Skip to content

Latest commit

 

History

History
395 lines (310 loc) · 11.8 KB

README.md

File metadata and controls

395 lines (310 loc) · 11.8 KB

Are We Compressed Yet?

This repository contains the arewecompressedyet.com website source code.

Running your own local copy of the website

To run a local copy, you will need a copy of node and npm.

You will need node, npm, etc. For Ubuntu:

sudo apt install nodejs npm nodejs-legacy libicu-dev

You'll also need tools to build AV1 or other codecs being tested:

sudo apt install yasm libtool

First, run the ./setup.sh script. It will create directories needed for AWCY to run.

Next, create a configuration file called config.json.

Here is an example config.json:

{ "channel": "#daalatest",
  "have_aws": false,
  "port": 3000,
  "rd_server_url": "http://localhost:4000"
}

You will also need a file called secret_key which contains the key needed to use the website.

echo 'fake_password_to_compile' > secret_key

These commands will create the configuration files and install the node.js modules that get used by awcy. Open a node command line and run the following:

mkdir runs
cd www
npm install
cd ..
npm install
npm run tsc
node generate_list.js
npm run start

To run the server, execute the run_awcy.bat file or run the following in your command line:

  node awcy_server.js

Now you can open localhost:3000 with your browser to see your local version of the website.

Setting up repositories

For the website to build codecs, you need local checkouts of every codec. While the ./setup.sh script automatically adds AV1, you can add your own. For example:

git clone https://aomedia.googlesource.com/aom av1
ln -s av1 av1-rt

Setting up rd_server

The AWCY web server manages the repositories and runs/ directory, and compiles and builds the codecs. Once this is done, it hands off the job of actually talking to all the AWS machines to rd_server.

To install rd_server, checkout the rd_tool repository in the same directory as awcy:

git clone https://github.com/tdaede/rd_tool.git

Then start the rd_server.py daemon:

./rd_server.py

The rd_server.py daemon listens on port 4000 by default.

More documentation on rd_server.py can be found in its README.

Setting up workers

Workers are Linux machines accessible over ssh. They need to be the same architecture as the AWCY server.

A worker needs a AWCY work root, which has a directory structure as follows:

dump_ciede2000/
daalatool/
slot0/
vmaf/

The slot* directores are created by AWCY for jobs.

daalatool must be pre-populated by a clone of the daala repo, with tools built (make tools):

git clone https://github.com/xiph/daala.git daalatool
cd daalatool
./autogen.sh
./configure
make
make tools

dump_ciede2000 must be populated by a built version of the dump_ciede2000 tool:

git clone https://github.com/KyleSiefring/dump_ciede2000
cd dump_ciede2000
cargo build --release

vmaf is a clone of the Netflix VMAF repository, also built.

git clone https://github.com/Netflix/vmaf.git
cd vmaf/libvmaf
meson build --buildtype release
ninja -C build

In addition, a copy of the test media must be accessible to the worker.

The workers should be configured on the main server with a machines.json file, which contains the host, user and port (for ssh). In addition, it contains the number of cores (which controlls the number of slots), plus the work root (containing the above mentioned directores) and the test media path. For example:

[
  {
    "host": "localhost",
    "user": "xiph",
    "cores": 8,
    "port": 22,
    "work_root": "/home/xiph/awcy_temp",
    "media_path": "/home/xiph/sets/"
  }
]

Each worker must have a unique work root - usually local storage on the worker. One worker is intended to be one physical machine or VM. For testing purposes, one machine can act as multiple workers, but the work_root must be independent. The media path is read-only and can be shared, e.g. via NFS.

The machines should be accesible via passwordless SSH authentication. If needed, the SSH_PRIVKEY_FILE environment variable can inform rd_server.py of a custom private key (.pem format) to use for logging into workers.

Run database format

The runs/ directory will contain all of the output files generated from a job. There is a info.json file that specifies what options were used by that particular run. Here is an example of an info.json file:

{"codec":"daala","commit":"","run_id":"2014-09-19T22-00-08.196Z","task":"video-1-short","nick":"AWCY","task_type":"video"}

There is also an output.txt file that contains the output of the rd_tool.py script.

