The project is made possible by volunteer contributors who have put in thousands of hours of their own time, and made the source code freely available under the Apache License 2.0.
๐ Grid 4 is under development and is a Release Candidate (RC)
Prereleases are happening on a regular basis to get early feedback. This means that all other Selenium components can be currently at a different RC or beta version (e.g. bindings on RC 1, and Docker images on prerelease RC 2).
Docker images for Grid 4 come with a handful of tags to simplify its usage, have a look at them in one of our releases
To get notifications of new prereleases, add yourself as a "Releases only" watcher.
Doubts? Questions? Get in touch through the different communication channels available in the Community section.
Looking for Grid 3? Head to the Selenium 3 branch. This branch will be having new browser releases until Grid 4 has had its major release.
Do you need help to use these Docker images? All the contact points for the different Selenium projects can be seen at: https://www.selenium.dev/support/
- Start a Docker container with Firefox
$ docker run -d -p 4444:4444 -p 7900:7900 --shm-size="2g" selenium/standalone-firefox:4.0.0-rc-2-prerelease-20210923
-
Point your WebDriver tests to http://localhost:4444 *
-
That's it!
-
(Optional) To see what is happening inside the container, head to http://localhost:7900 (password is
secret
).
- Grid 3 used "/wd/hub", while it should also work, it is no longer required
More details about visualising the container activity, check the Debugging section.
โ๏ธ When executing docker run
for an image that contains a browser please use
the flag --shm-size=2g
to use the host's shared memory.
โ๏ธ Always use a Docker image with a full tag to pin a specific browser and Grid version. See Tagging Conventions for details.
$ docker run -d -p 4444:4444 --shm-size="2g" selenium/standalone-firefox:4.0.0-rc-2-prerelease-20210923
$ docker run -d -p 4444:4444 --shm-size="2g" selenium/standalone-chrome:4.0.0-rc-2-prerelease-20210923
$ docker run -d -p 4444:4444 --shm-size="2g" selenium/standalone-edge:4.0.0-rc-2-prerelease-20210923
Note: Only one Standalone container can run on port 4444
at the same time.
There are different ways to run the images and create a Grid with a Hub and Nodes, check the following options.
The Hub and Nodes will be created in the same network and they will recognize each other by their container name. A Docker network needs to be created as a first step.
$ docker network create grid
$ docker run -d -p 4442-4444:4442-4444 --net grid --name selenium-hub selenium/hub:4.0.0-rc-2-prerelease-20210923
$ docker run -d --net grid -e SE_EVENT_BUS_HOST=selenium-hub \
--shm-size="2g" \
-e SE_EVENT_BUS_PUBLISH_PORT=4442 \
-e SE_EVENT_BUS_SUBSCRIBE_PORT=4443 \
selenium/node-chrome:4.0.0-rc-2-prerelease-20210923
$ docker run -d --net grid -e SE_EVENT_BUS_HOST=selenium-hub \
--shm-size="2g" \
-e SE_EVENT_BUS_PUBLISH_PORT=4442 \
-e SE_EVENT_BUS_SUBSCRIBE_PORT=4443 \
selenium/node-edge:4.0.0-rc-2-prerelease-20210923
$ docker run -d --net grid -e SE_EVENT_BUS_HOST=selenium-hub \
--shm-size="2g" \
-e SE_EVENT_BUS_PUBLISH_PORT=4442 \
-e SE_EVENT_BUS_SUBSCRIBE_PORT=4443 \
selenium/node-firefox:4.0.0-rc-2-prerelease-20210923
When you are done using the Grid, and the containers have exited, the network can be removed with the following command:
# Removes the grid network
$ docker network rm grid
The Hub and Nodes will be created on different machines/VMs, they need to know each other's IPs to communicate properly.
