ananke/internal/cluster/orchestrator_critical_endpoint_additional_test.go

package cluster

import (
	"context"
	"errors"
	"io"
	"log"
	"path/filepath"
	"strings"
	"testing"
	"time"

	"scm.bstein.dev/bstein/ananke/internal/config"
	"scm.bstein.dev/bstein/ananke/internal/execx"
	"scm.bstein.dev/bstein/ananke/internal/state"
)

// TestCriticalEndpointHelpers runs one orchestration or CLI step.
// Signature: TestCriticalEndpointHelpers(t *testing.T).
// Why: covers critical endpoint parsing and readiness checks that gate startup completion.
func TestCriticalEndpointHelpers(t *testing.T) {
	cfg := config.Config{
		Startup: config.Startup{
			CriticalServiceEndpoints: []string{"monitoring/victoria-metrics-single-server"},
		},
	}
	orch := buildOrchestratorWithStubs(t, cfg, []commandStub{
		{match: matchContains("kubectl", "get endpoints victoria-metrics-single-server"), out: "10.42.0.10\n10.42.0.11\n"},
	})
	ok, detail, ns, svc, err := orch.criticalServiceEndpointsReady(context.Background())
	if err != nil || !ok {
		t.Fatalf("expected criticalServiceEndpointsReady success, got ok=%v detail=%q ns=%q svc=%q err=%v", ok, detail, ns, svc, err)
	}
	if detail != "services=1" {
		t.Fatalf("unexpected readiness detail: %q", detail)
	}
	gotNS, gotSvc, err := parseCriticalServiceEndpoint("monitoring/victoria-metrics-single-server")
	if err != nil || gotNS != "monitoring" || gotSvc != "victoria-metrics-single-server" {
		t.Fatalf("unexpected parse result ns=%q svc=%q err=%v", gotNS, gotSvc, err)
	}
	if _, _, err := parseCriticalServiceEndpoint("invalid"); err == nil {
		t.Fatalf("expected parseCriticalServiceEndpoint error")
	}
}

// TestCriticalEndpointAutoHealWorkflow runs one orchestration or CLI step.
// Signature: TestCriticalEndpointAutoHealWorkflow(t *testing.T).
// Why: covers endpoint-zero recovery where startup heals workload replicas before succeeding.
func TestCriticalEndpointAutoHealWorkflow(t *testing.T) {
	cfg := config.Config{
		Startup: config.Startup{
			CriticalServiceEndpointWaitSec: 2,
			CriticalServiceEndpointPollSec: 1,
			CriticalServiceEndpoints:       []string{"monitoring/victoria-metrics-single-server"},
		},
		State: config.State{
			Dir:            t.TempDir(),
			ReportsDir:     filepath.Join(t.TempDir(), "reports"),
			RunHistoryPath: filepath.Join(t.TempDir(), "runs.json"),
		},
	}
	orch := &Orchestrator{
		cfg:    cfg,
		runner: &execx.Runner{},
		store:  state.New(cfg.State.RunHistoryPath),
		log:    log.New(io.Discard, "", 0),
	}

	endpointChecks := 0
	dispatch := func(_ context.Context, _ time.Duration, name string, args ...string) (string, error) {
		joined := name + " " + strings.Join(args, " ")
		if strings.Contains(joined, "get endpoints victoria-metrics-single-server") {
			endpointChecks++
			if endpointChecks == 1 {
				return "", nil
			}
			return "10.42.0.10\n", nil
		}
		if strings.Contains(joined, "scale deployment victoria-metrics-single-server") {
			return "", errors.New(`Error from server (NotFound): deployments.apps "victoria-metrics-single-server" not found`)
		}
		if strings.Contains(joined, "scale statefulset victoria-metrics-single-server") {
			return "", nil
		}
		if strings.Contains(joined, "rollout status statefulset/victoria-metrics-single-server") {
			return "statefulset rolled out", nil
		}
		return "", nil
	}
	orch.runOverride = dispatch
	orch.runSensitiveOverride = dispatch

	if err := orch.waitForCriticalServiceEndpoints(context.Background()); err != nil {
		t.Fatalf("waitForCriticalServiceEndpoints failed: %v", err)
	}
	if endpointChecks < 2 {
		t.Fatalf("expected repeated endpoint checks, got %d", endpointChecks)
	}
}