hubot-chatops-rpc

Easily add RPC endpoints to your hubot.

Chatops RPC is an extraction of years of experience with Chatops at GitHub. It's a simple protocol based on JSON-RPC over HTTPS. Chatops RPC pushes all of the code save parsing to a server for processing. This means that chat commands can be written and tested entirely server-side, without any glue code in hubot/node.

This module is the client in the Chatops RPC protocol. Hubot stores a list of servers and the endpoints they offer in his brain.

For more information, see the protocol description

Usage

To add a chatops RPC server, use:

.rpc add https://example.com/_chatops --prefix example

This will add https://example.com/_chatops to the list of endpoints. Hubot will poll this endpoint every 10 seconds for commands. All commands from this endpoint will require a prefix of example.

To remove a server:

.rpc remove https://example.com/_chatops

To see a server's response:

.rpc debug https://example.com/_chatops

Commands from this endpoint will be prefixed with example. Assume the endpoint exposes an endpoint with the regex /ping/. The command will be accessible with:

.example ping

Hubot will parse long-form arguments, such as --argument1 foo bar, and send them to the server as additional arguments. These can be added to any command, so .example ping --reason just because we feel like it will call the ping method with an extra argument of { reason: "just because we feel like it"}.

Protocol Description

CRPC is a client-server protocol; Hubot is a client. Servers expose an endpoint listing available methods. Each endpoint provides a regex to fire on and a relative URL path to execute it.

Chatops RPC pushes a lot of complexity to clients. This is a design decision, intended to keep the burden of creating new chat commands in existing systems as low as possible.

Listing Commands

A CRPC service listing is an endpoint that exposes JSON including the following fields:

• namespace: A globally unique namespace for these commands. Clients can use this to uniquely identify this endpoint. A namespace should be a slug of the form /[a-Z0-9\-_]+/.
• help: Optional: Overall help for this namespace, if a client chooses to provide help
• error_response: Optional: A message to present when this endpoint returns an error. This can direct users to next steps when the server fails.
• methods: A mapping of named operations to their metadata.
• version: The version of ChatOps RPC protocol to use, currently version 3

Each key in the methods hash will be a string name. Each name should be a slug of the form /[a-Z0-9]\-_]+/. Clients can use these method names to uniquely identify methods within a namespace. Each name shall point to an object with the following fields:

• regex: A string regular expression source used to execute the command. This regular expression should use named capture groups of the form (?<parameter_name>.+).
• path: A path, relative to the listing URL, to execute the command.
• params: A list of available named parameters for this command.
• help: Optional: User help for a given command.

Each server is assumed to be given a prefix, which the client will handle prepending to a command's regex source. Clients can use the namespace as a default prefix if they wish, but servers may not demand a particular prefix. Chatops RPC clients should require whitespace after the prefix, so a command with a regex like /ping/ with a prefix of test would match on test ping.

Executing Commands

CRPC clients use the listings to create a listing of available commands. When a chat message matches a command's regex matcher, the CRPC client creates a method invocation. A method invocation is a JSON object with the following fields:

• user: A slug username corresponding to to the command giver's GitHub login.
• room_id: A slug room name where the command originated.
• method: The method name, without namespace, of the matching regex.
• params: A mapping of parameter names to matches extracted from named capture groups in the command's regex. Parameters that are empty or null should not be passed.

The JSON object is posted to the path associated with the command from the listing of commands. CRPC servers should assume that parameters in the params hash are under user control, but trust that the user and room_id to be correct.

CRPC servers must produce a response JSON object with the following fields:

• result: A string to be displayed in the originating chat room.

CRPC may optionally include the following fields in a response JSON object for use in situations where richer results can be displayed. Clients will optionally utilize some or all of the extra information to provide an enhanced response, but it is important that result be sufficient on its own.

• title: The title text for the response
• title_link: Optional URL to link the title text to
• color: Hex color for the message, to indicate status/group e.g. "ddeeaa'
• buttons: An array of button objects
  • label: The text to display on the button
  • image_url: An image URL to display as the button, will generally take precedence
  • command: The command to use when the button is clicked
• image_url: An image URL to be included with the response

CRPC may also produce error JSON according to the JSON-RPC spec, consisting of an object containing an error object with a message string. This is sometimes helpful for clients that make a distinction between failed and successful commands, such as a terminal. CRPC point of view. CRPC clients should still parse these error messages.

