Files
easyai-ai-gateway/apps/api/internal/capacitycontroller/kubernetes_test.go
T
wangbo 60d16b11ba fix(capacity): 忽略未就绪的 Worker 节点
香港站点保留了已离线的旧 control-plane 节点标签,容量控制器仍尝试读取其 metrics,导致 404 并阻断 readiness。容量采集现在只纳入 Ready=True 的候选节点,实际 Worker Pod 所在的就绪节点继续参与容量计算。

验证:容量控制器聚焦测试与 API 全量 go test 均通过,并覆盖同站点 NotReady 旧节点与 Ready Worker 节点并存场景。
2026-08-04 20:54:25 +08:00

173 lines
6.8 KiB
Go

package capacitycontroller
import (
"context"
"net/http"
"net/http/httptest"
"os"
"path/filepath"
"strings"
"testing"
)
func TestKubernetesClientReadsSiteAndMutatesOnlyWorkerScale(t *testing.T) {
var patches []string
server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, request *http.Request) {
if request.Header.Get("Authorization") != "Bearer test-token" {
http.Error(w, "unauthorized", http.StatusUnauthorized)
return
}
switch {
case request.Method == http.MethodGet && strings.Contains(request.URL.Path, "/deployments/easyai-worker-hongkong"):
_, _ = w.Write([]byte(`{
"spec":{"replicas":1,"template":{"spec":{"containers":[{
"name":"worker","image":"registry.invalid/gateway@sha256:abc",
"env":[{"name":"AI_GATEWAY_REVISION","value":"release-sha"}],
"resources":{"requests":{"memory":"512Mi","cpu":"250m"}}
}]}}}
}`))
case request.Method == http.MethodGet && request.URL.Path == "/api/v1/nodes":
if request.URL.Query().Get("labelSelector") != "easyai.io/site=hongkong,easyai.io/worker=true" {
http.Error(w, "unexpected Worker node selector", http.StatusBadRequest)
return
}
_, _ = w.Write([]byte(`{"items":[
{"metadata":{"name":"easyai-hongkong"},"status":{
"allocatable":{"memory":"8Gi","cpu":"4"},
"conditions":[{"type":"Ready","status":"False"}]
}},
{"metadata":{"name":"easyai-hongkong-worker-2"},"status":{
"allocatable":{"memory":"8Gi","cpu":"4"},
"conditions":[{"type":"Ready","status":"True"},{"type":"MemoryPressure","status":"False"}]
}}]}`))
case request.Method == http.MethodGet && strings.Contains(request.URL.Path, "/api/v1/namespaces/easyai/pods"):
_, _ = w.Write([]byte(`{"items":[{"metadata":{"name":"worker-hongkong-1"},"spec":{"nodeName":"easyai-hongkong-worker-2"}}]}`))
case request.Method == http.MethodGet && strings.Contains(request.URL.Path, "/metrics.k8s.io/") &&
strings.HasSuffix(request.URL.Path, "/nodes/easyai-hongkong-worker-2"):
_, _ = w.Write([]byte(`{"usage":{"memory":"3Gi","cpu":"500m"}}`))
case request.Method == http.MethodGet && strings.Contains(request.URL.Path, "/metrics.k8s.io/") &&
strings.HasSuffix(request.URL.Path, "/nodes/easyai-hongkong"):
t.Error("NotReady node metrics must not be queried")
http.Error(w, "NotReady node", http.StatusInternalServerError)
case request.Method == http.MethodGet && strings.Contains(request.URL.Path, "/metrics.k8s.io/") &&
strings.HasSuffix(request.URL.Path, "/pods"):
_, _ = w.Write([]byte(`{"items":[{
"metadata":{"name":"worker-hongkong-1"},
"containers":[{"usage":{"memory":"384Mi","cpu":"125m"}}]
}]}`))
case request.Method == http.MethodPatch:
patches = append(patches, request.URL.Path)
w.Header().Set("Content-Type", "application/json")
_, _ = w.Write([]byte(`{}`))
default:
http.NotFound(w, request)
}
}))
defer server.Close()
tokenFile := filepath.Join(t.TempDir(), "token")
if err := os.WriteFile(tokenFile, []byte("test-token"), 0o600); err != nil {
t.Fatal(err)
}
client, err := NewKubernetesClient(KubernetesConfig{
Namespace: "easyai", APIServer: server.URL, TokenFile: tokenFile, HTTPClient: server.Client(),
})
if err != nil {
t.Fatal(err)
}
state, err := client.SiteState(context.Background(), "hongkong")
