coop-cloud/docs.coopcloud.tech
coop-cloud
/
docs.coopcloud.tech
4
1
Fork 14
Code Issues Pull Requests 2 Releases Activity
docs.coopcloud.tech/docs/intro/faq.md

15 KiB
Raw Blame History

title
Frequently asked questions

What is the Co-op Cloud?

Co-op Cloud aims to make hosting libre software apps simple for small service providers such as tech co-operatives who are looking to standardise around an open, transparent and scalable infrastructure. It uses the latest container technologies and configurations are shared into the commons for the benefit of all.

Who is behind the project?

The project was started by workers at Autonomic which is a worker-owned co-operative. We provide technologies and infrastructure to empower users to make a positive impact on the world. We're using Co-op Cloud in production, amongst other systems.

Why Co-op Cloud?

Pros

  • 👍 Thin "ease of use" layer on top of already standardised tooling
  • 👍 Extremely modular
  • 👍 Collective commons based configuration via public git repos
  • 👍 Focussed on hosting providers
  • 👍 Uses upstream published containers (no duplication on packaging)
  • 👍 Now and always libre software
  • 👍 Command line focussed
  • 👍 Horizontal and vertical scaling

Cons

  • 👎 Still a very young project
  • 👎 Limited availability of well tested apps
  • 👎 Requires command line knowledge to use
  • 👎 Currently x86 only (see this FAQ question for more)

Why start another project?

Please read our initial project announcement post for more on this.

Also see our strategy page.

How do I make a recipe for (package) an app?

See "Package your first recipe" for more.

What about $alternative?

We have various technical critiques of other similar projects which are already up-and-running in the ecosystem, as they don't necessarily meet our needs as a small tech co-op. However, Co-op Cloud isn't meant to be a replacement for these other projects.

Here is a short overview of the pros/cons we see, in relation to our goals and needs.

Cloudron

Pros

  • 👍 Decent web interface for app, domain & user management.
  • 👍 Large library of apps.
  • 👍 Built-in SSO using LDAP, which is compatible with more apps and often has a better user interface than OAuth.
  • 👍 apps are actively maintained by the Cloudron team.

Cons

  • 👎 Moving away from open source. The core is now proprietary software.
  • 👎 libre tier has a single app limit.
  • 👎 Based on Docker images, not stacks, so multi-process apps (e.g. parsoid visual editor for Mediawiki) are a non-starter.
  • 👎 Difficult to extend apps.
  • 👎 Only supported on Ubuntu LTS.
  • 👎 Upstreams libre software communities aren't involved in packaging.
  • 👎 Limited to vertical scaling.
  • 👎 Tension between needs of hosting provider and non-technical user.
  • 👎 LDAP introduces security problems - one vulnerable app can expose a user's password for all apps.
  • 👎 Bit of a black box.

YunoHost

Pros

  • 👍 Lovely web interface for app, domain & user management.
  • 👍 Bigger library of apps.
  • 👍 Awesome backup / deploy / restore continuous integration testing.
  • 👍 Supports hosting apps in subdirectories as well as subdomains.
  • 👍 Doesn't require a public-facing IP.
  • 👍 Supports system-wide mutualisation of resources for apps (e.g. sharing databases by default)

Cons

  • 👎 Upstream libre software communities aren't involved in packaging.
  • 👎 Uninstalling apps leaves growing cruft.
  • 👎 Limited to vertical scaling.
  • 👎 Not intended for use by hosting providers.

Caprover

Pros

  • 👍 Bigger library of apps.
  • 👍 Easy set-up using a DigitalOcean one-click app.
  • 👍 Works without a domain name or a public IP, in non-HTTPS mode (good for homeservers).
  • 👍 Deploy any app with a docker-compose.yml file as a "One Click App" via the web interface.
  • 👍 Multi-node (multi-server) set-up works by default.

Cons

  • 👎 Single-file app definition format, difficult to tweak using entrypoint scripts.
  • 👎 Nginx instead of Traefik for load-balancing.
  • 👎 Command-line client requires NodeJS / npm.
  • 👎 Requires 512MB RAM for a single app.
  • 👎 Backup/restore is "experimental", and doesn't currently help with backing up Docker volumes.
  • 👎 Exposes its bespoke management interface to the internet via HTTPS by default.

Ansible

Pros

  • 👍 Includes server creation and bootstrapping.

Cons

  • 👎 Upstream libre software communities aren't publishing Ansible roles
  • 👎 Lots of manual work involved in things like app isolation, backups, updates

Kubernetes

Pros

  • 👍 Helm charts are available for some key apps already.
  • 👍 Scale all the things.

Cons

  • 👎 Too big -- requires 3rd party tools to run a single-node instance.
  • 👎 Not suitable for a small to mid size hosting provider.

Docker-compose

Pros

Doing it Manually (Old School)

Pros

  • 👍 Simple - just follow upstream instructions to install and update.

Cons

  • 👎 Loads of manual work required for app isolation and backups.
  • 👎 Array of sysadmin skills required to install and maintain apps.
  • 👎 Hard to share configurations into the commons.
  • 👎 No idea who has done what change when.

Which technologies are used?

The core technologies of Co-op Cloud are libre software and enjoy wide adoption across software developer communities.

Why containers?

We use containers because so many libre software communities choose to use them. They are already writing and using Docker files and Docker-compose definitions for their development and production environments.

We can directly re-use this good packaging work and contribute back by helping maintain their in-repository files. We meet them where they are at and do not create a new packaging format or duplicate effort.