Hub - Machine/VM 1
$ docker run -d -p 4442-4444:4442-4444 --name selenium-hub selenium/hub:4.0.0-rc-2-prerelease-20210923
Node Chrome - Machine/VM 2
$ docker run -d -p 5555:5555 \
--shm-size="2g" \
-e SE_EVENT_BUS_HOST=<ip-from-machine-1> \
-e SE_EVENT_BUS_PUBLISH_PORT=4442 \
-e SE_EVENT_BUS_SUBSCRIBE_PORT=4443 \
-e SE_NODE_HOST=<ip-from-machine-2> \
selenium/node-chrome:4.0.0-rc-2-prerelease-20210923
Node Edge - Machine/VM 3
$ docker run -d -p 5555:5555 \
--shm-size="2g" \
-e SE_EVENT_BUS_HOST=<ip-from-machine-1> \
-e SE_EVENT_BUS_PUBLISH_PORT=4442 \
-e SE_EVENT_BUS_SUBSCRIBE_PORT=4443 \
-e SE_NODE_HOST=<ip-from-machine-3> \
selenium/node-edge:4.0.0-rc-2-prerelease-20210923
Node Firefox - Machine/VM 4
$ docker run -d -p 5555:5555 \
--shm-size="2g" \
-e SE_EVENT_BUS_HOST=<ip-from-machine-1> \
-e SE_EVENT_BUS_PUBLISH_PORT=4442 \
-e SE_EVENT_BUS_SUBSCRIBE_PORT=4443 \
-e SE_NODE_HOST=<ip-from-machine-4> \
selenium/node-firefox:4.0.0-rc-2-prerelease-20210923
Docker Compose is the simplest way to start a Grid. Use the linked resources below, save them locally, and check the execution instructions on top of each file.
To stop the Grid and cleanup the created containers, run docker-compose down
.
It is possible to start a Selenium Grid with all its components apart. For simplicity, only an example with docker-compose will be provided. Save the file locally, and check the execution instructions on top of it.
docker-compose-v3-full-grid.yml
Tests execution can be recorded by using the selenium/video:ffmpeg-4.3.1-20210923
Docker image. One container is needed per each container where a browser is running. This means if you are
running 5 Nodes/Standalone containers, you will need 5 video containers, the mapping is 1-1.
Currently, the only way to do this mapping is manually (either starting the containers manually, or through
docker-compose
). We are iterating on this process and probably this setup will be more simple in the future.
The video Docker image we provide is based on the ffmpeg Ubuntu image provided by the jrottenberg/ffmpeg project, thank you for providing this image and simplifying our work ๐
Notes:
- If you have questions or feedback, please use the community contact points shown here.
- Please report any bugs through GitHub issues, and provide all the information requested on the template.
- Video recording for headless browsers is not supported.
- Video recording tends to use considerable amounts of CPU. Normally you should estimate 1CPU per video container, and 1 CPU per browser container.
- Videos are stored in the
/videos
directory inside the video container. Map a local directory to get the videos. - If you are running more than one video container, be sure to overwrite the video file name through the
FILE_NAME
environment variable to avoid unexpected results.
This example shows how to start the containers manually:
$ docker network create grid
$ docker run -d -p 4444:4444 -p 6900:5900 --net grid --name selenium --shm-size="2g" selenium/standalone-chrome:4.0.0-rc-2-prerelease-20210923
$ docker run -d --net grid --name video -v /tmp/videos:/videos selenium/video:ffmpeg-4.3.1-20210923
# Run your tests
$ docker stop video && docker rm video
$ docker stop selenium && docker rm selenium
After the containers are stopped and removed, you should see a video file on your machine's /tmp/videos
directory.
Here is an example using a Hub and a few Nodes:
Grid 4 has the ability to start Docker containers on demand, this means that it starts a Docker container in the background for each new session request, the test gets executed there, and when the test completes, the container gets thrown away.
This execution mode can be used either in the Standalone or Node roles. The "dynamic" execution mode needs to be told what Docker images to use when the containers get started. Additionally, the Grid needs to know the URI of the Docker daemon.
You can save this file locally and name it, for example, config.toml
.
