set the default format for the value type you use most by setting the default_format property on the connection object, either during construction or afterwards:

c = Couchbase.connect(bucket='default', default_format=FMT_UTF8)


c.default_format = FMT_PICKLE


In this chapter we build on the foundations of the Getting Started guide and build a simple web application. Make sure you have the beer-sample bucket installed, because we’ll be using it. The sample application will allow you to edit and manage various beers and breweries.

The sample application is not entirely complete, and there are some features that remain to be implemented. Implementing them is an exercise for the reader.

The full source code for the sample application is available through couchbaselabs on GitHub.

Note that the sample application provides more content than described in this tutorial, but it should be simple to navigate while reading this tutorial.


  • Ensure you have Flask installed. You can either install it via your distribution or use pip install Flask.

  • Download Couchbase Server and install it. Make sure to install the beer-sample dataset when you run the wizard because this tutorial application works with it.

  • Clone the repository and cd into the directory:

    shell> git clone git://
    Cloning into ‘beersample-python’‘
    shell> cd beersample-python
  • Some views need to be set up. You can set up the views manually via the Web UI, or invoke the script located in the beersample-python directory.

    In the beer design document, create a view called by_name :

    function (doc, meta) {
         if (doc.type && doc.type == "beer") {
             emit(, null);

    Create a design document called brewery and add a view called by_name :

    function (doc, meta) {
          if (doc.type && doc.type == "brewery") {
              emit(, null);
  • Invoke the script: 

    shell> python
    * Running on
    * Restarting with reloader
  • Navigate to localhost:5000 and enjoy the application!


In this section we’ll talk a bit about setting up your directory layout and adding some views in the server before we start dealing with the Python SDK and Flask itself.

Project Setup

Create a project directory named beer :

shell> mkdir beer
shell> cd beer
shell> mkdir templates
shell> mkdir templates/beer
shell> mkdir templates/brewery
shell> mkdir static
shell> mkdir static/js
shell> mkdir static/css

Showing your directory contents displays something like this:

shell> find. -type d

To make the application look pretty, we’re incorporating jQuery and Twitter Bootstrap. You can either download the libraries and put them in their appropriate css and js directories (under static ), or clone the project repository and use it from there. If you followed the Quickstart steps, you already have the files in your beersample-python directory. Either way, make sure you have the following files in place:

  • static/css/beersample.css

  • static/css/bootstrap.min.css (the minified twitter bootstrap library)

  • static/css/bootstrap-responsive.min.css (the minified responsive layout classes from bootstrap)

  • static/js/beersample.js

  • static/js/jquery.min.js (the jQuery javascript library)

From here on, you should have a bare bones web application configured that has all the dependencies included. We’ll now move on and configure the beer-sample bucket the way we need it.

Preparing the Views

The beer-sample bucket comes with a small set of predefined views, but to make our application function correctly we need some more. This is also a good chance to explore the view management possibilities inside the Web-UI.

Because we want to list beers and breweries by name, we need to define one view for each. Head over to the Web-UI and click on the Views menu. Select beer-sample from the drop-down list to switch to the correct bucket. Now click on Development Views and then Create Development View to define your first view. You need to give it the name of both the design document and the actual view. Insert the following names:

  • Design Document Name: _design/dev_beer

  • View Name: by_name

The next step is to define the map and (optional) reduce functions. In our examples, we won’t use the reduce functions at all but you can play around and see what happens. Insert the following map function (that’s JavaScript) and click Save.

function (doc, meta) {
  if(doc.type && doc.type == "beer") {
    emit(, null);

Every map function takes the full document ( doc ) and its associated metadata ( meta ) as the arguments. You are then free to inspect this data and emit a result when you want to have it in your index. In our case, we emit the name of the beer ( ) when the document both has a type field and the type is beer. We don’t need to emit a value — that’s why we are using null here. It’s always advisable to keep the index as small as possible. Resist the urge to include the full document through emit(, doc), because it will increase the size of your view indexes. If you need to access the full document (or large parts), then use include_docs in the query method, which returns ViewRow objects together with their documents. You can also call cb.get(row.docid) to get the individual doc for a single row. The resulting retrieval of the document might be slightly out of sync with your view, but it will be fast and efficient.

Now we need to define a view for our breweries. You already know how to do this — here is all the information you need to create a brewery view:

  • Design Document Name: _design/dev_brewery

  • View Name: by_name

  • Map Function:

    function (doc, meta) {
       if(doc.type && doc.type == "brewery") {
         emit(, null);

The final step is to push the design documents in production. While the design documents are in development, the index is applied only on the local node. Because we want to have the index on the whole dataset, click the Publish button on both design documents (and accept any pop-up windows that warn you about overriding the old design documents).

For more information about using views for indexing and querying from Couchbase Server, see the following helpful resources in the Couchbase Server Manual :

Structure of the Flask Application

We’ll be showing bits and pieces of the web app as it pertains to specific sections. The entire file is less than 300 lines long, and you can inspect it by looking into the file in the repository.

First, our imports. We need some extra imports to be able to handle exceptions properly and let us build better view queries. (imports)

from collections import namedtuple
import json

from Flask import Flask, request, redirect, abort, render_template

from couchbase import Couchbase
from couchbase.exceptions import KeyExistsError, NotFoundError
from couchbase.views.iterator import RowProcessor
from couchbase.views.params import UNSPEC, Query

Then, we want to set some constants for our application: (configuration)

DATABASE = 'beer-sample' HOST = 'localhost' ENTRIES_PER_PAGE = 30

The ENTRIES_PER_PAGE variable is used later on to configure how many beers and breweries to show in the search results.

Now, we’re ready to create our Flask application instance: (creating the application)

app = Flask(__name__, static_url_path='')

The first line creates a new Flask application. The first argument is the module in which the application is defined. Because we’re using only a single file as our application, we can use __name__, which expands to the name of the current file being executed (minus the .py suffix).

The second argument instructs Flask to treat unrouted URLs as being requests for files located in the static directory we created earlier. This allows our templates to load the required .js and .css files.

The second line creates a configuration object for our app. The argument is the name of the module to scan for configuration directives. Flask scans this module for variable names in UPPER_CASE and places them in the app.config dictionary.

Then, define a function to give us a database connection: (generating a Connection object)

def connect_db():
    return Couchbase.connect(

db = connect_db()

You already know how to connect to a Couchbase cluster, so we’ll skip the explanation here.

The module-level db variable is set to be the Connection object. In larger applications this is not a good idea, but we can get away with it here because this is a simple app.

The Welcome Page

The first route we will implement is that of the welcome page, that is, the page that is displayed when someone goes to the root of your site. Because there is no Couchbase interaction involved, we just tell Flask to render the template. (welcome page)

def welcome():
    return render_template('welcome.html')

app.add_url_rule('/welcome', view_func=welcome)

The welcome.html template is actually a Jinja template inside the templates directory. It looks like this:


{% extends "layout.html" %}
{% block body %}
<div class="span6">
  <div class="span12">
    <h4>Browse all Beers</h4>
    <a href="/beers" class="btn btn-warning">Show me all beers</a>
    <hr />
  <div class="span12">
    <h4>Browse all Breweries</h4>
    <a href="/breweries" class="btn btn-info">Take me to the breweries</a>
<div class="span6">
<div class="span6">
  <div class="span12">
    <h4>About this App</h4>
    <p>Welcome to Couchbase!</p>
    <p>This application helps you to get started on application
        development with Couchbase. It shows how to create, update and
        delete documents and how to work with JSON documents.</p>

{% endblock %}

The template simply provides some links to the brewery and beer pages (which are shown later).

An interesting thing about this template is that it “inherits” from the common layout.html template. All pages in the beer app have a common header and footer to them — with only their body differing. This is the layout.html template.



<html lang="en">
    <meta charset="utf-8">
    <title>Couchbase Python Beer Sample</title>
    <meta name="viewport" content="width=device-width, initial-scale=1.0">
    <meta name="description" content="The Couchbase Java Beer-Sample App">
    <meta name="author" content="Couchbase, Inc. 2013">

    <link href="/css/bootstrap.min.css" rel="stylesheet">
    <link href="/css/beersample.css" rel="stylesheet">
    <link href="/css/bootstrap-responsive.min.css" rel="stylesheet">

    <!-- HTML5 shim, for IE6-8 support of HTML5 elements -->
    <!--[if lt IE 9]>
      <script src=""></script>

    <div class="container-narrow">
      <div class="masthead">
        <ul class="nav nav-pills pull-right">
          <li><a href="/welcome">Home</a></li>
          <li><a href="/beers">Beers</a></li>
          <li><a href="/breweries">Breweries</a></li>
        <h2 class="muted">Couchbase Beer Sample</h2>
      <div class="row-fluid">
        <div class="span12">
            {% block body %}{% endblock %}
      <div class="footer">
        <p>&copy; Couchbase, Inc. 2013</p>
    <script src="/js/jquery.min.js"></script>
    <script src="/js/bootstrap.min.js"></script>
    <script src="/js/beersample.js"></script>

If you start your app now, you should be able to navigate to localhost:5000 and see the welcome page. You’ll get a 404 error if you try to visit any links though - this is because we haven’t implemented them yet. Let’s do that now!

Managing Beers

In this section we’ll show the construction of the web app with respect to managing beers. We’ll be able to list, inspect, edit, create, search, and delete beers.

Showing Beers

Now we’re finally getting into the cooler stuff of this tutorial. First, we’ll implement several classes for our pages to use. (custom Beer row class and processing)

class Beer(object):
    def __init__(self, id, name, doc=None): = id = name
        self.brewery = None
        self.doc = doc

    def __getattr__(self, name):
        if not self.doc:
            return ""
        return self.doc.get(name, "")

class BeerListRowProcessor(object):
    This is the row processor for listing all beers (with their brewery IDs).
    def handle_rows(self, rows, connection, include_docs):
        ret = []
        by_docids = {}

        for r in rows:
            b = Beer(r['id'], r['key'])
            by_docids[] = b

        keys_to_fetch = [ for x in ret ]
        docs = connection.get_multi(keys_to_fetch, quiet=True)

        for beer_id, doc in docs.items():
            if not doc.success:

            beer = by_docids[beer_id]
            beer.brewery_id = doc.value['brewery_id']

        return ret

First, we declare a simple Beer object. This app isn’t too fancy and we could’ve just used a simple dict. However, it allows us to demonstrate the use of the RowProcessor interface.

In the beer listing page, we want to display each beer along with a link to the brewery that produces it. However, we’ve defined the beer/by_name view to return only the name of the beer. To obtain the brewery, we need to fetch each beer document and examine it. The document contains the Brewery ID that we need later.

The BeerListRowProcessor is an implementation of the RowProcessor interface that operates on the returned view rows.

For each raw JSON row, it creates a new Beer object. The first argument is the document ID, which is used to provide a link to display more information about the beer. The second argument is the name of the beer itself, which we use in the beer list on the webpage.

We also create a local variable called by_docids that allows us to get a Beer object by its document ID.

After we’ve created all the beers, we create a list of document IDs to fetch by using list comprehension. We pass this list to get_multi (passing quiet=True, because there might be some inconsistencies between the view indexes and the actual documents).

While we could have made this simpler by performing an individual get on each, that would be less efficient in terms of network usage.

Now that we have the beer documents, it’s time to set each beer’s brewery_id to its relevant value.

We first check to see that each document was successful in being retrieved; then we look up the corresponding Beer object by getting it from the by_docids dictionary using the beer_id as the key.

Then, we extract the brewery_id field from the document and place it into the Beer object.

Finally, we return the list of populated beers. The View object (returned by the query function) now yields results as we iterate over it.

Before we forget, let’s put this all together: (showing beer listings)

def beers():
    rp = BeerListRowProcessor()
    rows = db.query("beer", "by_name",

    return render_template('beer/index.html', results=rows)

We tell Flask to route requests to /beers to this function. We create an instance of the BeerListRowProcessor function we just defined.

