- Spin down cluster
- Deploy new version of server code google.
- Push new docker images to :latest
- Check that lab docker image is set correctly.
- Go to sleep
- Check that datahub is starting the right image.
- Spin up cluster.
Make the DSMP cluster fetch new images:
for node in 03 04 05 06 07 08 09 10 11 12; do for tags in v60 22fa-lab-1 latest; do for image in stevenjswanson/cse142l-service-dsmlp stevenjswanson/cse142l-runner; do ssh sjswanson@its-ieng6maas-$node.ucsd.edu sudo docker pull $image:$tags;done;done&done
docker swarm join --token SWMTKN-1-0fqztdaagojadjzw3wcqfqksyp81duidygog9h7xblnj44ejdn-aozma6gikjgwa3ski5gjxbrys 172.17.77.30:2377
sudo docker swarm join --token SWMTKN-1-0fqztdaagojadjzw3wcqfqksyp81duidygog9h7xblnj44ejdn-aozma6gikjgwa3ski5gjxbrys 172.17.77.30:2377;done
sudo docker node promote its-ieng6maas-11.ucsd.edu its-ieng6maas-10.ucsd.edu
headnode: ssh sjswanson@its-ieng6maas-12.ucsd.edu
To start a service on the cluster, do
. d i
cse142 -n dev --replicas 10 --service --name runner-service --image stevenjswanson/cse142l-swanson-service:latest bash -c "cse142 --no-http cluster runner"
and the put the command at the end of the output in quotes:
docker service create --replicas 10 --replicas-max-per-node 1 --env GOOGLE_CLOUD_PROJECT --env DJR_CLUSTER --env DJR_DOCKER_SCRATCH --env CLOUD_MODE --env DJR_SERVER --env DJR_JOB_TYPE --env CSE142L_RUNNER_DOCKER_IMAGE --env EMULATION_DIR --env CLOUD_NAMESPACE --env SECRETS_DIRECTORY --env PACKET_PROJECT_ID --env DOCKER_USERNAME --env DOCKER_ACCESS_TOKEN --env GOOGLE_CREDENTIALS_FILE --env GITHUB_OAUTH_TOKEN --env THIS_DOCKER_CONTAINER=runner-service --env HOME=/root --mount type=bind,source=/var/run/docker.sock,dst=/var/run/docker.sock --mount type=bind,source=/tmp,dst=/tmp --env GOOGLE_APPLICATION_CREDENTIALS=/cse142L/CSE141pp-Config/secrets/cse142l-dev-c775b40fa9bf.json --workdir /cse142L --name runner-service stevenjswanson/cse142l-swanson-service:latest bash -c 'cse142 --no-http cluster runner'
You can do this with a simple ssh command.
sudo docker run -it --env GOOGLE_CLOUD_PROJECT --env DJR_CLUSTER --env DJR_DOCKER_SCRATCH --env CLOUD_MODE --env DJR_SERVER --env DJR_JOB_TYPE --env CSE142L_RUNNER_DOCKER_IMAGE --env EMULATION_DIR --env CLOUD_NAMESPACE --env SECRETS_DIRECTORY --env PACKET_PROJECT_ID --env DOCKER_USERNAME --env DOCKER_ACCESS_TOKEN --env GOOGLE_CREDENTIALS_FILE --env GITHUB_OAUTH_TOKEN --env THIS_DOCKER_CONTAINER=runner-service --env HOME=/root --mount type=bind,source=/var/run/docker.sock,dst=/var/run/docker.sock --mount type=bind,source=/tmp,dst=/tmp --env GOOGLE_APPLICATION_CREDENTIALS=/cse142L/CSE141pp-Config/secrets/cse142l-dev-c775b40fa9bf.json --workdir /cse142L --name runner-service -w /cse142L/ stevenjswanson/cse142l-swanson-service:latest bash --rcfile /cse142L/env.sh -c '. env.sh; cse142 --no-http -v cluster runner'
Start the runner service on the dsmlp cluster:
sudo docker service create --replicas 10 --env GOOGLE_CLOUD_PROJECT --env DJR_CLUSTER --env DJR_DOCKER_SCRATCH --env CLOUD_MODE --env DJR_SERVER --env DJR_JOB_TYPE --env CSE142L_RUNNER_DOCKER_IMAGE --env EMULATION_DIR --env CLOUD_NAMESPACE --env SECRETS_DIRECTORY --env PACKET_PROJECT_ID --env DOCKER_USERNAME --env DOCKER_ACCESS_TOKEN --env GOOGLE_CREDENTIALS_FILE --env GITHUB_OAUTH_TOKEN --env THIS_DOCKER_CONTAINER=runner-service --env HOME=/root --mount type=bind,source=/var/run/docker.sock,dst=/var/run/docker.sock --mount type=bind,source=/tmp,dst=/tmp --env GOOGLE_APPLICATION_CREDENTIALS=/cse142L/CSE141pp-Config/secrets/cse142l-dev-c775b40fa9bf.json --workdir /cse142L --name runner-service -w /cse142L/ stevenjswanson/cse142l-service-dsmlp:latest bash -c '. env.sh; cse142 --no-http -v cluster runner'
