feat(fleet): add service type enum and infra scope

Replace the free-form service_type string with a ServiceType enum (ccms,
ccmc, ccb, cces, ccsa, ccnc, ccem). The short code is the canonical value
stored in the DB, the config-tree directory name and a NATS discovery
subject token; Description() gives the human-readable name.

Add a second scope alongside per-node agents: ScopeInfra covers
cluster-independent monitoring infrastructure services, which are
identified by (hostname, service_type) and carry an empty cluster.
InfraRegistry is the ScopeInfra sibling of Registry and shares the same
table, state machine and config-revision handshake. It deliberately
exposes no StartSweep: MarkStale ages rows by last_heartbeat regardless
of scope, so a single sweep goroutine per process covers both scopes.

On the repository side ServiceDB gains a Scope column and the two
queries the discovery publisher and infra enumeration need, ListActive()
and ListByScope(). ResetConnection() now also resets the fleet
repository singleton so tests get a fresh handle.

Migration 13 is edited in place rather than superseded by a new
migration: the service table is unreleased, so no existing deployment
has it yet.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

internal/fleet/registry.go

@@ -21,6 +21,7 @@ import (
 	"encoding/hex"
 	"encoding/json"
 	"errors"
+	"fmt"
 	"time"
 
 	"github.com/ClusterCockpit/cc-backend/internal/repository"
@@ -36,7 +37,7 @@ var ErrUnknownInstance = errors.New("fleet: unknown or deregistered instance id"
 type Service struct {
 	Cluster        string
 	Hostname       string
-	ServiceType    string
+	ServiceType    ServiceType
 	InstanceID     string
 	State          string
 	RegisteredAt   time.Time
@@ -50,7 +51,7 @@ type Service struct {
 type RegistrationRequest struct {
 	Cluster     string
 	Hostname    string
-	ServiceType string
+	ServiceType ServiceType
 	MetaData    map[string]string
 }
 
@@ -85,8 +86,11 @@ func NewRegistry(staleAfter time.Duration) *Registry {
 // instance_id plus the config_revision it currently has on record (0 for a
 // never-before-seen service, so the caller knows to pull its initial config).
 func (r *Registry) Register(req RegistrationRequest) (*Registration, error) {
-	if req.Cluster == "" || req.Hostname == "" || req.ServiceType == "" {
-		return nil, errors.New("fleet: cluster, hostname and service_type are required")
+	if req.Cluster == "" || req.Hostname == "" {
+		return nil, errors.New("fleet: cluster and hostname are required")
+	}
+	if !req.ServiceType.Valid() {
+		return nil, fmt.Errorf("fleet: unknown service_type %q", req.ServiceType)
 	}
 
 	instanceID, err := generateInstanceID()
@@ -102,8 +106,9 @@ func (r *Registry) Register(req RegistrationRequest) (*Registration, error) {
 	svc := &repository.ServiceDB{
 		Cluster:      req.Cluster,
 		Hostname:     req.Hostname,
-		ServiceType:  req.ServiceType,
+		ServiceType:  string(req.ServiceType),
 		InstanceID:   instanceID,
+		Scope:        ScopeCluster,
 		RegisteredAt: time.Now().Unix(),
 		MetaData:     metaJSON,
 	}
@@ -239,7 +244,7 @@ func toService(row *repository.ServiceDB) (*Service, error) {
 	svc := &Service{
 		Cluster:        row.Cluster,
 		Hostname:       row.Hostname,
-		ServiceType:    row.ServiceType,
+		ServiceType:    ServiceType(row.ServiceType),
 		InstanceID:     row.InstanceID,
 		State:          row.State,
 		RegisteredAt:   time.Unix(row.RegisteredAt, 0),