We then execute a view query using the query method, passing it the name of the design and view ( beer and by_name, respsectively).

We set the limit directive to the aforementioned ENTRIES_PER_PAGE directive, to avoid flooding a single webpage with many results.

We finally tell the query method to use our own BeerListRowProcessor for processing the results.

We then direct the template engine to render the beer/index.html template, setting the template variable rows to the iterable returned by the query function.

Here is the beer/index.html template:


{% extends "layout.html" %}
{% block body %}

<h3>Browse Beers</h3>
<form class="navbar-search pull-left">
    <input id="beer-search" type="text" class="search-query" placeholder="Search for Beers">

<table id="beer-table" class="table table-striped">
        {% for beer in results %}
            <td><a href="/beers/show/{{}}">{{}}</a></td>
            <td><a href="/breweries/show/{{beer.brewery_id}}">To Brewery</a></td>
                <a class="btn btn-small btn-warning" href="/beers/edit/{{}}">Edit</a>
                <a class="btn btn-small btn-danger" href="/beers/delete/{{}}">Delete</a>
        {% endfor %}

    <a class="btn btn-small btn-success" href="/beers/create">Add Beer</a>

{% endblock %}

We’re using Jinja {% for %} blocks to iterate and emit a fragment of HTML for each Beer object returned by the query.

Navigate to localhost:5000/beers, to see a listing of beers. Each beer has To Brewery, Edit, and Delete buttons.

On the bottom of the page, you can also see an Add Beer button, which allows you to define new beers.

Let’s implement the Delete button next!

Deleting Beers

Due to the simplicity of Couchbase and Flask, we can implement a single method to delete both beers and breweries. (deleting a beer)

def delete_object(otype, id):
        return redirect('/welcome')

    except NotFoundError:
        return "No such {0} '{1}'".format(otype, id), 404

Here we tell Flask to route any URL that has as its second component the string delete to this method. The paths in <angle brackets> are routing tokens that Flask passes to the handler as arguments. This means that URLs such as /beers/delete/foobar and /foo/delete/whatever are all routed here.

When we get an ID, we try to delete it by using the delete method in a try block. If successful, we redirect to the welcome page, but if the key does not exist, we return with an error message and a 404 status code.

You can now access this page by going to localhost:5000/beers/delete/nonexistent and get a 404 error. Or you can delete a beer by clicking on one of the Delete buttons in the /beers page!

If you find that a beer is still displayed after you click the delete button, you can refresh the browser page to verify that the beer has been deleted.

Another way to verify that a beer has been deleted is by clicking the delete button again and getting a 404 error.

Displaying Beers

Here we demonstrate how you can display the beers. In this case, we display a page showing all the fields and values of a given beer. (showing a single beer)

def show_beer(beer_id):
    doc = db.get(beer_id, quiet=True)
    if not doc.success:
        return "No such beer {0}".format(beer_id), 404

    return render_template(
        beer=Beer(beer_id, doc.value['name'], doc.value))

Like for the delete action, we first check to see that the beer exists. We are passed the beer ID as the last part of the URL - this is passed to us as the beer_id.

In order to display the information for the given beer ID, we simply call the connection’s get method with the beer_id argument. We also pass the quiet parameter so that we don’t receive an exception if the beer does not exist.

We then check to see that the success property of the returned Result object is true. If it isn’t we return an HTTP 404 error.

If the beer exists, we construct a new Beer object; passing it the ID and the name field within the value dictionary.

We then pass this beer to the templates/beer/show.html template which we’ll show here:


{% extends "layout.html" %}
{% block body %}

{% set display = beer.doc %}
{% set brewery_id = display['brewery_id'] %}

<h3>Show Details for Beer "{{}}"</h3>
<table class="table table-striped">
            <td><a href="/breweries/show/{{brewery_id}}">{{brewery_id}}</a></td>
        {% for k, v in display.items() if not k == "brewery_id" %}
        {% endfor %}

<a class="btn btn-medium btn-warning"
<a class="btn btn-medium btn-danger"

{% endblock %}

Here we make the display variable in a special {% set %} directive. This makes dealing with the rest of the code simpler.

The next thing we do is extract the brewery_id, and create a special entry with a link pointing to the page to display the actual brewery.

Then we is iterate over the rest of the fields (omitting the brewery ID), printing out the key and value of each.

Finally, we provide links at the bottom to Edit and Delete the beer.

Editing Beers (beer edit page)

def normalize_beer_fields(form):
    doc = {}
    for k, v in form.items():
        name_base, fieldname = k.split('_', 1)
        if name_base != 'beer':

        doc[fieldname] = v

    if not 'name' in doc or not doc['name']:
        return (None, ("Must have name", 400))

    if not 'brewery_id' in doc or not doc['brewery_id']:
        return (None, ("Must have brewery ID", 400))

    if not db.get(doc['brewery_id'], quiet=True).success:
        return (None,
                ("Brewery ID {0} not found".format(doc['brewery_id']), 400))

    return doc, None

@app.route('/beers/edit/<beer>', methods=['GET'])
def edit_beer_display(beer):
    bdoc = db.get(beer, quiet=True)
    if not bdoc.success:
        return "No Such Beer", 404

    return render_template('beer/edit.html',
                           beer=Beer(beer, bdoc.value['name'], bdoc.value),

@app.route('/beers/edit/<beer>', methods=['POST'])
def edit_beer_submit(beer):
    doc, err = normalize_beer_fields(request.form)

    if not doc:
        return err

    db.set(beer, doc)
    return redirect('/beers/show/' + beer)

We define two handlers for editing. The first handler is the GET method for /beers/edit/<beer>, which displays a nice HTML form that we can use to edit the beer. It passes the following parameters to the template: the Beer object and a Boolean that indicates this is not a new beer (because the same template is also used for the Create Beer form).

The second handler is the POST method, which validates the input. The post handler calls the normalize_beer_fields function, which converts the form fields into properly formed names for the beer document, checks to see that the beer has a valid name, and checks to see that a brewery_id is specified and that it indeed exists. If all the checks pass, the function returns a tuple of ( doc, None ). The POST handler checks whether the second element of the returned tuple is false. If it is not false, then it’s an error code, and the first element is the error message. Otherwise, the first element is the document. It then sets the document in Couchbase by using the set method.

The template is rather wordy because we enumerate all the possible fields with a nice description.


{% extends "layout.html" %}
{% block body %}

{% if is_create %}
<h3>Create Beer</h3>
{% else %}
<h3>Editing {{}}</h3>
{% endif %}

<form method="post" action="">
    <legend>General Info</legend>
    <div class="span12">
      <div class="span6">
        <input type="text" name="beer_name" placeholder="The name of the beer." value="{{}}">

        <input type="text" name="beer_description" placeholder="A short description." value="{{beer.description}}">
      <div class="span6">
        <input type="text" name="beer_style" placeholder="Bitter? Sweet? Hoppy?" value="{{}}">

        <input type="text" name="beer_category" placeholder="Ale? Stout? Lager?" value="{{beer.category}}">
    <div class="span12">
      <div class="span6">
        <label>Alcohol (ABV)</label>
        <input type="text" name="beer_abv" placeholder="The beer's ABV" value="{{beer.abv}}">

        <label>Biterness (IBU)</label>
        <input type="text" name="beer_ibu" placeholder="The beer's IBU" value="{{beer.ibu}}">
      <div class="span6">
        <label>Beer Color (SRM)</label>
        <input type="text" name="beer_srm" placeholder="The beer's SRM" value="{{beer.srm}}">

        <label>Universal Product Code (UPC)</label>
        <input type="text" name="beer_upc" placeholder="The beer's UPC" value="{{beer.upc}}">
    <div class="span12">
      <div class="span6">
        <input type="text" name="beer_brewery_id" placeholder="The brewery" value="{{beer.brewery_id}}">
  <div class="form-actions">
      <button type="submit" class="btn btn-primary">Save changes</button>

{% endblock %}

The template first checks the is_create variable. If it’s False, then we’re editing an existing beer, and the caption is filled with that name. Otherwise, it’s titled as Create Beer.

Creating Beers

Creating beers is largely the same as editing beers: (create beer page)

def create_beer_display():
    return render_template('beer/edit.html', beer=Beer('', ''), is_create=True)

@app.route('/beers/create', methods=['POST'])
def create_beer_submit():
    doc, err = normalize_beer_fields(request.form)

    if not doc:
        return err

    id = '{0}-{1}'.format(doc['brewery_id'],
                          doc['name'].replace(' ', '_').lower())

        db.add(id, doc)
        return redirect('/beers/show/' + id)

    except KeyExistsError:
        return "Beer already exists!", 400

Here we display the same form as the one for editing beers, except we set the is_create parameter to True, and pass an empty Beer object. This is necessary because the template still tries to populate the form fields with existing values.

In the POST handler, we call normalize_beer_field as above when editing beers.

Because we’re creating a new beer, we use the add method instead. This raisew an exception if the beer already exists. We catch this and display it to the user.

If everything went well, the user is redirected to the beer display page for the newly created beer.

Searching Beers

In the beer listing page above, you might have noticed a search box at the top. We can use it to dynamically filter our table based on user input. We’ll use Javascript at the client layer to perform the querying and filtering, and views with range queries at the server (Flask) layer to return the results.

Before we implement the Python-level search method, we need to put the following in the static/js/beersample.js file (if it’s not there already) to listen on search box changes and update the table with the resulting JSON (which is returned from the search method):

static/js/beersample.js (snippet)

$(document).ready(function() {

     * AJAX Beer Search Filter
    $("#beer-search").keyup(function() {
       var content = $("#beer-search").val();
       if(content.length >= 0) {
           $.getJSON("/beers/search", {"value": content}, function(data) {
               $("#beer-table tbody tr").remove();
               for(var i=0;i<data.length;i++) {
                   var html = "<tr>";
                   html += "<td><a href=\"/beers/show/"+data[i].id+"\">"+data[i].name+"</a></td>";
                   html += "<td><a href=\"/breweries/show/"+data[i].brewery+"\">To Brewery</a></td>";
                   html += "<td>";
                   html += "<a class=\"btn btn-small btn-warning\" href=\"/beers/edit/"+data[i].id+"\">Edit</a>\n";
                   html += "<a class=\"btn btn-small btn-danger\" href=\"/beers/delete/"+data[i].id+"\">Delete</a>";
                   html += "</td>";
                   html += "</tr>";
                   $("#beer-table tbody").append(html);

The code waits for keyup events on the search field, and if they happen, it issues an AJAX query on the search function within the app. The search handler computes the result (using views) and returns it as JSON. The JavaScript then clears the table, iterates over the results, and creates new rows.

The search handler looks like this: (ajax search response)

def return_search_json(ret):
    response = app.make_response(json.dumps(ret))
    response.headers['Content-Type'] = 'application/json'
    return response

def beer_search():
    value = request.args.get('value')
    q = Query()
    q.mapkey_range = [value, value + Query.STRING_RANGE_END]
    q.limit = ENTRIES_PER_PAGE

    ret = []

    rp = BeerListRowProcessor()
    res = db.query("beer", "by_name",

    for beer in res:
        ret.append({'id' :,
                    'name' :,
                    'brewery' : beer.brewery_id})

    return return_search_json(ret)

The beer_search function first extracts the user input by examining the query string from the request.

It then creates a Query object. The Query object’s mapkey_range property is set to a list of two elements; the first is the user input, and the second is the user input with the magic STRING_RANGE_END string appended to it. This form of range indicates that all keys that start with the user input ( value ) are returned. If we just provided a single element, the results would also contain matches that are lexically greater than the user input; if we just provided the same value for the second and first elements, only items that match the string exactly are returned.

The special STRING_RANGE_END is actually a u"\u0FFF" UTF-8 character, which for the view engine means “end here.” You need to get used to it a bit, but it’s actually very neat and efficient.

