metis/pkg/service/cluster_test.go

package service

import (
	"encoding/json"
	"errors"
	"net/http"
	"net/http/httptest"
	"os"
	"path/filepath"
	"strings"
	"testing"
)

func TestInClusterKubeClientMissingEnv(t *testing.T) {
	t.Setenv("KUBERNETES_SERVICE_HOST", "")
	t.Setenv("KUBERNETES_SERVICE_PORT", "")
	if _, err := inClusterKubeClient(); err == nil {
		t.Fatal("expected inClusterKubeClient error without env")
	}
}

func TestKubeClientAndPodHelpers(t *testing.T) {
	kube := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
		switch {
		case r.Method == http.MethodGet && r.URL.Path == "/api/v1/nodes":
			_ = json.NewEncoder(w).Encode(map[string]any{
				"items": []any{
					map[string]any{
						"metadata": map[string]any{"name": "b", "labels": map[string]string{"kubernetes.io/arch": "arm64", "node-role.kubernetes.io/worker": "true"}},
						"spec":     map[string]any{"unschedulable": false},
					},
					map[string]any{
						"metadata": map[string]any{"name": "a", "labels": map[string]string{"kubernetes.io/arch": "arm64", "node-role.kubernetes.io/worker": "true"}},
						"spec":     map[string]any{"unschedulable": false},
					},
				},
			})
		case r.Method == http.MethodPost && strings.Contains(r.URL.Path, "/pods"):
			w.WriteHeader(http.StatusCreated)
		case r.Method == http.MethodDelete && strings.Contains(r.URL.Path, "/nodes/"):
			w.WriteHeader(http.StatusNotFound)
		case r.Method == http.MethodGet && strings.Contains(r.URL.Path, "/pods/") && strings.HasSuffix(r.URL.Path, "/log"):
			http.Error(w, "proxy error from 127.0.0.1:6443", http.StatusBadGateway)
		case r.Method == http.MethodGet && strings.Contains(r.URL.Path, "/pods/"):
			_ = json.NewEncoder(w).Encode(map[string]any{
				"metadata": map[string]any{"name": filepath.Base(r.URL.Path)},
				"status": map[string]any{
					"phase":   "Failed",
					"reason":  "CrashLoopBackOff",
					"message": "boom",
					"containerStatuses": []any{
						map[string]any{
							"state": map[string]any{
								"waiting":    map[string]any{"reason": "ImagePullBackOff", "message": "pulling"},
								"terminated": map[string]any{"reason": "Completed", "message": "done"},
							},
						},
					},
				},
			})
		default:
			http.NotFound(w, r)
		}
	}))
	defer kube.Close()

	client := kubeClientFactoryForURL(kube.URL, kube.Client())
	if err := client.jsonRequest(http.MethodGet, "/api/v1/nodes", nil, &map[string]any{}); err != nil {
		t.Fatalf("jsonRequest: %v", err)
	}
	if err := client.deleteRequest("/api/v1/nodes/a"); err != nil {
		t.Fatalf("deleteRequest 404 should be nil: %v", err)
	}
	if err := client.jsonRequest(http.MethodGet, "/missing", nil, &map[string]any{}); err == nil {
		t.Fatal("expected jsonRequest failure on 404")
	}

	origFactory := kubeClientFactory
	kubeClientFactory = func() (*kubeClient, error) {
		return client, nil
	}
	t.Cleanup(func() { kubeClientFactory = origFactory })
	nodes := clusterNodes()
	if len(nodes) != 2 || nodes[0].Name != "a" {
		t.Fatalf("clusterNodes sort mismatch: %#v", nodes)
	}

	app := newTestApp(t)
	app.settings.Namespace = "maintenance"
	app.settings.RunnerImageARM64 = "runner:arm64"
	state, err := app.remotePodState(client, "metis-build-test")
	if err != nil {
		t.Fatalf("remotePodState: %v", err)
	}
	if state.Reason != "Completed" || state.Message != "done" {
		t.Fatalf("expected terminated state override, got %#v", state)
	}
	if _, err := app.remotePodLogs(client, "metis-build-test"); err == nil || !strings.Contains(err.Error(), "could not reach the node kubelet log endpoint") {
		t.Fatalf("expected kubelet log endpoint error, got %v", err)
	}

	if _, err := app.runRemotePod("job-1", "metis-fail-test", map[string]any{}); err == nil {
		t.Fatal("expected runRemotePod failure")
	}
	if _, err := app.ensureDevice("titan-22", "missing"); err == nil {
		t.Fatal("expected ensureDevice missing target to fail")
	}
}

func TestDeleteNodeObjectFallback(t *testing.T) {
	tmp := t.TempDir()
	kubectl := filepath.Join(tmp, "kubectl")
	if err := os.WriteFile(kubectl, []byte("#!/usr/bin/env bash\nset -eu\nprintf '%s' \"$*\" > \""+filepath.Join(tmp, "kubectl.args")+"\"\n"), 0o755); err != nil {
		t.Fatal(err)
	}
	t.Setenv("PATH", tmp+string(os.PathListSeparator)+os.Getenv("PATH"))
	origFactory := kubeClientFactory
	kubeClientFactory = func() (*kubeClient, error) { return nil, errors.New("offline") }
	t.Cleanup(func() { kubeClientFactory = origFactory })
	if err := deleteNodeObject("titan-15"); err != nil {
		t.Fatalf("deleteNodeObject fallback: %v", err)
	}
}

func TestClusterActiveRemotePodLoadsCountsOnlyLivePods(t *testing.T) {
	kube := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
		switch {
		case r.Method == http.MethodGet && r.URL.Path == "/api/v1/namespaces/maintenance/pods":
			if got := r.URL.Query().Get("labelSelector"); got != "app=metis-remote,metis-run=build" {
				t.Fatalf("unexpected labelSelector %q", got)
			}
			_ = json.NewEncoder(w).Encode(map[string]any{
				"items": []any{
					map[string]any{
						"metadata": map[string]any{"labels": map[string]string{"app": "metis-remote", "metis-run": "build"}},
						"spec":     map[string]any{"nodeName": "titan-04"},
						"status":   map[string]any{"phase": "Running"},
					},
					map[string]any{
						"metadata": map[string]any{"labels": map[string]string{"app": "metis-remote", "metis-run": "build"}},
						"spec":     map[string]any{"nodeName": "titan-04"},
						"status":   map[string]any{"phase": "Pending"},
					},
					map[string]any{
						"metadata": map[string]any{"labels": map[string]string{"app": "metis-remote", "metis-run": "build"}},
						"spec":     map[string]any{"nodeName": "titan-05"},
						"status":   map[string]any{"phase": "Succeeded"},
					},
				},
			})
		default:
			http.NotFound(w, r)
		}
	}))
	defer kube.Close()

	origFactory := kubeClientFactory
	kubeClientFactory = func() (*kubeClient, error) {
		return kubeClientFactoryForURL(kube.URL, kube.Client()), nil
	}
	t.Cleanup(func() { kubeClientFactory = origFactory })

	loads := clusterActiveRemotePodLoads("maintenance", "build")
	if loads["titan-04"] != 2 {
		t.Fatalf("expected titan-04 load 2, got %#v", loads)
	}
	if _, ok := loads["titan-05"]; ok {
		t.Fatalf("expected succeeded pod to be ignored, got %#v", loads)
	}
}