Running a Bokeh server



To make this guide easier to follow, consider familiarizing yourself with some of the core concepts of Bokeh in the section Defining Key Concepts.

Bokeh server makes it easy to create interactive web applications that connect front-end UI events to running Python code.

Bokeh creates high-level Python models, such as plots, ranges, axes, and glyphs, and then converts these objects to JSON to pass them to its client library, BokehJS. For more information on the latter, see BokehJS.

This flexible and decoupled design offers some advantages. For instance, it is easy to have other languages, such as R or Scala, drive Bokeh plots and visualizations in the browser.

However, keeping these models in sync between the Python environment and the browser would provide further powerful capabilities:

  • respond to UI and tool events in the browser with computations or queries using the full power of Python

  • automatically push server-side updates to the UI elements such as widgets or plots in the browser

  • use periodic, timeout, and asynchronous callbacks to drive streaming updates

This is where the Bokeh server comes into play:

The primary purpose of the Bokeh server is to synchronize data between the underlying Python environment and the BokehJS library running in the browser.

Here’s a simple example from that illustrates this behavior.

Manipulating the UI controls communicates new values to the backend via Bokeh server. This also triggers callbacks that update the plots with the input in real time.

Use case scenarios

Consider a few different scenarios when you might want to use the Bokeh server.

Local or individual use

You might want to use the Bokeh server for exploratory data analysis, possibly in a Jupyter notebook, or for a small app that you and your colleagues can run locally.

The Bokeh server is very convenient here, allowing for quick and simple deployment through effective use of Bokeh server applications. For more detail, see Building Bokeh applications.

Creating deployable applications

You might also want to use the Bokeh server to publish interactive data visualizations and applications to a wider audience, say, on the internet or an internal company network. The Bokeh server also suits this usage well, but you might want to first consult the following:

Shared publishing

Both of the scenarios above involve one person making applications on the server, either for personal use or for consumption by a larger audience.

While it is possible for several people to publish different applications to the same server, this does not make for a good use case because hosted applications can execute arbitrary Python code. This raises process isolation and security concerns and makes this kind of shared tenancy prohibitive.

One way to support this kind of multi-application environment with multiple users is to build up infrastructure that can run a Bokeh server for each app or at least for each user. The Bokeh project or a third party might create a public service for this kind of usage in the future but such developments are beyond the scope this documentation.

Another possibility is to have one app that can access data and other artifacts published by many different people, possibly with access controls. This sort of scenario is possible with the Bokeh server, but often involves integrating it with other web application frameworks.

Building Bokeh applications

By far the most flexible way to create interactive data visualizations with the Bokeh server is to create Bokeh applications and serve them with the bokeh serve command. The Bokeh server then uses the application code to create sessions and documents for all connecting browsers.


The Bokeh server (left) uses the application code to create Bokeh documents. Every new connection from a browser (right) results in the server creating a new document just for that session.

The Bokeh server executes the application code with every new connection and creates a new Bokeh document, syncing it to the browser. The application code also sets up the callbacks that should run whenever properties, such as widget values, change.

You can provide the application code in several ways.

Single module format

Consider the following complete example.


from random import random

from bokeh.layouts import column
from bokeh.models import Button
from bokeh.palettes import RdYlBu3
from bokeh.plotting import figure, curdoc

# create a plot and style its properties
p = figure(x_range=(0, 100), y_range=(0, 100), toolbar_location=None)
p.border_fill_color = 'black'
p.background_fill_color = 'black'
p.outline_line_color = None
p.grid.grid_line_color = None