We re-use our BeerListRowProcessor class to filter the results here (because the data required is the same as that of the beer listing ( beer/index.html ) page.

However we need to return a JSON array of

{ "id" : "beer_id", "name" : "beer_name", "brewery" : "the_brewery_id" }

so we need to convert the rows into JSON first. This is done by the return_search_json function.

Now your search box should work nicely.

Managing Breweries

While this is implemented in the repository above, it is left as an exercise to the reader to work out some more details.

Wrapping Up

The tutorial presents an easy approach to start a web application with Couchbase Server as the underlying data source. If you want to dig a little bit deeper, the full source code in the couchbaselabs repository on GitHub has more code to learn from. This code might be extended and updated from time to time.

Of course, this is only the starting point for Couchbase, but together with the Getting Started Guide, you should now be well equipped to start exploring Couchbase Server on your own. Have fun working with Couchbase!

Food For Thought

Some things are not implemented in the example. Here’s some ideas for features you might add while learning to use the SDK:

  • When deleting a brewery, ensure it has no beers dependent on it.

  • Provide a search where one can query beers belonging to a given brewery.

  • Handle concurrent updates to a beer and/or brewery.

  • Implement a like feature, where one can like a beer or a brewery; likewise, they can unlike one as well!

Using the APIs

This section only gives an introduction to the available APIs. The actual API reference features more options for each of the APIs described here and is always be more up-to-date than the documentation here.

This section goes a level lower than the Getting Started guide and features the aspects of the APIs offered by the SDK.


You can connect to a bucket using the simple Couchbase.connect() class method:

from couchbase import Couchbase

client = Couchbase.connect(bucket='default')

Multiple Nodes

Sometimes it is beneficial to let the client know beforehand about multiple nodes. For example, when you have a cluster with several nodes and some nodes might not be up, if you pass only one node the client’s constructor might raise an exception. Instead, you can pass multiple nodes as a list so that the constructor tries each node until it gets a successful connection or the timeout is reached.

Using Multiple Nodes

c = Couchbase.connect(
    host=['', '', '']


The client uses timeouts so that your application does not wait too long if the cluster is overloaded or there are connectivity issues. By default, this timeout value is 2.5 seconds.

You can adjust this value by setting it in the constructor.:

c = Couchbase.connect(bucket='default', timeout=5.5)

Or by setting the timeout property:

c.timeout = 4.2

SASL Buckets

If your bucket is password protected, you can pass the Simple Authentication and Security Layer (SASL) password by using the password keyword parameter in the constructor:

c = Couchbase.connect(bucket='default', password='s3cr3t')


Threads will be discussed later on in more detail, but the Connection object is fully thread-safe out of the box by default. You can tune some parameters that sacrifice thread-safety for performance.

API Return Values

Before we discuss the individual sections of the API, we’ll discuss the common return value, which is the Result object.

Typically subclasses of this object are returned appropriate for the operation executed.

All Result objects have the following properties:

  • success

    A boolean property that indicates whether this operation was successful.

  • rc

    This is the low level return code as received from the underlying libcouchbase layer. This is 0 on success and nonzero on failure. Typically this is useful on operations in which quiet was set to True. Normally you’d use it like this:

    result = client.get("key", quiet=True)
     if not result.success:
         print "Got error code", result.rc
  • __str__

    While this isn’t really a property, printing out the result object will yield interesting metadata to aid in debugging this particular operation.

Storing Data

Setting Values

These methods, if successful, set the value of their respective keys. If they fail, they raise an exception (and are not affected by the quiet property).

These methods can accept a format property, which indicates the format in which the value will be stored on the server, as well as a ttl property which indicates the lifetime of the value. After the ttl lifetime is reached, the value is deleted from the server.

Starting from version 1.1.0, these methods can also accept persist_to and replicate_to arguments, specifying the synchronization of application code with replication and persistence events. See the API documentation for more details.

  • client.set(key, value, **kwargs)

    Sets the key unconditionally.

  • client.add(key, value, **kwargs)

    Sets the key to the specified value, but only if the key does not already exists (an exception is raised otherwise).

  • client.replace(key, value, **kwargs)

    Replaces an existing key with a new value. This raises an exception if the key does not already exist.

Arithmetic and Counter Operations

These methods operate on 64-bit integer counters. They provide efficient mutation and retrieval of values. You can use these in place of the set family of methods when working with numeric values. For example:

Using set

key = "counter"
    result = c.get("counter")
    c.set(key, result.value + 1)
except KeyNotFoundError:
    c.add(key, 10)

Using incr

key = "counter"
c.incr(key, initial=10)

These methods accept the ttl argument to set the expiration time for their values, as well as an amount value that indicates by what amount to modify their values. Additionally, an initial keyword argument is available to provide the default value for the counter if it does not yet exist. If an initial argument is not provided and the key does not exist, an exception is raised.

The value for the counter stored must either not exist (if initial is used) or should be a “Number,” that is, a textual representation of an integer.

If using the default FMT_JSON, then your integers are already compliant.

If the existing value is not already a number, the server raises an exception (specifically, a DeltaBadvalError ).

Arithmetic methods return a ValueResult object (subclass of Result ). The value property can be used to obtain the current value of the counter.

  • c.incr(key, amount=1, ttl=0)

    Increments the value stored under the key.

  • c.decr(key, amount=1, ttl=0)

    Decrements the value stored under the key. In this case, amount is how much to subtract from the key.

Append and Prepend Operations

These operations act on the stored values and append or prepend additional data to it. They treat existing values as strings and only work if the existing stored data is a string (that is, FMT_UTF8 or FMT_BYTES ).

The format argument is still available, but the value must be either FMT_UTF8 or FMT_BYTES. If not specified, it defaults to FMT_UTF8. Otherwise, they are part of the set family of methods:

c.set("greeting", "Hello", format=FMT_UTF8)
c.append("greeting", " World!")
c.get("greeting").value == "Hello World!"
c.prepend("greeting", "Why, ")
c.get("greeting").value == "Why, Hello World!"
  • c.append(key, data_to_append, **kwargs)

    Appends data to an existing value.

  • c.prepend(key, data_to_prepend, **kwargs)

    Prepends data to an existing value.

Ensure that you only append or prepend to values that were initially stored as FMT_UTF8 or FMT_BYTES. It does not make sense to append to a JSON or pickle string.


c.set("a_dict", { "key for" : "a dictionary" })

The key a_dict now looks like this on the server:

Now, prepend the following to it:

c.prepend("a dict", "blah blah blah")

The value for a_dict looks like this now:

Now, when you try to get it back, you see this happen:

>>> c.get("a_dict")
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "couchbase/", line 376, in get
    return _Base.get(self, key, ttl, quiet, replica, no_format)
  File "/usr/lib/python2.7/json/", line 338, in loads
    return _default_decoder.decode(s)
  File "/usr/lib/python2.7/json/", line 365, in decode
    obj, end = self.raw_decode(s, idx=_w(s, 0).end())
  File "/usr/lib/python2.7/json/", line 383, in raw_decode
    raise ValueError("No JSON object could be decoded")
    <Failed to decode bytes, Results=1, inner_cause=No JSON object could be decoded,
    C Source=(src/convert.c,215), OBJ='blah blah blah{"key for": "a dictionary"}'>

Expiration Operations

This consists of a single method that is used to update the expiration time of a given key. It is passed two arguments, a key and an expiration time. If the expiration time is greater than zero, the key receives the new expiration time.

The expiration time can be expressed as either an offset in seconds or a Unix time stamp. If the value of the expiration time is smaller than 60*60*24*30 (that is, one month in seconds) it is considered to be an offset in seconds. Larger values are considered to be Unix time stamps.

If the expiration time is zero, then any existing expiration time is cleared and the value remains stored indefinitely unless explicitly deleted or updated with expiration at a later time.

This is a lightweight means by which to ensure entities “stay alive” without the overhead of having to reset their value or fetch them.

  • c.touch(key, ttl)

    Updates the given key with the specified ttl.

Deleting Data

  • client.delete(key, quiet=False)

    Removes a key from the server. If quiet is specified, an exception is not raised if the key does not exist.

Retrieving Data

  • client.get(key, quiet=False, ttl=0)

    Retrieves a key from the server. If the key does not exist, an exception is raised if the key does not exist and quiet is set to False.

If ttl is specified, this also modifies, in-situ, the expiration time of the key when retrieving it. This is also known as Get and Touch.

This returns a ValueResult object (subclass of Result ) that can be used to obtain the actual value via the value property.

Locking Data/Ensuring Consistency

In production deployments, it is possible that you will have more than a single instance of your application trying to modify the same key. In this case a race condition happens in which a modification one instance has made is immediately overidden.

Consider this code:

def add_friend(user_id, friend):
    result = c.get("user_id-" + user_id)
    result.value['friends'][friend] = { 'added' : time.time() }
    c.set("user_id-" + user_id, result.value)

In this case, friends is a dictionary of friends the user has added, with the keys being the friend IDs, and the values being the time when they were added.

When the friend has been added to the dictionary, the document is stored again on the server.

Assume that two users add the same friend at the same time, in this case there is a race condition where one version of the friends dict ultimately wins.

Couchbase provides two means by which to solve for this problem. The first is called Opportunistic Locking and the second is called Pessimistic Locking.

Both forms of locking involve using a CAS value. This value indicates the state of a document at a specific time. Whenever a document is modified, this value changes. The contents of this value are not significant to the application, however it can be used to ensure consistency. You can pass the CAS of the value as it is known to the application and have the server make the operation fail if the current (server-side) CAS value differs.

Opportunistic Locking

The opportunistic locking functionality can be employed by using the cas keyword argument to the set family of methods.

Note that the cas value itself may be obtained by inspecting the cas property of the Result object returned by any of the API functions.

We can now modify the add_friend example so that it handles concurrent modifications gracefully:

def add_friend(user_id, friend):

    while True:
        result = c.get("user_id-" + user_id)
        result.value['friends'][friend] = { 'added' : time.time() }

            c.set("user_id-" + user_id, result.value, cas=result.cas)

        except KeyExistsError:
            print "It seems someone tried to modify our user at the same time!"
            print "Trying again"

This is called opportunistic locking, because if the CAS is not modified during the first loop, the operation succeeds without any additional steps.

Pessimistic Locking

Pessimistic locking is useful for highly contended resources; that is, if the key being accessed has a high likelihood of being contended. While this method may be more complex, it is much more efficient for such resources.

We can use pessimistic locking by employing the lock and unlock functions.

The lock method locks the key on the server for a specified amount of time. After the key is locked, further attempts to access the key without passing the proper CAS fail with a TemporaryFailureError exception until the key is either unlocked or the lock timeout is reached.

  • c.lock(key, ttl=0)

    This has the same behavior as get (that is, it returns the value on the server), but the ttl argument now indicates how long the lock should be held for. By default, the server-side lock timeout is used, which is 15 seconds.

    Returns a ValueResult.

  • c.unlock(key, cas)

    Unlocks the key. The key must have been previously locked and must have been locked with the specified cas. The cas value can be obtained from the Result object’s cas property.

Calling any of the set methods with a valid CAS implicitly unlocks the key, and thus makes an explicit call to unlock unnecessary — calling unlock on a key that is not currently locked raises an exception.

We can rewrite our add_friend example using the lock functions:

def add_friend(user_id, friend):
    while True:
            result = c.lock("user_id-" + user_id)

        except TemporaryFailError:
            # Someone else has locked the key..
        result.value['friends'][friend] = { 'added' : time.time() }
        c.set("user_id-" + user_id, result.value, cas=result.cas)

        # We want to unlock if anything happens, rather than waiting
        # for it to time out
        c.unlock(result.key, result.cas)

        # then, raise the exception

When To Use Optimistic Or Pessimistic Locking

Optimistic locking is more convenient and sometimes more familiar to users. Additionally, it does not require an explicit unlock phase.

