Revision History

Summary

The Persistence Plugin Architecture is intended to address decoupling persistence implementations from the EdgeX Foundry code base. The core-data, core-metadata, export-client, support-logging and support-notifications services all currently rely on database persistence. Today (Delhi release), persistence implementations are part and parcel of the EdgeX Golang Services repo. As a result, adding a new persistence implementation or updating an existing implementation requires all team members to have knowledge of every supported persistence platform as well as access to an installation of each in order to fully test any changes. New capabilities must be added to every persistence implementation before they can be called complete. This is not scalable. For a future (Edinburgh) release, the goal is to move persistence implementations into their own repos so they can be maintained independently of the EdgeX code base. 

...

Within each service, a configuration value will drive which persistence plugin to load, similar to what we do now. This value can be the same across all services or even different for every service should a customer feel that one or another platform is best suited to the role the service plays in their environment. Each service defines an interface for the necessary database operations (core-data, core-metadata) and then a vendor's plugin must conform to and support that interface.

User Stories

1. As a persistence implementation provider (a provider), I want a mechanism so that I can add a new persistence implementation to the EdgeX platform.
2. As a provider, I want a mechanism for sustaining a persistence implementation so that I do not rely on EdgeX platform releases.
3. As a provider, I want a mechanism for configuring my implementation without knowing the rest of the EdgeX platform configuration.
4. As a provider, I want the EdgeX platform build scripts to consolidate my build-time and run-time requirements so that I can maintain my implementation independently of the EdgeX platform.
5. As a provider, I want an API contract so that I know what to implement.
6. As a provider, I want an API contract so that I can unit test.
7. As a provider, I want EdgeX platform testing to be able to certify implementation so that EdgeX platform users will have the confidence to use my implementation.
8. As an EdgeX platform maintainer (a maintainer), I want a default persistence implementation so that I can maintain CI/CT.
9. As a maintainer, I want providers to create reasonable defaults so that I can add them into CI/CT.
10. As an EdgeX platform user (a user), I want to select one or more persistence implementations for use by the various EdgeX services.
11. As a user, I want the option to run the platform along with the selected persistence implementation(s) directly on the OS.
12. As a user, I want the option to run the platform along with the selected persistence implementation(s) in a containerized deployment.

Design

The design has two major roles: the provider role and the platform role.

Provider

...

The provider maintains their plugin in their own repository along with the build scripts to create the plugin, the EdgeX build fragments needed by the platform build, and the EdgeX Compose file fragments needed to create the EdgeX Compose file.

Repo Organization

The provider repo follows the Go standard project layout. In addition, providers can are encouraged to maintain and distribute prebuilt versions binaries of their plugin. The `.so` lives in `pkg

...

Platform Responsibility

...

Note NAME and VERSION must match across plugin, compose file fragment, and configuration file.

Platform

The platform role is further divided into config time and run time. During config time, the configuration of the platform drives the consolidation of the Docker Compose file, possibly downloading the plugin, and possibly downloading the persistence configuration file. It is then assumed at run time, the compose file has mounted the volume where the persistence configuration file lives and the plugin is in a path known by the startup scripts.

Config Time

At config time, the plugin, the Docker Compose file fragment, and the persistence configuration file may need to be downloaded from the repo. This is obtained from the service `.toml` file or by querying Consul.

[Databases]
  [Databases.Primary]
  Host = 'localhost'
  Name = 'coredata'
  Username = 'scott'
  Password = 'tiger'
  Port = 1234
  Timeout = 5000
  Type = 'NAME.so'
  Version = '1.0.0'
  PluginSource = 'https://github.com/some-provider/edgex-plugin/blob/master/pkg'

In addition to overloading the Type key, two additional keys are added. The plugin, compose file fragment, and configuration file are copied to `cmd/SERVICE/plugins`.

The compose file fragment is merged into the Docker Compose file for the service. If there is a configuration file it is mounted as part of the compose file.

...

  some-provider:
    image: some-provider:alpine
    ports:
      - "1234:1234"
    container_name: edgex-some-provider
    hostname: edgex-some-provider
    networks:
      - edgex-network
    volumes:
      - data:/data
    depends_on:
      - volume

Acceptance Criteria

...

data
    depends_on:
      - volume

Run Time

The assembled Docker Compose file will start the persistence service, if needed. When started, the EdgeX service attempts to use the Databases.Primary.Type key to reference a known builtin database. When the value is not recognized it will now attempt to load a plugin.

Acceptance Criteria

MVP

• Provider implementations in their own repo
• EdgeX platform config time scripts to create Docker Compose file
• Updated black box tests

v.Next

• Provider certification
• Provider included in platform CI/CT

Reference Patterns

The basis of the requirements came out of looking at how others have solved this challenge. Also, note Vladimir Vivien's helpful article on "Writing Modular Go Programs with Plugins".

Azure IoT Edge

Azure IoT Edge is organized around independently configured modules (aka containers) that are grouped to form a solution. Each module is described independently and then combined into a solution with a deployment manifest. For example, the RedisEdge module is described using the following module.json. 

...