Skip to content

Commit 08e8af2

Browse files
committed
add HTTPRoute manifests and update profile
Signed-off-by: samzong <[email protected]>
1 parent 7fd4057 commit 08e8af2

File tree

2 files changed

+51
-6
lines changed

2 files changed

+51
-6
lines changed
Lines changed: 49 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,49 @@
1+
apiVersion: gateway.networking.k8s.io/v1
2+
kind: HTTPRoute
3+
metadata:
4+
name: vsr-llama8b-svc
5+
namespace: default
6+
spec:
7+
parentRefs:
8+
- group: gateway.networking.k8s.io
9+
kind: Gateway
10+
name: inference-gateway
11+
rules:
12+
- backendRefs:
13+
- name: vllm-llama3-8b-instruct-epp
14+
port: 9002
15+
matches:
16+
- path:
17+
type: PathPrefix
18+
value: /
19+
headers:
20+
- type: Exact
21+
name: x-selected-model
22+
value: llama3-8b
23+
timeouts:
24+
request: 300s
25+
---
26+
apiVersion: gateway.networking.k8s.io/v1
27+
kind: HTTPRoute
28+
metadata:
29+
name: vsr-phi4-mini-svc
30+
namespace: default
31+
spec:
32+
parentRefs:
33+
- group: gateway.networking.k8s.io
34+
kind: Gateway
35+
name: inference-gateway
36+
rules:
37+
- backendRefs:
38+
- name: vllm-phi4-mini-epp
39+
port: 9002
40+
matches:
41+
- path:
42+
type: PathPrefix
43+
value: /
44+
headers:
45+
- type: Exact
46+
name: x-selected-model
47+
value: phi4-mini
48+
timeouts:
49+
request: 300s

e2e/profiles/llm-d/profile.go

Lines changed: 2 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -88,8 +88,7 @@ func (p *Profile) Setup(ctx context.Context, opts *framework.SetupOptions) error
8888

8989
func (p *Profile) Teardown(ctx context.Context, opts *framework.TeardownOptions) error {
9090
p.verbose = opts.Verbose
91-
_ = p.kubectlDelete(ctx, "deploy/kubernetes/llmd-base/httproute-llama-pool.yaml")
92-
_ = p.kubectlDelete(ctx, "deploy/kubernetes/llmd-base/httproute-phi4-pool.yaml")
91+
_ = p.kubectlDelete(ctx, "e2e/profiles/llm-d/manifests/httproute-services.yaml")
9392
_ = p.kubectlDelete(ctx, "deploy/kubernetes/llmd-base/dest-rule-epp-llama.yaml")
9493
_ = p.kubectlDelete(ctx, "deploy/kubernetes/llmd-base/dest-rule-epp-phi4.yaml")
9594
_ = p.kubectlDelete(ctx, "deploy/kubernetes/llmd-base/inferencepool-llama.yaml")
@@ -223,10 +222,7 @@ func (p *Profile) deployGatewayRoutes(ctx context.Context) error {
223222
if err := p.kubectlApply(ctx, "deploy/kubernetes/istio/gateway.yaml"); err != nil {
224223
return err
225224
}
226-
if err := p.kubectlApply(ctx, "deploy/kubernetes/llmd-base/httproute-llama-pool.yaml"); err != nil {
227-
return err
228-
}
229-
if err := p.kubectlApply(ctx, "deploy/kubernetes/llmd-base/httproute-phi4-pool.yaml"); err != nil {
225+
if err := p.kubectlApply(ctx, "e2e/profiles/llm-d/manifests/httproute-services.yaml"); err != nil {
230226
return err
231227
}
232228
if err := p.kubectlApply(ctx, "deploy/kubernetes/istio/destinationrule.yaml"); err != nil {

0 commit comments

Comments
 (0)