However, during a CAS mismatch, the full value is still sent to the server in the case of opportunistic locking. For highly contended resources this has impacts on network I/O, because the value must be sent multiple times before it is actually stored.

Pessimistic locking does not retrieve its value unless the operation was successful, however.

Working With Views

This section provides a bit more information on how to work with views from the Python SDK. If you are new to views, you should read the chapter about views in the Couchbase Server Manual.

To use views, you must have already set up design documents containing one or more view queries you have defined. You can execute these queries from the Python SDK and retrieve their results.

You can define views via either the Couchbase Server web interface or thePython SDK design document management functions.

Couchbase Server comes with two pre-defined sample buckets which can be installed from the “Sample Buckets” section in the “Settings” pane.

The basic interface for views is such

client.query(design_name, view_name) Which returns an iterable object which yields ViewRow objects.

ViewRow objects are simple named tuples with the following fields:

  • vr.key

    The key emitted by the view’s map function (i.e. first argument to emit.

  • vr.value

    The value emitted by the view’s map function (i.e. second argument to emit ).


    The document ID of this row. The ID can be passed to get and set.

  • vr.doc

    A Result object containing the actual document, if the query method was passed the include_docs directive.

The object returned by query is a class that defines an __iter__ (and thus does not have a __len__ or items() method). You can convert it to a list by using list comprehension :

rows_as_list = [ c.query("beer", "brewery_beers") ]

You can also pass options to the query method. The list of available options is documented in the Query class in the API documentation.

from couchbase.views.params import Query
             mapkey_range = ["abbaye", "abbaye" + Query.STRING_RANGE_END],

The include_docs directive can be used to fetch the documents along with each ViewRow object. Note that while it is possible to simply call c.get(, the client handles the include_docs directive by actually performing a batched ( get_multi ) operation.

You can also pass options for the server itself to handle. These options can be passed as either an encoded query string, a list of key-value parameters, or a Query object.

Using encoded query strings

client.query("beer", "brewery_beers", query="limit=3&skip=1&stale=false")

Note that this is the most efficient way to pass options because they do not need to be re-encoded for each invocation.

However, it is impossible for the SDK to verify the inputs and thus it is suggested you only use a raw string once your query has been refined and optimized.

Using key-value pairs

client.query("beer", "brewery_beers", limit=3, skip=1, stale=False)

This allows simple and idiomatic construction of query options.

Using a Query object

from couchbase.views.params import Query

q = Query
q.limit = 3
q.skip = 1
q.stale = False
client.query("beer", "brewery_beers", query=q)

The Query object makes it simple to programmatically construct a query, and provides the most maintainable option. When using key-value pairs, the SDK actually converts them to a Query object before processing.

Query objects also have named properties, making query construction easy to integrate if using an IDE with code completion.

Common View Parameters

Here are some common parameters used for views. They are available either as keyword options to the query method, or as properties on the Query object

Server Parameters

  • mapkey_range = [ "start", "end" ]

    Sets the start and end key range for keys emitted by the map function.

  • startkey = "start"

    Sets the start key.

  • endkey = "end"

    Sets the end key.

  • descending = True

    Inverts the default sort order.

  • stale = False

    Possible values are True, False, or the string update_after.

  • limit = 10

    Limits the number of rows returned by the query.

query Method Options

These are only available as options to the query method, and should not be used on the Query object.

  • include_docs = True

    Fetches corresponding documents along with each row.

  • streaming = True

    Fetches results incrementally. Don’t buffer all results in memory at once.


Often, view results can be large. By default the client reads all the results into memory and then returns an iterator over that result set. You can change this behavior by specifying the streaming option to the query method. When used, results are fetched incrementally.

Using streaming does not have any impact on how the rows are returned.

Design Document Management

The Python Couchbase SDK provides means by which you can manage design documents; including all phases of design document development. You can:

  • Create a development design.

  • Publish a development design to a production design.

  • Retrieve a design document.

  • Delete a design document.

Note that all design creation methods take a syncwait argument, which is an optional amount of time to wait for the operation to be complete. By default the server (and thus the SDK) only schedule a design document operation. This means that if you try to use the view right after you created it, you might get an error because the operation has not yet completed. Using the syncwait parameter polls for this many seconds and either returns successfully or raises an exception.

An additional argument that can be provided is the use_devmode parameter. If on, the name of the design will be prepended with dev_ (if it does not already start with it).

All these operations return an HttpResult object which contains the decoded JSON payload in its value property.

  • c.design_create(name, design, use_devmode=True, syncwait=0)

    Creates a new design document. name is the name of the design document (for example, "beer" ). design is either a Python dictionary representing the structure of the design or a valid string (that is, encoded JSON) to be passed to the server.

  • c.design_get(name, use_devmode=True)

    Retrieves the design document.

  • c.design_publish(name, syncwait=0)

    Converts a development-mode view into a production mode view. This is equivalent to pressing the Publish button on the web UI.

  • c.design_delete(name, use_devmode=True)

    Deletes a design document.

Twisted Interface

twisted is a popular asynchronous network framework for Python. Version 1.1.1 supports twisted natively through the txcouchbase module.

The twisted interface builds on the synchronous interface mentioned in the previous chapter.

Note that the Twisted interface is currently considered to be evolving in terms of stability and interface. As such, the couchbase.experimental feature must be explicitly imported and enabled in order to use.


from couchbase.experimental import enable as enable_experimental
from txcouchbase.connection import Connection as TxCouchbase

client = TxCouchbase(bucket='default')

The TxCouchbase constructor will return a new Connection object (actually a subclass of the synchronous connection class) which you can then use to perform operations.

You can invoke the client’s .connect() method which will actually return a Deferred object which can be used to receive notification of the connection or its failure.

class Thing(object):
    def __init__(self):
        self.client = TxCouchbase(bucket='default')
        d = self.client.connect()

    def on_connect_error(self, err):
        print "Got error", err
        # Handle it, it's a normal Failure object

    def on_connect_success(self):
        print "Couchbase Connected!"

thing = Thing()

Initial Connection Errors

It is currently recommended to call the ._close() method on the connection instance if a connection error is detected. Otherwise the program may busy loop until the Connection object is automatically garbage collected.

Performing Operations

The model for the twisted operation interface is simple. Where the synchronous API returns a Result class, the twisted interface returns a Deferred which will invoke its callback with the received Result once it is available.

client = TxCouchbase(bucket='default')

def on_get(result):
    print "Got result", result

def on_set(result):
    print "Set item", result

client.set("foo", "bar").addCallback(on_set)
Set item OperationResult<RC=0x0, Key=u'foo', CAS=0xbd3671590b5a0100>
Got result ValueResult<RC=0x0, Key=u'foo', Value=u'bar', CAS=0xbd3671590b5a0100, Flags=0x0>

To handle errors, simply add the error callback via the returned Deferred object’s addErrback method.

def on_get_miss(err, key):
    print "Couldn't fetch", key

client = TxCouchbase(bucket='default')
client.delete('foo', quiet=True)
client.get('foo').addErrback(on_get_miss, 'foo')

Naming Of Multi Operations

The Connection object supports usage of both underscore_names and camelCase for the multi API methods, thus cb.getMulti() is the same as cb.get_multi. This is specific to the twisted interface where camelCase names are prevalent and conventional.

Ordering Of Operations

If your operations need to be executed in a certain order, then keep in mind that operations are performed in order only in respect to operations on the same key. In the above example we can be sure that the key will always be removed before it is deleted. However if operations on different keys require a particular form of ordering, it is best to wait until the first operation completes and then invoke the next operation via the received callback.

Querying Views

Since the synchronous API uses iterators and generators performing network I/O on demand, the non-blocking API for querying views from Twisted is slightly different. The interface does support the same Query object used in the synchronous client.

Two methods are provided to query views. The first one is the simple queryAll method which passes all the rows from the result over to a single callback, e.g.

def on_view_rows(rows):
    for row in rows:
        print row

client = TxCouchbase(bucket='beer-sample')
d = client.queryAll("beer", "brewery_beers", limit=5)
ViewRow(key=[u'21st_amendment_brewery_cafe'], value=None, docid=u'21st_amendment_brewery_cafe', doc=None)
ViewRow(key=[u'21st_amendment_brewery_cafe', u'21st_amendment_brewery_cafe-21a_ipa'], value=None, docid=u'21st_amendment_brewery_cafe-21a_ipa', doc=None)
ViewRow(key=[u'21st_amendment_brewery_cafe', u'21st_amendment_brewery_cafe-563_stout'], value=None, docid=u'21st_amendment_brewery_cafe-563_stout', doc=None)
ViewRow(key=[u'21st_amendment_brewery_cafe', u'21st_amendment_brewery_cafe-amendment_pale_ale'], value=None, docid=u'21st_amendment_brewery_cafe-amendment_pale_ale', doc=None)
ViewRow(key=[u'21st_amendment_brewery_cafe', u'21st_amendment_brewery_cafe-bitter_american'], value=None, docid=u'21st_amendment_brewery_cafe-bitter_american', doc=None)

queryAll returns a Deferred. Its callback is invoked with an iterable of all rows and its errback is invoked with an error, if any.

All the additional keyword arguments supported in either the synchronous query method or the Query object are supported with the exception of include_docs.

Getting Streaming Results

You can also use the queryEx method to receive streaming results. This has the same effect in setting streaming=True in the synchronous API.

The difference between queryAll and queryEx is that the former buffers all the rows in memory before invoking the callback whereas the latter invokes the callback with rows as they arrive on the network.

In order to receive streaming results, you must first extend the couchbase.async.view.AsyncViewBase object and then implement the relevant methods needed to handle the streaming events.

from twisted.internet import reactor

from couchbase.experimental import enable as enable_experimental

from txcouchbase.connection import Connection as TxCouchbase
from couchbase.exceptions import CouchbaseError
from couchbase.async.view import AsyncViewBase

class MyRowHandler(AsyncViewBase):
    def __init__(self, *args, **kwargs):
        self.__times_called = 0
        self.__received_rows = 0
        super(MyRowHandler, self).__init__(*args, **kwargs)

    def on_rows(self, rows):
        self.__times_called += 1
        self.__received_rows += len([r for r in rows])

    def on_error(self, ex):
        # ex is an Exception, *not* a Failure
        print "Got error while querying view", ex

    def on_done(self):
        print ("Callback invoked {0} times. Total of {1} rows received"
               .format(self.__times_called, self.__received_rows))
        print "Indexed Rows: ", self.indexed_rows

client = TxCouchbase(bucket='beer-sample')
client.queryEx(MyRowHandler, "beer", "brewery_beers")
Callback invoked 218 times. Total of 7303 rows received
Indexed Rows:  7303

queryEx receives a class (not an instance!) of the row handler which it then instantiates with the relevant parameters.

Here the on_rows method is called with an iterable of rows multiple times until no more rows are available. The on_done method is called when there are no more rows remaining.

As the AsyncViewBase object itself is a subclass of View we can also see how many total rows were indexed by inspecting the indexed_rows property.


While the previous examples showed the client returning convenient Deferred objects, in reality Deferreds incur a significant performance overhead as they must support an entire callback chain as well as asynchronous suspend/resume functionality. See the Deferred source code for the full details.

If performance is critical and you only care about receiving error/success notifications and results without the overhead of a callback chain, you may use the txcouchbase.connection.TxAsyncConnection object (which is a base of the txcouchbase.connection.Connection class).

Using the TxAsyncConnection with the AsyncResult objects instead of Deferred objects yield a 3.5x performance benefit.

from twisted.internet import reactor

from couchbase.experimental import enable as enable_experimental
from txcouchbase.connection import TxAsyncConnection as TxAsync

client = TxAsync(bucket='default')

def on_connected(*args):
    ft_set = client.set("foo", "bar")
    def on_set(rr):
        print "Stored item", rr
        def on_get(rr2):
            print "Got item", rr2
        ft_get = client.get("foo")
        ft_get.callback = on_get
    ft_set.callback = on_set


For key value operations, this class returns an AsyncResult object which contains raw properties of callback and errback. These properties function a bit differently than they do in a deferred.