# add a text renderer to the plot (no data yet)
r = p.text(x=[], y=[], text=[], text_color=[], text_font_size="26px",
           text_baseline="middle", text_align="center")

i = 0

ds = r.data_source

# create a callback that adds a number in a random location
def callback():
    global i

    # BEST PRACTICE --- update .data in one step with a new dict
    new_data = dict()
    new_data['x'] =['x'] + [random()*70 + 15]
    new_data['y'] =['y'] + [random()*70 + 15]
    new_data['text_color'] =['text_color'] + [RdYlBu3[i%3]]
    new_data['text'] =['text'] + [str(i)] = new_data

    i = i + 1

# add a button widget and configure with the call back
button = Button(label="Press Me")

# put the button and plot in a layout and add to the document
curdoc().add_root(column(button, p))

The code above doesn’t specify any output or connection method. It is a simple script that creates and updates objects. The bokeh command line tool lets you specify output options after processing your data. You could, for example, run bokeh json to get a JSON-serialized version of the app. However, to run the app on a Bokeh server, use the following command:

bokeh serve --show

The --show option will cause your default browser to open a new tab at the address of the running application, which in this case is:


If you have only one application, the server root will redirect to it. Otherwise, you will see an index of all applications running on the server root:


You can disable this index with the --disable-index option. Likewise, you can disable redirecting with the --disable-index-redirect option.

In addition to creating Bokeh applications from single Python files, you can also create applications from directories.

Directory format

You can create Bokeh apps by creating and populating a filesystem directory with application files. To start an application in a directory named myapp, you could execute bokeh serve as follows:

bokeh serve --show myapp

This directory must contain a file that constructs a document for the Bokeh server to serve:


The following is the directory app structure that the Bokeh server is familiar with:


Some of the files and subdirectories above are optional.

  • An file that marks this directory as a package. You can make imports relative to the package, such as from . import mymod and from .mymod import func.

  • A file that lets you declare an optional function to process HTTP requests and return a dictionary of items that the session token includes as described in Request handler hooks.

  • A file that lets you trigger optional callbacks at different stages of application execution as described in Lifecycle hooks and Request handler hooks.

  • A static subdirectory that you can use to serve static resources associated with this application.

  • A theme.yaml file where you can declare default attributes for Bokeh to apply to model types.

  • A templates subdirectory with an index.html Jinja template file. The directory may contain additional Jinja templates for index.html to refer to. The template should have the same parameters as the FILE template. For more information, see Customizing the application’s Jinja template.

When executing your, the Bokeh server ensures that the standard __file__ module attribute works as you would expect. So you can include data files or custom user-defined models in your directory however you like.

Bokeh also adds the application directory sys.path to facilitate importing of Python modules in the application directory. However, if an is in the directory, you can use the app as a package as well as make standard package-relative imports.

Here’s an example of a more developed directory tree:

   |    +---things.csv
   |    +---custom.js
   |    +---css
   |    |    +---special.css
   |    |
   |    +---images
   |    |    +---foo.png
   |    |    +---bar.png
   |    |
   |    +---js
   |        +---special.js
   |    +---index.html

In this case, your code might be similar to the following:

from os.path import dirname, join
from .helpers import load_data

load_data(join(dirname(__file__), 'data', 'things.csv'))

The code to load a JavaScript implementation for a custom model from models/custom.js is also similar.

Customizing the application’s Jinja template

The Directory format section mentions that you can override the default Jinja template, which the Bokeh server uses to generate user-facing HTML.

This lets you use CSS and JavaScript to tweak the way the application appears in the browser.

For more details on how Jinja templating works, see the Jinja project documentation.

Embedding figures in the template

To reference a Bokeh figure in the templated code, you need to set its name attribute and add the figure to the current document root in the main thread of your Bokeh app, that is

from bokeh.plotting import curdoc

# templates can refer to a configured name value
plot = figure(name="bokeh_jinja_figure")


You can then use that name in the corresponding Jinja template to reference the figure via the roots template parameter as follows:

{% extends base %}

{% block contents %}
    {{ embed(roots.bokeh_jinja_figure) }}
{% endblock %}

Defining custom variables

You can pass custom variables to the template with the curdoc().template_variables dictionary as follows:

# set a new single key/value pair
curdoc().template_variables["user_id"] = user_id

# or update multiple pairs at once
curdoc().template_variables.update(first_name="Mary", last_name="Jones")

You can then reference the variables in the corresponding Jinja template.

