Best practices for nodes configuration syntonization

Applications & Usage
1

In my system, different nodes share some identical configuration parameters that rarely change (approximately once per month) but could change while the system is running. It is not critical to restart a node during its reconfiguration. However, it is crucial that all nodes remain synchronized when the system is in operation mode.

I am exploring the best options to implement this behavior. Here are my options, along with some pros and cons from my perspective:

  • uavcan/register

    • Option 1: Populate identical configuration parameters in each node’s registry.
      • Pros:
        • Utilizes standard functionality.
      • Cons:
        • Requires special synchronization logic (and maybe tool), especially when adding new nodes to an already configured system in operation mode.
    • Option 2: Store identical configuration parameters in a central node registry from which other nodes read.
      • Pros:
        • Utilizes standard functionality.
        • Simplifies synchronization logic, but still requires some notifications to other nodes about configuration updates when the system is in operation mode.
      • Cons:
        • Overuse standard functionality.

  • Provide a dedicated service for reading identical configuration parameters

    • Pros:
      • Similar to the Option 2 of uavcan/register but does not overuse its functionality.

  • Periodically publish identical configuration parameters from a central node

    • Pros:
      • Straightforward and organic implementation.
    • Cons:
      • Almost all of the utilized bandwidth is senseless because it sends similar data.

  • Publish identical configuration parameters from a central node only when a new node appears or a configuration parameter changes

    • Pros:
      • Uses only the necessary amount of bandwidth.
      • Synchronization logic is straightforward.
    • Cons:
      • Still requires synchronization logic.

I may be missing some other options.

Currently, I am leaning towards the last option with implementation in a separate node. However, this approach (and actually all others) seems to deviate from the distributed system paradigm.

I am a bit confused by this task and would be very grateful for any valuable feedback or suggestions on this topic.