Additionally, operations must only be performed once the initial callback has been invoked.

If an error takes place, the errback field’s callback will be invoked with the signature of (opres, exc_type, exc_value, exc_traceback).

Gevent Interface

gevent is a coroutine library for Python. It allows multiple thread-like objects (eventlets) to run in seeming parallelism within a single Python application (and within a single Python thread).

In order for applications to take full benefit of gevent, any form of I/O must be done in a cooperative manner (i.e. it must be gevent-aware). While running non-gevent-aware libraries will not cause any malfunction within the application, it will reduce performance as other eventlets will not be able to run while a non-gevent-aware I/O operation is taking place.

The normal synchronous API may be used without modification for gevent. However it is not gevent-aware.

Gevent aware client

With version 1.1.1, a specifically gevent-aware library was introduced allowing the full performance benefits of gevent with couchbase. Currently both gevent pre-release (e.g. 0.13) and 1.0.x release versions are known to work.

from couchbase import experimental
from gcouchbase.connection import GConnection

cb = GConnection(bucket='default')
cb.set("foo", "bar")

In order to use the gcouchbase library, you must first enable the experimental client features. The gcouchbase package is considered experimental because it relies on undocumented internals of the gevent library.


The gcouchbase API is exactly the same as the normal couchbase API (and uses the exact same codebase) with the following exceptions:

  • The gcouchbase.connection.GConnection class must be used
  • The include_docs option in views is not supported.

PyPy Support

PyPy is an alternative Python interpreter which claims significant speed benefit over CPython. It attains this speed boost by using JIT (just-in-time) compilation among other things.

Built-In SDK Support Functionality

The mainline supported Couchbase Python SDK is a CPython extension using CPython’s native API to deliver a highly efficient and highly stable library. Unfortunately as a result it means that it will not work properly on Python implementations which do not support the CPython API.

By means of something called cpyext PyPy offers rudimentary alpha-level support for libraries requiring the CPython API. The Couchbase Python SDK will properly compile against cpyext without modification and some of its basic functionality may be available. Note however that because of the stability level of cpyext, your application may randomly crash or leak.

Native PyPy Support

As an alternative, an external project is available ( which provides a pure-python implementation of the Couchbase SDK, utilizing the cffi library to interface with libcouchbase.

The CFFI-based implementation is the recommended way to use the Couchbase SDK in a PyPy environment. It provides a subset of functionality as the mainline implementation and exposes an identical API.

Currently using the CFFI-implementation itself requires that the mainline SDK be installed as well.

See the project’s homepage for installation instructions and more details on its supported featureset.

Advanced Usage

This covers advanced topics and builds on the Using the APIs section.

Batched (Bulk) Operations

Most API functions have both single and multi-key (batched) variants. The batched variant has the same name as the single-key variant, but its method name has _multi appended to it.

The batched operations are significantly quicker and more efficient, especially when dealing with many small values, because they allow pipelining of requests and responses, saving on network latency.

Batched operations tend to accept an iterable of keys (or a dict of keys, depending on the method) and return a dictionary of the following format:

c.foo_multi(["key1", "key2", "key3"])

    "key1" : FooResult(...),
    "key2" : FooResult(...),
    "key3" : FooResult(...)

Exceptions in Batched Operations

Sometimes a single key in a batched operation fails, resulting in an exception. It is still possible to retrieve the full result set of the failed batched operation by using the all_results property of the thrown exception (assuming it is of type CouchbaseError ).

c.set("foo", "foo value")

        "foo" : "foo value",
        "bar" : "bar value",
        "baz" : "baz value"
except CouchbaseError as exc:
    for k, res in exc.all_results.items():
        if res.success:
            # Handle successful operation
            print "Key {0} failed with error code {1}".format(k, res.rc)
            print "Exception {0} would have been thrown".format(

Using With and Without Threads

The Connection object by default is thread safe. To do so, it uses internal locks and explicitly locks and unlocks the Python GIL to ensure that a fatal error is not thrown by the application.

The locking and unlocking has a slight performance impact, with the guarantee that things will not crash if an application is using threads.

If you are not using threads in your application (at all), you can pass the unlock_gil=False option to the connect method like so:

c = Couchbase.connect(unlock_gil=False, bucket='default')

This disables all locking/unlocking (not to be confused with the lock and unlock features that operate on keys in the server) functionality. If your application does use threads, those threads will be blocked while the Connection object waits for the server to respond.

In addition to locking and unlocking the GIL, upon entry to each function the Connection object locks itself (using the equivalent of Lock.acquire ) and unlocks itself after it leaves. This is to ensure that multiple threads are not using the same Connection object at once; and thus access is serialized.

You may disable this behavior with the following two options:

  • Don’t lock at all

    The Connection object is not locked at all. If your application tries to use the Connection object from multiple threads at once, strange errors might happen and your program will eventually core dump.

    If you’re sure you’re not going to use it from more than one thread, you can use the lockmode = LOCKMODE_NONE in the constructor:

    from couchbase import Couchbase, LOCKMODE_NONE
     c = Couchbase.connect(bucket=‘default’, lockmode=LOCKMODE_NONE)
  • Throw an exception if concurrent access is detected

    This is helpful for debugging an application where multiple threads should not be accessing the Connection object (but for some reason, they are). You can use the lockmode = LOCKMODE_EXC for this:

    from couchbase import Couchbase, LOCKMODE_EXC
     c = Couchbase.connect(bucket=‘default’, lockmode=LOCKMODE_EXC)

The default lockmode is couchbase.LOCKMODE_WAIT, which waits silently if concurrent access is detected.

Custom Encodings/Conversions

While the Python SDK offers numerous options for converting your data to be suitable for storing on the server, it might sometimes not be enough. For this, the Transcoder interface is used.

The Transcoder interface allows you to define an object that is called with each value together with the format arguments passed to it.

Formats and Flags

The value passed for the format parameter is actually a flag that is stored on the server. Each key has a small amount of metadata which is stored along with it on the server. The Python SDK stores the format value to the metadata when you store a value (using set ) and then reads it when retrieving the value (using get ). If the flag is equal to FMT_JSON, then it attempts to decode it as JSON; if the flag is equal to FMT_PICKLE, then it attempts to decode it as Pickle; and so on.

You may pass the FMT_AUTO flag to set, in which case it will guess the suitable format for your data based on the Python datatype being used.

Additionally, you can pass the no_format option to get to suppress automatic conversion of the value to a native Python type (the bytes type will be used instead). This may be useful when conversion is not needed, or if other issues arise when decoding a format (for example, a Pickle version mismatch).

Custom Transcoder Objects

You can write a custom transcoder that allows Zlib compression. Here’s a snippet:

import zlib

from couchbase import Couchbase
from couchbase.transcoder import Transcoder
from couchbase import FMT_MASK, FMT_BYTES

# We'll define our own flag.

class ZlibTranscoder(Transcoder):
    def encode_value(self, value, format):
        converted, flags = super(ZlibTranscoder, self).encode_value(value, format & FMT_MASK)
        if (format & FMT_ZLIB):
            flags |= FMT_ZLIB
            converted = zlib.compress(converted)
        return (converted, flags)

    def decode_value(self, value, flags):
        if (flags & FMT_ZLIB):
            value = zlib.decompress(value)
            flags &= FMT_MASK
        return super(ZlibTranscoder, self).decode_value(value, flags)

In the example, the ZlibTranscoder class is defined as a subclass of the provided couchbase.transcoder.Transcoder class. The latter is a wrapper class that defaults to use the default conversion methods in the SDK (note that the library does not use any Transcoder object by default, but the provided one wraps the built-in converters.

For encode_value we are passed the user-specified value, which is any Python object, and a format value, which also can be any valid Python object (though the default transcoder accepts only the FMT_JSON, FMT_UTF8, FMT_BYTES and FMT_PICKLE values.

We define an additional format flag called FMT_ZLIB. We make this one higher than FMT_MASK, which is the bitmask for the built-in formatting flags.

In encode_value we first call our parent’s encode_value (passing only the relevant bits of the format ) and receive the converted value and output flags back (in reality, output flags will typically be the same as the format flags).

Then we convert the already-converted value and compress it as zlib. We then AND the flag with our FMT_ZLIB bit, and return it. The value and flag returned from the encode_value method are stored as is on the server.

We do the converse when reading data back from the server in decode_value. In this method we are passed the value as it is stored on the server, along with the numeric flags as they are stored in the key’s metadata. We check to see first whether there is any special FMT_ZLIB flag applied, and if so, decompress the data and strip those bits from the flag. Then we dispatch it to the default decode_value to handle any further encapsulation formats.

This can all be used like so from Python:

# assuming the ZlibTrancoder class is defined above

c = Couchbase.connect(transcoder=ZlibTranscoder(), bucket='default')
c.set("foo", b"long value" * 1000, format=FMT_BYTES|FMT_ZLIB)

Bypassing Conversion

If you are having difficulties with reading some value from the server (possibly because it was stored using a different client with different flag semantics) then you can disable conversion (when retrieving) entirely by using the Connection object’s data_passthrough property. This is a Boolean, and when enabled does not reconvert the value (that is, it does not call decode_value but simply interprets the value as a sequence of bytes and returns them as part of the Result object’s value property.

c.set("a_dict", {"foo":"bar"})
c.data_passthrough = True
c.get("a_dict").value == b'{"foo":"bar"}'

Item API

The Item API introduced in version 1.1 introduces a more streamlined way to deal with Couchbase documents and key-value entries.

The Item API represents a persistent object in your application which may be passed by your application to any of the SDK _multi APIs. When this object is passed, the result of the operation is stored within the Item object itself, rather than creating a new result object.

This allows developers to avoid some common boilerplate methods such as converting a Result object to their internal application structure. In fact, the Item object itself is a subclass of ValueResult

All methods accepting Items should be passed a collection of items. These collections must be the special classes of either an ItemOptionDict or an ItemSequence. The former can be used to specify additional arguments for each individual item (such as CAS, format, and the like), while the latter may be used if no specific options are required.

from pprint import pprint

from couchbase import Couchbase, FMT_UTF8
from couchbase.items import Item, ItemSequence, ItemOptionDict

# The Item API allows subclassing.
class MyItem(Item):
    def print_value(self):
        pprint(self.value, indent=4)

cb = Couchbase.connect(bucket='default')

# Create blank item
itm = MyItem()

# Must set a key before we do anything with it
itm.key = "simple_key"

# We'll first store it
itm.value = "Hello World"

# Now, we need a sequence for the item.
itmcoll = ItemSequence([itm])
cb.set_multi(itmcoll, format=FMT_UTF8)


# Let's append some data to it. We use the special 'append_items' method
# However, append requires a 'fragment' option for each item, so we need
# to make 'itmcoll' an ItemOptionsDict

itmcoll = ItemOptionDict()
itmcoll.add(itm, fragment=" ..again")
cb.append_items(itmcoll, format=FMT_UTF8)

# Let's get the key using a normal 'get' this time
print cb.get(itm.key).value


'Hello World'
'Hello World ..again'
Hello World .. again

Persistence/Durability Requirements

Persistence requirements are provided to allow an application developer to ensure that a given mutation operation has been performed and propagated to more than one node in the cluster. By default, operations are considered successful if they are stored in the RAM of a single node. Using the persist_to and replicate_to arguments in the set family of methods you may specify stricted requirements, for example to replicate to at least two nodes, and to have it persisted on at least one node’s disk.

Note that Couchbase will always eventually store each item on all of its associated replicas, in both disk and memory. The persistence requirements are there as a means for the application developer to synchronize his/her code so that the application does not continue until those requirements are met.