internal/fleet/services.go

@@ -0,0 +1,160 @@
+// Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
+// All rights reserved. This file is part of cc-backend.
+// Use of this source code is governed by a MIT-style
+// license that can be found in the LICENSE file.
+
+package fleet
+
+import (
+	"errors"
+	"fmt"
+	"time"
+
+	"github.com/ClusterCockpit/cc-backend/internal/repository"
+	cclog "github.com/ClusterCockpit/cc-lib/v2/ccLogger"
+)
+
+// Scope values distinguish the two kinds of fleet member persisted in the
+// shared "service" table:
+//
+//   - ScopeCluster: per-node agents, identified by (cluster, hostname,
+//     service_type). This is the default and what Registry issues.
+//   - ScopeInfra: cluster-independent monitoring infrastructure services
+//     (metric stores, collectors, gateways). They are not tied to a single
+//     cluster, so their cluster column is empty and identity is
+//     (hostname, service_type). InfraRegistry issues these.
+const (
+	ScopeCluster = "cluster"
+	ScopeInfra   = "infra"
+)
+
+// InfraRegistrationRequest is what a cluster-independent monitoring
+// infrastructure service posts to the REST registration endpoint. Unlike
+// RegistrationRequest it has no Cluster field: these services span clusters.
+type InfraRegistrationRequest struct {
+	Hostname    string
+	ServiceType ServiceType
+	MetaData    map[string]string
+}
+
+// InfraRegistry is the business-logic layer for cluster-independent monitoring
+// infrastructure services. It is the ScopeInfra sibling of Registry and shares
+// the same FleetRepository and "service" table; identity issuance, the
+// pending/active/stale/deregistered state machine and the config-revision
+// handshake are identical. The only differences are that registration takes no
+// cluster and discovery is by scope rather than by cluster.
+//
+// Staleness: MarkStale (driven by StartSweep) ages any service by its
+// last_heartbeat regardless of scope, so infra rows are already covered by the
+// sweep. Do NOT start a second sweep goroutine here — exactly one StartSweep
+// across the whole process is sufficient. InfraRegistry therefore deliberately
+// exposes no StartSweep.
+type InfraRegistry struct {
+	repo *repository.FleetRepository
+}
+
+// NewInfraRegistry returns an InfraRegistry backed by the singleton
+// FleetRepository.
+func NewInfraRegistry() *InfraRegistry {
+	return &InfraRegistry{
+		repo: repository.GetFleetRepository(),
+	}
+}
+
+// Register upserts a cluster-independent service's identity by
+// (hostname, service_type) and returns a freshly issued instance_id plus the
+// config_revision currently on record (0 for a never-before-seen service, so
+// the caller knows to pull its initial config).
+func (r *InfraRegistry) Register(req InfraRegistrationRequest) (*Registration, error) {
+	if req.Hostname == "" {
+		return nil, errors.New("fleet: hostname is required")
+	}
+	if !req.ServiceType.Valid() {
+		return nil, fmt.Errorf("fleet: unknown service_type %q", req.ServiceType)
+	}
+
+	instanceID, err := generateInstanceID()
+	if err != nil {
+		return nil, err
+	}
+
+	metaJSON, err := marshalMeta(req.MetaData)
+	if err != nil {
+		return nil, err
+	}
+
+	svc := &repository.ServiceDB{
+		Cluster:      "",
+		Hostname:     req.Hostname,
+		ServiceType:  string(req.ServiceType),
+		InstanceID:   instanceID,
+		Scope:        ScopeInfra,
+		RegisteredAt: time.Now().Unix(),
+		MetaData:     metaJSON,
+	}
+
+	id, err := r.repo.RegisterService(svc)
+	if err != nil {
+		return nil, err
+	}
+
+	stored, err := r.repo.GetByID(id)
+	if err != nil {
+		return nil, err
+	}
+
+	cclog.Infof("fleet: registered infra %s/%s as instance '%s'", req.Hostname, req.ServiceType, instanceID)
+	return &Registration{InstanceID: instanceID, ConfigRevision: stored.ConfigRevision}, nil
+}
+
+// Heartbeat refreshes liveness for an already-registered infra instance. It is
+// a no-op for unknown or deregistered instance IDs — see the package doc for
+// why that must hold when this is reachable over NATS.
+func (r *InfraRegistry) Heartbeat(instanceID string, at time.Time) error {
+	affected, err := r.repo.Heartbeat(instanceID, at.Unix())
+	if err != nil {
+		return err
+	}
+	if affected == 0 {
+		return ErrUnknownInstance
+	}
+	return nil
+}
+
+// Deregister marks an infra instance as deregistered. Idempotent.
+func (r *InfraRegistry) Deregister(instanceID string) error {
+	return r.repo.Deregister(instanceID)
+}
+
+// AckConfig records that instanceID has pulled configRevision. Called by the
+// REST config-pull handler after it serves the config payload.
+func (r *InfraRegistry) AckConfig(instanceID string, configRevision int64) error {
+	return r.repo.SetConfigRevision(instanceID, configRevision)
+}
+
+// Get returns a single infra service by instance_id.
+func (r *InfraRegistry) Get(instanceID string) (*Service, error) {
+	svc, err := r.repo.GetByInstanceID(instanceID)
+	if err != nil {
+		return nil, err
+	}
+	return toService(svc)
+}
+
+// List returns all registered cluster-independent infrastructure services.
+func (r *InfraRegistry) List() ([]*Service, error) {
+	rows, err := r.repo.ListByScope(ScopeInfra)
+	if err != nil {
+		return nil, err
+	}
+
+	services := make([]*Service, 0, len(rows))
+	for _, row := range rows {
+		svc, err := toService(row)
+		if err != nil {
+			return nil, err
+		}
+		services = append(services, svc)
+	}
+	return services, nil
+}

