Data Flow#
The data flow in Edge Infrastructure Manager can be broken down into declarative and discovery operations. This two fold nature of the APIs shows that the controlled resources are a mix of declarative and discovered abstractions. Not all the managed resources are modeled using this approach and there are resources represented using a pure declarative model that provides top-down policy.
The following image summarizes at high level the data flow in Edge Infrastructure Manager in terms of the main components (see High-Level Architecture) and the interactions between them. The following sections will provide a more detailed description of the data flows.
Reconciliation of User Intents#
Stage 1: Input: - Resource requests submitted via APIs or command-line tools by the user. User intents are conveyed through the resources using a desired state field.
Stage 2: Processing and storage: - Data are sent and preprocessed in Inventory. An event is generated and the listening clients are getting notified.
Stage 3: Reconciling upon the received stimulus: - Resource Managers are listening to the Inventory events and are reconciling the user intents with the actual state of the infrastructure.
Stage 4: Execution: - Resource Managers are sending commands to the Edge Node agents to perform the actual work on the edge infrastructure. On success the current state is updated in Inventory to match the desired state which stops future reconciliation of this resource.
On errors new attempts might be re-scheduled by the Resource Manager if the errors are recoverable. Otherwise, the system is left in an error state and the requests will not be further reconciled until the user does not intervene.
Top-down Requests#
Stage 1: Input: - Resource requests submitted via APIs or command-line tools by the user.
Stage 2: Processing and storage: - Data are sent and preprocessed in Inventory. An event is generated and the listening clients are getting notified.
Stage 3: External notification and consumption: - Other Resource Managers upon external stimulus might consume this data and act accordingly.
Bottom-up Discovery#
Stage 1: Input: - Resource Manager pushes data “Bottom Up” from the Edge into Inventory.
Stage 2: Processing and storage: - Data are sent and preprocessed in Inventory. An event is generated and the listening clients are getting notified.
Stage 3: Reconciling upon the received stimulus: - Resource Managers are listening to the Inventory events and might react accordingly
Stage 4: User notification*: - Users are notified of the changes and the new data that have been learned from the infrastructure;