{% extends base %}

{% block contents %}
    <p> Hello {{ user_id }}, AKA '{{ last_name }}, {{ first_name }}'! </p>
{% endblock %}

Accessing HTTP requests

When creating a session for an application, Bokeh makes the session context available as curdoc().session_context. The most useful function of the session context is to make the Tornado HTTP request object available to the application as session_context.request. HTTP requests are not available directly because of an incompatibility with --num-procs. Instead, only the arguments attribute is available in full and only a subset of cookies and headers allowed by the --include-headers, --exclude-headers, --include-cookies, and --exclude-cookies parameters is available. Attempting to access any other attribute on a request results in an error.

You can enable additional request attributes as described in Request handler hooks.

The following code accesses the request arguments to provide a value for the variable N that could, for example, control the number of plot points.

# request.arguments is a dict that maps argument names to lists of strings,
# for example, the query string ?N=10 results in {'N': [b'10']}

args = curdoc().session_context.request.arguments

  N = int(args.get('N')[0])
  N = 200


The request object makes inspecting values, such as arguments, easy. However, calling any of the Tornado methods, such as finish(), or writing directly to request.connection is unsupported and results in undefined behavior.

Request handler hooks

To provide additional information where full Tornado HTTP requests may not be available, you can define a custom handler hook.

To do so, create an app in directory format and include a file called in the directory. This file must include a process_request function.

def process_request(request):
    '''If present, this function executes when an HTTP request arrives.'''
    return {}

The process then passes Tornado HTTP requests to the handler, which returns a dictionary for curdoc().session_context.token_payload. This lets you work around some of the --num-procs issues and provide additional information.

Callbacks and events

Before jumping into callbacks and events specifically in the context of the Bokeh server, it’s worth discussing different use cases for callbacks in general.

JavaScript callbacks in the browser

Whether you are using the Bokeh server or not, you can create callbacks that execute in the browser with CustomJS and other methods. For more information and examples, see JavaScript callbacks.

CustomJS callbacks never execute Python code, not even if you convert a Python callback into JavaScript. CustomJS callbacks only execute inside the browser’s JavaScript interpreter, which means that they can only interact with JavaScript data and functions, such as BokehJS models.

Python callbacks with Jupyter interactors

When working with Jupyter notebooks, you can use Jupyter interactors to quickly create simple GUI forms. Updates to GUI widgets trigger Python callbacks that execute in the Python kernel of Jupyter. It is often useful to have these callbacks call push_notebook() to push updates to displayed plots. For more information, see Jupyter interactors.


You can push plot updates from Python to BokehJS with push_notebook(). For two-way communication, embed a Bokeh server in the notebook. For example, this lets range and selection updates trigger Python callbacks. For further details, see examples/howto/server_embed/notebook_embed.ipynb

Updating from threads

You can make blocking computations in separate threads. However, you must schedule document updates via a next tick callback. This callback executes as soon as possible with the next iteration of the Tornado event loop and automatically acquires necessary locks to safely update the document state.


The ONLY safe operations to perform on a document from a different thread are add_next_tick_callback() and remove_next_tick_callback()

Remember, direct updates to the document state issuing from another thread, whether through other document methods or setting of Bokeh model properties, risk data and protocol corruption.

To allow all threads access to the same document, save a local copy of curdoc(). The example below illustrates this process.

from functools import partial
from random import random
from threading import Thread
import time

from bokeh.models import ColumnDataSource
from bokeh.plotting import curdoc, figure

from tornado import gen

# only modify from a Bokeh session callback
source = ColumnDataSource(data=dict(x=[0], y=[0]))

# This is important! Save curdoc() to make sure all threads
# see the same document.
doc = curdoc()

def update(x, y):[x], y=[y]))

def blocking_task():
    while True:
        # do some blocking computation
        x, y = random(), random()

        # but update the document from a callback
        doc.add_next_tick_callback(partial(update, x=x, y=y))

p = figure(x_range=[0, 1], y_range=[0,1])
l ='x', y='y', source=source)


thread = Thread(target=blocking_task)

To see this example in action, save the above code to a Python file, for example,, and then execute the following command:

bokeh serve --show


There is currently no locking around adding next tick callbacks to documents. Bokeh should have a more fine-grained locking for callback methods in the future, but for now it is best to have each thread add no more than one callback to the document.

