Load Balancing

12 minute read

Envoy load balancing is a way of distributing traffic between multiple hosts within a single upstream cluster in order to effectively make use of available resources.

Envoy Gateway supports the following load balancing policies:

Round Robin: a simple policy in which each available upstream host is selected in round robin order.
Random: load balancer selects a random available host.
Least Request: load balancer uses different algorithms depending on whether hosts have the same or different weights.
Consistent Hash: load balancer implements consistent hashing to upstream hosts.

Envoy Gateway introduces a new CRD called BackendTrafficPolicy that allows the user to describe their desired load balancing polices. This instantiated resource can be linked to a Gateway, HTTPRoute or GRPCRoute resource. If loadBalancer is not specified in BackendTrafficPolicy, the default load balancing policy is Least Request.

Prerequisites

Install Envoy Gateway

Follow the steps from the Quickstart task to install Envoy Gateway and the example manifest. Before proceeding, you should be able to query the example backend using HTTP.

Verify the Gateway status:

kubectl get gateway/eg -o yaml

egctl x status gateway -v

For better testing the load balancer, you can add more hosts in upstream cluster by increasing the replicas of one deployment:

kubectl patch deployment backend -n default -p '{"spec": {"replicas": 4}}'

Install the hey load testing tool

Install the Hey CLI tool, this tool will be used to generate load and measure response times.

Follow the installation instruction from the Hey project docs.

Round Robin

This example will create a Load Balancer with Round Robin policy via BackendTrafficPolicy.

cat <<EOF | kubectl apply -f -
apiVersion: gateway.envoyproxy.io/v1alpha1
kind: BackendTrafficPolicy
metadata:
  name: round-robin-policy
  namespace: default
spec:
  targetRefs:
    - group: gateway.networking.k8s.io
      kind: HTTPRoute
      name: round-robin-route
  loadBalancer:
    type: RoundRobin
---
apiVersion: gateway.networking.k8s.io/v1
kind: HTTPRoute
metadata:
  name: round-robin-route
  namespace: default
spec:
  parentRefs:
    - name: eg
  hostnames:
    - "www.example.com"
  rules:
    - matches:
        - path:
            type: PathPrefix
            value: /round
      backendRefs:
        - name: backend
          port: 3000
EOF

Save and apply the following resource to your cluster:

---
apiVersion: gateway.envoyproxy.io/v1alpha1
kind: BackendTrafficPolicy
metadata:
  name: round-robin-policy
  namespace: default
spec:
  targetRefs:
    - group: gateway.networking.k8s.io
      kind: HTTPRoute
      name: round-robin-route
  loadBalancer:
    type: RoundRobin
---
apiVersion: gateway.networking.k8s.io/v1
kind: HTTPRoute
metadata:
  name: round-robin-route
  namespace: default
spec:
  parentRefs:
    - name: eg
  hostnames:
    - "www.example.com"
  rules:
    - matches:
        - path:
            type: PathPrefix
            value: /round
      backendRefs:
        - name: backend
          port: 3000

The hey tool will be used to generate 100 concurrent requests.

hey -n 100 -c 100 -host "www.example.com" http://${GATEWAY_HOST}/round

Summary:
  Total:	0.0487 secs
  Slowest:	0.0440 secs
  Fastest:	0.0181 secs
  Average:	0.0307 secs
  Requests/sec:	2053.1676

  Total data:	50500 bytes
  Size/request:	505 bytes

Response time histogram:
  0.018 [1]	    |■■
  0.021 [2]  	|■■■■
  0.023 [10]	|■■■■■■■■■■■■■■■■■■■■■■
  0.026 [16]	|■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■
  0.028 [7]  	|■■■■■■■■■■■■■■■■
  0.031 [10]	|■■■■■■■■■■■■■■■■■■■■■■
  0.034 [17]	|■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■
  0.036 [18]	|■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■
  0.039 [11]	|■■■■■■■■■■■■■■■■■■■■■■■■
  0.041 [6] 	|■■■■■■■■■■■■■
  0.044 [2]	    |■■■■

As a result, you can see all available upstream hosts receive traffics evenly.

kubectl get pods -l app=backend --no-headers -o custom-columns=":metadata.name" | while read -r pod; do echo "$pod: received $(($(kubectl logs $pod | wc -l) - 2)) requests"; done

backend-69fcff487f-2gfp7: received 26 requests
backend-69fcff487f-69g8c: received 25 requests
backend-69fcff487f-bqwpr: received 24 requests
backend-69fcff487f-kbn8l: received 25 requests

You should note that this results may vary, the output here is for reference purpose only.

