In this document, you will learn the basic concept of an upstream object in APISIX and why you would want to use it. You will be introduced to a few relevant features, including load balancing, service discovery, and upstream health checking.
Explore additional resources at the end for more information on related topics.
An upstream object in APISIX is a logical abstraction of a set containing one or more upstream addresses. It is required in routes or services to specify where requests flow to and how they are distributed.
There are different ways to configure upstream addresses, one of which is to explicitly configure them in every route/service. Here is an example of such a configuration in routes, where the same upstream address is repeated across three different routes:
This approach, however, leads to data redundancies and poses a risk for update anomalies (i.e. inconsistent data resulting from a partial update).
A better approach is to extract the repetitive upstream configurations into a separate upstream object and embed its
upstream_id into routes/services, such as the following:
As you can probably see, large-scale systems with many routes/services would benefit significantly from configuring identical groups of upstream addresses in upstream objects, reducing redundant information and operational costs.
An important configuration option in upstreams is the load balancing algorithm.
APISIX offers four load-balancing algorithm options to choose from:
roundrobin- weighted round robin
chash- consistent hashing
ewma- exponentially weighted moving average
least_conn- least connections
You can also build your own load balancing algorithm by adding the source file under
apisix/balancer and importing it with
require("apisix.balancer.your_balancer_name"), where needed.
While it is straightforward to figure upstream addresses statically, in microservice-based architectures, upstream addresses are often dynamically assigned and therefore, changed, during autoscaling, failures, and updates. Static configurations are less than ideal in this case.
Service discovery comes to rescue. It describes the process of automatically detecting the available upstream services, keeping their addresses in a database (called a service registry) for others to reference. That way, an API gateway can always fetch the latest list of upstream addresses through the registry, ensuring all requests are forwarded to healthy upstream nodes.
APISIX supports integrations with many service registries, such as Consul, Eureka, Nacos, Kubernetes service discovery, and more.
For more details about how to integrate with third-party service registries, please see Service Discovery (coming soon).
Upstream Health Checking
APISIX provides active and passive health checking options to probe if upstream nodes are online (a.k.a. healthy). Unhealthy upstream nodes will be ignored until they recover and are deemed healthy again.
Upstream health checking can be configured in the
checks parameter in an upstream object.
More details about how to configure upstream health checking will be provided in Active and Passive Health Checking (coming soon).
- Getting Started - Load Balancing