Updating from unlocked callbacks

Normally Bokeh session callbacks recursively lock the document until all future work they initiate is completed. However, you may want to drive blocking computations from callbacks using Tornado’s ThreadPoolExecutor in an asynchronous callback. This requires that you use the without_document_lock() decorator to suppress the normal locking behavior.

As with the thread example above, all actions that update document state must go through a next tick callback.

The following example demonstrates an application that drives a blocking computation from one unlocked Bokeh session callback. It yields to a blocking function that runs on the thread pool executor and then updates with a next tick callback. The example also updates the state simply from a standard locked session callback with a different update rate.

from functools import partial
import time

from concurrent.futures import ThreadPoolExecutor
from tornado import gen

from bokeh.document import without_document_lock
from bokeh.models import ColumnDataSource
from bokeh.plotting import curdoc, figure

source = ColumnDataSource(data=dict(x=[0], y=[0], color=["blue"]))

i = 0

doc = curdoc()

executor = ThreadPoolExecutor(max_workers=2)

def blocking_task(i):
    return i

# the unlocked callback uses this locked callback to safely update
def locked_update(i):[['x'][-1]+1], y=[i], color=["blue"]))

# this unlocked callback will not prevent other session callbacks from
# executing while it is running
def unlocked_task():
    global i
    i += 1
    res = yield executor.submit(blocking_task, i)
    doc.add_next_tick_callback(partial(locked_update, i=res))

def update():[['x'][-1]+1], y=[i], color=["red"]))

p = figure(x_range=[0, 100], y_range=[0,20])
l ='x', y='y', color='color', source=source)

doc.add_periodic_callback(unlocked_task, 1000)
doc.add_periodic_callback(update, 200)

As before, you can run this example by saving to a Python file and running bokeh serve on it.

Lifecycle hooks

You may want to execute code at specific points of server or session runtime. For instance, if you are using a Bokeh server with a Django server, you need to call django.setup() for each Bokeh server to properly initialize Django for use by Bokeh application code.

Bokeh enables this through a set of lifecycle hooks. To use these hooks, create your application in directory format and include a designated file called in the directory. In this file you can include any or all of the following conventionally named functions:

def on_server_loaded(server_context):
    # If present, this function executes when the server starts.

def on_server_unloaded(server_context):
    # If present, this function executes when the server shuts down.

def on_session_created(session_context):
    # If present, this function executes when the server creates a session.

def on_session_destroyed(session_context):
    # If present, this function executes when the server closes a session.

You can also define on_session_destroyed lifecycle hooks directly on the Document being served. This makes it easy to clean up after a user closes a session by performing such actions as database connection shutdown without the need to bundle a separate file. To declare such a callback, define a function and register it with the Document.on_session_destroyed method:

doc = Document()

def cleanup_session(session_context):
    # This function executes when the user closes the session.


Besides the lifecycle hooks above, you may also define request hooks to access the HTTP requests your users make. For further information, see Request handler hooks.

Embedding Bokeh server as a library

It can be useful to embed the Bokeh Server in a larger Tornado application, or a Jupyter notebook, and use the already existing Tornado IOloop. Here is the basis for integration of Bokeh in such a scenario:

from bokeh.server.server import Server

server = Server(
    bokeh_applications,  # list of Bokeh applications
    io_loop=loop,        # Tornado IOLoop
    **server_kwargs      # port, num_procs, etc.

# start timers and services and immediately return

You can also create and control an IOLoop directly. This can be useful when creating standalone “normal” Python scripts that serve Bokeh apps or embedding a Bokeh application in a framework like Flask or Django without having to run a separate Bokeh server process. You can find some examples of this technique in the examples directory:

Also note that every command line argument for bokeh serve has a corresponding keyword argument for Server. For instance, using the --allow-websocket-origin command line argument is equivalent to passing allow_websocket_origin as a parameter.

Connecting with bokeh.client

You can directly interact with the Bokeh server via a client API, which you can use to make modifications to Bokeh documents in existing sessions on a Bokeh server.


