Tag Archives: Technical

Fixed Cross-Site Scripting Vulnerability on Zoomapper App

On 9 November 2020, a security researcher notified us of a cross-site scripting (XSS) vulnerability on our zoomapper application. This service hosts tile sets that are used to render maps for a small number of other Zooniverse applications, but is not connected to any critical Zooniverse infrastructure. This XSS vulnerability could have allowed users to execute malicious code on the zoomapper application in the browser.

We were able to remediate the vulnerability within hours of the report by disabling the browser GUI for zoomapper (see PR #6). The GUI had been turned on by default for the zoomapper app, but is not necessary to fulfill the app’s intended role.

Additional notes on the incident:

  • The vulnerability existed since the app was first deployed on September 15th 2020.
  • The vulnerability was located in the underlying Tileserver-GL dependency.
  • No Zooniverse user or project data was vulnerable or exposed by this vulnerability.

We’d like to thank Rachit Verma (@b43kd00r) for bringing this issue to our attention and for following responsible disclosure by reporting it to us in private, as requested on our security page.

Caesar Subject Rule Effect Vulnerability Report

In the beginning of April 2020, we were notified that subjects from one Zooniverse project were appearing in the subject set of a separate project where they did not belong. In our investigation of the issue, our team determined that this behavior was being caused by a Caesar configuration mistake that used an incorrect Subject Set ID. Project owners using Caesar were able to create Subject Rule Effects that added subjects to collections or subject sets, even without proper subject set editing permissions. We have rectified the issue surrounding Subject Rule Effects and eliminated this vulnerability, and would like to share the details for anyone who is interested.

The issue was raised by project lead James Perry (@JamesPerry), who reported that subjects that didn’t belong to his project were appearing in his subject sets.  Due to a mistyped subject set ID in a Caesar `add_to_subject_set` effect for an unrelated project, that Subject Rule Effect was sending subjects from that project to one of James’s subject sets instead of the correct target.

Our immediate course of action was to fix the project impacted by the vulnerability, and push out a temporary code fix to prevent the vulnerability from being exploited. 

  1. To fix the affected project, we updated the incorrect subject set id for the project that was incorrectly sending subjects to the wrong project and removed the unwanted subjects from the set. 
  2. On April 3rd we deployed a temporary code fix to disable Subject Rule Effect creation and modification for all but admin users (see PR #1109). This change was communicated to affected teams that were most impacted by the change, and teams that reached out after seeing our notification banner or encountering a Caesar interface error.

On May 15th we pushed out a permanent fix that checked the user has permissions to send data to the target subject set or collection. Specifically, the updated validation code checks that the user has update permissions on the project the subject set or collection is linked to. (see PRs #1115, #1129 and #1131). 

For anyone running their own hosted copy of Caesar, we recommend pulling these changes as soon as you’re able.

Panoptes CLI 1.1 now available

I recently released version 1.1 of the Panoptes CLI – the command-line interface for managing Zooniverse projects. This update includes some exciting new features. Here are the highlights.

You can install the update by running pip install -U panoptescli. Any bugs or issues should be raised via GitHub. See the changelog for the full list of changes.

Resuming failed subject uploads

This one adds what is probably the CLI’s most requested feature: the ability to resume a failed upload from where it left off, without duplicating subjects or requiring manual changes to the manifest. I hope this will be a huge help to researchers, especially when uploading large manifests.

If the upload fails for any reason – whether that’s an issue with our systems, a problem with your internet connection, a bug in the CLI itself, or if you just decide to stop the upload by pressing ctl-c – the CLI will detect that there was a problem and will ask you if you want to be able to resume the upload later. If you say yes, it will then save a new manifest in YAML format containing the remaining upload queue along with all of the upload’s command line options. Then to resume, you just start a new upload with the YAML manifest instead of the original CSV.

Multithreaded subject uploads

Uploading new subjects can often take a long time. The total upload time depends not only on your internet connection speed, but also on the time it takes for the CLI to talk to the Panoptes API. Creating a new subject typically requires the CLI to make two HTTP requests: one to create the subject and one to upload the subject’s media (the image, or video, or whatever). If the subject has multiple images then that only increases the number of requests. Plus subjects need to be linked to the subject set; this happens in batches, but it can still add up to a lot of requests for large uploads. If you’re uploading 10,000 subjects for example, that means the CLI has to make a minimum of 20,000 requests (probably more), and each of those requests includes some overhead where the CLI is waiting for the server to respond, which is all basically wasted time.

Luckily the Panoptes CLI 1.1 gets around that, by taking advantage of the multithreading features of the Panoptes Client for Python which were released earlier this year. Now, those 20,000 requests will happen five at a time, so for example three of them can be sending data while two of them are waiting for the server, meaning your internet connection is fully utilised the whole time and no time is wasted. In my testing, this substantially sped up subject uploads, potentially saving hours of your time.

