vida-nyu / reproserver

A web application reproducing ReproZip packages in the cloud.

ReproServer

An online service to run ReproZip bundles in the cloud. No need to install reprounzip on your machine, point ReproServer to a RPZ file or upload one, and interact with the reproduced environment from your browser.

Additionally, ReproServer can capture remote web assets referenced by a web application that has been packaged as a RPZ file. That way, you will always be able to get a consistent reproduction of the RPZ bundle, even if those remote assets disappear.

How to run this with Tilt

Make sure you have checked out the submodule with git submodule init && git submodule update

You will need Tilt, kubectl, and a cluster with a local registry (that you can set up with ctlptl).

For example, create a local cluster with:

minikube start --kubernetes-version=1.22.2 --driver=docker --nodes=1 --container-runtime=docker --ports=8000:30808

Install the ingress controller using:

kubectl apply -f k8s/nginx-ingress.k8s-1.22.yml

Start the application for development using:

tilt up

You can then open http://localhost:8000/ in your browser. Tilt will automatically rebuild images and update Kubernetes as you make changes.

How to run this with docker-compose

You will need Docker and docker-compose.

• Make sure you have checked out the submodule with git submodule init && git submodule update
• Copy env.dist to .env (you probably don't need to change the settings)
• Start services by running docker-compose up -d --build
  • Alternatively, use the development mode (insecure, but displays debug info and autoreloads): docker-compose -f docker-compose.dev.yml up -d --build
• Open localhost:8000 in your browser

How to stop it: docker-compose down -v

reproserver's Issues

Output page: text description

2: Minor Usability Problem
Adding text descriptions like "You can see the log and output files from the research, which you can view or download to your machine. You can also share the link for other people to reproduce your research. " would be helpful.

Output files: no back button or exit button

2: Minor Usability Problem
There is no back button or exit button to go back to the output files page when the files are clicked.

Test on CI

Originally opened 2019-11-17 15:47 EST by @remram44

Find a way to run this on CI, and run some sort of integration tests.

Upload page: Order

2: Minor Usability Problem
I would change the order of the upload page to allow users to learn how to use it first

1. How to use this site
2. Examples
3. Select a package

to upload

Better feedback that things are happening

Originally opened 2020-01-16 16:15 EST by @remram44

Output page: "Go back" button instead of clicking the file name

2: Minor Usability Problem
Currently, the file name in the heading “Package bechdel.rpz, run s67hd” is clickable but It is unclear what clicking it will do.
Instead of making it clickable, adding another button that says something like "Go back and modify>" would make it more clear that users can go back and change the input.

Usage limits to prevent abuse

Originally opened 2017-05-19 14:44 EDT by @remram44

• Set CPU/RAM usage limits
  • This can be done via Kubernetes, setting limit to DinD container in runner pod
  • Or via the Dind daemon, having the runner set a limit
• Set runtime limit
  • Have the runner spawn a thread that kills the container after a while
• Set disk space limit
  • Docker doesn't seem to support this?
• Careful of GZIP bombs
  • The builder could check the unpacked size

Limit data movement during build

Originally opened 2019-09-11 16:59 EDT by @remram44

The server is made slower by the need to move data around.

• RPZ is received from upload / downloaded from repo
• uploaded to S3
• downloaded from S3 to builder container
• uploaded to Docker-in-Docker in builder pod
• then image is uploaded to registry

Upload button should not be clickable if file is empty

3: Major Usability Problem
https://staging.server.reprozip.org/web
Upload button is clickable even if the file is not selected and leads to an error page

Cache repository APIs

There's no need to query the repositories every single time you visit a /reproduce/<repo>/<path> URL. A very simple caching mechanism would probably go a long way here.

Top nav: unclear what page user is on

3: Major Usability Problem
There is no indication at the top navigation bar to let the users know what page they're on. A small dot under the menu items would make it clear.

