I haven't posted in a while, but things have been going on. I thought I'd post about some of the more interesting aspects. I've recently began a fairly regular contracting deal with an interesting company, and I'll have to be a little vauge on some aspects, because of NDAs and such.
During one of my usual nights of aiding the Pythoners of #python on irc.freenode.net, I was discussing a project someone was trying to complete and the long debate about various routes that could be taken led to me being contracted for the job, which I've had fun with. I've been contracted to build a Fuse package, which uses Amazon's S3 as its storage mechanism. It is a fun system to work with, because of its simplicity and challenging limitations. For example, all operations are redundant, but non-atomic, because the same data could be changed at the same time, and its unpredictable how it would propogate across their redundant network. Mostly this hasn't been an issue, because you have almost the same trust in file locks on a local system anyway, and the only issues have been how to ensure integrity within directory entries and file node chains.
This aspect of the work is to be released under the GPL upon completion, and hopefully I can factory out some things that will be useful for other uses of the S3, which I've developed for the use in this project. I'll try to factor out modules for the following features:
- Define classes that represent a type of data stored in an S3 entry
- Easily define meta-attributes, with coercers and defaults
- Unique IDs generated for entries
- "Sub-Entry" concept, where one entry is owned by another
- Caching of data both in disk and over memcache, with an open API to implement other cache-types, like local-memory caches, or even other web services.
- Node entries, which can span data across multiple entries for more efficient (and cost effective) reads and writes that do not involve the entire data buffer.
- Test facilities for both BitBucket (a Python S3 access package I use) and Python-MemCached, which I use for offline testing. Both mirror all the functionalty (read: most) of the related projects, so they can be tested against without actual network use.
If you are interested, look forward to the S3 packages I'll wrapping up this weekend. Hopefully, someone will find them useful.