Adding and removing lists of subjects to and from subject sets

Often project owners need to add large numbers of existing subjects to a new set, or remove subjects from their current set. It was possible to do this with the previous version of the CLI by passing subject IDs on the command-line, but it was often difficult to modify large numbers of subjects this way (it was possible with xargs on Linux and macOS, but this isn’t the most intuitive way to do it).

Now, there’s a new option to pass a list of IDs in a text file rather than having to specify IDs on the command-line. (The old way is still there too if you prefer to do it that way!) Just produce a text file containing the relevant subject IDs, one per line. If you already have the subject information in a spreadsheet, exporting a CSV file with just the subject ID column will produce the right file (just make sure it only contains the one column).

For example, if you have a file called subject_ids.csv containing the following:


You can run:

panoptes subject-set add-subjects -f subject_ids.csv 1357

to add subjects 1234, 5678, and 9012 to subject set 1357.


Edited 29 November 2019: Fixed typo in pip command for upgrade.

The Zooniverse is Now Powered by Kubernetes

We recently finished the first stage in a pretty big change to our web hosting infrastructure. We’ve moved all of our smaller backend services (everything except Panoptes, Ouroboros, and frontend code) into a Kubernetes cluster. I’m pretty excited about this change, so I wanted to share what we’ve done and what we’ll be doing next.

Kubernetes is what’s called a container orchestration system, which is a system that lets us run applications on a cluster of servers without having to worry about which specific server each thing is running on. There are a few different products out there that do this sort of thing, and prior to this we were using Docker Swarm. We didn’t find Docker Swarm to be a great fit for us, but we’re really pleased with Kubernetes and what it’s letting us do.

As a result of moving to Kubernetes, we’ve been able to fully automate the process of updating our server-side apps when we make changes to the code. This automation is important, because it means that the process of deploying updated code is no longer a bottleneck in our development process – it means that any member of our team can easily deploy changes, even in components they haven’t worked on before. This smooths out our development process and it should make our jobs a little easier, meaning we can more easily focus on the job of building the Zooniverse without our infrastructure getting in the way.

Not only has Kubernetes made it easier for us to automate things, but we’ve also found it to be a lot more reliable. So much so, in fact, that we’re now planning to move all of our web services into a Kubernetes cluster, including Panoptes and our main HTTP frontend servers. This is the part I’m really excited about! By making this change, we’ll be making our infrastructure a lot simpler to manage while also saving money by using our cloud computing resources more efficiently (since the cluster’s resources are pooled for everything to share). That should obviously be a huge win, because it will leave more time and money for everything else we do.

Watch this space for updates as we make more improvements to our infrastructure over the coming months!

Panoptes Client for Python 1.1

I’ve just released version 1.1 of the Panoptes Client for Python. The changelog has a full list of what’s new, but there are a few things I wanted to highlight, the first two of which will make it substantially faster to create new subjects:

  • Multithreaded media uploads – the client will automatically use several threads to upload media when you first save a new subject. So, for example, if you create a subject which has three images they will all upload simultaneously (up to five simultaneous uploads, then it will queue them).
  • Multithreaded subject creation – you can also simultaneously create the subjects themselves. That means if you’re creating, say, a thousand subjects, the client can queue them all and create up to five of them simultaneously. This works in conjunction with the media uploads, using one combined queue for the subject creation and the media uploads, to avoid overloading the network and to make sure the subject creation doesn’t get too far ahead of the uploads. This one isn’t automatic – you’ll need to create your subjects with the new SubjectSet.async_saves() context manager to take advantage of it.
  • Retries for all GET requests – we’re quite proud of how reliable the Zooniverse platform is, but sometimes server-side errors do happen. The client will now automatically retry all GET requests (i.e. the ones that don’t modify any data) if an error occurs, improving reliability.
  • Retries for batch linking operations – similar to above, the client will retry any add/remove operations via the new LinkCollection class, which handles linking groups of objects (i.e. subjects to a subject set, subjects to a collection, etc.). This means you should see far fewer failures when linking thousands of subjects to a subject set, for example.
  • Context manager for multiple connections – the Panoptes class can now act as a context manager, providing a safe way to perform operations as multiple users (for example, in a web app).

You can install the update by running pip install -U panoptes-client. Any bugs or issues should be raised via GitHub.