See the API reference for more information on how to use these parameters.

Replicas and Data Availability

Sometimes it may be useful to retrieve the value for an item, even if the value may not be the most recent version present on the server. Such a scenario may happen for example when trying to retrieve a “tag cloud” entry or similar, where the presence of any data is more important than no data at all.

Data outages can happen for brief periods of time in between the phyiscal disruption of a node (i.e. the node loses immediate network connectivity) and the period of time the cluster waits for auto-failover (if it is enabled).

If your cluster is configured with support for one or more replicas, then any item stored will typically be available on a replica, even if the original vBucket master is unavailable. However, there is a slight chance that the master stored a newer version of the item immediately before it went down and before it had a chance to replicate the item to the replicas.

To retrieve an item from a replica, simply use the replica boolean parameter in get or get_multi. Typically this is done in a try-except loop, where the try block attempts to fetch a key through normal means, and the except block attempts to use the replica:

    rv = cb.get(key)
except CouchbaseError:
    rv = cb.get(key, replica=True)

There is also the rget method which is able to get an item from a specific replica. See the API documentation for more details

Logging and Debugging

This section explains how to uncover bugs in your application (or in the SDK itself).


To debug anything, you must be able to identify in which domain a problem is found. Specifically, the following components participate in typical Couchbase operations:

  • Couchbase Server

    This is the server itself, which stores your data. Errors can happen here if your data does not exist, or if there are connectivity issues with one or more nodes in the server. Note that while Couchbase Server is scalable and fault tolerant, there are naturally some conditions that would cause failures (for example, if all nodes are unreachable).

  • libcouchbase

    This is the underlying layer that handles network communication and protocol handling between a client and a Couchbase node. Network connectivity issues tend to happen here.

  • Python C Extension Layer

    This is the C code that provides the bulk of the SDK. It interfaces with the libcouchbase component, creates Result objects, performs input validation, and encoding and decoding of keys and values.

  • Python Layer

    This is written in pure python. For simple key-value operations these normally just dispatch to the C layer. Most of the view option and row code is handled here as well, with the C layer just performing the lower level network handling.

Exception Handling

When something goes wrong, an exception of CouchbaseError is typically thrown. The exception object contains a lot of information that can be used to find out what went wrong.

from couchbase import Couchbase
from couchbase.exceptions import CouchbaseError

c = Couchbase.connect(bucket='default')
    # Will fail because 'object' is not JSON-serializable
    c.set("key", object())
except CouchbaseError as e:
    print e

Printing the exception object typically produces something like this:

# line breaks inserted for clarity

<Couldn't encode value,
    inner_cause=<object object at 0x7f873cf220d0> is not JSON serializable,
    C Source=(src/convert.c,131),
    OBJ=<object object at 0x7f873cf220d0>

The exception object consists of the following properties:

  • message

    This is the message, if any, that indicates what went wrong. It is always a string.

    >>> e.message
     "Couldn’t encode value"
  • inner_cause

    If this exception is triggered by another exception, this field contains it. In the above example, we see the exception

    >>> e.inner_cause
     TypeError(‘<object object at 0x7f873cf220d0> is not JSON serializable’,)
  • csrc_info

    If present, contains the source code information where the exception was raised. This is only present for exceptions raised from within the C extension.

    >>> e.csrc_info
     (‘src/convert.c’, 131)
  • objextra

    Contains the Python object that likely caused the exception. If present, it means the object was of an invalid type or format.

    >>> e.objextra
     <object object at 0x7f873cf220d0>

Application Crashes

Because this is a C extension, some fatal errors might result in an application crash. On Unix-based systems, these typically look like this:

python: src/callbacks.c:132: get_common_objects: Assertion `PyDict_Contains((PyObject*)*mres, hkey) == 0' failed.

Or simply:

Segmentation Fault

While the actual cause might be in the application code or in the SDK itself, there is often less information available in debugging it.

The SDK should never crash under normal circumstances, and any application crash ultimately indicates a bug in the SDK itself (invalid user input should result in a Python exception being thrown).

To better help us fix the SDK, a C backtrace is needed. To generate a helpful backtrace, Python must be available with debugging symbols (this can be done by installing python-dbg or python-debuginfo from your distribution. Likewise, libcouchbase itself must also be installed with debugging symbols (this can be done by installing libcouchbase2-dbg or libcouchbase2-debuginfo, depending on your distribution).

You also need gdb (this is also available on any distribution).

When you have the desired debugging symbols, invoke gdb as follows. For this example, we assume python is a Python interpreter, and is a script that can trigger the crash.

shell> gdb --args python
GNU gdb (GDB) 7.4.1-debian
Copyright (C) 2012 Free Software Foundation, Inc.
License GPLv3+: GNU GPL version 3 or later <>
This is free software: you are free to change and redistribute it.
There is NO WARRANTY, to the extent permitted by law.  Type "show copying"
and "show warranty" for details.
This GDB was configured as "x86_64-linux-gnu".
For bug reporting instructions, please see:
Reading symbols from /usr/bin/python...Reading symbols from /usr/lib/debug/usr/bin/python2.7...done.

This brings you to the gdb prompt. Run the program by typing r and then pressing enter.

(gdb) r
Starting program: /usr/bin/python
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib/x86_64-linux-gnu/".
python: src/callbacks.c:132: get_common_objects: Assertion `PyDict_Contains((PyObject*)*mres, hkey) == 0' failed.

Program received signal SIGABRT, Aborted.
0x00007ffff6fc9475 in *__GI_raise (sig=<optimized out>) at ../nptl/sysdeps/unix/sysv/linux/raise.c:64
64      ../nptl/sysdeps/unix/sysv/linux/raise.c: No such file or directory.

Debugging an already-running application

Often in the case of web servers, it is difficult to invoke the script directly. In this case, you need to debug an already-running application. This can be done with gdb by determining the process ID of the already-running process. In this case, you can attach gdb to the running process like so:

shell> gdb -p 29342
(gdb) continue

Once gdb is attached, you can type continue (instead of r ) to continue the application.

This shows us that an application crashed. When this happens, gdb will print the location of the crash. This is not enough, however as we need the full trace of the crash. To do this, type bt and then enter to obtain the trace:

(gdb) bt
#0  0x00007ffff6fc9475 in *__GI_raise (sig=<optimized out>)
    at ../nptl/sysdeps/unix/sysv/linux/raise.c:64
#1  0x00007ffff6fcc6f0 in *__GI_abort () at abort.c:92
#2  0x00007ffff6fc2621 in *__GI___assert_fail (assertion=assertion@entry=
    0x7ffff67f6f68 "PyDict_Contains((PyObject*)*mres, hkey) == 0",
    file=<optimized out>, file@entry=0x7ffff67f6e0d "src/callbacks.c",
    line=line@entry=132, function=function@entry=
    0x7ffff67f6fe0 "get_common_objects") at assert.c:81
#3  0x00007ffff67f000c in get_common_objects (cookie=<optimized out>,
    key=<optimized out>, nkey=<optimized out>, err=err@entry=LCB_KEY_ENOENT,
    conn=conn@entry=0x7fffffffd328, res=res@entry=0x7fffffffd330,
    restype=restype@entry=2, mres=mres@entry=0x7fffffffd338)
    at src/callbacks.c:132
#4  0x00007ffff67f0623 in get_callback (instance=<optimized out>,
    cookie=<optimized out>, err=LCB_KEY_ENOENT, resp=0x7fffffffd3e0)
    at src/callbacks.c:216
#5  0x00007ffff65cf861 in lcb_server_purge_implicit_responses ()
   from /sources/libcouchbase/inst/lib/
#6  0x00007ffff65d0f1b in lcb_proto_parse_single ()
   from /sources/libcouchbase/inst/lib/
#7  0x00007ffff65cfef5 in lcb_server_v0_event_handler ()
   from /sources/libcouchbase/inst/lib/
#8  0x00007ffff58b9ccc in event_base_loop ()
   from /usr/lib/x86_64-linux-gnu/
#9  0x00007ffff65d50f0 in lcb_wait ()
---Type <return> to continue, or q <return> to quit---

Python traces can be rather long; continue pressing enter until the last line ( --Type <return>... ) is no longer present.

After you have a backtrace, send the information (along with the script to reproduce, if possible) to your desired support venue.

You can also debug a crash using Valgrind, but the process is significantly more involved and requires a slightly modified build of Python. See Contributing for more details.


This section describes how you can contribute to this SDK.

General Information

The latest source code for the Python SDK can be found on github. It is located at

If you want to contribute to the C extension itself, it might be worthwhile to use a debug build of Python.

Compiling Python From Source

You can skip this section if you do not intend to contribute to the C part of the SDK.

The instructions here have been tested on Python 2.6.7 and Python 3.2.4. They will likely work for any version of Python.

To generate a debug build of python, you need to compile it from source. To do this, you need to modify some Python source files as instructed in the Misc/README.valgrind file within the Python source distribution.

Additionally, if you want to have your Python be useful for installing other packages (for example, nose ), you need to have pip and distribute installed. These themselves depend on several core modules that might not be built by default on some systems.

The Modules/Setup file can be modified using the following diff as a guidline:

--- ../../tmp/Python-2.6.7/Modules/Setup.dist   2008-11-27 02:15:12.000000000 -0800
+++ Setup.dist  2013-05-15 15:58:30.559170619 -0700
@@ -162,7 +162,7 @@
 # it, depending on your system -- see the GNU readline instructions.
 # It's okay for this to be a shared library, too.

-#readline readline.c -lreadline -ltermcap
+readline readline.c -lreadline -ltermcap

 # Modules that should always be present (non UNIX dependent):
@@ -215,6 +215,7 @@
 #_ssl _ssl.c \
 #      -DUSE_SSL -I$(SSL)/include -I$(SSL)/include/openssl \
 #      -L$(SSL)/lib -lssl -lcrypto
+_ssl _ssl.c -DUSE_SSL -lssl -lcrypto

 # The crypt module is now disabled by default because it breaks builds
 # on many systems (where -lcrypt is needed), e.g. Linux (I believe).
@@ -248,14 +249,14 @@
 # Message-Digest Algorithm, described in RFC 1321.  The necessary files
 # md5.c and md5.h are included here.

-#_md5 md5module.c md5.c
+_md5 md5module.c md5.c

 # The _sha module implements the SHA checksum algorithms.
 # (NIST's Secure Hash Algorithms.)
-#_sha shamodule.c
-#_sha256 sha256module.c
-#_sha512 sha512module.c
+_sha shamodule.c
+_sha256 sha256module.c
+_sha512 sha512module.c

 # SGI IRIX specific modules -- off by default.
@@ -460,7 +461,7 @@
 # Andrew Kuchling's zlib module.
 # This require zlib 1.1.3 (or later).
 # See
-#zlib zlibmodule.c -I$(prefix)/include -L$(exec_prefix)/lib -lz
+zlib zlibmodule.c -I$(prefix)/include -L$(exec_prefix)/lib -lz

 # Interface to the Expat XML parser

Note that on some distributions (specifically Debian) you might get a build failure when building the ssl module. If so, you likely need to modify the Modules/_ssl.c file like so:

--- ../../tmp/Python-2.6.7/Modules/_ssl.c       2010-08-03 11:50:32.000000000 -0700
+++ _ssl.c      2013-05-15 15:58:03.471170217 -0700
@@ -302,8 +302,6 @@
         self->ctx = SSL_CTX_new(TLSv1_method()); /* Set up context */
     else if (proto_version == PY_SSL_VERSION_SSL3)
         self->ctx = SSL_CTX_new(SSLv3_method()); /* Set up context */
-    else if (proto_version == PY_SSL_VERSION_SSL2)
-        self->ctx = SSL_CTX_new(SSLv2_method()); /* Set up context */
     else if (proto_version == PY_SSL_VERSION_SSL23)
         self->ctx = SSL_CTX_new(SSLv23_method()); /* Set up context */

After the source tree is prepared, you can do something like:

shell>./configure --without-pymalloc --prefix=/source/pythons/py267 shell> make install

Running Tests

If you’ve made changes to the library, you need to run the test suite to ensure that nothing broke with your changes.

