stream video to backend

[elianiva]

nothing works yet

Here are some brief notes explaning how it works:

Frontend:

• Requests permission when a user enters the /coding-test page.

  I have a question regarding this step, though. What if the user rejects the
  permission request? Do we ask them again? or do we simply don't allow them
  to go to the /coding-test page until they allow it?

• Starts the video stream without audio in 360p and record it using MediaRecorder.

  We can't do compression so we're forced to use a low resolution (I think 360p is sensible)
  MediaRecorder will splice the video to 1 second / chunk. We can increase this if we want.
  I chose 1s because it's very small so it can be sent directly to the server.

• Push the chunk into a queue inside a web worker.

  Just in case the internet isn't fast enough (which is very unlikely, it's like ~20-40kb per chunk,
  but who knows). We'll store the chunks inside a queue and post them sequentially to backend.

Backend:

• Added Minio client as a singleton to ServiceCollection.

• Made a VideoController to handle post request from client. Currently the path is POST /video.

• The controller is fat since I'm not sure how to refactor the uploader logic into its own service.

• It has the same authentication rules as ResumeExam in the event hub.

• It will create a minio bucket if it doesn't exist.

• Upload the file to Minio bucket using stream.

  We don't want to wait the backend to download the file and then upload it to Minio.
  That's too slow and inefficient.

• The file is stored using time-started_time-ended.webm format.

  The reason why it's like this is because the file was chunked and
  the browser's MediaRecorder API only attach Matroska header to the first chunk.

  Possible solution is we can pluck the header and shove it to the next chunks, but I think
  it's better for a separate PR since we'll only need this for video compression.

  I've explained this here and here.
  I also attached some comments in the code.
  We can then join the video chunks based on their timestamps using an automated worker
  or just do it manually using ffmpeg. Whichever works better.

[aldy505]

I have a question regarding this step, though. What if the user rejects the permission request? Do we ask them again? or do we simply don't allow them to go to the /coding-test page until they allow it?

yes they can't continue. but don't repeatedly ask the permission.

[aldy505]

I haven't had the time to review the code. I just read the description for now.

[ronnygunawan]

Tech debt: fat controller