Fixed Cross-Site Scripting Vulnerability on Project Page’s External Links

We recently fixed a security vulnerability that existed in the external/social links (e.g. to Twitter) of projects. Prior to this fix, it was possible for project owners to do two things: 1. create external links that ran malicious JavaScript code if they were clicked (e.g. allowing attackers to capture a user’s login session), and 2. create a link to a malicious website disguised as a “legitimate” link to a known Social website (e.g. a Twitter link that actually directed users to the spoof website “twiitteerrer.com”). Our patches fix both issues, and a follow up investigation revealed that there is no indication this vulnerability was exploited by anyone.

The security issue was discovered on 11 Dec 2018 during an internal security check. The first patch in the series (addressing the major JavaScript injection vulnerability) was deployed within 6 hours, and the final patch (addressing the relatively less harmful issue of spoof-able social links) deployed 2 days later.

The fixes for this vulnerability are contained in pull requests #5141, #5142, and #5148 of the Panoptes Front End project on GitHub. Anyone running their own hosted copy of this should pull these changes as soon as possible.

Additional notes on our investigation are as follows:

  • The vulnerability was introduced on 14 May 2015, in pull request 324.
  • Custom links for projects and organisations allowed project builders to cause a user’s browser to execute arbitrary javascript by entering URLs like javascript: alert('oh no'); if the user clicked on that link, the javascript would run.
  • Malicious javascript executed this way could do whatever it wanted on the site, i.e. it could have stolen logged-in users’ API tokens, logged users out and captured their passwords when re-logging in.
  • Theoretically, passwords may have been have been exposed if malicious Javascript captured them on login, though this would only impact users that click malicious links. Emails may have been exposed, notably if an admin user account was breached.
  • However, we audited the database but could find no evidence (other than our own tests) of this having been done by project owners.
  • Our current solution is to sanitise all external/social links – both when taking input from users and when rendering them on webpages – and only allowing standard website URLs to pass.

As a side effect of our fixes, project owners are now unable to add non-standard website URLs to their project’s external links – for example, https://example.com continues to work fine, but mailto:hello@example.com no longer does.

We apologise for any concern this issue may have caused.

Zooniverse Data Aggregation

Hi all, I am Coleman Krawczyk and for the past year I have been working on tools to help Zooniverse research teams work with their data exports.  The current version of the code (v1.3.0) supports data aggregation for nearly all the project builder task types, and support will be added for the remaining task types in the coming months.

What does this code do?

This code provides tools to allow research teams to process and aggregate classifications made on their project, or in other words, this code calculates the consensus answer for a given subject based on the volunteer classifications.  

The code is written in python, but it can be run completely using three command line scripts (no python knowledge needed) and a project’s data exports.


The first script is the uses a project’s workflow data export to auto-configure what extractors and reducers (see below) should be run for each task in the workflow.  This produces a series of `yaml` configuration files with reasonable default values selected.


Next the extraction script takes the classification data export and flattens it into a series of `csv` files, one for each unique task type, that only contain the data needed for the reduction process.  Although the code tries its best to produce completely “flat” data tables, this is not always possible, so more complex tasks (e.g. drawing tasks) have structured data for some columns.


The final script takes the results of the data extraction and combine them into a single consensus result for each subject and each task (e.g. vote counts, clustered shapes, etc…).  For more complex tasks (e.g. drawing tasks) the reducer’s configuration file accepts parameters to help tune the aggregation algorithms to best work with the data at hand.

A full example using these scripts can be found in the documentation.

Future for this code

At the moment this code is provided in its “offline” form, but we testing ways for this aggregation to be run “live” on a Zooniverse project.  When that system is finished a research team will be able to enter their configuration parameters directly in the project builder, a server will run the aggregation code, and the extracted or reduced `csv` files will be made available for download.

Panoptes Client for Python 1.0.3

Hot on the heels of last week’s update, I’ve just released version 1.0.3 of the Python Panoptes Client, which fixes a bug introduced in the previous release. If you encounter a TypeError when you try to create subjects, please update to this new version and that should fix it.

This release also updates the default client ID that is used to identify the client to the Panoptes API. This is to ensure that each of our API clients is using a unique ID.

As before, you can install the update by running pip install -U panoptes-client.

Why you should use Docker in your research

Last month I gave a talk at the Wetton Workshop in Oxford. Unlike the other talks that week, mine wasn’t about astronomy. I was talking about Docker – a useful tool which has become popular among people who run web services. We use it for practically everything here, and it’s pretty clear that researchers would find it useful if only more of them used it. That’s especially true in fields like astronomy, where a lot of people write their own code to process and analyse their data. If after reading this post you think you’d like to give Docker a try and you’d like some help getting started, just get in touch and I’ll be happy to help.