Random

This example will create a Load Balancer with Random policy via BackendTrafficPolicy.

cat <<EOF | kubectl apply -f -
apiVersion: gateway.envoyproxy.io/v1alpha1
kind: BackendTrafficPolicy
metadata:
  name: random-policy
  namespace: default
spec:
  targetRefs:
    - group: gateway.networking.k8s.io
      kind: HTTPRoute
      name: random-route
  loadBalancer:
    type: Random
---
apiVersion: gateway.networking.k8s.io/v1
kind: HTTPRoute
metadata:
  name: random-route
  namespace: default
spec:
  parentRefs:
    - name: eg
  hostnames:
    - "www.example.com"
  rules:
    - matches:
        - path:
            type: PathPrefix
            value: /random
      backendRefs:
        - name: backend
          port: 3000
EOF

Save and apply the following resource to your cluster:

---
apiVersion: gateway.envoyproxy.io/v1alpha1
kind: BackendTrafficPolicy
metadata:
  name: random-policy
  namespace: default
spec:
  targetRefs:
    - group: gateway.networking.k8s.io
      kind: HTTPRoute
      name: random-route
  loadBalancer:
    type: Random
---
apiVersion: gateway.networking.k8s.io/v1
kind: HTTPRoute
metadata:
  name: random-route
  namespace: default
spec:
  parentRefs:
    - name: eg
  hostnames:
    - "www.example.com"
  rules:
    - matches:
        - path:
            type: PathPrefix
            value: /random
      backendRefs:
        - name: backend
          port: 3000

The hey tool will be used to generate 1000 concurrent requests.

hey -n 1000 -c 100 -host "www.example.com" http://${GATEWAY_HOST}/random

Summary:
  Total:	0.2624 secs
  Slowest:	0.0851 secs
  Fastest:	0.0007 secs
  Average:	0.0179 secs
  Requests/sec:	3811.3020

  Total data:	506000 bytes
  Size/request:	506 bytes

Response time histogram:
  0.001 [1] 	|
  0.009 [421]	|■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■
  0.018 [219]	|■■■■■■■■■■■■■■■■■■■■■
  0.026 [118]	|■■■■■■■■■■■
  0.034 [64]	|■■■■■■
  0.043 [73]	|■■■■■■■
  0.051 [41]	|■■■■
  0.060 [22]	|■■
  0.068 [19]	|■■
  0.077 [13]	|■
  0.085 [9] 	|■

As a result, you can see all available upstream hosts receive traffics randomly.

kubectl get pods -l app=backend --no-headers -o custom-columns=":metadata.name" | while read -r pod; do echo "$pod: received $(($(kubectl logs $pod | wc -l) - 2)) requests"; done

backend-69fcff487f-bf6lm: received 246 requests
backend-69fcff487f-gwmqk: received 256 requests
backend-69fcff487f-mzngr: received 230 requests
backend-69fcff487f-xghqq: received 268 requests

You should note that this results may vary, the output here is for reference purpose only.

Least Request

This example will create a Load Balancer with Least Request policy via BackendTrafficPolicy.

cat <<EOF | kubectl apply -f -
apiVersion: gateway.envoyproxy.io/v1alpha1
kind: BackendTrafficPolicy
metadata:
  name: least-request-policy
  namespace: default
spec:
  targetRefs:
    - group: gateway.networking.k8s.io
      kind: HTTPRoute
      name: least-request-route
  loadBalancer:
    type: LeastRequest
---
apiVersion: gateway.networking.k8s.io/v1
kind: HTTPRoute
metadata:
  name: least-request-route
  namespace: default
spec:
  parentRefs:
    - name: eg
  hostnames:
    - "www.example.com"
  rules:
    - matches:
        - path:
            type: PathPrefix
            value: /least
      backendRefs:
        - name: backend
          port: 3000
EOF

Save and apply the following resource to your cluster:

---
apiVersion: gateway.envoyproxy.io/v1alpha1
kind: BackendTrafficPolicy
metadata:
  name: least-request-policy
  namespace: default
spec:
  targetRefs:
    - group: gateway.networking.k8s.io
      kind: HTTPRoute
      name: least-request-route
  loadBalancer:
    type: LeastRequest
---
apiVersion: gateway.networking.k8s.io/v1
kind: HTTPRoute
metadata:
  name: least-request-route
  namespace: default
spec:
  parentRefs:
    - name: eg
  hostnames:
    - "www.example.com"
  rules:
    - matches:
        - path:
            type: PathPrefix
            value: /least
      backendRefs:
        - name: backend
          port: 3000

The hey tool will be used to generate 100 concurrent requests.

