System Architecture

Overview

The system should be made up of a few services, that are small, and horizontally scalable.

The system should be the following: - Scalable - Fault Tolerant - Use other OpenStack projects as first class citizens - For the GSLB appliance drivers the default should be Designate - For regional pool members Neutron LBaaS V2 should be the default

Note

This is for the MVP. Post MVP we need to also be able to run as a global service.

Overview Diagram

sysarch/sysarch-diagram-overview.dot

Services

service name	deployment model	purpose	notes
kosmos-api	multiple	Configuration of GSLBs
kosmos-conductor	multiple - single region	Database Access
kosmos-status-checker	multiple - global	Check status of endpoints
kosmos-engine	multiple - single region	Business Logic
GSLB Appliance	multiple - global (depending on appliance used)	Applicance that is used by Kosmos to direct traffic to the correct endpoints.	This is Designate in the reference Implementation
Endpoints	multiple - multiple regions	The destination for the traffic that is being load balanced by Kosmos	This will be LBaaS V2 Load Balancers by default, but could also be IPs or Hardware Load Balancers

kosmos-api

The API would be a WSGI service that implements the API Spec previously approved.

It should follow the standard OpenStack patterns (use Keystone / oslo.middleware / oslo.context)

kosmos-conductor

This will act as a single point of access to the DB, allowing for consistent data validation.

As results are fed in from status-checkers, then engine will decide if an endpoint is suitable to be included in the pool. As it adds and removes nodes, it will update both the database, and the GSLB appliance, via the plugin. (e.g. if using the Designate plugin it would remove the endpoint's IP from the record)

Warning

The logging component described below may be cut in the MVP

There is a logging component that could be a separate DB, a timeseries DB, a columnar DB, elastic search or other storage system This will store a section of history for the pools, to allow users to see when / why endpoints were added / removed from the load balancer.

sysarch/sysarch-diagram-conductor.dot

kosmos-status-checker

This is a worker style service that will run status checks on the defined endpoints.

This will have a plugin interface to allow for more checks to the loaded by different plugins, or for custom checks to be written by the deployer (using standard OpenStack plugin patterns)

For example, any pool-members / endpoints that are LBaaS instances will not call into the VIP, but will call the LBaaS V2 API to get the health of the regional members.

sysarch/sysarch-diagram-status-checker.dot

kosmos-engine

This is where all the business logic resides. This service will consume status results and decide if an endpoint should be removed / added. This service will then use the plugin loaded for the GSLB backend to orchestrate this.

The plugins will allow different types of GSLB appliances / services to be used, and will use standard OpenStack plugin patterns.

sysarch/sysarch-diagram-engine.dot

Entity Relationship Diagram

sysarch/erd-diagram.dot

Example

This is an example flow of information, when we are using Designate and checking the health info of a Neutron LBaaS load balancer. The plugin components are excluded for clarity, but would be between the "Status Check" and "Engine" components.

sysarch/example-flow.diag

Implementation

Assignee(s)

Primary assignee:: kosmos-drivers

Milestones

Target Milestone for completion:: M-2

7.7 KiB Raw Blame History