I’m going to give a brief outline of what Docker is and why it’s useful, but first let’s set the scene. You’re trying to run a script in Python that needs a particular version of NumPy. You install that version but it doesn’t seem to work. Or you already have a different version installed for another project and can’t change it. Or the version it needs is really old and isn’t available to download anymore. You spend hours installing different combinations of packages and eventually you get it working, but you’re not sure exactly what fixed it and you couldn’t repeat the same steps in the future if you wanted to exactly reproduce the environment you’re now working in. 

Many projects require an interconnected web of dependencies, so there are a lot of things that can go wrong when you’re trying to get everything set up. There are a few tools that can help with some of these problems. For Python you can use virtual environments or Anaconda. Some languages install dependencies in the project directory to avoid conflicts, which can cause its own problems. None of that helps when the right versions of packages are simply not available any more, though, and none of those options makes it easy to just download and run your code without a lot of tedious setup. Especially if the person downloading it isn’t already familiar with Python, for example.

If people who download your code today can struggle to get it running, how will it be years from now when the version of NumPy you used isn’t around anymore and the current version is incompatible? That’s if there even is a current version after so many years. Maybe people won’t even be using Python then.

Luckily there is now a solution to all of this, and it’s called software containers. Software containers are a way of packaging applications into their own self-contained environment. Everything you need to run the application is bundled up with the application itself, and it is isolated from the rest of the operating system when it runs. You don’t need to install this and that, upgrade some other thing, check the phase of the moon, and hold your breath to get someone’s code running. You just run one command and whether the application was built with Python, Ruby, Java, or some other thing you’ve never heard of, it will run as expected. No setup required!

Docker is the most well-known way of running containers on your computer. There are other options, such as Kubernetes, but I’m only going to talk about Docker here.

Using containers could seriously improve the reproducibility of your research. If you bundle up your code and data in a Docker image, and publish that image alongside your papers, anyone in the world will be able to re-run your code and get the same results with almost no effort. That includes yourself a few years from now, when you don’t remember how your code works and half of its dependencies aren’t available to install any more.

There is a growing movement for researchers to publish not just their results, but also their raw data and the code they used to process it. Containers are the perfect mechanism for publishing both of those together. A search of arXiv shows there have only been 40 mentions of Docker in papers across all fields in the past year. For comparison there have been 474 papers which mention Python, many of which (possibly most, but I haven’t counted) are presenting scripts and modules created by the authors. That’s without even mentioning other programming languages. This is a missed opportunity, given how much easier it would be to run all this code if the authors provided Docker images. (Some of those authors might provide Docker images without mentioning it in the paper, but that number will be small.)

Docker itself is open source, and all the core file formats and designs are standardised by the Open Container Initiative. Besides Docker, other OCI members include tech giants such as Amazon, Facebook, Microsoft, Google, and lots of others. The technology is designed to be future proof and it isn’t going away, and you won’t be locked into any one vendor’s products by using it. If you package your software in a Docker container you can be reasonably certain it will still run years, or decades, from now. You can install Docker for free by downloading the community edition.

So how might Docker fit into your workday? Your development cycle will probably look something like this: First you’ll probably outline an initial version of the code, and then write a Dockerfile containing the instructions for installing the dependencies and running the code. Then it’s basically the same as what you’d normally do. As you’re working on the code, you’d iterate by building an image and then running that image as a container to test it. (With more advanced usage you can often avoid building a new image every time you run it, by mounting the working directory into the container at runtime.) Once the code is ready you can make it available by publishing the Docker image.

There are three approaches to publishing the image: push the image to the Docker Hub or another Docker registry, publish the Dockerfile along with your code, or export the image as a tar file and upload that somewhere. Obviously these aren’t mutually exclusive. You should do at least the first two, and it’s probably also wise to publish the tar file wherever you’d normally publish your data.


The Docker Hub is a free registry for images, so it’s a good place to upload your images so that other Docker users can find them. It’s also where you’ll find a wide selection of ready-built Docker images, both created by the Docker project themselves and created by other users. We at the Zooniverse publish all of the Docker images we use for our own work on the Docker Hub, and it’s an important part of how we manage our web services infrastructure. There are images for many major programming languages and operating system environments.

There are also a few packages which will allow you to run containers in high performance computing environments. Two popular ones are Singularity and Shifter. These will allow you to develop locally using Docker, and then convert your Docker image to run on your HPC cluster. That means the environment it runs in on the cluster will be identical to your development environment, so you won’t run into any surprises when it’s time to run it. Talk to your institution’s IT/HPC people to find out what options are available to you.

Hopefully I’ve made the case for using Docker (or containers in general) for your research. Check out the Docker getting started guide to find out more, and as I said at the beginning, if you’re thinking of using Docker in your research and you want a hand getting started, feel free to get in touch with me and I’ll be happy to help you.