The course's lab runner provide allow students to run short-running jobs on bare-metal servers in the cloud. The basic flow is this:
- Students edit their source code.
- They submit a command to run in the cloud.
- The tools zip up their local files, ship them to the cloud, run the command, collect the results, and ship them back.
This underlying system is built on four cloud-based services:
- A data base (Google Cloud Datastore)
- A blob store (Google Cloud Storage)
- A publish/subscribe system (Google Cloud Pub/Sub)
- A web-based REST API for manipulating the above (Running in Google App Engine).
The main interface to the system is the cse142 command line tool. It takes
numerous subcommands to manipulate the above resources and submit jobs.
User authenticate to the system using their UCSD Google credentials (i.e, either their @eng or @ucsd.edu email addresses.)
You'll need docker installed. We have a document about doing that here:
For linux, Followed this guide to install: https://docs.docker.com/engine/install/ubuntu/#install-using-the-repository And this guide to run docker as a non-root user: https://docs.docker.com/engine/install/linux-postinstall/
If you are running on your own machine, this is not necessary
Get the IP address of the devel server from Steve, we'll call it IP.
You have access to the root account on the server, but you shouln't use it. Create and use your own account below, it prevents us from getting in eachother's way.
Do
ssh root@IP
to log in as root.
Create your account and allow it to sudo and use docker:
adduser <your @ucsd or @eng email without the @ part>
adduser <your username> sudo
adduser <your username> docker
Then, install your ssh public key (It seems to require and RSA key: id_rsa.pub rather than id_dsa.pub):
su <your username>
mkdir ~/.ssh
cat >> ~/.ssh/authorized_keys
And then paste in your public key.
Then set perms on your .ssh directory:
chmod go-rwx -R ~/.ssh
Logout and then back in:
ssh <username>@IP
To set up your development environment initially, first do
git clone --recurse-submodules git@github.com:NVSL/CSE141pp-Root.git
This clones the root repo and several sub-repos.
You should get something like:
Submodule path 'CSE141pp-Config': checked out 'fa0e4ebb3959df2a30c8a49570b0fed346bd2604'
Submodule path 'CSE141pp-DJR': checked out 'b297949fd62dbb6c8e3e897c928a8acdf449df2d'
Submodule path 'CSE141pp-LabPython': checked out '991e709477ea0a187c01c39a8c2fd7b8e3a0afd8'
Submodule path 'CSE141pp-SimpleCNN': checked out '3656b5c5213434d437106f61eb0aaeb92dc7cbe8'
Submodule path 'CSE141pp-Tool-Moneta': checked out '73213d9872f80f057d20e0527ebfd11381e3088a'
Submodule path 'CSE141pp-Tool-Moneta-Pin': checked out 'ccad791816593cbff21d8599c3925fd1c0677c57'
Submodule path 'cse141pp-archlab': checked out 'c5e2e711fbc59740bb6e14d19a3f9a4f2785524e'
You may need to do install python 3.9's venv package. On linux you do it like this:
sudo apt install python3.9-venv python3.8-venv
Next build install the tools we use to build the docker images:
cd CSE141pp-Root
. env.sh # this initializes some environment variables and sets your PATH
make bootstrap # this creates a python virtual environment to hold the tools we use to invoke docker properly.
