Colin Walters
1ec7c30408
We'll always have "bare" mode for keeping files-as-hardlinks as root. But "archive" was my second attempt at a format for non-root file storage, used by the gnome-ostree buildsystem which runs as non-root. It was really handy to have a "tar" like mode where I can create tarballs as a user, that contain files owned by root for example. The "archive" mode stored content files as two pieces in the filesystem; ".file" contained metadata, and ".filecontent" was the actual content, uncompressed. The nice thing about this was that to check out a tree as non-root, you could just hardlink into the repo. However, archive was fairly bad for serving via HTTP; it required *two* HTTP requests per content object, greatly magnifing the already inefficient fetch process. So "archive-z2" was introduced. To allow gnome-ostree to still check out trees as a user, the "uncompressed-object-cache" was introduced, and that's how things have been working for a while. So we should just be able to kill this code. Specifically note just how much better the stage_object() function became. https://bugzilla.gnome.org/show_bug.cgi?id=706057
Repository design
At the heart of OSTree is the repository. It's very similar to git, with the idea of content-addressed storage. However, OSTree is designed to store operating system binaries, not source code. There are several consequences to this. The key difference as compared to git is that the OSTree definition of "content" includes key Unix metadata such as owner uid/gid, as well as all extended attributes.
Essentially OSTree is designed so that if two files have the same OSTree checksum, it's safe to replace them with a hard link. This fundamental design means that an OSTree repository imposes negligible overhead. In contrast, a git repository stores copies of zlib-compressed data.
Key differences versus git
- As mentioned above, extended attributes and owner uid/gid are versioned
- Optimized for Unix hardlinks between repository and checkout
- SHA256 instead of SHA1
- Support for empty directories
Binary files
While this is still in planning, I plan to heavily optimize OSTree for versioning ELF operating systems. In industry jargon, this would be "content-aware storage".
Trimming history
OSTree will also be optimized to trim intermediate history; in theory one can regenerate binaries from corresponding (git) source code, so we don't need to keep all possible builds over time.
MILESTONE 1
- Basic pack files (like git)
MILESTONE 2
- Store checksums as ay
- Drop version/metadata from tree/dirmeta objects
- Add index size to superindex, pack size to index
- So pull can calculate how much we need to download
- Split pack files into metadata/data
- pull: Extract all we can from each packfile one at a time, then delete it
- Restructure repository so that links can be generated as a cache; i.e. objects/raw, pack files are now the canonical
- For files, checksum combination of metadata variant + raw data
- i.e. there is only OSTREE_OBJECT_TYPE_FILE (again)
MILESTONE 3
- Drop archive/raw distinction - archive repositories always generate packfiles per commit
- Include git packv4 ideas:
- metadata packfiles have string dictionary (tree filenames and checksums)
- data packfiles match up similar objects
- Rolling checksums for partitioning large files? Kernel debuginfo
- Improved pack clustering
- file fingerprinting?
- ELF-x86 aware deltas
Related work in storage
git: http://git-scm.com/ Venti: http://plan9.bell-labs.com/magic/man2html/6/venti Elephant FS: http://www.hpl.hp.com/personal/Alistair_Veitch/papers/elephant-hotos/index.html
Compression
xdelta: http://xdelta.org/ Bsdiff: http://www.daemonology.net/bsdiff/ xz: http://tukaani.org/xz/