Typically, web browsers connect to the Bokeh server, but you can make a connection from Python by using the bokeh.client module.

This can be useful, for example, to make user-specific customizations to a Bokeh app that is embedded by another web framework, such as Flask or Django. In the following example, a Flask endpoint embeds a “sliders” app already running on the server but changes the plot title before passing the output to the user.

from flask import Flask, render_template

from bokeh.client import pull_session
from bokeh.embed import server_session

app = Flask(__name__)

@app.route('/', methods=['GET'])
def bkapp_page():

    with pull_session(url="http://localhost:5006/sliders") as session:

        # update or customize that session
        session.document.roots[0].children[1].title.text = "Special sliders for a specific user!"

        # generate a script to load the customized session
        script = server_session(, url='http://localhost:5006/sliders')

        # use the script in the rendered page
        return render_template("embed.html", script=script, template="Flask")

if __name__ == '__main__':

Deployment scenarios

To make your application into a user-friendly service, you’ll have to deploy your work. This subsection explores various aspects of deployment.

Standalone Bokeh server

You can have the Bokeh server running on a network for users to interact with your app directly. This can be a simple solution for local network deployment, provided the capabilities of the hardware running the server match your app requirements and the expected number of users.

However, if you have authentication, scaling, or uptime requirements, you’ll have to consider more sophisticated deployment configurations.

SSH tunnels

To run a standalone instance of the Bokeh server on a host with restricted access, use SSH to “tunnel” to the server.

In the simplest scenario, the user accesses the Bokeh server from another location, such as a laptop with no intermediary machines.

Run the server as usual on the remote host.

bokeh server

Next, issue the following command on the local machine to establish an SSH tunnel to the remote host:

ssh -NfL localhost:5006:localhost:5006

Replace user with your username on the remote host and with the hostname or IP address of the system hosting the Bokeh server. The remote system may prompt you for login credentials. Once you are connected, you will be able to navigate to localhost:5006 as though the Bokeh server were running on the local machine.

A slightly more complicated scenario involves a gateway between the server and the local machine. In that situation, a reverse tunnel must be established from the server to the gateway with another tunnel connecting the gateway with the local machine.

Issue the following commands on the remote host where the Bokeh server will be running:

nohup bokeh server &
ssh -NfR 5006:localhost:5006

Replace user with your username on the gateway and with the hostname or IP address of the gateway. The gateway may prompt you for login credentials.

To setup the tunnel between the local machine and the gateway, run the following command on the local machine:

ssh -NfL localhost:5006:localhost:5006

Again, replace user with your username on the gateway and with the hostname or IP address of the gateway.

You should now be able to access the Bokeh server from the local machine by navigating to localhost:5006. You can even set up client connections from a Jupyter notebook running on the local machine.


We intend to expand this section with more guidance for other tools and configurations. If you have experience with other web deployment scenarios and wish to contribute your knowledge here, please contact us on

SSL termination

You can configure the Bokeh server to terminate SSL connections and serve secure HTTPS and WSS sessions directly. To do so, you’ll have to supply the --ssl-certfile argument with the value of the path to a single PEM file containing a certificate as well as any number of CA certificates needed to establish the certificate’s authenticity.

bokeh serve --ssl-certfile /path/to/cert.pem

You can also supply a path to the certificate file by setting the environment variable BOKEH_SSL_CERTFILE.

If the private key is stored separately, you can supply its location by setting the --ssl-keyfile command line argument or by setting the BOKEH_SSL_KEYFILE environment variable. If the private key requires a password, supply it by setting the BOKEH_SSL_PASSWORD environment variable.

Alternatively, you may wish to run a Bokeh server behind a proxy and have the proxy terminate SSL connections. See the next subsection for details.

Basic reverse proxy setup

To serve a web application to the general internet, you may wish to host your app on an internal network and proxy connections to it through some dedicated HTTP server. This subsection provides guidance on how to configure some common reverse proxies.


