.arcfrees your architecture from infra and vendor cruft so you can focus on the real business logic of your app. Ship only the code that matters, iterate faster and enjoy unprecedented availability guarantees.
The iron age of compute began with racked physical servers. Early cloud compute evolved past physical servers into virtual machines. Virtual machines eventually gave way to containers and quickly containers have given rise to cloud functions. Each cycle has taught new lessons in software architecture and this most recent iteration brings new challenges.
- Config and tooling is designed for the last generation of metaphors
- AWS is massive and overwhelming with many similar, but not the same, products with divergent interfaces between interlocking services
- Deep proprietary knowledge is required to configure and maintain common infrastructure primitives
- Configuration and infrastructure can drift, leaving systems in states that are difficult to repeat / reproduce, and thus scale
- Painful manifest files; JSON is difficult to read, has no comments, and unforgiving to edit; YAML isn't much better and especially worse with deeply nested statements
Some of these problems have been tamed with infrastructure as code, creating repeatable and reproducible systems. The tradeoff is you are committing AWS configuration knowledge into your revision control systems.
.arc views infrastructure as a build artifact. And we prefer to not check build artifacts in with our code.
The .arc file
architect defines a high level vendor agnostic plaintext format,
.arc, as a manifest file to solve the specific problems laid out above.
- Focus on defining your app architecture with a subset of service primitives as high level definitions
npm scriptsto generate local code, configure, provision, and deploy cloud infrastructure from the
- You can still safely use the console tactically to access and administer primitives defined in
- The format, parser, and tooling are completely open to extension
.arcis also entirely portable between cloud vendors. (However no ports to clouds other than AWS have been made as of today.)
The .arc format
.arc format follows a few simple rules:
- Comments start with
- Sections start with
- Everything between sections becomes instructions for generating AWS infrastructure
.arc files are made up of the following sections:
@appdefines the application namespace
@htmlsection defines html routes
@jsondefines json routes
@eventsdefines application events you can publish and subscribe to
@slackdefines Slack API endpoints
@staticdefines S3 buckets for static assets
@scheduleddefines functions that run on a schedule
@tablesdefines DynamoDB database tables and trigger functions for them
@indexesdefines table global secondary indexes
This is a complete
.arc file example:
# .arc @app hello @html get / post /likes @json get /likes @events hit-counter @scheduled daily-affirmation rate(1 day) @tables likes likeID *String update Lambda @indexes likes date *String
npm run create in the same directory as the
.arc file above generates the following function code:
/ |-- src | |-- html | | |-- get-index/ | | `-- get-likes/ | |-- json | | `-- get-likes/ | |-- events | | `-- hit-counter/ | |-- scheduled | | `-- daily-affirmation/ | `-- tables | `-- likes-update/ |-- .arc `-- package.json
The code was also immediately deployed to the cloud in isolated
.arc format is terse, easy to read, and quickly learnable to author. The expressions in a
.arc file unlock the formerly complex tasks of cloud infrastructure provisioning, deployment, and orchestration.
Implementing principles and practices
architect follows many of the principles pioneered by agile and championed by devops. Versioning infrastructure. Tight feedback loops for dev, while maintaining isolation between stages. Systems that are consistent, inspectable, transparent, and extensible.
Architecture as text
.arcmanifest file defines architecture elements as plainly as possible in text
- Nesting and syntax is deliberately constrained
- And you can add comments!
Repeatable and consistent builds
arc-createonly creates, and never destroys; it skips if what it was to generate already exists
- Per above,
arc-createis intended to be run and re-run as your system changes and grows
- Use the AWS console to administer (i.e. remove infrastructure) or script destructive actions yourself;
Delivery is isolated from deployment
arc-sandboxallows you to work locally and offline from the cloud
productionas first class concepts
- Deployment to
productionrequires an extra manual step of setting
Extensible and flexible systems
arc-parseris open, and
@sectionsit does not know
.arcwith your own
@sectionsand workflows with