Autoscaling selenium grid on kubernetes with video recording

[prashanth-volvocars]

Description

Autoscale selenium browser nodes running in kubernetes based on the request pending in session queue using KEDA. It also includes ability to automatically record videos using ffmpeg and capture network, browser logs.
The recorded videos and logs are stored under <session_id> name and can be uploaded directly to S3.

Motivation and Context

Auto scaling selenium grid was a problem that was pending to be solved for long. So i took it up when there was a requirement at current my work place. KEDA seemed to the best candidate for the Job and i wrote a new scalar for Selenium Grid a year ago. I would like to have this enabled by default in our charts so everyone could use it.

Types of changes

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to change)

Checklist

• I have read the contributing document.
• My change requires a change to the documentation.
• I have updated the documentation accordingly.
• I have added tests to cover my changes.
• All new and existing tests passed.

[CLAassistant]

All committers have signed the CLA.

[jamesmortensen]

Thank you for submitting the pull request regarding issue #1688 Please note it may take us some time to find someone who can review the changes.

[prashanth-volvocars]

Sure thanks @jamesmortensen for the swift response. Hope to see this getting merged. Also there is a similar issue in the selenium project SeleniumHQ/selenium#9845 on the same topic

[mhnaeem]

@prashanth-volvocars Thank you for this work, my team is looking for an auto scaling solution and this PR would be great to have.

Personal opinion below - please wait for reply from contributors before making any changes
I do have one concern and perhaps @jamesmortensen or @diemol could chime in. Is it possible to make this video upload process cloud agnostic, since a lot of users won't be using AWS S3 buckets. Perhaps there is some way to let the user decide how the videos get processed or uploaded after being captured instead of the selenium repo handling the upload process which also comes at a cost of installing aws cli in every docker image.

Maybe we can have two PRs addressing the two separate parts - Autoscaling and Video Capture.

[jamesmortensen]

@mhnaeem Thanks for jumping in. It does seems like we have two separate PR's here, and I think it would be easier to evaluate them if we could see them as separate PR's. @prashanth-volvocars what do you think?

---

Here's my feedback for the video recording portion of the PR:

Regarding the changes in NodeBase, I don't see any major issues with installing the AWS CLI. We do already install components that not everyone uses. For instance, for those who don't use VNC or noVNC, those components, including xvfb and fluxbox, are installed in the container images and end up doing nothing. Do we know how much extra space it takes to install the AWS CLI? If it's not trivial, then perhaps there are other ideas?

Regarding adding other cloud providers, I don't think we need to be concerned about that right now, I think we could merge with what we have. At the same time, we could make some space for it with a few abstract changes. For example, start-s3-uploader.sh could be encapsulated inside start-uploader.sh and then start-uploader.sh would either call start-s3-uploader.sh or start-gcp-uploader.sh depending on the environment variables. This would allow other providers to be easily added without disturbing other logic. If someone wanted to upload videos to Backblaze B2, then they'd add an implementation for start-b2-uploader.sh and open a pull request. We could potentially even script it in a way that the Dockerfile doesn't need to be modified for someone to add another provider.

What do you both think?

[prashanth-volvocars]

Hello @jamesmortensen

Yeah sure we can have them as two separate PR's.

For the video upload to cloud provider. We can change it as per your suggestion to use start-uploader.sh and inside it call a scipt specific to a cloud provider. I will make these changes when i find some time and i will update the PR.

[jamesmortensen]

Sounds great. :)

[diemol]

I apologize for the delay in a proper review. In general, yes, it'd be nice to have to separate PRs. Basically because this one is adding ffmpeg to the Node images, and that is a huge change how we could handle video recording.

I prefer to have a PR that only focuses on the Keda integration, and when we are there, we can discuss the video recording approach.

[prashanth-volvocars]

@diemol @jamesmortensen

Closing this in favour of #1714