One very common HTTP and reverse-proxying server is Nginx. Here’s an example of a server configuration stanza:

server {
    listen 80 default_server;
    server_name _;

    access_log  /tmp/bokeh.access.log;
    error_log   /tmp/bokeh.error.log debug;

    location / {
        proxy_set_header Upgrade $http_upgrade;
        proxy_set_header Connection "upgrade";
        proxy_http_version 1.1;
        proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
        proxy_set_header Host $host:$server_port;
        proxy_buffering off;


The above server block sets up Nginx to proxy incoming connections to on port 80 over to on port 5100. To work in this configuration, you need to use some of the command line options to configure the Bokeh server. In particular, use --port to have the Bokeh server listen on port 5100.

bokeh serve --port 5100

The basic server block above does not configure any special handling for static resources, such as Bokeh JS and CSS files. This means that the Bokeh server serves these files directly.

Although this is a viable option, it requires that the Bokeh server do extra work that is better handled with Nginx. To serve static assets with Nginx, add the following sub-block to the code above, substituting the path to your static assets for /path/to/bokeh/server/static:

location /static {
    alias /path/to/bokeh/server/static;

Make sure that the account running Nginx has permissions to access Bokeh resources. Alternatively, you can copy the resources to a global static directory during the deployment.

To communicate cookies and headers across processes, Bokeh may include this information in a JSON web token, sending it via a WebSocket. In certain cases this token can grow very large causing Nginx to drop the request. You may have to work around this by overriding the default Nginx setting large_client_header_buffers:

large_client_header_buffers 4 24k;


Another common HTTP server and proxy is Apache. Here is an example configuration for a Bokeh server running behind Apache:

<VirtualHost *:80>
    ServerName localhost

    CustomLog "/path/to/logs/access_log" combined
    ErrorLog "/path/to/logs/error_log"

    ProxyPreserveHost On
    ProxyPass /myapp/ws ws://
    ProxyPassReverse /myapp/ws ws://

    ProxyPass /myapp
    ProxyPassReverse /myapp

    <Directory />
        Require all granted
        Options -Indexes

    Alias /static /path/to/bokeh/server/static
    <Directory /path/to/bokeh/server/static>
        # directives to effect the static directory
        Options +Indexes


The above configuration aliases /static to the location of the Bokeh static resources directory. However, it is also possible (and probably preferable) to copy the static resources to whatever standard location for static files you configure for Apache as part of the deployment.

You may also need to enable some modules for the above configuration:

a2enmod proxy
a2enmod proxy_http
a2enmod proxy_wstunnel
apache2ctl restart

Depending on your system, you may have to use sudo to run the above.

As before, run the Bokeh server with the following command:

bokeh serve --port 5100

Reverse proxying with Nginx and SSL

To deploy a Bokeh server behind an SSL-terminated Nginx proxy, you’ll need a few additional customizations. In particular, you’ll have to configure the Bokeh server with the --use-xheaders flag.

bokeh serve --port 5100 --use-xheaders

The --use-xheaders flag causes Bokeh to override the remote IP and URI scheme/protocol for all requests with X-Real-Ip, X-Forwarded-For, X-Scheme, and X-Forwarded-Proto headers when they are available.

You’ll also have to customize Nginx. In particular, you have to configure Nginx to send X-Forwarded-Proto headers and use SSL termination. Optionally, you may want to redirect all HTTP traffic to HTTPS.

The complete details of this configuration, such as how and where to install SSL certificates and keys, varies by platform and the following is only a reference nginx.conf setup:

# redirect HTTP traffic to HTTPS (optional)
server {
    listen      80;
    return      301 https://$server_name$request_uri;

server {
    listen      443 default_server;

    # adds Strict-Transport-Security to prevent man-in-the-middle attacks
    add_header Strict-Transport-Security "max-age=31536000";

    ssl on;

    # SSL installation details vary by platform
    ssl_certificate /etc/ssl/certs/my-ssl-bundle.crt;
    ssl_certificate_key /etc/ssl/private/my_ssl.key;

