What is the best way to use libValkka to build an RTSP application?

[karma0]

General Question

I'm struggling to get started.

I've skimmed the tutorials for relevant pieces to my work, I've perused the valkka-examples project, and I feel ready to get started. I am trying to figure out what I needed to do to get up and running with a simple development environment.

I have come to understand, given the really great documentation and examples within the ecosystem of Valkka projects, that I need to install Valkka Core. This at first glance didn't seem right, since I thought that I would be installing a Python package, but makes sense given that the core library leverages C++ heavily. Unfortunately, it seems that using virtual environments and installing via pip is not an ideal approach.

In installation of Valkka Core, I seem to hit a snag - PPAs aren't really supported by Ubuntu any longer, and this is the approach in the instructions. So, I decided to pivot and to try to get up and running in a docker container in order to test out the examples and ultimately develop my application there.

Since I would rather not have to modify my installation of Ubuntu to get this running, and running in production will likely be in a container, Docker seems like an ideal solution to building my application, but the README seems to be outdated. I tried initially building just using the Dockerfile for my platform, but after failing I realized the helper scripts are there to help. When I use the tags and versions in the README, they fail, and the build script has a different name.

I decided to modify the files, but I'm getting the same errors (vaapi requested but not found) in any attempt to build. I also attempted to pull the container up on Docker Hub but it failed to run.

---

If I want to dig in and play with libValkka and use the examples in order to stream RTSP feeds to disk on a headless system and spit out occasional snapshot images, what would be the best approach to get started?

FWIW, I'd be happy do submit some bugs and PRs, but I'm feeling a little overwhelmed with all of the challenges that I'm encountering and I would rather only follow best practices and see if I hit any challenges there first.

[elsampsa]

Installing libValkka

If you are not into PPAs, I recently added a ci/cd pipeline that produces deb packages:

https://github.com/elsampsa/valkka-core/releases

After using sudo dpkg -i packagename.deb you still need to do sudo apt-get install -fy to pull the dependencies.

note for myself: need to add this into the main documentation page

Using PPAs or those packages just mentioned, doesn't really "harm" your ubuntu installation: they just pull a number of dependencies (see the last "depends" clause in here) and then install libValkka.

All that installation is done into the "system-wide" python installation directory usr/lib/python/python3 (seen by all users)

Now, if you use virtualenv or you have installed packages into userspace (i.e. as normal/non-sudo user with pip install) then those packages take preference over the "system-wide" directory, say, the numpy you have in your virtualenv, etc. has preference over the one in usr/lib/python/python3. You can check what actually happens with the quicktest.py script.

So no problem there.

Docker

Docker images are as of today, outdated and I need to build a ci/cd pipeline for them (similar as I did for the deb package) asap

note for myself: do this asap

Compilation problem

VAAPI acceleration was recently added into libValkka. I wanted to add it as a separate deb package, but eventually went for the "easy" solution of using the existing ffmpeg/libav infra for using VAAPI - so now it is part of valkka-core and giving all kinds of dependency problems: https://github.com/elsampsa/valkka-core/issues/9

Please do file a separate ticket on the problems you encounter

note for myself: test for compilation in arm, bitch!

rtsp app

You want to read rtsp stream(s), write it to disk and spit out occasional images & do this in docker. The filterchains would look roughly like this:

live branch
    decoding branch
        push decoded frames through shmem server (1)
    frag-mp4 branch
        push frag-mp4 packets through shmem server (2)

Then in your (separate) python multiprocesses (MP) read (1) and (2). In MP(1), use any standard tool to decode a frame occasionally to jpg/png (PIL.Image for example). In MP(2) dump the received frag-mp4's to the disk.

A word of warning: you need to master & understand python multiprocessing (as per my tutorials).

For similar stuff, this is the most up-to-date example: https://github.com/elsampsa/valkka-streamer This is not so up-to-date, but maybe simpler one: https://github.com/elsampsa/valkka-examples/tree/master/example_projects/basic

Please remember to use the H264 substream of your IP cameras as of today, libValkka is still not using H265.

[karma0]

Thank you! This is immensely helpful!

It seems that I had misread the notification from apt-add-repository in my late-night shenanigans:

NOTE: PPA will soon be deprecated in the favor of .deb packages from https://github.com/elsampsa/valkka-core

This, followed by a 404 on apt update, which is more likely because I'm on the 23.04 release of Ubuntu:

Err:17 https://ppa.launchpadcontent.net/sampsa-riikonen/valkka/ubuntu lunar Release
  404  Not Found [IP: 185.125.190.52 443]

I'll see if I can get the deb package installed, and go from there. Thanks again!

[elsampsa]

yeah, I only provide packages for ubuntu LTS distros.

[elsampsa]

ok, now you can find fresh docker images in the dockerhub as well - there is now a ci/cd pipeline that constructs docker images automatically