[docker]
# Configs have a mapping between the Docker image to use and the capabilities that need to be matched to
# start a container with the given image.
configs = [
"selenium/standalone-firefox:4.0.0-rc-2-prerelease-20210923", "{\"browserName\": \"firefox\"}",
"selenium/standalone-chrome:4.0.0-rc-2-prerelease-20210923", "{\"browserName\": \"chrome\"}",
"selenium/standalone-edge:4.0.0-rc-2-prerelease-20210923", "{\"browserName\": \"MicrosoftEdge\"}"
]
# URL for connecting to the docker daemon
# host.docker.internal works for macOS and Windows.
# Linux could use --net=host in the `docker run` instruction or 172.17.0.1 in the URI below.
# To have Docker listening through tcp on macOS, install socat and run the following command
# socat -4 TCP-LISTEN:2375,fork UNIX-CONNECT:/var/run/docker.sock
url = "http://host.docker.internal:2375"
# Docker imagee used for video recording
video-image = "selenium/video:ffmpeg-4.3.1-20210923"
# Uncomment the following section if you are running the node on a separate VM
# Fill out the placeholders with appropriate values
#[server]
#host = <ip-from-node-machine>
#port = <port-from-node-machine>
This can be expanded to a full Grid deployment, all components deployed individually. The overall idea is to have the Hub in one virtual machine, and each of the Nodes in separate and more powerful virtual machines.
$ docker network create grid
$ docker run -d -p 4442-4444:4442-4444 --net grid --name selenium-hub selenium/hub:4.0.0-rc-2-prerelease-20210923
$ docker run -d --net grid -e SE_EVENT_BUS_HOST=selenium-hub \
-e SE_EVENT_BUS_PUBLISH_PORT=4442 \
-e SE_EVENT_BUS_SUBSCRIBE_PORT=4443 \
-v ${PWD}/config.toml:/opt/bin/config.toml \
-v /path/on/your/host/machine:/opt/selenium/assets \
selenium/node-docker:4.0.0-rc-2-prerelease-20210923
To have the assets saved on your host, please mount your host path to /opt/selenium/assets
.
When you are done using the Grid, and the containers have exited, the network can be removed with the following command:
# Removes the grid network
$ docker network rm grid
docker run --rm --name selenium-docker -p 4444:4444 \
-v ${PWD}/config.toml:/opt/bin/config.toml \
-v /path/on/your/host/machine:/opt/selenium/assets \
selenium/standalone-docker:4.0.0-rc-2-prerelease-20210923
To have the assets saved on your host, please mount your host path to /opt/selenium/assets
.
Here is an example using a Hub and a Node:
docker-compose-v3-dynamic-grid.yml
To record your WebDriver session, you need to add a se:options
section to
your capabilities and inside it, configure the desired settings, for example:
{
"browserName": "firefox",
"platformName": "linux",
"se:recordVideo": "true",
"se:timeZone": "US/Pacific",
"se:screenResolution": "1920x1080"
}
After running a test, check the path you mounted to the Docker container,
(/path/on/your/host/machine
), and you should see videos and session information.
Here are the steps to deploy the Grid 4 to a Kubernetes cluster.
# Deploying all the grid components to kubernetes
$ kubectl apply -f k8s-deployment-full-grid.yaml
# Exposing the router
$ kubectl expose deployment selenium-router-deployment --type=NodePort --port=4444
# Get the router URL to access the grid from outside K8s cluster
$ minikube service selenium-router-deployment --url
# To list all the Grid componenets
$ kubectl get all -l component=selenium-grid-4
Check out the Kubernetes examples on how to deploy selenium hub and nodes on a Kubernetes cluster.
You can pass SE_OPTS
variable with additional commandline parameters for starting a hub or a node.
$ docker run -d -p 4444:4444 -e SE_OPTS="--log-level FINE" --name selenium-hub selenium/hub:4.0.0-rc-2-prerelease-20210923
You can pass JAVA_OPTS
environment variable to java process.