After each run, a cache file called list.json file is generated by the generate_list.js script. This contains all of the info.json files, as an ordered list. This should probably be replaced by a "real" database at some point.

Docker support

AWCY can be started as a docker container for local usage or development.

Build (server)

The docker image is pretty large (3.3 GB), as it includes:

  • build dependencies for all codecs to be built (gcc/g++, autotools, make, rust, ...)
  • build dependencies for the WebUI (nodejs, emscripten, ...)
  • runtime dependencies for build and analysis scripts (python, daalatool, ...)

Note: this image has not been optimized for size yet, and may be split in the future.

To build the image:

docker build --tag xiph/awcy:latest .

Build (worker)

Another dockerfile (Dockerfile.worker) is available to create worker nodes.

To build the worker image:

docker build --file Dockerfile.worker --tag xiph/awcy-worker:latest .

Run (all-in-one)

You can run an all-in-one container which will start:

  • awcy server
  • rd_server
  • sshd (to emulate a local rd_tool worker)

In addition, a git clone of every codecs' source is going to be done from upstream at initial startup.

Configuration is driven by a few environment variables (with their default values):

  • CONFIG_DIR=/data/conf: configuration files, sqlite database, generate SSH keys, set list, ...
  • CODECS_SRC_DIR=/data/src: where codecs' sources are going to be cloned (and compiled)
  • RUNS_DST_DIR=/data/runs: rd_tool jobs runs output
  • WORK_DIR=/data/work: rd_tool job workdir directory
  • MEDIAS_SRC_DIR=/data/media: directory containing media files (get them from https://media.xiph.org/sets/)
  • LOCAL_WORKER_ENABLED=false: if a local worker need to be started and configured
  • LOCAL_WORKER_SLOTS=$(nproc): number of slots for the local worker
  • IRC_CHANNEL=none: IRC notifications target channel (none means disabled)
  • AWCY_API_KEY=awcy_api_key: WebUI/API key
  • AWCY_SERVER_PORT=3000: awcy server listening port
  • RD_SERVER_PORT=4000: rd_server listening port

Simple run example (all data will be contained on the docker host in /tmp/awcy):

docker run \
	-it \
	--rm \
	--name xiph-awcy \
	--publish 3000:3000 \
	--volume /tmp/awcy:/data \
	--volume ${HOME}/xiph-media-files:/media \
	--env MEDIAS_SRC_DIR=/media \
	--env AWCY_API_KEY=complicated_password \
	--env LOCAL_WORKER_ENABLED=true \
	--env LOCAL_WORKER_SLOTS=4 \
	xiph/awcy:latest

Output:

--2019-01-16 22:46:11--  https://people.xiph.org/~tdaede/sets/subset1-monochrome/Claudette_14_july_2003_1920Z.y4m
Resolving people.xiph.org (people.xiph.org)... 140.211.15.28, 2001:470:eb26:54::1
Connecting to people.xiph.org (people.xiph.org)|140.211.15.28|:443... connected.
HTTP request sent, awaiting response... 200 OK
Length: 1485050 (1.4M) [application/octet-stream]
Saving to: ‘/media/awcy-builder-quicktest/test_frame.y4m’

/media/awcy-builder-quicktest/test_frame.y4m         100%[=====================================================================================================================>]   1.42M  1.14MB/s    in 1.2s

2019-01-16 22:46:13 (1.14 MB/s) - ‘/media/awcy-builder-quicktest/test_frame.y4m’ saved [1485050/1485050]

Cloning into '/data/src/av1'...
remote: Sending approximately 132.14 MiB ...
remote: Counting objects: 8, done
remote: Finding sources: 100% (5/5)
remote: Total 193273 (delta 156479), reused 193273 (delta 156479)
Receiving objects: 100% (193273/193273), 132.11 MiB | 2.33 MiB/s, done.
Resolving deltas: 100% (156479/156479), done.
Cloning into '/data/src/daala'...
remote: Enumerating objects: 12671, done.
remote: Total 12671 (delta 0), reused 0 (delta 0), pack-reused 12671
Receiving objects: 100% (12671/12671), 8.70 MiB | 2.37 MiB/s, done.
Resolving deltas: 100% (9442/9442), done.
Cloning into '/data/src/rav1e'...
remote: Enumerating objects: 1, done.
remote: Counting objects: 100% (1/1), done.
remote: Total 4426 (delta 0), reused 0 (delta 0), pack-reused 4425
Receiving objects: 100% (4426/4426), 2.61 MiB | 2.33 MiB/s, done.
Resolving deltas: 100% (3108/3108), done.
Cloning into '/data/src/thor'...