internal/fleet/services_test.go

@@ -0,0 +1,127 @@
+// Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
+// All rights reserved. This file is part of cc-backend.
+// Use of this source code is governed by a MIT-style
+// license that can be found in the LICENSE file.
+
+package fleet
+
+import (
+	"errors"
+	"path/filepath"
+	"testing"
+	"time"
+
+	"github.com/ClusterCockpit/cc-backend/internal/repository"
+	cclog "github.com/ClusterCockpit/cc-lib/v2/ccLogger"
+	_ "github.com/mattn/go-sqlite3"
+)
+
+// setupDB migrates a fresh temp database to the current schema and wires the
+// repository singletons to it, isolated per test.
+func setupDB(t *testing.T) {
+	t.Helper()
+	cclog.Init("warn", true)
+
+	dbfile := filepath.Join(t.TempDir(), "fleet.db")
+	if err := repository.ResetConnection(); err != nil {
+		t.Fatal(err)
+	}
+	if err := repository.MigrateDB(dbfile); err != nil {
+		t.Fatal(err)
+	}
+	repository.Connect(dbfile)
+	t.Cleanup(func() { repository.ResetConnection() })
+}
+
+func TestInfraRegistry(t *testing.T) {
+	setupDB(t)
+	reg := NewInfraRegistry()
+
+	t.Run("register requires hostname and a valid service_type", func(t *testing.T) {
+		if _, err := reg.Register(InfraRegistrationRequest{ServiceType: ServiceTypeMetricStore}); err == nil {
+			t.Fatal("expected error for missing hostname")
+		}
+		if _, err := reg.Register(InfraRegistrationRequest{Hostname: "ms01"}); err == nil {
+			t.Fatal("expected error for missing service_type")
+		}
+		if _, err := reg.Register(InfraRegistrationRequest{Hostname: "ms01", ServiceType: "bogus"}); err == nil {
+			t.Fatal("expected error for unknown service_type")
+		}
+	})
+
+	var instanceID string
+	t.Run("register issues identity with revision 0", func(t *testing.T) {
+		r, err := reg.Register(InfraRegistrationRequest{
+			Hostname: "ms01", ServiceType: ServiceTypeMetricStore,
+			MetaData: map[string]string{"version": "1.2.3"},
+		})
+		if err != nil {
+			t.Fatal(err)
+		}
+		if r.InstanceID == "" || r.ConfigRevision != 0 {
+			t.Fatalf("unexpected registration: %+v", r)
+		}
+		instanceID = r.InstanceID
+	})
+
+	t.Run("get returns infra scope and empty cluster", func(t *testing.T) {
+		svc, err := reg.Get(instanceID)
+		if err != nil {
+			t.Fatal(err)
+		}
+		if svc.Cluster != "" || svc.ServiceType != ServiceTypeMetricStore || svc.State != "pending" {
+			t.Fatalf("unexpected service: %+v", svc)
+		}
+		if svc.MetaData["version"] != "1.2.3" {
+			t.Fatalf("metadata not round-tripped: %+v", svc.MetaData)
+		}
+	})
+
+	t.Run("heartbeat activates known, rejects unknown", func(t *testing.T) {
+		if err := reg.Heartbeat(instanceID, time.Unix(5000, 0)); err != nil {
+			t.Fatal(err)
+		}
+		svc, _ := reg.Get(instanceID)
+		if svc.State != "active" || svc.LastHeartbeat == nil {
+			t.Fatalf("heartbeat did not activate: %+v", svc)
+		}
+		if err := reg.Heartbeat("unknown", time.Unix(5000, 0)); !errors.Is(err, ErrUnknownInstance) {
+			t.Fatalf("want ErrUnknownInstance, got %v", err)
+		}
+	})
+
+	t.Run("list returns only infra services", func(t *testing.T) {
+		// A cluster-scope agent must not appear in the infra listing.
+		cReg := NewRegistry(time.Hour)
+		if _, err := cReg.Register(RegistrationRequest{
+			Cluster: "fritz", Hostname: "node01", ServiceType: ServiceTypeCollector,
+		}); err != nil {
+			t.Fatal(err)
+		}
+
+		list, err := reg.List()
+		if err != nil {
+			t.Fatal(err)
+		}
+		if len(list) != 1 || list[0].ServiceType != ServiceTypeMetricStore {
+			t.Fatalf("unexpected infra list: %+v", list)
+		}
+	})
+
+	t.Run("ack config records revision, deregister is terminal", func(t *testing.T) {
+		if err := reg.AckConfig(instanceID, 99); err != nil {
+			t.Fatal(err)
+		}
+		svc, _ := reg.Get(instanceID)
+		if svc.ConfigRevision != 99 {
+			t.Fatalf("want revision 99, got %d", svc.ConfigRevision)
+		}
+
+		if err := reg.Deregister(instanceID); err != nil {
+			t.Fatal(err)
+		}
+		if err := reg.Heartbeat(instanceID, time.Unix(6000, 0)); !errors.Is(err, ErrUnknownInstance) {
+			t.Fatalf("heartbeat after deregister should fail: %v", err)
+		}
+	})
+}