$ docker run -d -p 4444:4444 -e JAVA_OPTS=-Xmx512m --name selenium-hub selenium/hub:4.0.0-rc-2-prerelease-20210923
The Nodes register themselves through the Event Bus. When the Grid is started in its typical Hub/Node setup, the Hub will be the one acting as the Event Bus, and when the Grid is started with all its five elements apart, the Event Bus will be running on its own.
In both cases, it is necessary to tell the Node where the Event Bus is, so it can register itself. That is
the purpose of the SE_EVENT_BUS_HOST
, SE_EVENT_BUS_PUBLISH_PORT
and SE_EVENT_BUS_SUBSCRIBE_PORT
environment
variables.
Here is an example with the default values of these environment variables:
$ docker run -d -e SE_EVENT_BUS_HOST=<event_bus_ip|event_bus_name> -e SE_EVENT_BUS_PUBLISH_PORT=4442 -e SE_EVENT_BUS_SUBSCRIBE_PORT=4443 --shm-size="2g" selenium/node-chrome:4.0.0-rc-2-prerelease-20210923
By default, nodes start with a screen resolution of 1360 x 1020 with a color depth of 24 bits and a dpi of 96.
These settings can be adjusted by specifying SCREEN_WIDTH
, SCREEN_HEIGHT
, SCREEN_DEPTH
, and/or SCREEN_DPI
environmental variables when starting the container.
docker run -d -e SCREEN_WIDTH=1366 -e SCREEN_HEIGHT=768 -e SCREEN_DEPTH=24 -e SCREEN_DPI=74 selenium/standalone-firefox
In some use cases you might need to set the Grid url to the Node, for example if you'd like to access the CDP endpoint. You
can do that through the SE_NODE_GRID_URL
environment variable. Setting this env var is needed if you want to see the live
view while sessions are executing.
Grid has a default session timeout of 300 seconds, where the session can be on a stale state until it is killed. You can use
SE_NODE_SESSION_TIMEOUT
to overwrite that value in seconds.
A new session request is placed in the Session Queue before it is processed, and the request sits in the queue until a matching slot is found across the registered Nodes. However, the new session request might timeout if no slot was found. By default, a request will stay in the queue up to 300 seconds before it a timeout is reached. In addition, an attempt to process the request is done every 5 seconds (by default).
It is possible to override those values through environment variables in the Hub and the SessionQueue (SE_SESSION_REQUEST_TIMEOUT
and SE_SESSION_RETRY_INTERVAL
). For example, a timeout of 500 seconds would be SE_SESSION_REQUEST_TIMEOUT=500
and a retry
interval of 2 seconds would be SE_SESSION_RETRY_INTERVAL=2
.
By default, only one session is configured to run per container through the SE_NODE_MAX_SESSIONS
environment variable. It is
possible to increase that number up to the maximum available processors, this is because more stability is achieved when one
container/browser has 1 CPU to run.
However, if you have measured performance and based on that, you think more sessions can be executed in each container, you can
override the maximum limit by setting both SE_NODE_MAX_SESSIONS
to a desired number and SE_NODE_OVERRIDE_MAX_SESSIONS
to
true
. Nevertheless, running more browser sessions than the available processors is not recommended since you will be overloading
the resources.
Overriding this setting has a undesired side effect when video recording is enabled, since more than one browser session might be captured in the same video.
Firefox, Chrome, When using headless mode, there's no need for the Xvfb server to be started.
To avoid starting the server you can set the START_XVFB
environment variable to false
(or any other value than true
), for example:
$ docker run -d --net grid -e SE_EVENT_BUS_HOST=selenium-hub -e SE_EVENT_BUS_PUBLISH_PORT=4442 -e SE_EVENT_BUS_SUBSCRIBE_PORT=4443 -e START_XVFB=false --shm-size="2g" selenium/node-chrome
For more information, see this GitHub issue.
