This repository is part of the Joyent Triton project. See the contribution guidelines and general documentation at the main Triton project page.
When a Triton server starts up, it will require a way of setting up the server, boot-strapping various sub-systems such as zpool creation, agents, etc, based on the configuration and role of the machine. The node agent Ur will listen on AMQP for scripts to execute.
It's important to note, however, that we will not be able to update the operating system live image, so the Ur agent must be reasonably future proof. To that end, it should be as simple as possible to then load additional agents with more logic at a later time.
The Ur agent connects to an AMQP broker, emits a message to signal it is listening and waits. The headnode will then send scripts to be executed on the Node (identified by said ID), with Ur replying back with the process return code and the script's captured STDOUT and STDERR.
- start up
- emit "start up" message
- receive messages with script to execute
- execute script
- reply with script exit code and captured stdout and stderr
- repeat
When a compute node starts up, Ur will be one of the first processes running. Once it has started, the Ur agent will post a message to AMQP. This message is simply to indicate, "Hey! I have started up and am ready to be told what to do".
At this point, the headnode will take note that this new node exists and will know a little bit about its configuration.
From this point on the headnode can use Ur for setting up the server, reading configuration etc.
- Ur starts up
- looks up broker details
- connects to broker
- emits start up message containing sysinfo payload
- headnode records node information
In these examples, should be the UUID of the machine as obtained from the "/usr/bin/sysinfo" script.
On start up, Ur will execute the sysinfo utility and publish a message to AMQP with this information.
==> routing-key: ur.startup.<node-id>
{
"Live Image": "20140905T202142Z",
"System Type": "SunOS",
...
}
Routing keys for Ur requests should contain a unique token, {{request-id}}, so that when we reply back we can re-use this token to indentify which command this was a reply to.
<== routing-key: ur.execute.<node-id>.<request-id>
{ type: "file"
, file: "scripts/pools.sh"
, timestamp: "2010-11-10T..."
}
<== routing-key: ur.execute.<node-id>.<request-id>
{ type: "script"
, script: "#!/bin/bash\necho hello world\nexit 0"
, timestamp: "2010-11-10T..."
}
On reply to an execute command:
==> routing-key: ur.execute-reply.<node-id>.<the-request-id>
{ exit_status: 0
, stdout: "hello world\n";
, stderr: ""
}
git clone git@github.com:joyent/sdc-ur-agent.git
cd sdc-ur-agent
git submodule update --init