The Timeless Stack?
The Timeless Stack is a software ecosystem that's about making computers reliable.
Now in html book form!
The latest docs are in the process of being ported to a html book format. You can read that at https://polydawn.github.io/ . (The source is in the "book" dir in this repo.)
What is "reliable"?
Reliable means the system changes when you want it to, and stays the same when not explicitly changed.
Change needs to be easy. Updates need to be simple. Simultaneously, control is paramount: unexpected change and unwanted updates need to be preventable.
Reliability also implies decentralization. Relying on a remote system is not reliable. Managing change control based on a network service is not reliable. Change control needs to be something that different systems and different people can own, individually, and needs to operate free-standing.
What is our edge?
To be reliable, the Timeless Stack builds everything around snapshots, immutable data, and reproducible transformations.
To do this, the Timeless Stack encompasses...
- handling snapshots of files
- building software in containers
- publishing releases that associate names and metadata to filesets snapshots
- and distributing packages of filesets that assemble into useful systems!
It's a lot of responsibilities. We've broken them up into several tools. Use the parts you need. Ditch the ones you don't. Each tool is API-driven and can be used in isolation, or glues together easily with the others.
At the foundation is Timeless Repeatr: a tool that defines a total sandbox for data processing and running programs, which works together with the concept of a Timeless Fileset in order to be reproducible anywhere, anywhen. Repeatr consumes a document we call a Formula: in keeping with the theme, it's all timeless. No dates; just Filesets.
Timeless Hitch records IDs for Timeless Filesets together with the Timeless Repeatr Formulas that produced them. Timeless Hitch is both a release management system, an update tracking database, and build instructions for everything it tracks: with Timeless Hitch, we finally have all the advantages of a source distribution and a binary distribution in the same system.
Jump down to the 'Tools' heading for an overview of the other core tools; but we recommend scanning the 'Concepts' heading first, for vocab.
|Data/Files||Fileset||A set of files and directories, including standard posix metadata. Nothing special :)|
|Data/Files||Ware||The "packed" form of a Fileset. Tarballs, git commits; many formats are defined. Wares are immutable and indexed by 'WareID'.|
|Data/Files||WareID||The hash identifying a Ware. Holding a WareID gives you an immutable reference to a Ware which you can unpack into a Fileset.|
|Formula||A complete definition of a computation. Pins all input filesystems by WareID. Runs in a sandbox. Running produces a 'runrecord'.|
|Formula.Inputs||A group of Filesets and where to mount them in a directory tree when setting up the formula and preparing to exec.|
|Formula.Action||A script to run inside the container when evaluating a formula.|
|Formula.Output||A list of paths to save when the formula's action is complete. Files under an output path will be exported as a Ware, and when the run is complete, the WareIDs are reported.|
|SetupHash||The hash of a 'formula'. This is the unique name of a computation! Use it as a primary key; memoize builds with it!|
|RunRecord||The result of running a 'formula'. Includes a 'results' map, and also metadata like timestamp and UID (thus, runrecords are always unique).|
|Results||The group of WareIDs produced when a 'formula' is evaluated -- one WareID for every 'output' path!|
|Pipelines of Compute||Module||A graph (a DAG) describing a series of proto-formulas: each "step" in the graph will be resolved to one formula, evaluated, and then its outputs can be used as inputs in other steps.|
||github||Repeatr evaluates formulas.
||github||R-I/O stands for Repeatable Input/Output: it's tooling for packing Filesets into Wares, mirroring Wares between storage systems, and unpacking Filesets fetched by WareID. Glue for getting files to and from other systems, in other words.|
||github||Stellar builds larger pipelines: use it to stich together several formulas by passing wares between them. Entire pipelines quickly rebuild incrementally, automatically memoizing any steps that haven't changed.|
||github||Hitch associates human-readable names and metadata to WareIDs. It's a release tracking system. Stellar also integrates these features; hitch is for making other manual alterations.|
||planned||The Timeless package manager: helps manage $PATH and environments to|
All parts of this stack are loosely coupled:
hitchcan be driven entirely by shell script to make releases or look up releases and metadata.
repeatrworks entirely with formulas. You can template formulas with
reppl, or with anything you like -- the API is simply json/yaml.
tlpkgis built to handle packages that play well in the Timeless ecosystem -- but there's nothing to say you can't use
See the ecosystem map for a big-picture layout of how the projects are connected.
||github||A collection of repeatr formulas. You can use this as reference materials. (All of our base image snapshots come from here!)|
How do I get started doing something Fun?
Check out the omnibus dir for an example of using Repeatr and Rio to build... Rio, Repeatr, and the rest of the Timeless Stack!
You'll need to grab the binary releases in order to bootstrap (or, build the individual tools from source).
This omnibus example uses basic bash string templating to compose pipelines just to keep things simple and make it utterly clear that you can compose objects for this API any way you want. You may want to use a more powerful language when building your own higher level pipelines. We're explicitly committed to supporting that; the Repeatr JSON API is meant to be used freely.
Why hashes? Why can't I just download "somefoo-vbar.baz.tar.gz"?
Repeatr and the Timeless Stack uses hashes for WareIDs for the same reason stores use SKUs and part numbers: consistency and reliability.
As an individual, you can go to a hardware store and ask for a 5mm screw. But you haven't really picked yet by saying that: the store staff will show you to an entire rack of screws.
If you're the manager of a factory, you order your screws by their part number, because you don't want to re-order screws halfway through a project and end up with a subtly different threading, or screws that are too long in some other dimension, and so forth. Similarly, as a hardware store manager, your organize your stocks and reorder supplies by their SKUs, because consistency is important.
Store bins may have labels of all sorts of descriptions in various precision to help you select which items you need, but the store itself is tracking inventory by SKUs, because a single, precise identifier is vastly simpler to work with.
Repeatr is the same: we use hashes to identify files and data because it's simple and it's consistent and it's immune to silent changes over time. End-to-end use of hashes is part of what makes the Timeless Stack... timeless.
Keeping in touch
If you'd like to follow the Timeless project and get occasional email about releases and development milestones:
- We have an (extremely) low-volume mailing list for announcements: firstname.lastname@example.org
If you'd like to discuss development, ask questions, or just lurk the community: