pulumi/docs/metadata.md

Mu Metadata Specification

This document contains a formal description of Mu's metadata. For more details on how this metdata is compiled when targeting various cloud providers, please refer to the companion design document.

Overview

The essential artifact a developer uses to create Stacks and Services is something we call a Mufile. It is conventionally named Mu.yaml, usually checked into the Workspace, and each single file describes a Stack. (Note that Stacks may internally contain other Stacks, however this is purely an implementation detail of how the Stack works.)

TODO(joe): declarative specification format for Clusters.

Although all examples are in YAML, it is perfectly valid to use JSON instead if that is more desirable.

Mu preprocesses all metadata files to substitute context values not known until runtime, such as configuration, arguments, and so on. The Go template syntax is used for this. Please refer to the API documentation for the context object (TODO(joe): do this) for details on what information is available.

Package Managament

Each Mufile begins with some standard "package manager"-like metadata, like name, version, description, and so on. As with most package managers, most of these elements are optional. For example:

name: elk
version: 1.0.1
description: A fully functioning ELK stack (Elasticsearch, Logstash, Kibana).
author: Joe Smith <joesmith@elk.com>
website: https://github.com/joesmith/elk

TODO(joe): full set of attributes.

In addition to basic metadata like this, any dependency packages must also be listed explicitly.

TODO(joe): finish this section.

Security

TODO(joe): we need the ability to override the default Role/ACLs/etc.

Stacks and Subclassing

A Stack may subclass any other Stack, specializing aspects of it as appropriate. This facilitates reuse. For instance, perhaps my company wishes to enforce that certain best practices and standards are adhered to, for all Stacks. Or imagine someone in the community has published a best-in-breed Node.js application blueprint, leveraging Express, MongoDB, and the ELK stack, and I merely want to plug in my own application logic and leverage the overall Stack.

To do this, reference another Stack's fully qualified name in the base property:

base: some/other/stack

From there, I can specify additional metadata, however we will have inherited everything from the base.

TODO(joe): what about mixins?

TODO(joe): get more specific about what can be overridden. Furthermore, what about "deletes"?

In addition to subclassing, a Stack may be marked abstract, indicating that it cannot be instantiated:

abstract: true

A non-abstract Stack must in fact have a non-zero number of Services, whereas an abstract one can omit Services entirely; this can be used, for example, to predefine certain non-Service metadata for subclassing Stacks.

APIs

Every Stack may choose to export one or more APIs. These APIs can be standard "unconstrained" network interfaces, such as "HTTP over port 80", or can take on a more structured form, like leveraging OpenAPI to declare a protocol interface. The benefits of declaring the full interfaces are that the RPC semantics are known to the system, facilitating advanced management capabilities such as diagnostics, monitoring, fuzzing, and self-documentation, in addition to RPC code- generation. This also adds a sort of "strong typing" to the connections between Services.

TODO(joe): articulate this section further; e.g., the metadata format, precise advantages, etc.

Stack Constructors

Each Stack can declare a set of properties that callers can set during creation:

properties:

This is a bag of property names to property values, each of which has the following:

• type: A property type, restricting the legal values.
• description: An optional long-form description of the property.
• default: A default value to be supplied if the caller doesn't supply one.
• optional: If true, this property can be omitted at creation time.
• readonly: If true, this property cannot be set on a resource after provisioning, without recreating it.
• perturbs: If true, this property can be set after provisioning, but doing so "perturbs" the live service.

For example:

properties:
    title:
        type: string
        description: The title of this thing.
        default: Anonymous

Types

The set of types a property may take on are "JSON-like". This includes simple primitives:

type: string
type: number
type: boolean
type: any

As well as array shapes utilizing them simply by appending a [] to any type:

type: string[]
type: number[]
type: boolean[]
type: any[]

And maps, in which the syntax is map[key]value, where key and value are any types:

type: map[string]string
type: map[number]any
// etc...

More complex types, in addition to names ones, can be created using a JSON Schema-like syntax. This may be done a number of ways, depending on what level of reuse you want for your custom types.

