Introduction

To build your own client, you must understand some of the internals of the AgentController8 and how you can send and receive data to/from it. This document can be used as a guide to build your own client to execute commands.

Pre Read

Please make sure you understand the command syntax.

Communication

All communication between the client and the agent-controller2 is done over redis. You send new commands to agent-controller by pushing some-commands to a queue. You wait for the answer by waiting on another queue. ect...

Sending commands to agent controller 2

all commands must be pushed to the cmds_queue queue. Just build the correct command structure as descriped in command syntax serialized it to json, and push it to cmds_queue. You can immediately see the command got received by the agentcontroller in the agentcontrller logs.

The agent-controller will make sure to deliver the task to the right agent(s).

Receiving response:

To wait for job results, the client must do a blocking pop from the cmds_queue_<cmd-id> queue. Once the agent controller receives a response from the agent it will immediately push it to this queue releasing the client.

Note that, since the client must do POP of the responses, another POP can get the client to wait forever. So you only do the second wait if you are expecting more results to be sent (fanout=true)

The agent controller will also store the command results(s) in a HSET named jobresult:<job-id> where each key in the dict is the gid:nid compination and the value is the job result. This is useful if you need to get the command results in a non-blocking manner. Also the values in the dict are stored forever and not poped, so you can read the results of the command any time in the future.

View command history:

All the received commands are also pushed to joblog queue. This one should only be used for tracking purposes only.