PMM-15360: Gate OpenManager's service on the switch

Implements Enabled() bool on om.Service, so every /v1/om/* RPC and the
scheduled topology collection refuse while OpenManager is off, via the
same generic gRPC-service-enabled interceptor BackupService and the
other preview features already use -- no proto or per-method changes
needed.

Adds IsAvailable(), a read against SEP's om_inventory app (the same
call the scheduled collection already makes), and wires it into
ChangeSettings so turning OpenManager on is refused unless the app is
actually reachable -- rather than accepting a setting that would do
nothing. Turning it on also kicks an immediate topology collection
(extracted into its own method: golangci-lint's gocognit check flags
ChangeSettings once this is inlined alongside Advisors' and
telemetry's own transition handling).

The generated swagger client's field names are OMEnabled/EnableOM here
(not OmEnabled/EnableOm, as in the settings-flag commit this stacks
on): this branch's api/Makefile, via PMM-15326-om-api already merged
into PMM-15326-om-backend, registers "om" as an additional-initialism
for the swagger client generator. That commit hasn't reached the epic
yet, so the settings-flag PR's own branch (based on the epic directly)
correctly generates the un-capitalized form; this is the make gen
output for this branch specifically.

Does not touch SEP's om_inventory app in any way -- turning the switch
off does not stop its periodic sweep. That's deferred to a follow-up.

Signed-off-by: Pawel Lebioda <pawel.lebioda@percona.com>

Author: plebioda

managed/cmd/pmm-managed/main.go

@@ -1065,6 +1065,15 @@ func main() { //nolint:gocognit,maintidx,cyclop
 		VMURL:         *victoriaMetricsURLF,
 	})
 
+	// Where SEP is, optional. Empty means OM builds its document from PMM's own inventory
+	// and metrics alone and records the probe source as disabled.
+	//
+	// Constructed here, ahead of serverParams below, so server.Server can hold it for the
+	// OpenManager enable/disable switch (Enabled gate, IsAvailable check).
+	omService := om.New(db, v1.NewAPI(vmClient), haService, logrus.WithField("component", "om"))
+	omService.WithProbeSource(*sepURLF, *sepTokenF)
+	prom.MustRegister(om.NewMetricsCollector(omService))
+
 	serverParams := &server.Params{
 		DB:                   db,
 		VMDB:                 vmdb,
@@ -1081,6 +1090,7 @@ func main() { //nolint:gocognit,maintidx,cyclop
 		HAService:            haService,
 		Nomad:                nomad,
 		QANClient:            qanClient,
+		OmService:            omService,
 	}
 
 	server, err := server.NewServer(serverParams)
@@ -1197,12 +1207,6 @@ func main() { //nolint:gocognit,maintidx,cyclop
 		return nil
 	}))
 
-	// Where SEP is, optional. Empty means OM builds its document from PMM's own inventory
-	// and metrics alone and records the probe source as disabled.
-	omService := om.New(db, v1.NewAPI(vmClient), haService, logrus.WithField("component", "om"))
-	omService.WithProbeSource(*sepURLF, *sepTokenF)
-	prom.MustRegister(om.NewMetricsCollector(omService))
-
 	// Leader-only, like every other periodic writer here. A collection persists a run and
 	// its snapshot and then prunes the shared history, so running it on every node of an
 	// HA cluster would have each node writing runs and pruning the others' -- and the

managed/services/om/service.go

@@ -157,6 +157,33 @@ func (s *Service) WithProbeSource(sepURL, token string) *Service {
 	return s
 }
 
+// Enabled returns true if OpenManager is enabled, so every /v1/om/* RPC and the
+// scheduled collection in Run refuse while it is off, via the same generic
+// gRPC-service-enabled interceptor BackupService and the other preview features use.
+func (s *Service) Enabled() bool {
+	settings, err := models.GetSettings(s.db)
+	if err != nil {
+		s.l.WithError(err).Error("can't get settings")
+		return false
+	}
+	return settings.IsOMEnabled()
+}
+
+// IsAvailable reports whether SEP's om_inventory app is configured and reachable.
+//
+// Used to gate turning OpenManager on: an admin flipping the switch with no inventory
+// app to talk to would enable a UI backed by a source that can never answer, with no
+// way to tell "off" from "broken" apart from reading logs. It does not drive anything
+// on SEP's side -- this is the same read every scheduled collection already performs
+// via probeSource.collect, just run once up front rather than waited out.
+func (s *Service) IsAvailable(ctx context.Context) bool {
+	if s.probe == nil || s.probe.app.client == nil {
+		return false
+	}
+	_, err := s.probe.fetch(ctx)
+	return err == nil
+}
+
 // GetTopology returns the whole MongoDB estate as one document.
 //
 // A pure read path: memory, then the stored snapshot, never a collection. Collection is
@@ -261,6 +288,9 @@ func (s *Service) Run(ctx context.Context) {
 		case <-ctx.Done():
 			return
 		case <-ticker.C:
+			if !s.Enabled() {
+				continue
+			}
 			_, err := s.discover(ctx)
 			if err != nil && ctx.Err() == nil {
 				s.l.Warnf("scheduled collection failed: %s", err)

managed/services/server/deps.go

@@ -20,6 +20,7 @@ import (
 	"net/url"
 	"time"
 