Next, you'll need to build the docker images locally with:
make
This will run for a while. You can then do
docker images
to see them (It'll have your user name in the image name):
REPOSITORY TAG IMAGE ID CREATED SIZE
stevenjswanson/cse142l-swanson-runner latest e3cb52c116d7 2 hours ago 12.2GB
stevenjswanson/cse142l-swanson-runner s21-dev e3cb52c116d7 2 hours ago 12.2GB
stevenjswanson/cse142l-swanson-dev latest c90dcd27a70e 2 hours ago 7.84GB
stevenjswanson/cse142l-swanson-dev s21-dev c90dcd27a70e 2 hours ago 7.84GB
stevenjswanson/cse142l-swanson-service latest d28b327b0ada 2 hours ago 7.84GB
stevenjswanson/cse142l-swanson-service s21-dev d28b327b0ada 2 hours ago 7.84GB
Then, to get yourself into a development docker container:
bin/cse142dev
This will drop you into /cse142L in the docker container which is mapped to
the root of your CSE141pp-Root directory. You can edit your files from
outside the container using the editor of your choice and the changes will be
reflected here.
If cse142dev find a .ssh directory in your home directory, it will create
an ssh-agent to store your ssh key and invokes ssh-add to add it. If
cse142dev asks for a password, this is why. It'll save you from having to
type your password over and over again.
The first time you do this, you'll need to run this command to set yourself up
to do development (after getting inside docker with cse142dev):
./setup.sh
to build everything from source. It will take a little while. After that,
invoking CSE141pp-Root/bin/cse142dev is all you'll need to do to get into
docker and start working.
From inside the docker container you can see information about the docker container with
whereami
Moneta is our memory trace visualizer. It uses a command line tool called
mtrace to generate traces and jupyter-notebook to view them. The process
below includes the installation of Moneta as well.
From inside docker, you can run jupyter-notebook --allow-root . to start up
jupyter notebook server. The output will look something like this
[I 22:38:48.814 NotebookApp] Writing notebook server cookie secret to /root/.local/share/jupyter/runtime/notebook_cookie_secret
[I 22:38:49.494 NotebookApp] JupyterLab extension loaded from /opt/conda/lib/python3.7/site-packages/jupyterlab
[I 22:38:49.494 NotebookApp] JupyterLab application directory is /opt/conda/share/jupyter/lab
[I 22:38:49.497 NotebookApp] Serving notebooks from local directory: /cse142L
[I 22:38:49.497 NotebookApp] The Jupyter Notebook is running at:
[I 22:38:49.497 NotebookApp] http://swanson-dev-192.168.4.21:8888/?token=9480933a7aae975a670f6293f8fb4f94582be04f07a123a0
[I 22:38:49.497 NotebookApp] or http://127.0.0.1:8888/?token=9480933a7aae975a670f6293f8fb4f94582be04f07a123a0
[I 22:38:49.498 NotebookApp] Use Control-C to stop this server and shut down all kernels (twice to skip confirmation).
[C 22:38:49.506 NotebookApp]
To access the notebook, open this file in a browser:
file:///root/.local/share/jupyter/runtime/nbserver-49-open.html
Or copy and paste one of these URLs:
http://swanson-dev-192.168.4.21:8888/?token=9480933a7aae975a670f6293f8fb4f94582be04f07a123a0
or http://127.0.0.1:8888/?token=9480933a7aae975a670f6293f8fb4f94582be04f07a123a0
If you are running on your own machine, you should be able to acces it via the 127.0.0.1:8888 url at the end.
