How to express targeting and configuration?

[jamesdobson]

Neurons need some access to configuration to tell them about their environment so they can perform their checks/mutations.
Also, a synapse may need a way to be targeted at something or be "about" something.

Use case: We have a TeamCity server that is shared across our org. Sometimes users come to us with a question: "why isn't my build running?" They have triggered it, but it doesn't run.

Let's say we are creating a "build_does_not_run" synapse. To be able to diagnose the issue, we need a few pieces of information:

• Target of the synapse: the build that won't run. Could be any of:
  • URL to a build instance that is in the queue
  • URL to a build configuration that we can't get to run
  • ID of a build configuration that we can't get to run
• Configuration information:
  • TeamCity server base URL
  • TeamCity server user and password or API token

With just this information, we could do several checks:

• Check for undefined build parameters on the target build
• Check for compatible agents defined that can run the target build
• Check for free compatible agents
• Check if using agent pools and compatible agents can be started and register successfully

How to collect the input data? I'm not sure what you had planned, perhaps something like:

• Neurons can declare their inputs, e.g. "teamcity.token", "teamcity.url". If two neurons declare an input with the same name, they share the information.
  • It would be nice if the neurons could document their inputs in natural language and provide other metadata about the inputs
  • Might have to introduce a "config" object on which the neurons could depend, e.g. the neurons in this use case would reference as their input a "teamcity" config object that defines "token" and "url" parameters. Then we could just have one definition of the "teamcity" config object.
• The user running cortex gets prompted for the set of inputs.
• Neurons and synapses can declare a "target" that is the thing they're checking (the build in this use case)
• Synapse can pass its target down to the neurons

[anoop2811]

I was envisioning that neurons should be completely independent of each other so as not to cause any pollution of data, but that comes at the cost of repeating things multiple times. The config is one route I would prefer and a key value pair declaration per yaml could be used. For advanced use cases maybe have a top level config that could be anchored in neurons that sort of give reusability, but maybe for the first iteration, we should try with complete isolation and then with resuable config. The config could also act as validation points for info which could come from upstream neurons are is manually provided. WDYT?