(...)

Generating SSH host keys
Generating public/private rsa key pair.
Your identification has been saved in /data/conf/awcy.pem.
Your public key has been saved in /data/conf/awcy.pem.pub.
The key fingerprint is:
SHA256:TJa3T7FZhByGZ3owJOtV/OXk75nyBJa8I7pq2BwNEaU xiph@4ee6bf6c8ffe
The key's randomart image is:
+---[RSA 2048]----+
|       .+o.++o.  |
|       ..++o*.  o|
|       E* o*...= |
|       * o..o=o o|
|        S ..+=  .|
|       . . o. o .|
|      + .  ..o oo|
|     . +  . ..oo.|
|      ...o.   o. |
+----[SHA256]-----+
STARTING SSHD SERVICE
STARTING AWCY SERVICE
STARTING RDTOOL SERVICE
AWCY server started! Open a browser at http://172.20.0.12:3000
AWCY server directory: /opt/app
Configuration directory: /data/conf
Codecs git repositories location: /data/src
Media samples directory: /media
Runs output directory: /data/runs

If you open your browser at http://localhost:3000 (or the one in the container output), you should see the AWCY WebUI.

On initial startup, the following configuration files are generated:

/tmp/awcy/conf/awcy.pem
/tmp/awcy/conf/awcy.pub
/tmp/awcy/conf/config.json
/tmp/awcy/conf/list.json
/tmp/awcy/conf/machines.json
/tmp/awcy/conf/secret_key
/tmp/awcy/conf/sets.json
/tmp/awcy/conf/subjective.sqlite3

Run (with workers)

You first need to start an AWCY server instance using the all-in-one instructions, but setting LOCAL_WORKER_ENABLED to false.

Next, run once instance of the worker container on each worker machine. FOr testing, you can also run multiple workers on your local machine. In a second terminal, start one or more worker instances:

docker run \
	-it \
	--detach \
	--name xiph-awcy-worker-1 \
	--volume ${HOME}/xiph-media-files:/media \
	--env SSH_PUBKEY="$(cat /tmp/awcy/conf/awcy.pub)"
	xiph/awcy-worker:latest

docker run \
	-it \
	--detach \
	--name xiph-awcy-worker-2 \
	--volume ${HOME}/xiph-media-files:/media \
	--env SSH_PUBKEY="$(cat /tmp/awcy/conf/awcy.pub)"
	xiph/awcy-worker:latest

...

Note that the given public key is the server-generated one.

Then stop the server instance by hitting CTRL+C in its terminal.

Get IP addresses of your worker nodes:

docker inspect xiph-awcy-worker-1 | jq -r '.[].NetworkSettings.IPAddress'
172.20.0.12

docker inspect xiph-awcy-worker-2 | jq -r '.[].NetworkSettings.IPAddress'
172.20.0.13
...

You can now edit the /tmp/awcy/conf/machines.json (paths should be left as-is):

cat >/tmp/awcy/conf/machines.json <<EOF
[
  {
    "host": "172.20.0.12",
    "user": "xiph",
    "cores": 8,
    "port": 22,
    "work_root": "/data/work",
    "media_path": "/media"
  },
  {
    "host": "172.20.0.13",
    "user": "xiph",
    "cores": 8,
    "port": 22,
    "work_root": "/data/work",
    "media_path": "/media"
  }
]
EOF

Then start the server again using the previously used command.

You should now have two (or more) workers accessible.

Samples

To make sure that everything is working properly, you can try a run using the awcy-builder-quicktest set, which runs in one minute or so.

Cleanup

You can stop the container by hitting CTRL+C.

If you have started your container using the above example, all data is contained in /tmp/awcy, so you can removed this directory from your host.