Co-op cloud proposes the idea of more direct coordination between distribution (app packagers) and production (developers) methods.

Aren't containers horrible from a security perspective?

It depends, just like any other technology and understanding of security. Yes, we've watched that CCC talk.

It's on us all as the libre software community to deliver secure software and we think one of the promises of Co-op Cloud is better cooperation with developers of the software (who favour containers as a publishing format) and packagers and hosters (who deliver the software to the end-user).

This means that we can patch our app containers directly in conversation with upstream app developers and work towards a culture of security around containers.

We definitely recommend using best-in-class security auditing tools like docker-bench-security, IDS systems like OSSEC, security profiles like Apparmor and hooking these into your existing monitoring, alert and update maintenance flows.

It's up to how you want to arrange your system. For example, Co-op Cloud also allows you to compartmentalise different apps onto different servers. You could stack a bunch of apps on one big server or you could deploy one app per server.

These are organisational concerns that Co-op Cloud can't solve for you which any software system will require. See this additional question for further information.

What is important to consider when running containers in production?

The Co-op Cloud uses containers as a fundamental building block. Therefore it is important to be aware of some general principles for container management in production environments. These are typically things that you will want to discuss within your co-op or democratic collective about how to prioritise and build up process for.

However, as the Co-op Cloud project is still very young, we're also still thinking about how we can make the platform itself mitigate problematic issues and make the maintenance of containers a more stable experience.

With that all in mind, here are some leading thoughts.

  • How do you install the Docker daemon itself on your systems and how do you manage upgrades? (system package, upstream Docker Inc. repository?)
  • How do you secure the Docker daemon from remote access (firewalls and system access controls).
  • How do you secure the Docker daemon socket within the swarm (locking the socket down, using things like a socket proxy)
  • How do you trust the upstream container registry (there are content trust mechanisms but it seems also useful to think about whether we need community registry infrastructure using tools like harbor or distribution. This involves a broader discussion with upstream communities.)
  • How do I audit my container security in an on-going process (IDS, OSSEC, Apparmor, etc.)
  • Can I run my containers with a non-root user setup?

And further reading on this topic:

Why use the Compose specification?

Every app packaged for the Co-op Cloud is described using a file format which uses the compose specification. It is important to note that we do not use the Docker compose tool itself to deploy apps using this format, instead we rely on Docker swarm.

The compose specification is a useful open standard for specifying libre software app deployments in one file. It appears to be the most accessible format for describing apps and this can be seen in the existence of tools like Kompose where the compose format is used as the day-to-day developer workflow format which is then translated into more complicated formats for deployment.

We are happy to see the compose specification emerging as a new open standard because that means we don't have to rely on Docker Inc. in the future - there will be more community tools available.

Why Docker Swarm?

While many have noted that "swarm is dead" it is in fact not dead. As detailed in the architecture overview page, swarm offers an appropriate feature set which allows us to support zero-down time upgrades, seamless app rollbacks, automatic deploy failure handling, scaling, hybrid cloud setups and maintain a decentralised design.

While the industry is bordering on a k8s obsession and the need to scale down a tool that was fundamentally built for massive scale, we are going with swarm because it is the tool most suitable for small technology.

We hope to see a container orchestrator tool that is not directly linked to a for-profit company emerge soon but for now, this is what we have.

If you want to learn more, see dockerswarm.rocks for a nice guide.

What licensing model do you use?

The Co-op Cloud is and will always be available under copyleft licenses.

Isn't running everything in containers inefficient?

It is true that if you install 3 apps and each one requires a MySQL database, then you will have 3 installations of MySQL on your system, running in containers.

Systems like YunoHost mutualise every part of the system for maximum resource efficiency - if there is a MySQL instance available on the system, then just make a new database there and share the MySQL instance instead of creating more.

However, as we see it, this creates a tight coupling between apps on the database level - running a migration on one app where you need to turn the database off takes down the other apps.

It's a balance, of course. In this project, we think that running multiple databases and maintaining more strict app isolation is worth the hit in resource efficiency.

It is easier to maintain and migrate going forward in relation to other apps and problems with apps typically have a smaller problem space - you know another app is not interfering with it because there is no interdependency.

It can also pay off when dealing with GDPR related issues and the need to have more stricter data layer separation.

How do I keep up-to-date with Docker Swarm developments?

Can I run Co-op Cloud on multiple servers?

Yes! Horizontal scaling is one of the ways Co-op Cloud can really shine. abra is designed to handle multiple servers from the first day. As long as you have a DNS entry pointing to your server then Co-op Cloud can serve apps (e.g. you can serve a wordpress1.mydomain.com from one server and a wordpress2.mydomain.com from another server) and abra handles this seamlessly.

Why only x86 support?

We would love to do ARM support and hope to get there! We've been testing this and ran into some issues. The TLDR; is that a lot of upstream libre app developer communities are not publishing container builds that support ARM. If they are, there are typically subtle differences in the conventions used to build the image as they are mostly done by community members and not directly taken on by the upstream project themselves. Since one of the core goals is to coordinate and reuse upstream packaging work, we see that ARM support requires a lot of organising and community engagement. Perhaps projects themselves will not want to take on this burden? It is not the role of the Co-op Cloud to set up an entire ARM publishing work flow at this moment in time. We see the benefits of supporting ARM and if you've got ideas / thoughts / approaches for how to make progress here, please get in touch.