Skip to content

Latest commit

 

History

History
446 lines (278 loc) · 18.3 KB

README.md

File metadata and controls

446 lines (278 loc) · 18.3 KB
Raw

Lib-Courier-JS

This library can be used to create an event bus or communication mediator to help maintain loose coupling between application components. It is inspired by backbone.radio but adds some additional patterns as well as adhering to the Node.js eventemitter conventions. It can be used for purely intra-service component communication as well as to wrap inter-service interfaces as desired. The main goal is to remove direct calls between components so developers can focus on their APIs rather than where they are organized.

Three messaging patterns are provided:

  • Request/Reply: Named requests are delivered to the registered replier from which the result is delivered back to the requester
  • Pub/Sub: Named events are delivered to registered subscribers
  • Push/Pull: (Not yet implemented) Named events are delivered to a registered puller

Additionally, a barebones framework is provided to organize handlers into "modules" which can be composed into a single running service instance.

NPM: https://www.npmjs.com/package/lib-courier-js

Contents

Couriers, Networks & Namespaces

While a single courier can be used to mediate messages in patterns, they are even more powerful when organized into a network. Unless explicitly specified otherwise, couriers are placed into a global, default network. Couriers can be linked together via a namespace prefix to allow messages to propagate depending on the pattern being used. A courier can be namespaced multiple times by multiple nodes.

Patterns

Request/Reply

In the request/reply pattern, requests are made to a specific name and are delivered to the first discovered replier. In a single courier, only one replier can be registered to a name at a time. The search for a replier spreads from the requester's courier in the following manner:

  • Local name replier
  • Local pattern replier
  • List all linked namespaces whose prefix matches the request name, slice off the prefix (unless retainPrefix is enabled), add to the search queue
  • List all linking couriers, add to the search queue
  • Get next item from the queue, repeat until handler is found or queue is empty

With each iteration, the search casts a wider net until a replier is found or produces an error. Since it looks at the requester's courier first, names in the local vicinity can use their local name, while inter-courier requests must be prefixed with the desired namespace relative. To avoid cycles, node are only visited once per name per search. Search results are cached until the network changes.

Repliers can be registered for a name or a pattern:

courier.reply('test-1', (ctx, data) => ...);
courier.replyPattern('^test.*', (ctx, name, data) => ...);

Repliers can use promises or callbacks and all are automatically promisified using bluebird. See the API docs below or the tests for details.

Pub/Sub

In the pub/sub patter, events are emitted with a specific name and are delivered to all registered listeners. The search for listeners spreads from the emitter's courier in the following manner:

  • Local name listeners
  • Local pattern listeners
  • List all linked namespaces, add to the search queue
  • List all linking couriers, prepend the current node's prefix (unless retainPrefix is enabled), add to the search queue
  • Get next item from the queue, repeat until the queue is empty

Local listeners can use the event's local name while linked couriers must prepend the expected emitter's namespace. To avoid discovering a subscriber multiple times, once the search traverses into a linked namespace it will not propagate back to the linker. Search results are cached until the network changes.

Listeners can be registered for a name or a pattern:

courier.on('test-1', (ctx, data) => ...);
courier.once('test-1', (ctx, data) => ...); // only called once
courier.onPattern('^test.*', (ctx, name, data) => ...);

Return values are ignored and errors thrown synchronously are swallowed (may be an option in the future).

Push/Pull

Not yet implemented

Services, Modules & Composition

One of the goals in creating this project was to encapsulate namespace functionality into modules/groups and be able to declaratively compose them into a service. Further, it was desired to be able to have different configurations depending on when and where the modules were being run.

For instance, when running locally during development it is much easier to manage a few composed instances vs dozens of instances split solely on functionality. But in testing or production, an entirely different arrangement might be appropriate. Some groups that pose no current performance limitations could be combined while others that require dedicated instances or need to be horizontally scaled could be run separately with no module changes required.

In an attempt to accomplish this goal, lib-courier-js provides a minimal framework for composing "modules" and managing their lifecycle. This framework accepts a root courier instance and a configuration of where to find modules, what namespaces to put them in and what configuration to provide to them. When combined with a configuration management solution like node-config, it enables a nice balance of development and operational flexibility.

A very basic service example can be found at /examples/sample-service and a few reusable modules are located in /lib/service/modules. More examples will be provided in the future, but it has been used privately to compose dozens of interdependent micro-services consisting of multiple modules each. They communicate directly via courier when running in the same instance, or via networked messaging modules when run separately, all with no changes to the modules themselves.

Provided Modules

DirectoryLoader

lib-courier-js/lib/service/modules/directory-loader

DirectoryLoader is a utility module that facilitates the organization of module code into directories for repliers & listeners. It is not intended to be used on its own, but rather extended or composed into application specific modules. It works by implementing registerHandlers to load repliers & listeners relative the module package. As such, it is necessary to specify the package's location, which is done via the config.root setting.