hey -n 100 -c 100 -host "www.example.com" http://${GATEWAY_HOST}/least

Summary:
  Total:	0.0489 secs
  Slowest:	0.0479 secs
  Fastest:	0.0054 secs
  Average:	0.0297 secs
  Requests/sec:	2045.9317

  Total data:	50500 bytes
  Size/request:	505 bytes

Response time histogram:
  0.005 [1] 	|■■
  0.010 [1] 	|■■
  0.014 [8] 	|■■■■■■■■■■■■■■■
  0.018 [6] 	|■■■■■■■■■■■
  0.022 [11]	|■■■■■■■■■■■■■■■■■■■■
  0.027 [7] 	|■■■■■■■■■■■■■
  0.031 [15]	|■■■■■■■■■■■■■■■■■■■■■■■■■■■
  0.035 [13]	|■■■■■■■■■■■■■■■■■■■■■■■■
  0.039 [22]	|■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■
  0.044 [12]	|■■■■■■■■■■■■■■■■■■■■■■
  0.048 [4] 	|■■■■■■■

As a result, you can see all available upstream hosts receive traffics randomly, and host backend-69fcff487f-6l2pw receives fewer requests than others.

kubectl get pods -l app=backend --no-headers -o custom-columns=":metadata.name" | while read -r pod; do echo "$pod: received $(($(kubectl logs $pod | wc -l) - 2)) requests"; done

backend-69fcff487f-59hvs: received 24 requests
backend-69fcff487f-6l2pw: received 19 requests
backend-69fcff487f-ktsx4: received 30 requests
backend-69fcff487f-nqxc7: received 27 requests

If you send one more requests to the ${GATEWAY_HOST}/least, you can tell that host backend-69fcff487f-6l2pw is very likely to get the attention of load balancer and receive this request.

backend-69fcff487f-59hvs: received 24 requests
backend-69fcff487f-6l2pw: received 20 requests
backend-69fcff487f-ktsx4: received 30 requests
backend-69fcff487f-nqxc7: received 27 requests

You should note that this results may vary, the output here is for reference purpose only.

Consistent Hash

This example will create a Load Balancer with Consistent Hash policy via BackendTrafficPolicy.

The underlying consistent hash algorithm that Envoy Gateway utilise is Maglev, and it can derive hash from following aspects:

SourceIP
Header
Cookie

They are also the supported value as consistent hash type.

Source IP

This example will create a Load Balancer with Source IP based Consistent Hash policy.

cat <<EOF | kubectl apply -f -
apiVersion: gateway.envoyproxy.io/v1alpha1
kind: BackendTrafficPolicy
metadata:
  name: source-ip-policy
  namespace: default
spec:
  targetRefs:
    - group: gateway.networking.k8s.io
      kind: HTTPRoute
      name: source-ip-route
  loadBalancer:
    type: ConsistentHash
    consistentHash:
      type: SourceIP
---
apiVersion: gateway.networking.k8s.io/v1
kind: HTTPRoute
metadata:
  name: source-ip-route
  namespace: default
spec:
  parentRefs:
    - name: eg
  hostnames:
    - "www.example.com"
  rules:
    - matches:
        - path:
            type: PathPrefix
            value: /source
      backendRefs:
        - name: backend
          port: 3000
EOF

Save and apply the following resource to your cluster:

---
apiVersion: gateway.envoyproxy.io/v1alpha1
kind: BackendTrafficPolicy
metadata:
  name: source-ip-policy
  namespace: default
spec:
  targetRefs:
    - group: gateway.networking.k8s.io
      kind: HTTPRoute
      name: source-ip-route
  loadBalancer:
    type: ConsistentHash
    consistentHash:
      type: SourceIP
---
apiVersion: gateway.networking.k8s.io/v1
kind: HTTPRoute
metadata:
  name: source-ip-route
  namespace: default
spec:
  parentRefs:
    - name: eg
  hostnames:
    - "www.example.com"
  rules:
    - matches:
        - path:
            type: PathPrefix
            value: /source
      backendRefs:
        - name: backend
          port: 3000

The hey tool will be used to generate 100 concurrent requests.

hey -n 100 -c 100 -host "www.example.com" http://${GATEWAY_HOST}/source

Summary:
  Total:	0.0539 secs
  Slowest:	0.0500 secs
  Fastest:	0.0198 secs
  Average:	0.0340 secs
  Requests/sec:	1856.5666