If you are on the devolpment server, replace 127.0.0.1 with the IP address you used to connect to the devel server.
cse142dev drops you into docker container that has priviliged, direct access
to the job runner above, so you don't have to authenticate to use it. You will
use this capability to create an account for yourself. From then on, you'll
authenticate and use it as a student would.
The command lines to create and edit users are similar to those you'll use to manipulate other data (e.g., jobs and labs).
First, check if everything is working (it may take a little bit. Direct access is slow.):
cse142 --no-http user list
The --no-http bypasses authentication. user means we are going to do
something to users. list means list them.
This should give you list of current users. It will at least include
swanson@eng.ucsd.edu and probably others.
To create your user with admin priviliges:
cse142 --no-http user create --email <your @ucsd.edu or @eng.ucsd.edu email> --name "<your full name>" --role "ADMIN"
Check to see that it worked:
cse142 --no-http user list
You should see yourself:
3341f94e-254d-4f58-9f5c-a825e067f9a7 sjswanson@ucsd.edu sjswanson@ucsd.edu ['USER']
cb531cf4-2aa3-4e48-8629-3c36ffadee4c swanson@eng.ucsd.edu swanson@eng.ucsd.edu ['ADMIN']
The fields are a unique ID, email, full name, and roles (for permissions).
Now you can authenticate:
cse142 login <your email>
and follow the instructions to login via a web browser.
You should then be able to
cse142 user list
You should see yourself again. If you made a mistake you can edit your account like so:
cse142 user update sjswanson@ucsd.edu --set name="Steven Swanson"
You can pass either your email or the id.
To see more detail about yourself do:
cse142 user list -l sjswanson@ucsd.edu
and get something like
{ 'created_time': datetime.datetime(2021, 7, 10, 5, 59, 55, 399152, tzinfo=tzutc()),
'email': 'sjswanson@ucsd.edu',
'id': '3341f94e-254d-4f58-9f5c-a825e067f9a7',
'last_authenticated_time': datetime.datetime(2021, 7, 10, 5, 59, 55, 399208, tzinfo=tzutc()),
'last_login_time': None,
'mood': '',
'name': 'Steven Swanson',
'namespace': 'default',
'project_id': 'cse142l-dev',
'roles': ['USER'],
'secret': ''}
The job runner is supposed to approximate the experience of running command line commands on a machine in the cloud. Each job runs in a fresh docker container that is nearly identical to the one you are working in but has fewer permissions.
Each job is run in the context of a "lab" that specifies which exact docker image to use. You can see the list of available labs with
cse142 lab list
You should get something like this:
76c847c9-c3fa-4d3b-b56e-87fcfab46edf test 2021-07-06 10:35:00-07:00 stevenjswanson/cse142l-runner:latest test2
The field are id, short-name, due date, docker image, and full name.
To run your first job:
mkdir t
cd t
cse142 job run --lab test "echo hello world"
You'll get something like this:
You are submitting a job for lab test2 (test).