    # enables all versions of TLS, but not the deprecated SSLv2 or v3
    ssl_protocols TLSv1 TLSv1.1 TLSv1.2;

    # disables all weak ciphers

    ssl_prefer_server_ciphers on;

    location / {
        proxy_set_header Upgrade $http_upgrade;
        proxy_set_header Connection "upgrade";
        proxy_http_version 1.1;
        proxy_set_header X-Forwarded-Proto $scheme;
        proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
        proxy_set_header Host $host:$server_port;
        proxy_buffering off;


This configuration will proxy all incoming HTTPS connections to over to a Bokeh server running internally on

Load balancing with Nginx

The Bokeh server is scalable by design. If you need more capacity, you can simply run additional servers. In this case, you’ll generally want to run all the Bokeh server instances behind a load balancer so that new connections are distributed among individual servers.


The Bokeh server is horizontally scalable. To add more capacity, you can run more servers behind a load balancer.

Nginx can help with load balancing. This section describes some of the basics of one possible configuration, but please also refer to the Nginx load balancer documentation. For instance, there are different strategies available for choosing what server to connect to next.

First, you need to add an upstream stanza to the Nginx configuration. This typically goes above the server stanza and looks something like the following:

upstream myapp {
    least_conn;                 # Use the least-connected strategy
    server;      # Bokeh Server 0
    server;      # Bokeh Server 1
    server;      # Bokeh Server 2
    server;      # Bokeh Server 3
    server;      # Bokeh Server 4
    server;      # Bokeh Server 5

The rest of the configuration uses the name myapp to refer to the above upstream stanza, which lists the internal connection information for six different Bokeh server instances (each running on a different port). You can run and list as many Bokeh servers as you need.

To run a Bokeh server instance, use commands similar to the following:

serve --port 5100
serve --port 5101

Next, in the location stanza for the Bokeh server, change the proxy_pass value to refer to the upstream stanza above. The code below uses proxy_pass http://myapp;.

server {

    location / {
        proxy_pass http://myapp;

        # all other settings unchanged
        proxy_set_header Upgrade $http_upgrade;
        proxy_set_header Connection "upgrade";
        proxy_http_version 1.1;
        proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
        proxy_set_header Host $host:$server_port;
        proxy_buffering off;



The Bokeh server itself does not have any facilities for authentication or authorization. However, you can configure the Bokeh server with an “auth provider” that hooks into Tornado’s underlying capabilities. For background information, see the Tornado docs for Authentication and security. The rest of this section assumes some familiarity with that material.

Auth module

You can configure the Bokeh server to only allow authenticated users to connect. To do so, provide a path to the module that implements the necessary functions on the command line.

bokeh serve --auth-module=/path/to/

Alternatively, you can set the BOKEH_AUTH_MODULE environment variable to this path.

The module must contain one of the following two functions that return the current user (or None):

def get_user(request_handler):

async def get_user_async(request_handler):

The module passes the function to the Tornado RequestHandler that can inspect cookies or request headers to determine the authenticated user. If there is no authenticated user, these functions should return None.

Additionally, the module must specify where to redirect unauthenticated users by including either:

  • a module attribute login_url and (optionally) a LoginHandler class

  • a function definition for get_login_url

login_url = "..."

class LoginHandler(RequestHandler):

def get_login_url(request_handler):

If the module provides a relative login_url, it can also provide an optional LoginHandler class, which the Bokeh server will incorporate automatically.

The get_login_url function is useful in cases where the login URL must vary based on the request, cookies, or other factors. You can also specify a LoginHandler when defining the get_url_function.

To define an endpoint for logging users out, you can also use optional logout_url and LogoutHandler parameters, similar to the login options.

If you don’t provide an authentication module, the configuration will not require any authentication to access Bokeh server endpoints.


The configuration executes the contents of the authentication module.

Secure cookies

If you want to use Tornado’s set_secure_cookie and get_secure_cookie functions in your auth module, you’ll have to set a cookie secret. To do so, use the BOKEH_COOKIE_SECRET environment variable.

export BOKEH_COOKIE_SECRET=<cookie secret value>

The value should be a long, random sequence of bytes.


By default, the Bokeh server will accept any incoming connections with an allowed WebSocket origin. If you specify a session ID, and a session with that ID already exists on the server, the server will connect to that session. Otherwise, the server will automatically create and use a new session.