To run the tests, you need to have the nose package installed (this may also work with the unittest module as well, but is less tested).

Additionally, you need a real cluster to test against. The test might modify the buckets specified, so be sure not to point it to a production server!

Note that the views test might fail if you have made changes to the beer-sample bucket.

To tell the test about your cluster setup, copy the file tests/test.ini.sample to tests/test.ini and modify as needed.

To run the tests, from within the root of the SDK source simply do:

shell> nosetest -v

Building Docs

You will need sphinx and numpydoc installed. Simply do:

shell> make -C docs html After you do this, the built HTML should be in docs/build/html, and you can begin browsing by opening docs/build/html/index.html in your browser.

Source Style Guidelines

For the Python code, a loose adherence to PEP-8 should be used. For the C extension code, a fairly more strict adherence to PEP-7 should be used.

These rules are meant to be broken; this just reflects some guidelines to use.

In general:

  • Use spaces, not tabs.

  • Lines should never be longer than 80 columns.

  • Code should be compatible with Python versions 2.6 up to the latest 3.x.


  • Doc strings should be readable by Sphinx.

  • Methods should not have more than three positional arguments.

  • Avoid using string literals in code.

    If a new object makes use of a dictionary, consider converting this dictionary to a proper Python object by using a namedtuple, and so on.

  • Avoid dependencies not in Python’s standard library.

    You can add conditional functionality depending on whether a specific library is installed.

  • Don’t use threads.

    While threads are a useful construct in application code, they do not belong in library code without good reason.


  • Use of goto is better than deeply nested blocks.

  • Return type and storage specifiers should be on their own line.

    For example, use the following format

    static PyObject*
    do_something(PyObject *self, PyObject *args, …)
     /** … **/

    rather than this format

    static PyObject *do_something(PyObject *self, PyObject *args)
      /** … **/
  • Code must compile with the following flags (for GCC or clang):

    -std=c89 -pedantic -Wall -Wextra -Werror -Wno-long-long -Wno-missing-field-initializers
  • Non-static functions must have a pycbc_ prefix.

  • Functions exposed as Python methods must be named using the following pattern: pycbc_<Object>_<Method>.

    In this pattern, <Object> is the name of the class in the SDK and <Method> is the name of the method. For example, if you add a get method to the Connection class, name the new method pycbc_Connection_get.

  • Code should be portable to Win32.

    Therefore, include only standard library headers and use PyOS_* functions when needed.

Release Notes

The following sections provide release notes for individual release versions of Couchbase Client Library Python. To browse or submit new issues, see Couchbase Client Library Python Issues Tracker.

Release Notes for Couchbase Python SDK 1.2.4 (10 October 2014)

This release provides some minor bug fixes for the 1.2 series.

Fixes in 1.2.4 * The design_publish method no longer deletes the old development mode view when it is published into a production mode view. This is to ensure consistent behavior between the web UI and the SDK.

Issues: PYCBC-259

  • Allow building on Apple’s OS X Python. This adds the /usr/local/ to the linker and preprocessor search paths if the script detects that it is being invoked as /usr/bin/python. Newer versions of OS X no longer automatically add /usr/local as a default search path.

Release Notes for Couchbase Python SDK 1.2.3 GA (2 September 2014)

This release provides some minor bug fixes for the 1.2 series.

Fixes in 1.2.3

  • Fix potential hang in Twisted integration module.

    Issues: PYCBC-257

  • Handle non-CouchbaseError exceptions thrown in callbacks and avoid crashes.

    Sometimes a non-CouchbaseError is thrown within operation callbacks, usually due to a bad encoding or other environmental issues (such as memory allocation failures). This should be delivered to the user and not crash the application.

    Issues: PYCBC-253

  • Fix various error message and documentation errata.

    Issues: PYCBC-254 PYCBC-252

Release Notes for Couchbase Python SDK 1.2.2 GA (1 July 2014)

This release provides some minor bug fixes for 1.2.1

New Features and Behavior Changes in 1.2.2

  • The _cntl() method now accepts a value_type argument. This can be used to further modify settings not exposed in the SDK itself. The value_type describes the underlying C type to be used for the setting.

    Issues: PYCBC-247

  • The libcouchbase library embedded with the Windows installer has been upgraded to 2.3.2

Fixes in 1.2.2

  • When publishing a design document, poll the newly published view after the older view has been deleted. This allows more reliability of the view being operational immediately after the design_publish() method returns.

  • Fixed potential freed memory access within the stats() method when using asynchronous frameworks like Twisted or gevent.

    Issues: PYCBC-248

Release Notes for Couchbase Python SDK 1.2.1 GA (5 June 2014)

This release provides many bug fixes for 1.2.0. This also includes added functionality via libcouchbase 2.3.1. For Windows distributions, the built in libcouchbase has been upgraded to version 2.3.1

New Features and Behavior Changes in 1.2.1

  • Use compact JSON encoding when serializing documents to JSON.

    The default Python json.dumps inserts white space around commas and other JSON tokens, while increasing readability when printing to the screen. This also inflates the value size stored inside the database and the transfer size when sending and receiving the item to and from the cluster. Version 1.2.1 now uses the separators argument to only use the tokens without their white space padding.

    Issues: PYCBC-231

  • Added master-only observe() method to efficiently check for the existence and CAS of an item.

    This feature utilizes functionality from libcouchbase 2.3.x and allows the usage of the observe() method to check for the existence of an item. In contrast to a normal observe that probes the item’s master and replica nodes resulting in multiple packets, the master-only observe contacts the master only. The return value for the observe() method is still the same, so the actual results are present inside the value field, which is an array. Because only one node is contacted there is only ever a single element in the array.

    result = cb.observe("key", master_only=True).value[0]
      print "CAS for item is {0}".format(result.cas)

    This feature requires libcouchbase 2.3.0 or later.

    Issues: PYCBC-225

  • Provide semantic base exception classes.

    This adds support for additional semantically-grouped base exceptions that may act as the superclasses of other exceptions thrown within the library. This means that applications are now able to catch the semantic base class and filter further exception handling based on the specified categories. This also allows the library to return more detailed exception codes in the future (for example, a possible ConnectionRefusedError rather than NetworkError) without breaking application-side expectations and try/except logic. The new base classes include CouchbaseNetworkError (for network-based failures), CouchbaseDataError (for CRUD failures), and CouchbaseTransientError (for errors that are likely to be corrected if the operation is retried).

    This feature requires _ibcouchbase 2.3.0 or later.

    Issues: PYCBC-241

  • Twisted and gevent Integration modules are no longer experimental.

    The inclusion of the couchbase.experimental module is no longer needed to enable the functionality of Twisted and gevent. Existing code importing the experimental module will still function and will still retain the full stability of the Twisted and gevent modules.

    Issues: PYCBC-221

  • New --batch parameter for benchmark script

    This option allows benchmarking using multi operations. If this value is set to more than 1, then n commands are batched.

  • Provide _cntl method to directly manipulate libcouchbase settings.

    This unsupported method sends proxy calls into lcb_cntl() to better help adjust and tune library settings when deemed necessary. This exists mainly as a path to debug and test certain situations, and to provide an upgrade path to newer versions of libcouchbase that contain features not yet directly exposed by this library. As the documentation and the method name suggest, this is not considered a public API and should not be used unless otherwise specified.

    Issues: PYCBC-224

Fixes in 1.2.1

  • Fix build failures when building in a subdirectory of a Git repository.

    Errors would be encountered when building in a subdirectory of a Git repository because the script would invoke git describe to determine the version being used. If the distribution directory was not a Git repository but the parent directory was a Git repository, git describe would output the parent project’s version information. This has been fixed in 1.2.1 where git describe is invoked only if the top-level .git directory exists within the distribution.

    Issues: PYCBC-220

  • Fix hanging in select module when no I/O was pending.

    This applies to the select module that is used when the enable_experiemental_gevent_support option is used (which should not be confused with gcouchbase). A symptom of this issue is that certain events would not be delivered and the application would be blocked.

  • Fix hang in txcouchbase when connection is dropped.

    A write event is delivered to libcouchbase when the connection is dropped, so that it can detect a connection error and close the socket. Previously this would manifest itself as infinitely hanging in the case of connection failures (or excessive timeouts) because events on the socket would no longer be delivered.

  • Correct erroneous conflation of persist_to and replicate_to in set() method.

    This fixes a typo that resulted in the meanings of these two parameters to be inverted, such that persist_to would mean how many nodes to replicate to, and replicate_to would mean how many nodes the item should be persisted on. This issue is manifest in unexpected timeouts.

    Issues: PYCBC-228 PYCBC-242

  • Prevent application core dump when a python exception is thrown in a user-defined transcoder.

    Previously if an exception was thrown in a user defined Transcoder object (for example, because conversion failed) this would not properly be caught by the library and would result in an application crash. This has been fixed and Python exceptions are now wrapped inside a CouchbaseError which is now propagated back to the application

    Issues: PYCBC-232

  • Fix truncated row results and crashes when invoking other methods during iteration over a View object, or when using include_docs.

    This fixes an erroneous assumption that the internal fetch method for the iterator (that is, _fetch()) was the only entry point in which new results would be returned. However if a get() or other Couchbase method was invoked, then any pending socket data would also be delivered to the callback. This has now been corrected to no longer assume that _fetch() is the only entry point. The library now relies on the completion callback from libcouchbase only.

    Issues: PYCBC-223 PYCBC-236

  • Rename conncache parameter to config_cache.

    This renames the constructor parameter for the Configuration Cache feature. The older name is still accepted but is deprecated.

    issues: PYCBC-221

Release Notes for Couchbase Python SDK 1.2.0 GA (7 January 2014)

This release provides support for additional asynchronous interfaces, adds some additional convenience features, and fixes various bugs related to views.

New Features and Behavior Changes in 1.2.0

  • Native gevent support

    This feature adds native gevent support by means of an additional gcouchbase package. This supersedes the gevent support in the 1.1.0 release. Gevent supported versions are 0.13 and 1.0.0

    Issues: PYCBC-207

  • Twisted Support

    This feature adds an asynchronous API for use with the Twisted I/O framework for Python. Support is provided via the txcouchbase package. All known recent versions of Twisted are supported.

    Issues: PYCBC-194

  • Durability Context Manager

    This feature provides a context manager to affect the durability settings for all operations executed in its scope. This can be used with the python with keyword as an alternative to manually specifying the persist_to and replicate_to parameters for each operation individually.

    Issues: PYCBC-201

  • Multioperation Pipeline

    This feature provides a batched operation context manager in which multiple operations of different types can be batched together for execution. This might improve performance as it reduces network latency by sending all scheduled operations after the context manager exits, rather than sending each operation individually. This has the same network performance benefits as using the *_multi methods, but without requiring that all operations be of the same type. The _multi operations have other API and performance benefits and should still be preferred over the pipeline context manager when possible. The pipeline manager itself can also contain _multi operations in its own right.

    Issues: PYCBC-211

Fixes in 1.2.0

  • Streaming view parser sometimes provides invalid JSON

    The streaming view parser sometimes drops the last } from the response, resulting in an error. This affects only query operations where streaming=True.

    Issues: PYCBC-206

  • Views queries malfunction if the mapkey_single or mapkey_multi fields are over 150 characters.

    This was caused by multiple bugs in the library, such as placing an artificial limit on the URI size to 150 characters and placing extra such parameters via a POST request.

    Issues: PYCBC-203 PYCBC-199

  • Durability options missing from delete() method

    The delete method was missing the processing of durability options (e.g. persist_to and replicate_to).