Creating job e4e80402-9217-4835-9e6b-56c7ef3b4977 0.00 0.00
Ready for submission. 1.27 1.27
Job e4e80402-9217-4835-9e6b-56c7ef3b4977 is in state 'PUBLISHED'. 0.56 1.83
Job e4e80402-9217-4835-9e6b-56c7ef3b4977 is in state 'PUBLISHED'. 1.08 2.90
Job e4e80402-9217-4835-9e6b-56c7ef3b4977 is in state 'RUNNING'. 1.07 3.97
Job e4e80402-9217-4835-9e6b-56c7ef3b4977 is in state 'RUNNING'. 1.06 5.03
Job e4e80402-9217-4835-9e6b-56c7ef3b4977 is in state 'DONE_RUNNING'. 1.06 6.09
Job e4e80402-9217-4835-9e6b-56c7ef3b4977 is in state 'DONE_RUNNING'. 1.14 7.24
Job e4e80402-9217-4835-9e6b-56c7ef3b4977 is in state 'DONE_RUNNING'. 1.07 8.31
Job e4e80402-9217-4835-9e6b-56c7ef3b4977 is in state 'DONE_RUNNING'. 1.06 9.36
Job e4e80402-9217-4835-9e6b-56c7ef3b4977 succeeded. 1.06 10.42
Writing results 1.00 11.42
hello world
Job Complete 0.55 11.97
It tracks the lifetime of the job. The numbers at the end are the latency since last update and total execution time. You can see the output of your job near the bottom.
You can list the jobs you've run with
cse142 job list
As an admin, you'll see all the jobs run on the system. It'll be a lot.
You can list the last job you ran with
cse142 job list LAST
or
cse142 job list -l LAST
The steps in a jobs execution are as follows:
- Create the job object.
- Recursively zip up most of the files in the current directory. ("Ready for submission")
- Upload them to the cloud.
- Submit the job for execution (state "PUBLISHED")
- One of the job runners pulls it (state "SCHEDULED", not shown here)
- Runs it (state "RUNNING")
- Zips up modified files (state "DONE_RUNNING")
- Upload the modified files to the cloud.
- Finish up ("succeeded")
All the while, your cse142 command is polling the job for updates. When it
sees it's succeeded, it downloads the zip file of updated files, and unzips it.
It also prints the stdout/stderr for the command ("hello world").
We created the t directory above to prevent the job for sucking in all the
source code in your directory. It takes a long time (and might fail).
Here are some things to try:
Create main.c:
#include<stdio.h>
void main() {
printf("Hello world\n");
}
Then:
cse142 job run --lab test 'gcc main.c -o main; ./main'
and then
gcc main.c -o main
cse142 job run --lab test './main'
or even
cse142 job run --lab test 'gcc main.c -o main'
./main
We would like the course tools to be as user-friendly as possible. So, please use them for developing the labs. IF you have feedback about how we can make the tools more pleasant to use, please let Professor Swanson know.
In particular, cse142 should meeting the following requirements
- The user should never get at python stack dump. If you get one, it's bug please send me the command line and the output.
- You shouldn't see "Command failed due to uncaught exception" errors either. They are a catchall to avoid stack dumps. If you can
- The error messages make sense. If you find one misleading or confusing, let me know.
- The use of color should be consistent.
You should clone labs in the labs directory.
You can get a list of all the labs (including old and usued labs) here: https://github.com/NVSL?q=CSE141pp-Lab&type=&language
For instance to clone the final project:
cd labs
git clone git@github.com:NVSL/CSE141pp-Lab-FinalProject.git
To get started working on lab, let's run the final project. It's here:
cd labs/CSE141pp-Lab-FinalProject/
You should be able to run the lab locally with
make
It'll take a little while. Ignore the make: runlab: Command not found
Then you can run it in the cloud with
cse142 job run --lab test 'make clean; make
Unless you are steve, You probably don't need to do this.
First, start a swarm:
docker swarm init --advertise-addr <MANAGER-IP>
Then start the runner service:
cse142 dev --service --name runner-service --image stevenjswanson/cse142l-service:latest bash -c "cse142 cluster runner"
The runner docker image is self-contained so it'll work as a student image running on campus servers via dsmlp.
Here's a command line to mimic what they do:
docker run -it -h $HOME -w $HOME --mount type=bind,source=$HOME,dst=/root --publish published=8889,target=8888 stevenjswanson/cse142l-swanson-runner:latest cse142-notebook
OUr images are based on the Jupyter notebook scipy image, but the published version use an old version of python, so we built it ourselves:
make docker-stacks