Skip to content

Architecture

DP³ is generic platform for data processing. It's currently used in systems for management of network devices in CESNET, but during development we focused on making DP³ as universal as possible.

This page describes the high-level architecture of DP³ and the individual components.

Data-points

The base unit of data that DP³ uses is called a data-point, which looks like this:

{
  "type": "ip", // (1)!
  "id": "192.168.0.1", // (2)!
  "attr": "open_ports", // (3)!
  "v": [22, 80, 443], // (4)!
  "t1": "2022-08-01T12:00:00", // (5)!
  "t2": "2022-08-01T12:10:00",
  "src": "open_ports_module" // (6)!
}
  1. A data-point's value belongs to a specific (user-defined) entity type, declared by the type.
  2. The exact entity is using its entity id in id.
  3. Each entity has multiple defined attributes, the attr field specifies the attribute of the data-point.
  4. Finally, the data-point's value is sent in the v field.
  5. Data-point validity interval is defined using the t1 and t2 field.
  6. To easily determine the data source of this data-point, you can optionally provide an identifier using the src field.

This example shows an example of an observations data-point (given it has a validity interval), to learn more about the different types of data-points, please see the API documentation.

Platform Architecture

DP3 architecture

DP³ architecture

The DP³ architecture as shown in the figure above consists of several components, where the DP³ provided components are shown in blue:

  • The HTTP API (built with Fast API) validates incoming data-points and sends them for processing to the task distribution queues. It also provides access to the database for web or scripts.
  • The task distribution is done using RabbitMQ queues, which distribute tasks between workers.
  • The main code of the platform runs in parallel worker processes. In the worker processes is a processing core, which performs all updates and communicates with core modules and application-specific secondary modules when appropriate.
  • Both the HTTP API and worker processes use the database API to access the entity database, currently implemented in MongoDB.
  • The generic web interface, or a set of generic components has been implemented as Grafana plugin. It's primary use-case is to help quickly and intuitively visualize data while retaining genericity. Ultimately, web interface is application-specific. However, we wanted to provide a quick start alternative. Implementation as Grafana plugin also implies option to combine DP³ with other data sources.

The application-specific components, shown in yellow-orange, are as following:

  • The entity configuration via yml files determines the entities and their attributes, together with the specifics of platform behavior on these entities. For details of entity configuration, please see the database entities configuration page.
  • The distinction between primary and secondary modules is such that primary modules send data-points into the system using the HTTP API, while secondary modules react to the data present in the system, e.g.: altering the data-flow in an application-specific manner, deriving additional data based on incoming data-points or performing data correlation on entity snapshots. For primary module implementation, the API documentation may be useful, also feel free to check out the dummy_sender script in /scripts/dummy_sender.py. Feel free to check out the secondary module API documentation, or you can also refer to the CallbackRegistrar code reference or check out the test modules in /modules/ or /tests/modules/.

Data flow

This section describes the data flow within the platform.

DP3 Data flow

DP³ Data flow

The above figure shows a zoomed in view of the worker-process from the architecture figure. Incoming Tasks, which carry data-points from the API, are passed to secondary module callbacks configured on new data point, or around entity creation. These modules may create additional data points or perform any other action. When all registered callbacks are processed, the resulting data is written to two collections: The data-point (DP) history collection, where the raw data-points are stored until archivation, and the profile history collection, where a document is stored for each entity id with the relevant history. You can find these collections in the database under the names {entity}#raw and {entity}#master.

DP³ periodically creates new profile snapshots, triggered by the Scheduler. Snapshots take the profile history, and compute the current value of the profile, reducing each attribute history to a single value. The snapshot creation frequency is configurable. Snapshots are created on a per-entity basis, but all linked entities are processed at the same time. This means that when snapshots are created, the registered snapshot callbacks can access any linked entities for their data correlation needs. After all the correlation callbacks are called, the snapshot is written to the profile snapshot collection, for which it can be accessed via the API. The collection is accessible under the name {entity}#snapshots.