Handlers are invoked with an extra argument in the first position which is the a reference to the module itself. This is for convenience and to avoid using this and dealing with calling context issues.

Handlers in the repliers/listeners directories can be structured in several patterns:

Function Export

The handler name will be the file name without the extension. To make a pattern handler, export pattern. A name cannot be exported in this case because it would conflict with the function's name, which is a JS thing. Exporting enabled = false will cause the handler to be skipped.

module.exports = function(mod, ctx, data) { ... }

module.exports = function(mod, ctx, name, data) { ... }
module.exports.pattern = '^test-.*$';
Handler Export

A name can optionally be exported, otherwise the handler name will be the file name without the extension. To make a pattern handler, export pattern. Exporting enabled = false will cause the handler to be skipped. Handlers are invoked with

module.exports.name = 'test';
module.exports.handler = function(mod, ctx, data) { ... }

module.exports.pattern = '^test-.*$';
module.exports.handler = function(mod, ctx, name, data) { ... }
Handlers Export

Exporting handlers allows the packaging of multiple handlers into a single file via KVP. Each value can be a function, in which case the name is the key, or an object where enabled, name, pattern and/or handler can be specified (omitting name/pattern uses the key as name). Exporting enabled = false disables all handlers. handlers can be further nested if desired.

module.exports.handlers = {
  test: function(mod, ctx, data) { ... },
  test_pattern: {
    pattern: '^test-.*$',
    handler: function(mod, ctx, name, data) { ... }
  }
}

Config:

  • root: Path to the directory containing replier & listener directories. Typically this is the module's package directory.
  • repliers: Name of the repliers directory. Default repliers
  • listeners: Name of the listeners directory. Default listeners

There are several hooks that can be overridden if desired.

#loadHandler()

Calls loadRepliers & loadListeners.

#loadRepliers()

Loads the code from the repliers directory and registers them according to the patterns above.

#loadNamedReplier(name, handler)

By default uses this.courier.reply to register the handler to the name, but this is often useful to wrap for logging, validation, etc purposes.

#loadPatternReplier(name, handler)

By default uses this.courier.replyPattern to register the handler to the pattern, but this is often useful to wrap for logging, validation, etc purposes.

#loadListeners()

Loads the code from the listeners directory and registers them according to the patterns above.

#loadNamedListener(name, handler)

By default uses this.courier.on to register the handler to the name, but this is often useful to wrap for logging, validation, etc purposes.

#loadPatternListener(name, handler)

By default uses this.courier.onPattern to register the handler to the pattern, but this is often useful to wrap for logging, validation, etc purposes.

ServiceInfo

lib-courier-js/lib/service/modules/service-info

EventLogger

lib-courier-js/lib/service/modules/event-logger

CannedReplier

lib-courier-js/lib/service/modules/canned-replier

API

Courier

new Courier([options])

Options:

  • id: Courier id or uuid if omitted
  • name: Courier name
  • network: Network instance or global if omitted

#registerNamespace(prefix[, courier[, opts]]) -> Courier

Links two couriers on the network with the given prefix. Automatically creates a courier if omitted. Returns the linked courier.

Options:

  • override: Whether to allow overriding a previously registered prefix
  • courierOpts: Options to pass to the auto-created courier if courier is not provided
  • retainPrefix: Whether to retain the prefix when searching for handlers

#request(ctx, name, data[, cb]) -> Promise

Alias: req

Searches the network for a replier. If cb is provided, the signature is cb(err, result). Invokes the handler with the provided context and data. If the handlers is a pattern replier, request name is also provided.

#reply(name, handler[, opts]) -> this

Registers a replier for a request name. Handler signature is handler(ctx, data[, cb]). If handler has a cb arg (arity is 3), it will expect to be called as cb(err, result). Handlers are automatically promisified and can either return a promise or return/throw synchronously.

Options:

  • override: Whether to allow overriding a previously registered name. Default false.

#replyPattern(pattern, handler[, opts]) -> this

Registers a replier for a request pattern. Handler signature is handler(ctx, name, data[, cb]). If handler has a cb arg (arity is 4), it will expect to be called as cb(err, result). Handlers are automatically promisified and can either return a promise or return/throw synchronously.

Options:

  • override: Whether to allow overriding a previously registered pattern. Default false.

#emit(ctx, name, data) -> this

Searches the network for listeners. Invokes each listener with the provided context and data. If the handler is a pattern listener, event name is also provided.

#on(name, handler[, opts]) -> this

Registers a listener for the event name. Handler signature is handler(ctx, data).

Options:

  • once: Whether to only invoke the handler once

#once(name, hander[, opts]) -> this

Shortcut for context.on(name, handler, {once: true}).

#onPattern(pattern, handler) -> this

Registers a listener for the event pattern. Handler signature is handler(ctx, name, data).

#createContext(...args) -> Context

Shortcut for creating a new Context.

