Sticky Load Balancing

The sticky load balancer selects a single service instance and keeps using it until it fails. Then, it selects another one.

It is possible to configure a backoff time to specify for how long a failing service instance should not be retried.

Precisely, the load balancer works as follows:

• if no service instance has been selected so far, select the first instance from the collection;
• else if the previously selected service instance has not failed, and is still available, return it;
• else return the first available service instance that has no recorded failure, if one exists;
• else, find the available instance for which the time since the last failure is the longest, and
  • if the backoff time since the failure passed, return it;
  • or, throw an NoAcceptableServiceInstanceFoundException as no acceptable instances are available.

Configuration

To use the sticky load service selection strategy, set the load balancer type to sticky:

stork standalonestork in quarkus

stork.my-service.service-discovery.type=...
stork.my-service.service-discovery...=...
stork.my-service.load-balancer.type=sticky

quarkus.stork.my-service.service-discovery.type=...
quarkus.stork.my-service.service-discovery...=...
quarkus.stork.my-service.load-balancer.type=sticky

The following attributes are supported:

Attribute	Mandatory	Default Value	Description
failure-backoff-time	No	0	After how much time, a service instance that has failed can be reused.