Clone the repo and from the project directory root you can build everything by running:
$ VERSION=local make build
If you need to configure environment variable in order to build the image (http proxy for instance),
simply set an environment variable BUILD_ARGS
that contains the additional variables to pass to the
docker context (this will only work with docker >= 1.9)
$ BUILD_ARGS="--build-arg http_proxy=http://acme:3128 --build-arg https_proxy=http://acme:3128" make build
Note: Omitting VERSION=local
will build the images with the released version but replacing the date for the
current one.
It is a good practice to check first if the Grid is up and ready to receive requests, this can be done by checking the /wd/hub/status
endpoint.
A Grid that is ready, composed by a hub and two nodes, could look like this:
{
"value": {
"ready": true,
"message": "Selenium Grid ready.",
"nodes": [
{
"id": "6c0a2c59-7e99-469d-bbfc-313dc638797c",
"uri": "http:\u002f\u002f172.19.0.3:5555",
"maxSessions": 4,
"stereotypes": [
{
"capabilities": {
"browserName": "firefox"
},
"count": 4
}
],
"sessions": [
]
},
{
"id": "26af3363-a0d8-4bd6-a854-2c7497ed64a4",
"uri": "http:\u002f\u002f172.19.0.4:5555",
"maxSessions": 4,
"stereotypes": [
{
"capabilities": {
"browserName": "chrome"
},
"count": 4
}
],
"sessions": [
]
}
]
}
}
The "ready": true
value indicates that the Grid is ready to receive requests. This status can be polled through a
script before running any test, or it can be added as a HEALTHCHECK
when the docker container is started.
Adding a HEALTHCHECK to the Grid
The script check-grid.sh, which is included in the images, can be used to poll the Grid status.
This example checks the status of the Grid every 15 seconds, it has a timeout of 30 seconds when the check is done,
and it retries up to 5 times until the container is marked as unhealthy. Please use adjusted values to fit your needs,
(if needed) replace the --host
and --port
parameters for the ones used in your environment.
$ docker network create grid
$ docker run -d -p 4444:4444 --net grid --name selenium-hub \
--health-cmd='/opt/bin/check-grid.sh --host 0.0.0.0 --port 4444' \
--health-interval=15s --health-timeout=30s --health-retries=5 \
selenium/hub:4.0.0-rc-2-prerelease-20210923
$ docker run -d --net grid -e HUB_HOST=selenium-hub --shm-size="2g" selenium/node-chrome:4.0.0-rc-2-prerelease-20210923
$ docker run -d --net grid -e HUB_HOST=selenium-hub --shm-size="2g" selenium/node-edge:4.0.0-rc-2-prerelease-20210923
$ docker run -d --net grid -e HUB_HOST=selenium-hub --shm-size="2g" selenium/node-firefox:4.0.0-rc-2-prerelease-20210923
Note: The \
line delimiter won't work on Windows based terminals, try either ^
or a backtick.
The container health status can be checked by doing docker ps
and verifying the (healthy)|(unhealthy)
status or by
inspecting it in the following way:
$ docker inspect --format='{{json .State.Health.Status}}' selenium-hub
"healthy"
A common problem known in docker is that a running container does not always mean that the application inside it is ready. A simple way to tackle this is by using a "wait-for-it" script, more information can be seen here.
The following script is an example of how this can be done using bash, but the same principle applies if you want to do this with the programming language used to write the tests.
#!/bin/bash
# wait-for-grid.sh
set -e
cmd="$@"
while ! curl -sSL "http://localhost:4444/wd/hub/status" 2>&1 \
| jq -r '.value.ready' 2>&1 | grep "true" >/dev/null; do
echo 'Waiting for the Grid'
sleep 1
done
>&2 echo "Selenium Grid is up - executing tests"
exec $cmd
Will require
jq
installed viaapt-get
, else the script will keep printingWaiting
without completing the execution.
Note: If needed, replace localhost
and 4444
for the correct values in your environment. Also, this script is polling indefinitely, you might want
to tweak it and establish a timeout.
