tmp

Signed-off-by: jkoberg <jkoberg@owncloud.com>

ocis/pkg/runtime/service/service.go

@@ -69,6 +69,9 @@ import (
 var (
 	// runset keeps track of which services to start supervised.
 	runset map[string]struct{}
+
+	// wait funcs run after the service group has been started.
+	_waitFuncs = []func(*ociscfg.Config) error{pingNats, pingGateway, nil, wait(time.Second), nil}
 )
 
 type serviceFuncMap map[string]func(*ociscfg.Config) suture.Service
@@ -107,7 +110,7 @@ func NewService(options ...Option) (*Service, error) {
 	globalCtx, cancelGlobal := context.WithCancel(context.Background())
 
 	s := &Service{
-		Services:   make([]serviceFuncMap, 5),
+		Services:   make([]serviceFuncMap, len(_waitFuncs)),
 		Additional: make(serviceFuncMap),
 		Log:        l,
 
@@ -123,20 +126,25 @@ func NewService(options ...Option) (*Service, error) {
 			s.Services[priority] = make(serviceFuncMap)
 		}
 		s.Services[priority][name] = NewSutureServiceBuilder(exec)
-		// s.ServicesRegistry[name] = NewSutureServiceBuilder(exec)
 	}
 
-	// start nats first - it is used as service registry
+	// nats is in priority group 0. It needs to start before all other services
 	reg(0, opts.Config.Nats.Service.Name, func(ctx context.Context, cfg *ociscfg.Config) error {
 		cfg.Nats.Context = ctx
 		cfg.Nats.Commons = cfg.Commons
 		return nats.Execute(cfg.Nats)
 	})
+
+	// gateway is in priority group 1. It needs to start before the reva services
 	reg(1, opts.Config.Gateway.Service.Name, func(ctx context.Context, cfg *ociscfg.Config) error {
 		cfg.Gateway.Context = ctx
 		cfg.Gateway.Commons = cfg.Commons
 		return gateway.Execute(cfg.Gateway)
 	})
+
+	// priority group 2 is empty for now
+
+	// most services are in priority group 3
 	reg(3, opts.Config.AppProvider.Service.Name, func(ctx context.Context, cfg *ociscfg.Config) error {
 		cfg.AppProvider.Context = ctx
 		cfg.AppProvider.Commons = cfg.Commons
@@ -277,38 +285,39 @@ func NewService(options ...Option) (*Service, error) {
 		cfg.Webfinger.Commons = cfg.Commons
 		return webfinger.Execute(cfg.Webfinger)
 	})
-
-	reg(4, opts.Config.Frontend.Service.Name, func(ctx context.Context, cfg *ociscfg.Config) error {
-		cfg.Frontend.Context = ctx
-		cfg.Frontend.Commons = cfg.Commons
-		return frontend.Execute(cfg.Frontend)
-	})
-	reg(4, opts.Config.IDP.Service.Name, func(ctx context.Context, cfg *ociscfg.Config) error {
+	reg(3, opts.Config.IDP.Service.Name, func(ctx context.Context, cfg *ociscfg.Config) error {
 		cfg.IDP.Context = ctx
 		cfg.IDP.Commons = cfg.Commons
 		return idp.Execute(cfg.IDP)
 	})
-	reg(4, opts.Config.Proxy.Service.Name, func(ctx context.Context, cfg *ociscfg.Config) error {
+	reg(3, opts.Config.Proxy.Service.Name, func(ctx context.Context, cfg *ociscfg.Config) error {
 		cfg.Proxy.Context = ctx
 		cfg.Proxy.Commons = cfg.Commons
 		return proxy.Execute(cfg.Proxy)
 	})
-	reg(4, opts.Config.Sharing.Service.Name, func(ctx context.Context, cfg *ociscfg.Config) error {
+	reg(3, opts.Config.Sharing.Service.Name, func(ctx context.Context, cfg *ociscfg.Config) error {
 		cfg.Sharing.Context = ctx
 		cfg.Sharing.Commons = cfg.Commons
 		return sharing.Execute(cfg.Sharing)
 	})
-	reg(4, opts.Config.SSE.Service.Name, func(ctx context.Context, cfg *ociscfg.Config) error {
+	reg(3, opts.Config.SSE.Service.Name, func(ctx context.Context, cfg *ociscfg.Config) error {
 		cfg.SSE.Context = ctx
 		cfg.SSE.Commons = cfg.Commons
 		return sse.Execute(cfg.SSE)
 	})
-	reg(4, opts.Config.OCM.Service.Name, func(ctx context.Context, cfg *ociscfg.Config) error {
+	reg(3, opts.Config.OCM.Service.Name, func(ctx context.Context, cfg *ociscfg.Config) error {
 		cfg.OCM.Context = ctx
 		cfg.OCM.Commons = cfg.Commons
 		return ocm.Execute(cfg.OCM)
 	})
 
+	// out of some unknown reason ci gets angry when frontend service starts in priority group 3
+	// this is not reproducible locally, it can be start when nats and gateway are already running
+	reg(4, opts.Config.Frontend.Service.Name, func(ctx context.Context, cfg *ociscfg.Config) error {
+		cfg.Frontend.Context = ctx
+		cfg.Frontend.Commons = cfg.Commons
+		return frontend.Execute(cfg.Frontend)
+	})
 	// populate optional services
 	areg := func(name string, exec func(context.Context, *ociscfg.Config) error) {
 		s.Additional[name] = NewSutureServiceBuilder(exec)
@@ -409,11 +418,10 @@ func Start(o ...Option) error {
 	// trap will block on halt channel for interruptions.
 	go trap(s, halt)
 
-	waiters := []func(*ociscfg.Config) error{pingNats, pingGateway, nil, nil, nil}
 	for i, service := range s.Services {
 		scheduleServiceTokens(s, service)
-		if waiters[i] != nil {
-			if err := waiters[i](s.cfg); err != nil {
+		if _waitFuncs[i] != nil {
+			if err := _waitFuncs[i](s.cfg); err != nil {
 				s.Log.Fatal().Err(err).Msg("could not connect to service")
 			}
 		}
@@ -513,11 +521,14 @@ func pingNats(cfg *ociscfg.Config) error {
 	return err
 }
 
-func pingGateway(cfg *ociscfg.Config) error {
-	// we need to get a gateway address from somewhere. We can use any one.
-	// address := cfg.Graph.Reva.Address
-	b := backoff.NewExponentialBackOff()
+func pingGateway(_ *ociscfg.Config) error {
+	// init grpc connection
 	_, err := ogrpc.NewClient()
+	if err != nil {
+		return err
+	}
+
+	b := backoff.NewExponentialBackOff()
 	o := func() error {
 		n := b.NextBackOff()
 		_, err := pool.GetGatewayServiceClient("com.owncloud.api.gateway")
@@ -530,3 +541,10 @@ func pingGateway(cfg *ociscfg.Config) error {
 	err = backoff.Retry(o, b)
 	return err
 }
+
+func wait(d time.Duration) func(cfg *ociscfg.Config) error {
+	return func(cfg *ociscfg.Config) error {
+		time.Sleep(d)
+		return nil
+	}
+}