perkeep/doc/overview.md

Camlistore Overview

Camlistore is your personal storage system for life.

Summary

The project began because I wanted to...

• ... store all my stuff forever, not worrying about deleting, or losing stuff.

• ... save stuff easily, and without categorizing it or choosing a location whenever I save it. I just want a data dumptruck that I can throw stuff at whenever.

• ... never lose anything because nothing can be overwritten (all blobs are content-addressable), and there's no delete support. (optional garbage collection coming later)

• be able to search for anything I once stored.

• be able to browse and visualize stuff I've stored.

• ... not always be forced into a POSIX-y filesystem model. That involves thinking of where to put stuff, and most the time I don't even want filenames. If I take a bunch of photos, those don't have filenames (or not good ones, and not unique). They just exist. They don't need a directory or a name. Likewise with blog posts, comments, likes, bookmarks, etc. They're just objects.

• ... have a POSIX-y filesystem when I want one. And it should all be logically available on my tiny laptop's SSD disk, even if my laptop's disk is miniscule compared to my entire repo. That is, there should actually be a caching virtual filesystem, not a daemon running rsync in the background. If I have to have a complete copy of my data locally, or I have to "choose which folders" to sync, that's broken.

• ... be able to synthesize POSIX-y filesystems from search queries over my higher-level objects. e.g. a "recent" directory of recent photos from my Android phone (this all works already in 0.1)

• Not write another CMS system, ever. Camlistore should be able to store and model any type of content, so it can just be a backend for other apps.

• ... have backups of all my social network content I created daily on other people's servers, to protect myself if my account is hijacked, the company goes evil, changes ownership, or goes out of business.

• ... have both a web UI and command-line tools, as well as a FUSE filesystem.

• ... be in control of my data, but also still be able to utilize big companies' infrastructure cloud products if desired.

• ... be able to share content with both technical and non-technical friends.

Most of this works as of the 0.1 release, and the rest and more is in progress.

Longer Answer

Throughout our life, we all continue to generate content, whether that's writing documents, taking photos, writing comments online, liking our friends' posts on social networks, etc. Our content is typically spread between a mix of different companies' servers ("The Cloud") and your own hardware (laptops, phones, etc). All of these things are prone to failure: companies go out of business, change ownership, or kill products. Personal harddrives fail, laptops and phones are dropped.

It would be nice if we were a bit more in control. At least, it would be nice if we had a reliable backup of all our content. Once we have all our content, it's then nice to search it, view it, and directly serve it or share it out to others (public or with select ACLs), regardless of the original host's policies.

Camlistore is a system to do all that.

While Camlistore can store files like a traditional filesystem (think: "directories", "files", "filenames"), its specialized in storing higher-level objects, which can represent anything.

In addition to an implementation, Camlistore is also a schema for how to represent many types of content. Much JSON is used.

Because every type of content in Camlistore is represented using content-addressable blobs (even metadata), it's impossible to "overwrite" things. It also means it's easy for Camlistore to sync in any direction between your devices and Camlistore storage servers, without versioning or conflict resolution issues.

Camlistore can represent both immutable information (like snapshots of filesystem trees), but can also represent mutable information. Mutable information is represented by storing immutable, timestamped, GPG-signed blobs representing a mutation request. The current state of an object is just the application of all mutation blobs up until that point in time. Thus all history is recorded and you can look at an object as it existed at any point in time, just by ignoring mutations after a certain point.

Despite using parts of the OpenPGP spec, users don't need to use the GnuPG tools or go to key signing events or anything dorky like that.

You are in control of your Camlistore server(s), whether you run your own copy or use a hosted version. In the latter case, you're at least logically in control, analagous to how you're in charge of your email (and it's your private repository of all your email), even if a big company runs your email for you. Of course, you can also store all your email in Camlistore too, but Gmail's interface and search is much better.

Responsible (or paranoid) users would set up their Camlistore servers to cross-replicate and mirror between different big companies' cloud platforms if they're not able to run their own servers between different geographical areas. (e.g. cross-replicating between different big disks stored within a family)

A Camlistore server comprises several parts, all of which are optional and can be turn on or off per-instance:

• Storage: the most basic part of a Camlistore server is storage. This is anything which can Get or Put a blob (named by its content-addressable digest), and enumerate those blobs, sorted by their digest. The only metadata a storage server needs to track per-blob is its size. (No other metadata is permitted, as it's stored elsewhere) Implementations are trivial and exist for local disk, Amazon S3, Google Storage, etc. They're also composable, so there exists "shard", "replica", "remote", "conditional", and "encrypt" (in-progress) storage targets, which layer upon others.

• Index: index is implemented in terms of the Storage interface, so can be synchronously or asynchronously replicated to from other storage types. Putting a blob indexes it, enumerating returns what has been indexed, and getting isn't supported. An abstraction within Camlistore similar to the storage abstractions means that any underlying system which can store keys & values and can scan in sorted order from a point can be used to store Camlistore's indexes. Implementations are likewise trivial and exist for memory (for development), SQLite, LevelDB, MySQL, Postgres, MongoDB, App Engine, etc. Dynamo and others would be trivial.

• Search: pointing Camlistore's search handlers at an index means you can search for your things. It's worth pointing out that you can lose your index at any time. If your database holding your index goes corrupt, just delete it all and re-replicate from your storage to your index: it'll be re-indexed and search will work again.

• User Interface: the web user interface lets you click around and view your content, and do searches. Of course, you could also just use the command-line tools or API.

Enough words for now. See the docs and code for more.

Last updated 2013-06-12