The simplest is to simply specify the type inline, as a sort of anonymous type:

type:
    base: string
    pattern: "[a-zA-Z0-9-]+"

Of course, types can be created from scratch, without depending on a base type:

type:
    properties:
        id:
            type: number
        name:
            type: string

Alternatively, we can declare a named schema type inside of the Mufile, and then reference it by name:

schema:
    private:
        literal:
            base: string
            pattern: "[a-zA-Z0-9-]+"
...
type: literal

This also has the advantage of being able to easily create arrays and maps of these types:

type: literal[]
type: map[string]literal

The final way to go about this is to create a stack whose sole purpose is to export schema types. The schema section can have a public notation to allow this stack to export types. Schema-only stacks must be marked abstract:

abstract: true
schema:
    public:
        literal:
            base: string
            pattern: "a-zA-Z0-9-]+"

From there, the usual stack referencing and package management concerns are consistent with what we've encountered.

Capability Types

The most interesting feature here is the ability to request a "capability", or reference to another Service. This provides a strongly typed and more formal way of expressing Service dependencies, in a way that the system can understand and leverage in its management of the system (like ensuring Services are created in the right order). It also eliminates some of the fragility of weakly typed and dynamic approaches, which can be prone to race conditions.

The most basic form is to use the special type service:

type: service

This is helpful, as it exposes a dependency to the system -- compared to dynamically discovering and depending on a name at runtime -- but it isn't perfect. The shape of the dependency is still opaque to the system.

Even better is to declare that we depend on a specific kind of Service, by specifying the fully qualified name of a Stack. In such a case, the system ensures an instance of this Stack type, or subclass, is provided:

type: ex/kvstore

This hypothetical Stack defines an API that can be used as a key-value store. Presumably we would find subclasses of it for etcd, Consul, Zookeeper, and others, which a caller is free to choose from at instantiation time.

Another example leverages the primitive mu/volume type to require a Service which can be mounted as a volume:

type: mu/volume

Note that anywhere inside of this Mufile, we may access the arguments supplied at Stack instantiation time using the Go template syntax mentioned earlier. For example, {{.args.tag.name}}.

Readonly and Perturbing Properties

A readonly property is one that cannot be changed after provisioning a resource without replacing it. This is often used by core "infrastructure" that cannot change certain properties after creation, for example, the data-center, virtual private cloud, or physical machine size. Although the tools allow you to change these, the mental model is that of creating a "new" object, and wiring up all of its dependencies all over again. As a result, the deployment process is more delicate, and may trigger a cascading recreation of many resources.

A perturbing property is one that can be changed after provisioning, but doing so requires perturbing the existing service in a way that may interrupt the live service. Modifying this isn't quite as impactful to the deployment process as modifying a readonly property, however it too must be treated with care.

TODO(joe): CloudFormation distinguishes between three modes: update w/out interruption, update w/ interruption, and replacement; I personally like the logical nature of readonly, however it's possible we should adopt something closer to it: http://docs.aws.amazon.com/AWSCloudFormation/latest/UserGuide/using-cfn-updating-stacks-update-behaviors.html.

Configuration

TODO(joe): write this section.

Services

After that comes the section that describes what Services make up this Stack:

services:

In this section is zero-to-many Services that are co-created with one another. Each Service has:

• A name, both for dynamic and static use.
• A type, which is just the name of a Stack to instantiate.
• A visibility governing whether consumers of this Stack have access to it or not.
• Any number of name/value properties, mapping to the Stack's settable properties.

Although these Services are co-created, they may reference one another. The references between each other forms a DAG and the system topologically sorts that DAG in order to determine the order in which to create and destroy Services. Notably there may be no cycles. By default, the system understands liveness and health (TODO(joe): how); as a result, the developer need not explicitly worry about races, liveness, or retries during Service creation.

Names

A Service's name can be set in one of two ways. The simplest is to use the "default", derived from the Stack type. For example, in the following metadata, the single Service has type nginx/nginx, gets a default name of nginx:

services:
    public:
        nginx/nginx:
            port: 80

Note that this is the latter part of the name; something called elasticsearch/kibana would get a name of kibana.