Examples

Here is an end-to-end transaction, sans authentication (see below):

CRPC client issues:

GET /_chatops HTTP/1.1
Accept: application/json

{
 "namespace": "deploy",
 "help": null,
 "version": 3,
 "error_response": "The server had an unexpected error. More information is perhaps available in the [error tracker](https://example.com)"
 "methods": {
   "options": {
     "help": "hubot deploy options <app> - List available environments for <app>",
     "regex": "options(?: (?<app>\\S+))?",
     "params": [
       "app"
     ],
     "path": "wcid"
   }
 }
}

The client will use the suggested namespace as a prefix, deploy. Thus, when the client receives a command matching .deploy options hubot, the CRPC client issues:

POST /_chatops/wcid HTTP/1.1
Accept: application/json
Content-type: application/json
Content-length: 77

{"user":"bhuga","method":"wcid","params":{"app": "hubot"},"room_id":"developer-experience"}

The CRPC server should respond with output like the following:

{"result":"Hubot is unlocked in production, you're free to deploy.\nHubot is unlocked in staging, you're free to deploy.\n"}

The CRPC client should output "Hubot is unlocked in production, you're free to deploy.\nHubot is unlocked in staging, you're free to deploy.\n" to the chat room. The client can optionally display the output intelligently if it contains newlines, links in formats like markdown, etc. It's strongly recommended that a client support markdown links if possible.

Authentication

Authenticating clients

Clients authenticate themselves to servers by signing requests with RS256 using a private key. Servers have a public key associated with clients and verify the signature with it.

By convention, a CRPC server should allow authentication with two secrets simultaneously to allow seamless token rolling.

Clients send three additional HTTP headers for authentication: Chatops-Nonce, Chatops-timestamp, and Chatops-Signature.

• Chatops-Nonce: A random, base64-encoded string unique to every chatops request. Servers can cache seen nonces and refuse to execute them a second time.
• Chatops-Timestamp: An ISO 8601 time signature in UTC, such as 2017-05-11T19:15:23Z.
• Chatops-Signature: The signature for this request.

The value to be signed is formed by concatenating the value of the full http path, followed by a newline character, followed by the contents of the nonce header, followed by a newline character, followed by the value of the timestamp header, followed by a newline character, followed by the entire HTTP post body, if any. For example, for a GET request with these headers:

Chatops-Nonce: abc123
Chatops-Timestamp: 2017-05-11T19:15:23Z

Sent to the following URL:

https://example.com/_chatops

The string to be signed is: https://example.com/_chatops\nabc123\n2017-05-11T19:15:23Z\n

For a request with the same headers and a POST body of {"method": "foo"}, the string to be signed is:

https://example.com/_chatops\nabc123.2017-05-11T19:15:23Z\n{"method": "foo"}

The signature header starts with the word Signature, followed by whitespace, followed by comma-separated key-value pairs separated by an =. Each value is closed with double quotes. Keys must be all lowercase.

• keyid: An implementation-specific key identifier that servers can use to determine which private key signed this request.
• signature: The base64-encoded RSA-SHA256 signature of the signing string.

An example signature header would be:

Chatops-Signature: Signature keyid="rsakey1",signature="<base64-encoded-signature>"

Authentication

CRPC must trust that a user is authenticated by the user parameter sent with every command. Individual servers may request a second authentication factor after receiving a command; this is beyond the scope of CRPC.

Authorization

CRPC servers are responsible for ensuring that the given user has the proper authorization to perform an operation.

Execution

Chatops RPC clients are expected to add a few niceties not covered by the wire protocol. This complexity is exported to clients to keep the burden of implementing new automation low.

• Regex anchoring. Clients should anchor regexes received from servers. If a command is exported as where can i deploy, it should not be triggered on tell me where i can deploy or where can i deploy, i'm bored.
• Prefixing. Different execution contexts may prefix commands, such as ., hubot, or another sigil.
• Help display systems. These are heavily context dependent. Servers provide text snippets about commands, but accessing and displaying them is up to the client.

These niceties are optional and context-dependent. Different clients may or may not implement them. But if any of these are required in any execution context, they should not be pushed to the server.

Protocol Changes

The version of the ChatopsRPC protocol in use by a server is given as the version field. If no version is returned, 3 is assumed.

Server Implementations

• chatops-controller makes it easy to add CRPC endpoints to rails applications.

Installation

In hubot project repo, run:

npm install hubot-chatops-rpc --save

Then add hubot-chatops-rpc to your external-scripts.json:

[
  "hubot-chatops-rpc"
]

Create a public/private key pair for authentication:

ssh-keygen -t rsa -b 4096 -f crpc

This will create two files, crpc and crpc.pub. Use the contents of the crpc file as an environment variable, CHATOPS_PRIVATE_KEY. crpc.pub contains a public key for use by servers.

NPM Module

https://www.npmjs.com/package/hubot-chatops-rpc