    Issues: PYCBC-195

  • Negative persist_to and replicate_to values not honored

    A negative value should indicate “use all available nodes”, but was not being honored as such in all code paths.

    Issues: PYCBC-200

Release Notes for Couchbase Python SDK 1.1.0 GA (1 October 2013)

This is the second release. This adds API additions and improvements on top of the 1.0.0 release.

New Features and Behavior Changes in 1.1.0

  • Durability requirements

    This feature allows the application to wait until keys have been stored to more than one node. This is done through the additional persist_to and replicate_to parameters in the set() family of methods. A standalone method, endure() is provided as well. These features use the new Durability API implemented in libcouchbase.

    Issues: PYCBC-37

  • Experimental gevent support

    gevent is a cooperative multitasking framework that allows the creation of greenlets, which are cooperatively scheduled entities. This feature allows the couchbase.Connection object to optionally use a different pure-python non-blocking I/O implementation that can cooperatively yield to other greenlets when needed.

  • Item API

    Item API allows an application to use its own objects and pass them into the various methods of couchbase.Connection where they will be populated with the relevant key status/value information. This allows an application developer to create custom subclasses of the Item object which can be seamlessly used across both the SDK and the application code.

    Issues: PYCBC-156

  • Automatic storage formatting via FMT_AUTO

    New format specifier that allows the SDK to guess the appropriate value storage format depending on the value type.

    Issues: PYCBC-157

  • no_format argument for get()

    Allows a value to be retrieved in its raw representation (that is, as it is stored on the server without any conversion applied). This feature is provided as an additional keyword argument, no_format, to the get() family of methods.

  • Replica Read

    Allow reading from a replica server. This can be used if a normal get() fails. This functionality is exposed via the replica parameter to the get() method and the rget() method.

    Issues: PYCBC-38

Fixes In 1.1.0

  • Minor documentation errata fixes

  • Add OBS_LOGICALLY_DELETED, which was referenced but not defined. Used by observe.

  • Fix potential crash if HTTP view request cannot be scheduled

  • Fix view queries with many keys. Previously this would return a server error indicating the URI was too long.

    Issues: PYCBC-193

  • lockmode parameter not propagated from Couchbase.connect constructor. This would result in the lockmode always being set to LOCKMODE_EXC despite it being overridden in the arguments.

    Issues: PYCBC-192

Known Issues in 1.1.0

  • syncwait parameter for view creation will time out if new design document only contains spatial views

    Issues: PYCBC-173

Release Notes for Couchbase Client Library Python 1.0.0 GA (1 July 2013)

This is the first general availability (GA) release.

New Features and Behavior Changes in 1.0.0

  • Enable Transcoder to be a class as well as an instance. Fixes some common misuses. If a class, a new instance is created.

    Issues : PYCBC-135

Fixes in 1.0.0

  • Creating and destroying many Couchbase instances crash program. This happened because of an extra refcount decrement.

    Issues : PYCBC-140

  • Observe crashes when used against clusters with replicas.

    Issues : PYCBC-146

Release Notes for Couchbase Client Library Python 1.0.0 Beta (17 June 2013)

This version builds upon previous APIs. It provides new APIs for querying views and managing design documents.

New Features and Behavior Changes in 1.0.0

  • New query method and paginated view iterator. View querying is now more efficient and streamlined with the new query method. The older _view method has been removed.

    Issues : PYCBC-7, PYCBC-9, PYCBC-27, PYCBC-41, PYCBC-83

  • Low-level observe command wraps the libcouchbase implementation.

    Issues : PYCBC-36

  • The Connection object can now be used safely across multiple threads by default. Tuning the behavior among multiple threads can be changed by the lockmode constructor option.

    Issues : PYCBC-125

  • An API to change default JSON and Pickle converters is provided. This allows the user to select a more performant implementation. Note that this affects only the JSON and Pickle conversion performed with key-value operations, not with views.

    Issues : PYCBC-124

  • A new set of design document methods provides simple methods for creating, fetching, and removing design documents. These replace the older _design method.

    Issues : PYCBC-64, PYCBC-68

Fixes in 1.0.0

  • The quiet parameter was not being used if passed in the constructor.

    Issues : PYCBC-136

  • Multi-arithmetic ( incr_multi, decr_multi ) does not change the value. These functions appeared to work and did not return an error, but did not actually modify the value.

    Issues : PYCBC-138

  • unlock_multi does not raise an exception on missing CAS. This has been fixed because unlock must have a CAS.

    Issues : PYCBC-

  • Crash when specifying host:port in a single string. This is still illegal, but it now raises an exception and does not crash.

    Issues : PYCBC-130

  • A more informative message is returned if an empty string is passed to get. Previously, it returned InvalidArgument without more information. The client now checks for this and displays more helpful information.

    Issues : PYCBC-131

  • Client crashes if an invalid hostname is passed. Certain variants of bad hostnames (such as those with illegal characters) will have libcouchbase return an error code the client cannot handle and will thus crash the program.

    Issues : PYCBC-128

  • Lock without TTL fails with an erroneous ‘Encoding Error’. The client now throws a more informative exception.

    Issues : PYCBC-132

  • Client crashes if duplicate keys are passed in *_multi methods.

    Issues : PYCBC-134

Release Notes for Couchbase Client Library Python 0.11.1 Beta (29 May 2013)

This version maintains the 0.10.0 API, while adding some new features and fixing some additional bugs.

New Features and Behavior Changes in 0.11.1

  • Basic HTTP/Views support (experimental): view and design methods are now available.

    Issues : PYCBC-105

  • Add bucket attribute to show bucket name.

    Issues : PYCBC-119

  • Allow Python 3 dict_keys objects to be passed into the multi_* methods.

    Issues : PYCBC-107

  • Add touch command.

    Issues : PYCBC-115

  • Add unlock and lock commands.

    Issues : PYCBC-61, PYCBC-104

  • User-defined transcoder classes can be used to override or supplement the default serialization of keys and values.

    Issues : PYCBC-92

  • Expose list of nodes

    Issues : PYCBC-106

  • Pretty print repr and str for common objects.

    Issues : PYCBC-112, PYCBC-118

Fixes in 0.11.1

  • Passing a negative time to live (TTL) now throws an exception. Previously, this resulted in an integer underflow.

    Issues : PYCBC-109

  • Memory leak on each storage operation.

    Issues : PYCBC-123

  • JSON format is now more efficient for Unicode values. Previously, JSON would be encoded in ASCII-safe mode. 0.11 allows encoding in UTF-8 mode.

    Issues : PYCBC-108

  • Memory leak on some exceptions. If bad arguments were received, the exception handling code would leak memory.

    Issues : PYCBC-111

Release Notes for Couchbase Client Library Python 0.10.0 Beta (16 May 2013)

This version rewrites the client again, still using libcouchbase, but this time using the native Python C API. Specifically this means:

  • Cython is no longer required for the build.

  • Code works on Visual Studio compilers (and can be used on Microsoft Windows).

Additionally, the API was changed with respect to return values. Currently most API functions return a Result (or a subclass thereof).

The rewrite is filed as PYCBC-103.

Known Issues in 0.10.0

  • Support for byte values.

    Issues : PYCBC-103

  • Formal specification for key types - keytypes should currently be a Python Unicode object.

    Issues : PYCBC-66, PYCBC-91

  • Client-side timeout can be set via the timeout property.

    Issues : PYCBC-58

Release Notes for Couchbase Client Library Python 0.9.0 Beta (29 April 2013)

This version is the initial rewrite of the Python library using libcouchbase. The installation now requires a C compiler and libcouchbase.

New Features and Behavior Changes in 0.9.0

  • Basic key-value functionality, including:

    • get

    • set

    • add

    • replace

    • delete

    • append

    • prepend

    • incr

    • decr

Fixes in 0.9.0

Known Issues in 0.9.0

  • Version 0.9.0 blocks all Python threads.

Release Notes for Couchbase Client Library Python 0.8.0 Beta (1 September 2012)

Known Issues in 0.8.0

  • View queries on authenticated buckets are not currently supported.

  • “id” values from view rows must be converted to strings to be used with Memcached API.

    view = bucket.view("design/beer/view/by_name")
     for row in view:
      id = row["id"].str()
      beer = bucket.get(id)
      #do something
  • Exception is thrown on key not found errors with unified client.


Release Notes for Couchbase Client Library Python 0.7.2 Beta (August 2012)

Fixes in 0.7.2

  • Install with PyPi failing due to missing version number, now fixed.

    Issues : PYCBC-51

  • Install dependencies via rather than including them in the source distribution

    Issues : PYCBC-52

  • Improve bucket creation defaults:

    • ramQuotaMB is now set to 100 by default

    • authType is set to sasl by default

    • Several assets are included for better error catching and reporting on invalid combinations or values.

    • Server-side validation is now being done to check for port duplicates, name taken, and similar checks.

    Issues : PYCBC-53

Release Notes for Couchbase Client Library Python 0.7.1 Beta (6 August 2012)

This is the latest release of the Couchbase Python SDK. It is written from the ground up based on the Couchbase C library, libcouchbase.

This release is considered beta software, use it at your own risk; let us know if you run into any problems, so we can fix them.

New Features and Behavior Changes in 0.7.1

  • SDK now installable via python install from source or via pip install couchbase.

Fixes in 0.7.1

  • Temporarily removing unimplemented multi-get until full implementation available. This will be re-addressed in PYCBC-49 in a future release

    Issues : PYCBC-49, PYCBC-49

Release Notes for Couchbase Client Library Python 0.7.0 Beta (6 August 2012)

This is the latest release of the Couchbase Python SDK. It is written from the ground up based on the Couchbase C library, libcouchbase.

This release is considered beta software, use it at your own risk; let us know if you run into any problems, so we can fix them.

New Features and Behavior Changes in 0.7.0

  • SDK now requires Python 2.6.

  • SDK can now handle server restarts/warmups. Can handle functioning Couchbase Server that is loading data from disk after restart.

  • Introduced VBucketAwareClient which extends MemcachedClient with Membase/Couchbase specific features.

Fixes in 0.7.0

  • Deprecated Server() in favor of Couchbase() for the unified client name

  • Added Apache License headers to all files

  • Globally, logging is no longer disabled; fixes PYCBC-31.

    Issues : PYCBC-31

  • Renamed VBucketAwareCouchbaseClient to CouchbaseClient.

  • Set() now works with integer values; fixes PYCBC-15.

    Issues : PYCBC-15

  • Added memcached level flush() command to unify client with other SDKs. Please note this only works with 1.8.0 without changing settings. See the release notes for Couchbase 1.8.1 and 2.0.0 for how to enable memcached flush().

    This operation is deprecated as of the 1.8.1 Couchbase Server, to prevent accidental, detrimental data loss. Use of this operation should be done only with extreme caution, and most likely only for test databases as it will delete, item by item, every persisted record as well as destroy all cached data.

    Third-party client testing tools may perform a flush_all() operation as part of their test scripts. Be aware of the scripts run by your testing tools and avoid triggering these test cases/operations unless you are certain they are being performed on your sample/test database.

    Inadvertent use of flush_all() on production databases, or other data stores you intend to use will result in permanent loss of data. Moreover the operation as applied to a large data store will take many hours to remove persisted records.

  • Deprecating getMulti for pep8-compliant multi-get.

  • method; fixes MB-5609.

    Issues : MB-5609, MB-5609

  • Deprecated get_view as it was a duplicate of view_results.

  • SDK now working with mixed clusters, including clusters with memcached type buckets.

  • Can now create memcached buckets

  • Set() now returns a proper status in the unified Couchbase() client 0.7.0.

  • Better handling of topology changes; fixes PYCBC-4.

    Issues : PYCBC-4, PYCBC-4

  • init_cluster function has been removed.

  • SDK can now created memcached buckets.

  • Greater than 50% of SDK covered by unit tests; fixes PYCBC-46.

    Issues : PYCBC-46