Let's say that the normal command to execute your tests is mvn clean test
. Here is a way to use the above script and execute your tests:
$ ./wait-for-grid.sh mvn clean test
Like this, the script will poll until the Grid is ready, and then your tests will start.
This project uses x11vnc as VNC server to allow users inspect what is happening inside the container. Users can connect to this server in two ways:
The VNC server is listening to port 5900, you can use a VNC client and connect to it. Feel free to map port 5900 to any free external port that you wish.
The internal 5900 port remains the same because that is the configured port for the VNC server running inside the container.
You can override it with the VNC_PORT
environment variable in case you want to use --net=host
.
Here is an example with the standalone images, the same concept applies to the node images.
$ docker run -d -p 4444:4444 -p 5900:5900 --shm-size="2g" selenium/standalone-chrome:4.0.0-rc-2-prerelease-20210923
$ docker run -d -p 4444:4444 -p 5901:5900 --shm-size="2g" selenium/standalone-edge:4.0.0-rc-2-prerelease-20210923
$ docker run -d -p 4445:4444 -p 5902:5900 --shm-size="2g" selenium/standalone-firefox:4.0.0-rc-2-prerelease-20210923
Then, you would use in your VNC client:
- Port 5900 to connect to the Chrome container
- Port 5901 to connect to the Edge container
- Port 5902 to connect to the Firefox container
If you get a prompt asking for a password, it is: secret
. If you wish to change this, you should either change
it in the /NodeBase/Dockerfile
and build the images yourself, or you can define a Docker image that derives from
the posted ones which reconfigures it:
#FROM selenium/node-chrome:4.0.0-rc-2-prerelease-20210923
#FROM selenium/node-edge:4.0.0-rc-2-prerelease-20210923
#FROM selenium/node-firefox:4.0.0-rc-2-prerelease-20210923
#Choose the FROM statement that works for you.
RUN x11vnc -storepasswd <your-password-here> /home/seluser/.vnc/passwd
If you want to run VNC without password authentication you can set the environment variable VNC_NO_PASSWORD=1
.
If you want to run VNC in view-only mode you can set the environment variable VNC_VIEW_ONLY=1
.
This project uses noVNC to allow users inspect visually container activity with their browser. This might come handy if you cannot install a VNC client on your machine. Port 7900 is used to start noVNC, so you will need to connect to that port with your browser.
Similarly to the previous section, feel free to map port 7900 to any free external port that you wish.
You can also override it with the NO_VNC_PORT
environment variable in case you want to use --net=host
.
Here is an example with the standalone images, the same concept applies to the node images.
$ docker run -d -p 4444:4444 -p 7900:7900 --shm-size="2g" selenium/standalone-chrome:4.0.0-rc-2-prerelease-20210923
$ docker run -d -p 4444:4444 -p 7901:7900 --shm-size="2g" selenium/standalone-edge:4.0.0-rc-2-prerelease-20210923
$ docker run -d -p 4445:4444 -p 7902:7900 --shm-size="2g" selenium/standalone-firefox:4.0.0-rc-2-prerelease-20210923
Then, you would use in your browser:
- http://localhost:7900/ to connect to the Chrome container
- http://localhost:7901/ to connect to the Edge container
- http://localhost:7902/ to connect to the Firefox container
If you get a prompt asking for a password, it is: secret
.
All output gets sent to stdout, so it can be inspected by running:
$ docker logs -f <container-id|container-name>
You can increase the log output by passing environment variable to the containers:
SE_OPTS="--log-level FINE"
Why is --shm-size 2g
necessary?
This is a known workaround to avoid the browser crashing inside a docker container, here are the documented issues for Chrome and Firefox. The shm size of 2gb is arbitrary but known to work well, your specific use case might need a different value, it is recommended to tune this value according to your needs.
If you see the following selenium exceptions:
Message: invalid argument: can't kill an exited process
or
Message: unknown error: Chrome failed to start: exited abnormally
The reason might be that you've set the START_XVFB
environment variable to "false", but forgot to
actually run Firefox, Chrome or Edge in headless mode.