+	omv1 "github.com/percona/pmm/api/om/v1"
 	serverv1 "github.com/percona/pmm/api/server/v1"
 	"github.com/percona/pmm/managed/models"
 )
@@ -117,3 +118,12 @@ type victoriaMetricsParams interface {
 type nomadService interface {
 	UpdateConfiguration(settings *models.Settings) error
 }
+
+// omService is a subset of methods of om.Service used by this package.
+// We use it instead of the real type to avoid a dependency cycle.
+type omService interface {
+	// IsAvailable reports whether SEP's OpenManager Inventory app is configured and
+	// reachable, gating whether OpenManager may be enabled.
+	IsAvailable(ctx context.Context) bool
+	TriggerTopologyCollection(ctx context.Context, req *omv1.TriggerTopologyCollectionRequest) (*omv1.TriggerTopologyCollectionResponse, error)
+}

managed/services/server/server.go

@@ -38,6 +38,7 @@ import (
 	"google.golang.org/protobuf/types/known/timestamppb"
 	"gopkg.in/reform.v1"
 
+	omv1 "github.com/percona/pmm/api/om/v1"
 	serverv1 "github.com/percona/pmm/api/server/v1"
 	"github.com/percona/pmm/managed/models"
 	"github.com/percona/pmm/managed/utils/distribution"
@@ -63,6 +64,7 @@ type Server struct {
 	haService            haService
 	updater              *Updater
 	nomad                nomadService
+	omService            omService
 
 	l *logrus.Entry
 
@@ -89,6 +91,7 @@ type Params struct {
 	Dus                  *distribution.Service
 	HAService            haService
 	Nomad                nomadService
+	OmService            omService
 }
 
 // NewServer returns new server for Server service.
@@ -114,6 +117,7 @@ func NewServer(params *Params) (*Server, error) {
 		updater:              params.Updater,
 		haService:            params.HAService,
 		nomad:                params.Nomad,
+		omService:            params.OmService,
 		l:                    logrus.WithField("component", "server"),
 		envSettings:          &models.ChangeSettingsParams{},
 	}
@@ -470,6 +474,14 @@ func (s *Server) validateChangeSettingsRequest(ctx context.Context, req *serverv
 		return status.Error(codes.FailedPrecondition, "Azure Discover is configured via PMM_ENABLE_AZURE_DISCOVER environment variable.")
 	}
 
+	if req.EnableOm != nil && s.envSettings.EnableOM != nil && *req.EnableOm != *s.envSettings.EnableOM {
+		return status.Error(codes.FailedPrecondition, "OpenManager is configured via PMM_ENABLE_OM environment variable.")
+	}
+
+	if req.EnableOm != nil && *req.EnableOm && s.omService != nil && !s.omService.IsAvailable(ctx) {
+		return status.Error(codes.FailedPrecondition, "OpenManager cannot be enabled: the OpenManager Inventory app is not available in SEP.")
+	}
+
 	if !canUpdateDurationSetting(metricsRes.GetHr().AsDuration(), s.envSettings.MetricsResolutions.HR) {
 		return status.Error(
 			codes.FailedPrecondition,
@@ -522,6 +534,7 @@ func (s *Server) ChangeSettings(ctx context.Context, req *serverv1.ChangeSetting
 			EnableBackupManagement: req.EnableBackupManagement,
 			EnableAccessControl:    req.EnableAccessControl,
 			EnableInternalPgQAN:    req.EnableInternalPgQan,
+			EnableOM:               req.EnableOm,
 			AdvisorsRunInterval: models.AdvisorsRunIntervals{
 				RareInterval:     advisorsRunInterval.GetRareInterval().AsDuration(),
 				StandardInterval: advisorsRunInterval.GetStandardInterval().AsDuration(),
@@ -613,11 +626,25 @@ func (s *Server) ChangeSettings(ctx context.Context, req *serverv1.ChangeSetting
 		}
 	}
 
+	s.triggerOMCollectionIfJustEnabled(ctx, oldSettings, newSettings)
+
 	return &serverv1.ChangeSettingsResponse{
 		Settings: s.convertSettings(newSettings, disableInternalPgQan),
 	}, nil
 }
 
+// triggerOMCollectionIfJustEnabled kicks a topology collection so OpenManager's page is
+// not empty on first view, instead of waiting out the next scheduled tick.
+func (s *Server) triggerOMCollectionIfJustEnabled(ctx context.Context, oldSettings, newSettings *models.Settings) {
+	if oldSettings.IsOMEnabled() || !newSettings.IsOMEnabled() || s.omService == nil {
+		return
+	}
+	_, err := s.omService.TriggerTopologyCollection(ctx, &omv1.TriggerTopologyCollectionRequest{})
+	if err != nil {
+		s.l.Warnf("failed to trigger OpenManager topology collection after enabling: %s", err)
+	}
+}
+
 func (s *Server) getInternalPgQANAgent(q *reform.Querier) (*models.Agent, error) {
 	agents, err := models.FindAgents(q, models.AgentFilters{
 		PMMAgentID: models.PMMServerAgentID,