  Total data:	50600 bytes
  Size/request:	506 bytes

Response time histogram:
  0.020 [1] 	|■■
  0.023 [5] 	|■■■■■■■■■■■
  0.026 [12]	|■■■■■■■■■■■■■■■■■■■■■■■■■■■
  0.029 [16]	|■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■
  0.032 [11]	|■■■■■■■■■■■■■■■■■■■■■■■■
  0.035 [7] 	|■■■■■■■■■■■■■■■■
  0.038 [8] 	|■■■■■■■■■■■■■■■■■■
  0.041 [18]	|■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■
  0.044 [15]	|■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■
  0.047 [4] 	|■■■■■■■■■
  0.050 [3] 	|■■■■■■■

As a result, you can see all traffics are routed to only one upstream host, since the client that send requests has the same source IP.

kubectl get pods -l app=backend --no-headers -o custom-columns=":metadata.name" | while read -r pod; do echo "$pod: received $(($(kubectl logs $pod | wc -l) - 2)) requests"; done

backend-69fcff487f-grzkj: received 0 requests
backend-69fcff487f-n4d8w: received 100 requests
backend-69fcff487f-tb7zx: received 0 requests
backend-69fcff487f-wbzpg: received 0 requests

You can try different client to send out these requests, the upstream host that receives traffics may vary.

This example will create a Load Balancer with Header based Consistent Hash policy.

cat <<EOF | kubectl apply -f -
apiVersion: gateway.envoyproxy.io/v1alpha1
kind: BackendTrafficPolicy
metadata:
  name: header-policy
  namespace: default
spec:
  targetRefs:
    - group: gateway.networking.k8s.io
      kind: HTTPRoute
      name: header-route
  loadBalancer:
    type: ConsistentHash
    consistentHash:
      type: Header
      header:
        name: FooBar
---
apiVersion: gateway.networking.k8s.io/v1
kind: HTTPRoute
metadata:
  name: header-route
  namespace: default
spec:
  parentRefs:
    - name: eg
  hostnames:
    - "www.example.com"
  rules:
    - matches:
        - path:
            type: PathPrefix
            value: /header
      backendRefs:
        - name: backend
          port: 3000
EOF

Save and apply the following resource to your cluster:

---
apiVersion: gateway.envoyproxy.io/v1alpha1
kind: BackendTrafficPolicy
metadata:
  name: header-policy
  namespace: default
spec:
  targetRefs:
    - group: gateway.networking.k8s.io
      kind: HTTPRoute
      name: header-route
  loadBalancer:
    type: ConsistentHash
    consistentHash:
      type: Header
      header:
        name: FooBar
---
apiVersion: gateway.networking.k8s.io/v1
kind: HTTPRoute
metadata:
  name: header-route
  namespace: default
spec:
  parentRefs:
    - name: eg
  hostnames:
    - "www.example.com"
  rules:
    - matches:
        - path:
            type: PathPrefix
            value: /header
      backendRefs:
        - name: backend
          port: 3000

The hey tool will be used to generate 100 concurrent requests.

hey -n 100 -c 100 -host "www.example.com" -H "FooBar: 1.2.3.4" http://${GATEWAY_HOST}/header

Summary:
  Total:	0.0579 secs
  Slowest:	0.0510 secs
  Fastest:	0.0323 secs
  Average:	0.0431 secs
  Requests/sec:	1728.6064

  Total data:	53800 bytes
  Size/request:	538 bytes

Response time histogram:
  0.032 [1] 	|■■
  0.034 [3] 	|■■■■■■
  0.036 [1] 	|■■
  0.038 [1] 	|■■
  0.040 [7] 	|■■■■■■■■■■■■■■
  0.042 [20]	|■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■
  0.044 [20]	|■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■
  0.045 [20]	|■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■
  0.047 [16]	|■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■
  0.049 [9] 	|■■■■■■■■■■■■■■■■■■
  0.051 [2] 	|■■■■

As a result, you can see all traffics are routed to only one upstream host, since the header of all requests are the same.

kubectl get pods -l app=backend --no-headers -o custom-columns=":metadata.name" | while read -r pod; do echo "$pod: received $(($(kubectl logs $pod | wc -l) - 2)) requests"; done

backend-69fcff487f-dvt9r: received 0 requests
backend-69fcff487f-f8qdl: received 100 requests
backend-69fcff487f-gnpm4: received 0 requests
backend-69fcff487f-t2pgm: received 0 requests

You can try to add different header to these requests, and the upstream host that receives traffics may vary. The following output happens when you use hey to send another 100 requests with header FooBar: 5.6.7.8.

backend-69fcff487f-dvt9r: received 0 requests
backend-69fcff487f-f8qdl: received 100 requests
backend-69fcff487f-gnpm4: received 100 requests
backend-69fcff487f-t2pgm: received 0 requests

This example will create a Load Balancer with Cookie based Consistent Hash policy.