If we wish instead to give this an explicit name, say www, we can do so using the type property:

services:
    public:
        www:
            type: nginx/nginx
            port: 80

A Service's name is visible at runtime (e.g., in logs, diagnostics commands, and so on), in addition to controlling how metadata cross-referenes that Service. All Services live within a Stack, which of course has a name. Inside of a Stack, this outer name becomes its Namespace. For instance, inside of a Stack named marapongo/mu, a Service named x has a fully qualified name (FQN) of marapongo/mu/x. Although we seldom need the FQN for references within a single Stack, they are sometimes needed for inter-Stack references, in addition to management activities.

Types

Each Service has a type, which is merely the name of another Stack. Most of the time this is the FQN, although for references to other Stacks defined within the same Mufile (more on that later), this can just be a simple name. During instantiation of that Service, a fresh instance of that Stack is created and bound to in place of this Service name.

Although there are obviously many opportunities for ecosystems of user-defined Stacks, and indeed a rich library offered by the Mu platofrm itself, we eventually bottom out on a core set of "primitive" constructs.

The primitive types are in the mu namespace and include:

• mu/container: A Docker container wrapped in Mu metadata.
• mu/gateway: An API gateway and/or load balancer, multiplexing requests onto multiple target Services.
• mu/lambda: A single lambda function ordinarily used for serverless/stateless scenarios.
• mu/event: An Event that may be used to Trigger the execution of another Service (commonly a Function).
• mu/volume: A volume stores data that can be mounted by another Service.
• mu/autoscaler: A Service that automatically multi-instances and scales some other target Service based on policy.

TODO(joe): link to exhaustive details on each of these. TODO(joe): consider a mu/job (e.g., ECS's RunTask); unclear on how this would differ from mu/lambda. TODO(joe): consider a mu/daemon type, similar to Kube's DaemonSet abstraction.

Although these types may look "magical", each one simply leverages an open extensibility capability in the platform. Most interesting tasks may be achieved by composing existing Stacks, however, particular cloud providers may choose to offer new primitive types for even richer functionality. This is done by defining a so-called intrinsic type.

Finally, note that a companion namespace, mu/x also exists, that offers more cloud-neutral platform abstractions.

Dependencies

We just saw that Service types can refer to other Stacks. That is done with a so-called StackRef, which is simply a name that contains multiple parts:

• An optional protocol (e.g., https://).
• An optional base URL (e.g., hub.mu.com/, github.com/, etc).
• A required namespace and/or name part (e.g., acmecorp/worker, aws/s3/bucket, etc).
• An optional @ followed by version number (e.g., @^1.0.6, @6f99088, @latest, etc).

If protocol and base URL are absent, Mu will default to https://hub.mu.com/. If the version is omitted, Mu will default to latest, which just means "tip"; in other words, the most recent version is used at compile-time.

For Workspaces containing multiple Stacks, it can be advantageous to omit version information from your Stacks, and instead place them into your workspace.yaml file's dependencies section. For example:

dependencies:
    aws/s3/bucket: ^1.0.6

This helps to manage version numbers centrally which can be especially convenient when upgrading. Any StackRefs missing version information will consult this workspace at compile-time. You may even pin an entire namespace this way:

dependencies:
    aws/...: ^1.0.6

Note that the compiled Mu.yaml will always contain pinned versions, so that it stands on its own.

Please refer to Mu Dependencies for more details on dependencies and how they are resolved.

Visibility

At this point, a new concept is introduced: visibility. Visibility works much like your favorite programming language, in that a Stack may declare that any of its Services are public or private. This impacts the accessibility of those Services to consumers of this Stack. A private Service is merely an implementation detail of the Stack, whereas a public one is actually part of its outward facing interface. This facilitates encapsulation.

For instance, perhaps we are leveraging an S3 bucket to store some data in one of our Services. That obviously shouldn't be of interest to consumers of our Stack. So, we split things accordingly:

services:
    private:
        aws/s3:
            bucket: images
    public:
        nginx/nginx:
            data: s3
            port: 80

In this example, S3 buckets are volumes; we create a private one and mount it in our public Nginx container.

Service Properties

We have already seen plenty of property setters. But supplying capabilities -- or references to other Services -- warrants a special mention.