Web Capture page: Add section

• Get rid of RPZ file and WACZ file input section and add a section under "Web Capture" instead
  We have the following information about your files:
  ReproZip bundle: filename.rpz (size in italic)
  WACZ file:
  Web Server:
  Current web archive bundle:
  Enter the hostname and port number if needed:

Support notebooks packed with reprozip-jupyter (kernel only)

Originally opened 2019-09-11 12:27 EDT by @remram44

• Spin up server outside of experiment container (probably in runner container directly)
• Execute kernels in experiment container (can probably run sleeping entrypoint, and run kernels with docker exec)
• Do this automatically when the RPZ looks like a reprozip-jupyter one (same as is done in reprounzip-qt)

Error page: no back button

3: Major Usability Problem
There should be a back button on the error page

s

Run the experiment page: adding input examples

2: Minor Usability Problem
It might be confusing for new users because it's unclear what to input. Examples, information overlay, or documentation would be helpful. (In the redesign, I added an information icon that opens a help overlay when clicked and users can read what they should input and what it means.)

Bring your own compute: GCP

Originally opened 2019-11-14 22:29 EST by @remram44

See also #36

GCP supports oauth2, so it's possible to have a "log in with Google" button and then have ReproServer start VM on your account.

I have the code to start VMs. Need to make that VM run the bundle now...

Also need to review GCP's terms of service.

Update to Bootstrap 5

The templates were made on the previous version of bootstrap and should be updated to bootstrap 4.

Output page: unclear what the ID means

2: Minor Usability Problem
“Package bechdel.rpz, run s67hd” unclear what s67hd means.

Access/mount data from repositories

Originally opened 2019-09-09 14:21 EDT by @remram44

Related to VIDA-NYU/reprozip#221.

Types of accepted files

3: Major Usability Problem
Adding a supported file type would be helpful for new user -both upload page and web capture page


s

Don't open issues here -- use the GitLab issue tracker if you can

View issues

Open a new issue

Run the experiment page: text description

2: Minor Usability Problem
Adding a text like "We have finished loading your .rpz file. Now, reproduce the work by clicking the “Run” button." would make it more clear.

Web capture page: text description

3: Major Usability Problem
Suggestion for text description on web capture page:

Select your own .rpz file, or try using one of the examples from the ReproZip-Examples repository, and click the "Upload" button!
From there, you can set parameters and (optionally) upload your own inputs, and then reproduce the work by clicking the "Run" button. You can then see the log and output files from the research, which you can view or download to your machine. You can also share the lin

k for other people to reproduce your research.

Pluggable execution

Originally opened 2017-08-17 14:29 EDT by @remram44

All that's needed for execution is a Docker daemon that can pull from our registry. Currently there is a Docker-in-Docker daemon running in the runner pod, but really that could be something else. It could also be possible to choose one of multiple possible backends depending on RPZ metadata and user selection (both on the 'reproduce' page and via user accounts).

Example execution backends:

• Another Docker server (give reproserver host + certificate, from docker-machine)
• AWS (give reproserver IAM credentials and it'll start an EC2 instance to run your experiments on)

It could also be possible to run this without Docker. ReproUnzip supports Vagrant, and Vagrant supports AWS

Share configured environments

Originally opened 2019-11-01 21:07 EDT by @remram44

It might be useful to share a link to an environment configured differently (e.g. different parameters and input files).

The input files are stored on S3 so stay accessible.

Alternatively we could make it possible to get the configuration from the results page.

Support for more data repositories - with a shared library?

Originally opened 2019-12-20 07:11 EST by @nuest

In the ReproServer-preprint you mention you want to support more data repositories. 💯 !

Looking at reproserver's code to download data from Zenodo and the one that repo2docker uses in its "contentprovider", I see a lot of similarities!

suppdata is a "DOI to data" package for R, which so far focused on supplemental data for papers' DOIs, but I'd like to extend it towards data repositories.

A background discussion is also here: ropensci-archive/doidata#1

What do you think about a generic "data download from DOI" package in Python that both ReproServer and repo2docker could use?

