Running Selenium on AWS Lambda


About a year and a half ago or so, AWS announced support for custom Lambda images. This unlocks a number of new advantages - namely the ability to create images that have complex install requiremnents prior to runtime. A perfect example of this is running Selenium on a Lambda to screenshot a website in-browser. One can quickly imagine the use cases (and related conveniences) of a Lambda-based Selenium screenshotting tool. In my case, I sought to take multiple screenshots of a mapping tool at multiple geographic locales every N minutes in perpetuity for QA purposes.


While AWS suggests that defining a local image and running a container to test your logic locally is straightforward (as in this blog post), it’s not in practice. In my case, I was able to build an image from a Dockerfile with Selenium and Chrome and, while it would run fine locally, it would fail to either initialize the chromedriver or would have the browser itself crash when running on AWS Lambda infrastructure.

Lesons learned

Two key lessons I learned from this effort:

  1. Chrome running headless simply does not appear to work at this time on Lambda - always use Chromium. Chrome attempts to initialize user accounts and profiles which cannot be created or written to as the AWS Lambda file system is read only. This creates a series of cascading issues (I suspect) that leads Chrome to close unexpectedly / in an unhandled manner. Thus, chromedriver is unable to initialize successfully and Selenium exits early without being able to actually get the target website.

  2. Installing the Chrome or Chromium browser via wget or curl versus apt-get seems to cause issues with how the binary is stored and made available in PATH. Even when the location of the browser binary was specified, there remained issues with browser initialization or getting the chromedriver to play nice with the browser. Ultimately, installing via apt-get circumvented these issues and resulted in a clean install and, in conjunction with 1 above, allowed successful headless browser operation within the Lambda infrastructure.

In both of the above cases, the Docker container ran fine with the AWS Runtime Interface Emulator locally - I was able to simulate Lambda invocations and they would complete successfully within the container. In such cases, I would then deploy to AWS infrastructure only to experience random chromedriver initialization failures during invocation. This was a deeply frustrating and time-consuming process.

Selenium initialization

The settings I used for Lambda use of Selenium with Chromium via the options class are as shown:

from import Options

options = Options()

# other parameters for running headless in Lambda I used

# INFO: update if binary location needs to be set
# options.binary_location = "/other/location"

Note the final line in the above code block. The binary location setting is what could be adjusted if one were to try and install the browser and driver manually as described above. However, I remained unsuccessful at this and ultimately went with apt-get to install the driver and browser.

Also note that the user-data-dir parameter was set to the tmp directory. Because Lambda has a read only file system, we need to ensure that any write activity is scoped specifically to the tmp directory which is the sole location that allows intermediate writes during a Lambda’s runtime (not preserved). Also keep this in mind when saving your screen captures to local during a Lambda invocation. Saved images can then be moved from tmp to S3 for future retrieval or use.

Initialization of the Selenium WebDriver occured by incorporating those setting options alongside the driver location:

# install location as a result of apt-get install
driver_path = "/usr/lib/chromium-browser/chromedriver"

# return this initialized driver


Without further ado, here’s the Dockerfile that I found successfully installed Selenium, Chromium, and the related Chromedriver:

FROM ubuntu:18.04

SHELL ["/bin/bash", "-c"]

RUN apt update && apt-get install -y software-properties-common
RUN apt update && add-apt-repository ppa:deadsnakes/ppa
RUN apt update && apt-get install -y \
    make \
    curl \
    python3.7 \
    python3.7-distutils \
    g++ \
    cmake \
    unzip \
    libcurl4-openssl-dev \
RUN apt-get update && apt-get install -y \
    fonts-liberation libappindicator3-1 libasound2 libatk-bridge2.0-0 \
    libnspr4 libnss3 lsb-release xdg-utils libxss1 libdbus-glib-1-2 \
    autoconf cmake curl libtool unzip wget \

# install chromedriver and google-chrome

RUN apt update && apt-get install -y chromium-browser chromium-chromedriver

# install amazon RIE for lambda testing

RUN curl -L -o aws-lambda-rie-x86_64 && \
    mv aws-lambda-rie-x86_64 /usr/local/bin/aws-lambda-rie && \
    chmod +x /usr/local/bin/aws-lambda-rie

# install pip and set up virtualenv

RUN curl -o /tmp/ && \
    python3.7 /tmp/ && \
    python3.7 -m pip install virtualenv

# generate working directory locations

RUN mkdir -p /code
ENV PATH="${PATH}:/code:/usr/lib"

COPY requirements.txt /code/requirements.txt
COPY /code/
COPY lib/ /code/lib/

RUN make install

ENTRYPOINT [ "sh", "/code/" ]

The Makefile install method simply installs the available requirements and creates a virtual environment to operate within:

    virtualenv --python=python3.7 venv

install: venv
    source venv/bin/activate &&\
    pip install -r requirements.txt

The similarly is adapted the suggested pattern from AWS Lambda’s docs which allows for a conditional RIE entrance that runs in “local” mode if certain OS environment variables are not present, otherwise it executes with the expectation that it is in AWS Lambda infrastructure:

if [ -z "${AWS_LAMBDA_RUNTIME_API}" ]; then
  exec /usr/local/bin/aws-lambda-rie /code/venv/bin/python -m awslambdaric lib.snapshot_lambdas.handler
  exec /code/venv/bin/python -m awslambdaric lib.snapshot_lambdas.handler

At this point, you should be able to create a .py file (in the above case, snapshot_lambdas) that contains a handler for receiving Lambda invocations (or simulated ones in local run cases).

Good luck screen capturing on Lambda!