Heartbeat upgrade

[nsoblath]

Introduction

I propose a multi-part upgrade to the heartbeat feature with three main components:

1. Adding a special first heartbeat message
2. Adding a special last heartbeat message
3. Adding to the payload

The main purpose of this proposal is to make it easier to know what services are present on the mesh. This could enable interactions between services. It would also make it easier to implement a system for tracking what services exist on the mesh.

Current status

The heartbeat is a simple alert message sent at regular intervals from a service. The routing key starts with heartbeat.[service name]. The payload is the name and id of the service. Note that the heartbeat is currently not described in the standard, but there is an issue to address this absence (#39). The heartbeat is implemented in dl-cpp's heartbeater mixin class. The routing key prefix comes from core, and the use is in service::listen().

Proposed additions

First heartbeat message

The first heartbeat sent should announce the existence of a new service in the mesh. The routing key should be heartbeat.[service name].hello (or something to that effect).

Why

This will allow a service to learn of the existence of another service on the mesh. For example, service A might be able to use service B for something if the latter is present. Or there could be a service that keeps track of other services on the mesh.

Last heartbeat message

The heartbeater should make a best effort to send a last heartbeat message when it's shutting down. The routing key should be heartbeat.[service name].goodbye (or something to that effect).

Why

This would allow a service to learn that another service is leaving the mesh. If service A is using service B, it would want to know if service B leaves the mesh so it can configure itself appropriately. If a service is tracking what services exist on the mesh, this is the most efficient way to update the status of a service that leaves the mesh (the only other alternatives are to watch for a lack of heartbeats, or to constantly ping).

Payload additions

The following should be added to the payload:

• heartbeat frequency

Why

This would allow a service that's keeping track of other services on the mesh to know when to worry about a lack of heartbeats from a service.

Implementation

The new functionality would be implemented in dl-cpp's heartbeater class. Specifically, it would all go in the execute() function. The new components of the routing keys for the hello and goodbye messages would go in core and be configurable.