Support websites packed with reprozip-web

Originally opened 2019-09-11 12:28 EDT by @remram44

• Run proxy
• Run warc player
• Get warc from RPZ to the player somehow (Have runner get it from RPZ in S3? Extract warc and store in separate bucket?)

Add canonical logo

Originally opened 2018-11-02 14:43 EDT by @VickySteeves

We should probably have a logo for the server similar to ReproZip.

While uploading: no visibility of system status

3: Major Usability Problem
After clicking the upload button, the browser starts loading but there is no indication of the status on the website. Adding a loading icon or an overlay would be helpful.

Accept non-repository URLs and turn them into an upload

Currently:

• Upload a file -> get a unique ID from ReproServer backed by the DB (example: /reproduce/r2ptb)
  • This URL doesn't persist if ReproServer data is lost
• Provide a URL to a supported repository -> use a unique ID backed by the repository (uses the ID on the repository) (example: input https://osf.io/5ke97 -> /reproduce/osf.io/5ke97)
  • This URL
• Provide a URL to anything else -> get an error "unrecognized URL"

I think if an unrecognized URL is provided, there should be a message explaining that it is not recognized as a persistent URL, but the file can still be used if it's a direct link to a RPZ. From there the user would be able to click a button to have ReproServer download the RPZ and create a unique ID for it, just like if it had been uploaded.

JSON API

Originally opened 2017-11-09 21:29 EST by @remram44

Current views:

• /upload is POST-ing a file, redirects
• /reproduce when build is not done: has HTML and JSON response
• /reproduce when build is done: only HTML, JSON should have parameter list
• /run is POST, gets form data, redirects. Should accept JSON
• /results: only HTML, should return JSON results

We should separate the JSON endpoints under /api/ or similar. Also, do we need a public API?

Design changes

• Add progress bar below "Run log" section after clicking both manual crawl and automated crawl (https://staging.server.reprozip.org/web/kyjcn?port_number=5000) It can be the same desigin as the upload package section on the home screen

Web capture: design changes

• For hostname and port number, shorten the input box so they only extends to the middle of the screen size.
• Put "Hostname:" and "Port number:" on top of the input boxes instead of left
• Next to "Hostname", put "(optional) Enter the hostname the server is expecting" in italic on top of the input box. Text color should be #757575
• Under the hostname input box, add "Example: localhost" in italic and #757575
• Next to "Port number", put "Enter the port number that the web server listens on" in italic and #757575
• Under the port number input box, add "Example: 3000" in italic and #757575

Link back to file on provider from reproduce_provider page

Originally opened 2020-01-16 16:18 EST by @remram44

When seeing the reproduction form for a file, and that file is from a provider, we should link back to the file on the provider.

How to scale ReproServer?

Originally opened 2019-11-14 22:36 EST by @remram44

Not a problem for now, but worth thinking about.

Running multiple web pods wouldn't work because specific ones would know about specific builds/runs.

Changing VM states to not trigger multiple builds might be good but still cause issues. Plus container restart would have all reproservers watching all running pods.

I think the best option is probably to keep one reproserver per cluster, and just run multiple clusters. This is also an opportunity for federation, e.g. there could be "nyu reproserver" and "umass reproserver" and links would point to one specific server and everything would work. It could also be used to have institutional instances providing better quotas to their own employees.

Show more details about RPZ package on its web page

Stream logs instead of polling

Originally opened 2019-09-11 17:12 EDT by @remram44

Now that we have an async web framework. Use a websocket or long-polling.

Adding another home page

Suggestion
It is unclear especially to nontechnical users what the whole process is. Making another home page that explains the process, and having an option to go to either web capture or upload would be helpful.

Add web capture description

Could you add the following paragraph instead of lorem ipsum on this page?
[https://staging.server.reprozip.org/web]

After generating a ReproZip bundle (.rpz) file that contains the server-side elements of your app, you can archive the client-side assets using Web Capture.
Upload your .rpz file below and start crawling.

It will look something like this