#toJSON() -> Object

Returns a simplified version of the courier containing:

  • id
  • name
  • namespaces: Map of prefix to courier id
  • reply: Map of request names to repliers
    • calls: Number of invocations
    • lastCalled: Timestamp of last invocation
  • replyPattern: Map of request patterns to repliers
    • calls: Number of invocations
    • lastCalled: Timestamp of last invocation
  • event: Map of event names to listeners
    • calls: Number of invocations
    • lastCalled: Timestamp of last invocation
    • once: Whether it was registered as a once listener
  • eventPattern: Map of event patterns to listeners
    • calls: Number of invocations
    • lastCalled: Timestamp of last invocation
  • requests: Map of requested names to stats
    • calls: Number of invocations
    • lastCalled: Timestamp of last invocation
  • events: Map of emitted names to stats
    • calls: Number of emissions
    • lastCalled: Timestamp of last emissions

Context

Sending messages via courier requires a context be provided. This intended to be the "context" for the request/event/message and while the courier and network have no opinions about its structure, it is typically a place to keep some metadata, logger & expiry. The Context class in lib-courier is likely sufficient for most purposes, but ultimately it is up the event source to provide a context.

new Context([data[, opts]])

The data argument is used to populate the data property as a DataManager.

Options:

  • id: Context id or uuid if omitted
  • logger: Logger object
  • scope_logger: Whether to automatically scope the logger on instantiation via #scopeLogger
  • logger_id_field: Field name to put id if scope_logger: true. Default ctx_id.
  • exp: Expiry timestamp
  • ttl: Duration (ms) until expiry. Used to set exp as now + ttl.

#child(padding=0) -> Context

Creates a new context with the same data & opts as the parent. If padding is provided, subtracts from the exp date. Child's id is automatically scoped as {{parent.id}}.{{child_idx}}.

#scopeLogger() -> this

Used internally if scope_logger is true. Default implementation assumes the logger has a child method that accepts a kvp argument to which it provided the context id as the logger_id_field key.

#isExpired() -> Boolean

If exp is set, returns whether it has been reached.

#get(...) -> Object

Shortcut for ctx.data.get(...).

#set(...) -> this

Shortcut for ctx.data.set(...);

#unset(...) -> this

Shortcut for ctx.data.unset(...);

DataManager

A helper class wrapping a simple kvp datastore. Core functionality uses lodash get, set, has, unset, keys utilities.

new DataManager([data])

#get(path) -> Object

#has(path) -> Boolean

#reset() -> this

#set(path, value) -> this

#unset(path) -> this

#inc(path[, n=1]) -> this

#dec(path[, n=1]) -> this

#keys() -> [string]

#toJSON() -> Object

Service

Located in /lib/service/index.js.

initialize(opts) -> Promise(this)

Initializes the service by calling its own methods in the following order.

  • validateConfig: Validates the provided configuration against the schema
  • preInstantiation: Hook
  • instantiateModules: Loads & instantiates modules
  • postInstantiation: Hook
  • preInitialization: Hook
  • initializeModules: Calls optional initialize method on all modules
  • postInitialization: Hook
  • preRegistration: Hook
  • registerHandlers: Calls optional registerHandlers method on all modules
  • registerSources: Calls optional registerSources method on all modules
  • postRegistration: Hook

Options:

  • config: Service configuration. See /lib/service/schema.json for the JSON schema.
  • logger: Logger object. Will be provided to modules on instantiation. Is minimally expected to have debug & info methods.`
  • courier: Root courier, created if omitted
  • require: Require function to use when loading modules
  • loadableModules: Map of preloaded module factories. Looks here first if provided before attempting to load.

terminate() -> Promise(this)

Calls optional terminate method on all modules.

Network

new Network([opts])

Options:

  • id: Network id or uuid if omitted

#registerCourier(courier) -> this

Registers a if it has not already been registered.

#registerNamespace(prefix, target, source[, opts]) -> this

Registers a connection between the source and target couriers.

Options:

  • override
  • retainPrefix

#findRequestHandler(name, id) -> handler

Searches the network for a handler for the given name starting at the courier id. Returns an object containing:

  • courier: Courier to which the handler is registered
  • handler: Handler object
  • name: Resolved request name when handler was found

#findEventHandlers() -> handlers

Searches the network for handlers for the given name starting at the courier id. Returns an list of objects containing:

  • courier: Courier to which the handler is registered
  • handler: Handler object
  • name: Resolved event name when handler was found

#toJSON() -> Object

Returns a simplified version of the network containing:

  • id
  • couriers: Map of id to courier.toJSON()
  • connections: List of links (namespaces) between couriers on the network
    • from: Courier id
    • to: Courier id
    • prefix: Namespace prefix
    • retainPrefix

Development

docker build -t lib-courier-js .
docker run --rm -it -v `pwd`:/opt/pkg -v /opt/pkg/node_modules lib-courier-js bash
BLUEBIRD_DEBUG=1 yarn run test

npm version x.y.z -m '<release notes>'
npm publish