envoy: Implement locality-based prioritized (affinitized) endpoint traffic routing

[draychev]

This issue is for the following sub-issues:

• Build a quick proof of concept of the idea described below Build a proof of concept of the idea locality-based routing using Envoy #3557
• Create a small design document
• Create sub-issues and link to this issue
• Implement (if feasible) the design

---

Context

We need to implement traffic routing such that traffic is first routed to the group of endpoints closest to the origin, then to those groups further and further away (next cluster, next zone, next region etc.)

Documentation:

• Locality weighted load balancing
• config.endpoint.v3.LocalityLbEndpoints

Take a look at the priority field. From the Envoy docs on the priority field:

Under usual circumstances, Envoy will only select endpoints for the highest priority (0).
In the event all endpoints for a particular priority are unavailable/unhealthy, Envoy will
fail over to selecting endpoints for the next highest priority group.

[steeling]

Also seems like the locality should be set more intelligently? https://www.envoyproxy.io/docs/envoy/latest/api-v3/config/core/v3/base.proto#envoy-v3-api-msg-config-core-v3-locality

[allenlsy]

Also take a look at this: https://www.envoyproxy.io/docs/envoy/latest/intro/arch_overview/upstream/load_balancing/zone_aware.html. I feel it is quite similar to locality based routing

[allenlsy]

Can I get a early review for this WIP mini design doc?

https://docs.google.com/document/d/1OxDPYkdffEJmraSMZpl83JOt3Kjxjfmoehp4T1ebpfs/edit#

[allenlsy]

I think we don't need to create sub-tasks for the implementation. Using just this task should be sufficient.

Before the implementation, there are two prerequisites:

• OSM Multicluster CRD is ready to be consumed
• OSM Multicluster service is completed, so that source service can reach destination service in all clusters.

Then the changes need to be made are:

• OSM controller watches MultiCluster resource changes
• Updates the endpoint and load_balancing_weight field of the Envoy CDS API response. This should be part of the catalog.

[allenlsy]

Sub tasks of this issue:

• Customizable traffic shifting threshold #4036 Customizable traffic shifting threshold
• Cluster priority assignment #4037 Customizable cluster priority

[michelleN]

Is this being put behind a feature flag?

[allenlsy]

Sorry I missed the message.

It's not behind feature flag. The figure does not affect single cluster scenario. But when the enableMulticlusterMode flag is turned on, this figure will start to work.