cat <<EOF | kubectl apply -f -
apiVersion: gateway.envoyproxy.io/v1alpha1
kind: BackendTrafficPolicy
metadata:
  name: cookie-policy
  namespace: default
spec:
  targetRefs:
    - group: gateway.networking.k8s.io
      kind: HTTPRoute
      name: cookie-route
  loadBalancer:
    type: ConsistentHash
    consistentHash:
      type: Cookie
      cookie:
        name: FooBar
        ttl: 60s
        attributes:
          SameSite: Strict
---
apiVersion: gateway.networking.k8s.io/v1
kind: HTTPRoute
metadata:
  name: cookie-route
  namespace: default
spec:
  parentRefs:
    - name: eg
  hostnames:
    - "www.example.com"
  rules:
    - matches:
        - path:
            type: PathPrefix
            value: /cookie
      backendRefs:
        - name: backend
          port: 3000
EOF

Save and apply the following resource to your cluster:

---
apiVersion: gateway.envoyproxy.io/v1alpha1
kind: BackendTrafficPolicy
metadata:
  name: cookie-policy
  namespace: default
spec:
  targetRefs:
    - group: gateway.networking.k8s.io
      kind: HTTPRoute
      name: cookie-route
  loadBalancer:
    type: ConsistentHash
    consistentHash:
      type: Cookie
      cookie:
        name: FooBar
        ttl: 60s
        attributes:
          SameSite: Strict
---
apiVersion: gateway.networking.k8s.io/v1
kind: HTTPRoute
metadata:
  name: cookie-route
  namespace: default
spec:
  parentRefs:
    - name: eg
  hostnames:
    - "www.example.com"
  rules:
    - matches:
        - path:
            type: PathPrefix
            value: /cookie
      backendRefs:
        - name: backend
          port: 3000

By sending 10 request with curl to the ${GATEWAY_HOST}/cookie, you can see that all requests got routed to only one upstream host, since they have same cookie setting.

for i in {1..10}; do curl -I --header "Host: www.example.com" --cookie "FooBar=1.2.3.4" http://${GATEWAY_HOST}/cookie ; sleep 1; done

kubectl get pods -l app=backend --no-headers -o custom-columns=":metadata.name" | while read -r pod; do echo "$pod: received $(($(kubectl logs $pod | wc -l) - 2)) requests"; done

backend-69fcff487f-5dxz9: received 0 requests
backend-69fcff487f-gpvl2: received 0 requests
backend-69fcff487f-pglgv: received 10 requests
backend-69fcff487f-qxr74: received 0 requests

You can try to set different cookie to these requests, the upstream host that receives traffics may vary. The following output happens when you use curl to send another 10 requests with cookie FooBar: 5.6.7.8.

backend-69fcff487f-dvt9r: received 0 requests
backend-69fcff487f-f8qdl: received 0 requests
backend-69fcff487f-gnpm4: received 10 requests
backend-69fcff487f-t2pgm: received 10 requests

If the cookie has not been set in one request, Envoy Gateway will auto-generate a cookie for this request according to the ttl and attributes field.

In this example, the following cookie will be generated (see set-cookie header in response) if sending a request without cookie:

curl -v --header "Host: www.example.com" http://${GATEWAY_HOST}/cookie

> GET /cookie HTTP/1.1
> Host: www.example.com
> User-Agent: curl/7.74.0
> Accept: */*
>
* Mark bundle as not supporting multiuse
< HTTP/1.1 200 OK
< content-type: application/json
< x-content-type-options: nosniff
< date: Fri, 19 Jul 2024 16:49:57 GMT
< content-length: 458
< set-cookie: FooBar="88358b9442700c56"; Max-Age=60; SameSite=Strict; HttpOnly
<
{
 "path": "/cookie",
 "host": "www.example.com",
 "method": "GET",
 "proto": "HTTP/1.1",
 "headers": {
  "Accept": [
   "*/*"
  ],
  "User-Agent": [
   "curl/7.74.0"
  ],
  "X-Envoy-Internal": [
   "true"
  ],
  "X-Forwarded-For": [
   "10.244.0.1"
  ],
  "X-Forwarded-Proto": [
   "http"
  ],
  "X-Request-Id": [
   "1adeaaf7-d45c-48c8-9a4d-eadbccb2fd50"
  ]
 },
 "namespace": "default",
 "ingress": "",
 "service": "",
 "pod": "backend-69fcff487f-5dxz9"

Feedback

Was this page helpful?

Glad to hear it! Please tell us how we can improve.

Sorry to hear that. Please tell us how we can improve.

Last modified December 20, 2024: chore: fix typo (#4958) (2a10d47)