if err != nil {
t.Fatal(err)
}
if state.CurrentReplicas != 1 || state.NodeName != "easyai-hongkong-worker-2" ||
state.WorkerRequestMemoryBytes != 512<<20 || state.WorkerRequestMilliCPU != 250 ||
state.AllocatableMemoryBytes != 8<<30 || state.UsedMemoryBytes != 3<<30 ||
state.Nodes[0].WorkerUsedMemoryBytes != 384<<20 || state.Nodes[0].WorkerUsedMilliCPU != 125 {
t.Fatalf("site state=%+v", state)
}
if err := client.SetPodDeletionCost(context.Background(), "worker-pod", -1000); err != nil {
t.Fatal(err)
}
if err := client.ScaleWorkerDeployment(context.Background(), "hongkong", 2); err != nil {
t.Fatal(err)
}
if len(patches) != 2 ||
!strings.Contains(patches[0], "/pods/worker-pod") ||
!strings.Contains(patches[1], "/deployments/easyai-worker-hongkong/scale") {
t.Fatalf("patches=%v", patches)
}
}
func TestKubernetesClientAllowsDisabledSiteWithoutWorkerNodes(t *testing.T) {
server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, request *http.Request) {
switch {
case request.Method == http.MethodGet && strings.Contains(request.URL.Path, "/deployments/easyai-worker-ningbo"):
_, _ = w.Write([]byte(`{
"spec":{"replicas":0,"template":{"spec":{"containers":[{
"name":"worker","image":"registry.invalid/gateway@sha256:abc",
"env":[{"name":"AI_GATEWAY_REVISION","value":"release-sha"}],
"resources":{"requests":{"memory":"1536Mi","cpu":"500m"}}
}]}}}
}`))
case request.Method == http.MethodGet && request.URL.Path == "/api/v1/nodes":
_, _ = w.Write([]byte(`{"items":[]}`))
default:
http.NotFound(w, request)
}
}))
defer server.Close()
tokenFile := filepath.Join(t.TempDir(), "token")
if err := os.WriteFile(tokenFile, []byte("test-token"), 0o600); err != nil {
t.Fatal(err)
}
client, err := NewKubernetesClient(KubernetesConfig{
Namespace: "easyai", APIServer: server.URL, TokenFile: tokenFile, HTTPClient: server.Client(),
})
if err != nil {
t.Fatal(err)
}
state, err := client.SiteState(context.Background(), "ningbo")
if err != nil {
t.Fatal(err)
}
if state.Site != "ningbo" || state.CurrentReplicas != 0 || len(state.Nodes) != 0 {
t.Fatalf("site state=%+v", state)
}
}
func TestKubernetesClientRejectsActiveSiteWithoutWorkerNodes(t *testing.T) {
server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, request *http.Request) {
switch {
case request.Method == http.MethodGet && strings.Contains(request.URL.Path, "/deployments/easyai-worker-ningbo"):
_, _ = w.Write([]byte(`{"spec":{"replicas":1,"template":{"spec":{"containers":[]}}}}`))
case request.Method == http.MethodGet && request.URL.Path == "/api/v1/nodes":
_, _ = w.Write([]byte(`{"items":[]}`))
default:
http.NotFound(w, request)
}
}))
defer server.Close()
tokenFile := filepath.Join(t.TempDir(), "token")
if err := os.WriteFile(tokenFile, []byte("test-token"), 0o600); err != nil {
t.Fatal(err)
}
client, err := NewKubernetesClient(KubernetesConfig{
Namespace: "easyai", APIServer: server.URL, TokenFile: tokenFile, HTTPClient: server.Client(),
})
if err != nil {
t.Fatal(err)
}
if _, err := client.SiteState(context.Background(), "ningbo"); err == nil ||
!strings.Contains(err.Error(), "site ningbo has no ready matching nodes") {
t.Fatalf("expected missing Worker node error, got %v", err)
}
}
func TestQuantityParsing(t *testing.T) {
if value, err := parseBinaryQuantity("1536Mi"); err != nil || value != 1536<<20 {
t.Fatalf("memory=%d err=%v", value, err)
}
if value, err := parseCPUQuantity("250000000n"); err != nil || value != 250 {
t.Fatalf("cpu=%d err=%v", value, err)
}
}