internal/fleet/servicetype.go

@@ -0,0 +1,85 @@
+// Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
+// All rights reserved. This file is part of cc-backend.
+// Use of this source code is governed by a MIT-style
+// license that can be found in the LICENSE file.
+
+package fleet
+
+// ServiceType is the kind of cc-* service a fleet member is. It is a
+// string-backed enum (mirroring schema.MonitoringState and the local
+// ScopeCluster/ScopeInfra constants): the short code is the canonical value
+// stored in the DB, used as the config-tree directory name, and used as a NATS
+// discovery-subject token. Description() gives the human-readable name.
+type ServiceType string
+
+const (
+	ServiceTypeMetricStore    ServiceType = "ccms" // cc-metric-store
+	ServiceTypeCollector      ServiceType = "ccmc" // cc-metric-collector
+	ServiceTypeBackend        ServiceType = "ccb"  // cc-backend
+	ServiceTypeEventStore     ServiceType = "cces" // cc-event-store
+	ServiceTypeSlurmAdapter   ServiceType = "ccsa" // cc-slurm-adapter
+	ServiceTypeNodeController ServiceType = "ccnc" // cc-node-controller
+	ServiceTypeEnergyManager  ServiceType = "ccem" // cc-energy-manager
+)
+
+// AllServiceTypes lists every valid service type. Order is stable so callers
+// that iterate (e.g. the discovery publisher) produce deterministic output.
+var AllServiceTypes = []ServiceType{
+	ServiceTypeMetricStore,
+	ServiceTypeCollector,
+	ServiceTypeBackend,
+	ServiceTypeEventStore,
+	ServiceTypeSlurmAdapter,
+	ServiceTypeNodeController,
+	ServiceTypeEnergyManager,
+}
+
+var serviceTypeDescriptions = map[ServiceType]string{
+	ServiceTypeMetricStore:    "cc-metric-store",
+	ServiceTypeCollector:      "cc-metric-collector",
+	ServiceTypeBackend:        "cc-backend",
+	ServiceTypeEventStore:     "cc-event-store",
+	ServiceTypeSlurmAdapter:   "cc-slurm-adapter",
+	ServiceTypeNodeController: "cc-node-controller",
+	ServiceTypeEnergyManager:  "cc-energy-manager",
+}
+
+// Valid reports whether t is a known service type.
+func (t ServiceType) Valid() bool {
+	_, ok := serviceTypeDescriptions[t]
+	return ok
+}
+
+// Description returns the full human-readable service name (e.g. "cc-metric-store"),
+// or the raw code if unknown.
+func (t ServiceType) Description() string {
+	if d, ok := serviceTypeDescriptions[t]; ok {
+		return d
+	}
+	return string(t)
+}
+
+// relevantProviders is the single source of truth for which provider service
+// types each consumer type needs to discover. Keep it here and edit in one
+// place.
+//
+// Confirmed universal edge: every service must reach cc-backend (ccb) to
+// register and pull its config, so ccb is relevant to all of them. Richer