If you are deploying an embedded Bokeh app within a large organization or to the wider internet, you may want to limit who can initiate sessions, and from where. Bokeh lets you manage session creation privileges.

WebSocket origin

When a Bokeh server receives an HTTP request, it immediately returns a script that initiates a WebSocket connection. All subsequent communication happens over the WebSocket.

To reduce the risk of cross-site misuse, the Bokeh server will only initiate WebSocket connections from the origins that are explicitly allowed. Requests with Origin headers that are not on the allowed list will generate HTTP 403 error responses.

By default, only localhost:5006 is allowed, making the following two invocations identical:

bokeh serve --show


bokeh serve --show --allow-websocket-origin=localhost:5006

Both of these open your default browser to the default application URL localhost:5006 and, since localhost:5006 is on the list of allowed WebSocket origins, the Bokeh server creates and displays a new session.

When you embed a Bokeh server in another web page with server_document() or server_session(), the Origin header for the request to the Bokeh server is the URL of the page that hosts your Bokeh content.

For example, if a user navigates to your page at, the origin header reported by the browser will be In this case, you’d typically restrict the Bokeh server to honoring only the requests that originate from the page, preventing other pages from embedding your Bokeh app without your knowledge.

You can do so by setting the --allow-websocket-origin command line argument as follows:

bokeh serve --show

This will prevent other sites from embedding your Bokeh application in their pages because requests from users viewing those pages will report a different origin than, causing the Bokeh server to reject them.


Bear in mind that this only prevents other web pages from embedding your Bokeh app without your knowledge.

If you require multiple allowed origins, you can pass multiple instances of --allow-websocket-origin on the command line.

You can also configure the Bokeh server to allow all connections regardless of origin:

bokeh serve --show --allow-websocket-origin='*'

This option is only suitable for testing, experimentation, and local notebook usage.

Signed session IDs

By default, the Bokeh server will automatically create new sessions for all new requests from allowed WebSocket origins, even if you provide no session ID.

When embedding a Bokeh app inside another web application, such as Flask or Django, make sure that only your web application is capable of generating viable requests to the Bokeh server, which you can configure to only create sessions with a cryptographically signed session ID.

First, use the bokeh secret command to create a secret to sign session IDs.

export BOKEH_SECRET_KEY=`bokeh secret`

Then set BOKEH_SIGN_SESSIONS to yes when starting the Bokeh server. You’ll typically also want to set the allowed WebSocket origin at this point.

BOKEH_SIGN_SESSIONS=yes bokeh serve

Then, in your web application, explicitly provide signed session IDs with generate_session_id:

from bokeh.util.token import generate_session_id

script = server_session(url='http://localhost:5006/bkapp',
return render_template("embed.html", script=script, template="Flask")

Make sure to set identical BOKEH_SECRET_KEY environment variables both for the Bokeh server and for the web app processes, such as Flask, Django, or any other tool you are using.


Signed session IDs serve as access tokens. As with any token system, security is predicated on keeping the token secret. You should also run the Bokeh server behind a proxy that terminates SSL connections, or configure the Bokeh server to terminate SSL directly. This lets you securely transmit session IDs to the client browsers.

XSRF cookies

Bokeh server can use Tornado’s cross-site request forgery protection. To turn this feature on, use the --enable-xsrf-cookies option or set the environment variable BOKEH_XSRF_COOKIES to yes.

With this setting, you’ll have to properly instrument all PUT, POST, and DELETE operations on custom and login handlers in order for them to function. Typically, this means adding the following code to all HTML form submission templates:

{% module xsrf_form_html() %}

For full details, see the Tornado documentation on XSRF Cookies.

Scaling the server

You can fork multiple server processes with the num-procs option. For example, run the following command to fork 3 processes:

bokeh serve --num-procs 3

Note that the forking operation happens in the underlying Tornado server. For further information, see the Tornado docs.

Further reading

Now that you are familiar with the concepts of running a Bokeh server, you may be interested in learning more about the internals of the Bokeh server in Server architecture.