First, imagine we are creating a Stack that asks for a service; its properties might look like:

name: acmecorp/factory
properties:
    worker:
        type: service

Now let's create a Stack that instantiates acmecorp/factory, providing a reference to its own worker Service:

name: my/factory
services:
    private:
        myworker:
            type: mu/container
            build: .
            port: 80
        acmecorp/factory:
            worker: myworker

Notice that we have set acmecorp/factory's worker factory to myworker. This is a reference to our very own myworker Service, instantiated in the section just prior. At runtime, this simply expands to a URL referring to our worker's public endpoint listening on port 80.

If multiple possible ports are available, an error will occur, and you will need to pick one explicitly. For example:

        myworker:
            type: mu/container
            build: .
            ports: [ 80, 8080 ]

In this example, we still want to bind to port 80, however the system has no idea which to choose. So we must say:

        acmecorp/factory:
            worker: myworker:80

A similar problem happens if we are passing another Service as the argument. For example, let's say that we are using a 3rd party worker Service, rather than our own. For instance:

        jazzcorp/jazzworker:
        acmecorp/factory:
            worker: jazzworker

This will of course work just fine, provided jazzworker has a public Service enpoint. For example:

name: jazzcorp/jazzworker
services:
    public:
        api:
            port: 80
            ..

Much like how the system picked port 80 when there was only one in the raw container example, the system knows how to pick the right public endpoint when there is just one (in this case, api on port 80). However, just as the system needed a hint when there were multiple possible ports, it may need one here too, such as in this example:

name: jazzcorp/jazzworker
services:
    public:
        api:
            port: 80
            ..
        admin:
            port: 8080
            ..

Now jazzworker has two public endpoints: api on port 80 and admin on port 8080. We must select one:

        jazzcorp/jazzworker:
        acmecorp/factory:
            worker: jazzworker:api

Because this is a weakly typed example, using the base service type, plus a container, the amount of typechecking performed is quite minimal. In fact, just about the only thing it does is build a DAG of dependencies, ensure they are cycle-free, and select the port. Strongly typed examples, like the ex/kvstore one mentioned earlier, work similarly, except that there is extra compile-time validation and more rugged selection of endpoints.

Nested Stacks

Another feature that comes in handy sometimes is the ability to create nested Stacks:

stacks:

Each nested Stack is very much like the Stack defined by any given Mufile, except that it is scoped, much like a nested/inner class in object-oriented languages. Doing this lets you subclass and/or multi-instance a single Stack as multiple Services inside of the same Mufile. For example, consider a container that will be multi-instanced:

stacks:
    private:
        common:
            type: mu/container
            image: acmecorp/great
            env:
                NAME: {{.meta.name}}-cluster
                DATA: false
                MASTER: false
                HTTP: false

Now that we've defined common, we can go ahead and create it, without needing to expose the Stack to clients:

services:
    private:
        data:
            type: common
            env:
                DATA: true
    public:
        master:
            type: common
            env:
                MASTER: true
        worker:
            type: common
            env:
                HTTP: true

All of these three Services -- one private and two public -- leverage the same acmecorp/great container image, and each one defines the same four set of environment variables. Each instance, however, overrides a different environment variable default value, to differentiate the roles as per the container's semantics.

Different scenarios call for subclassing versus composition, and the Mu system supports both in a first class way.

TODO(joe): we need to decide whether you can export public Stacks for public consumption. At this point, my stance is that you must create an entirely different Stack to do that. This keeps things simple for the time being.

Target-Specific Metadata

Although for the most part, metadata strives to be cloud provider-agnostic, there are two ways in which it may not be. First, some Stack types are available only on a particular cloud, like aws/s3/bucket (and any that transitively reference this). Attempting to cross-deploy Stacks referencing such things will fail at compile-time, for obvious reasons. Second, some metadata can be cloud provider-specific. For example, even if we are creating a Service that is logically independent from any given cloud, like a Load Balancer, we may wish to provider cloud-specific settings. Those appear in a special metadata section and are marked in such a way that erroneous deployments fail at compile-time.

More details on target-specific Stacks and metadata settings are provided below in the relevant sections.