+// peer edges (a collector wanting the metric store, an energy manager wanting
+// the node controller, …) are left commented for an operator to enable once the
+// concrete topology is settled — they are intentionally not assumed here.
+var relevantProviders = map[ServiceType][]ServiceType{
+	ServiceTypeMetricStore:    {ServiceTypeBackend},
+	ServiceTypeCollector:      {ServiceTypeBackend /*, ServiceTypeMetricStore */},
+	ServiceTypeEventStore:     {ServiceTypeBackend},
+	ServiceTypeSlurmAdapter:   {ServiceTypeBackend},
+	ServiceTypeNodeController: {ServiceTypeBackend},
+	ServiceTypeEnergyManager:  {ServiceTypeBackend /*, ServiceTypeMetricStore, ServiceTypeNodeController */},
+	ServiceTypeBackend:        {}, // ccb discovers no peers by default
+}
+
+// RelevantProviders returns the provider service types that consumer should
+// discover. The returned slice must not be mutated by callers.
+func RelevantProviders(consumer ServiceType) []ServiceType {
+	return relevantProviders[consumer]
+}

internal/fleet/servicetype_test.go

@@ -0,0 +1,54 @@
+// Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
+// All rights reserved. This file is part of cc-backend.
+// Use of this source code is governed by a MIT-style
+// license that can be found in the LICENSE file.
+
+package fleet
+
+import "testing"
+
+func TestServiceTypeValid(t *testing.T) {
+	for _, st := range AllServiceTypes {
+		if !st.Valid() {
+			t.Errorf("%q should be valid", st)
+		}
+	}
+	for _, bad := range []ServiceType{"", "bogus", "metric-store", "CCMS"} {
+		if bad.Valid() {
+			t.Errorf("%q should be invalid", bad)
+		}
+	}
+}
+
+func TestServiceTypeDescription(t *testing.T) {
+	if got := ServiceTypeMetricStore.Description(); got != "cc-metric-store" {
+		t.Errorf("ccms description = %q, want cc-metric-store", got)
+	}
+	// Unknown code falls back to the raw string.
+	if got := ServiceType("bogus").Description(); got != "bogus" {
+		t.Errorf("unknown description = %q, want bogus", got)
+	}
+}
+
+func TestRelevantProviders(t *testing.T) {
+	// Confirmed universal edge: every non-ccb service must discover ccb.
+	for _, st := range AllServiceTypes {
+		if st == ServiceTypeBackend {
+			continue
+		}
+		rel := RelevantProviders(st)
+		found := false
+		for _, p := range rel {
+			if p == ServiceTypeBackend {
+				found = true
+			}
+		}
+		if !found {
+			t.Errorf("%q must have ccb as a relevant provider, got %v", st, rel)
+		}
+	}
+	// ccb discovers no peers by default.
+	if rel := RelevantProviders(ServiceTypeBackend); len(rel) != 0 {
+		t.Errorf("ccb should have no relevant providers, got %v", rel)
+	}
+}