Merge pull request #2475 from keboola/petr-hosek-PAT-1091

hosekpeter · web-flow · commit 42a560cf6fd6 · 2025-12-01T10:37:23.000+01:00
Add restart disabled handling and wakeup overrides
diff --git a/.github/workflows/release-service-apps-proxy.yml b/.github/workflows/release-service-apps-proxy.yml
@@ -6,6 +6,7 @@ on:
     tags:
       - 'production-apps-proxy-v**'
       - 'dev-apps-proxy-v**'
+      - 'canary-*-apps-proxy-v**'
 
 env:
   # DockerHub login
diff --git a/.github/workflows/release-service-stream.yml b/.github/workflows/release-service-stream.yml
@@ -6,6 +6,7 @@ on:
     tags:
       - 'production-stream-v**'
       - 'dev-stream-v**'
+      - 'canary-*-stream-v**'
 
 env:
   # DockerHub login
diff --git a/.github/workflows/release-service-templates.yml b/.github/workflows/release-service-templates.yml
@@ -6,6 +6,7 @@ on:
     tags:
       - 'production-templates-api-v**'
       - 'dev-templates-api-v**'
+      - 'canary-*-templates-api-v**'
 
 env:
   # DockerHub login
diff --git a/internal/pkg/service/appsproxy/dataapps/api/error.go b/internal/pkg/service/appsproxy/dataapps/api/error.go
@@ -5,10 +5,15 @@ import (
 	"net/http"
 )
 
+const (
+	ContextCodeRestartDisabled = "apps.restartDisabled"
+)
+
 // Error represents the structure of Sandboxes API error.
 type Error struct {
-	Message     string `json:"error"`
-	ExceptionID string `json:"exceptionId"`
+	Message     string         `json:"error"`
+	ExceptionID string         `json:"exceptionId"`
+	Context     map[string]any `json:"context,omitempty"`
 	request     *http.Request
 	response    *http.Response
 }
@@ -32,6 +37,14 @@ func (e *Error) ErrorExceptionID() string {
 	return e.ExceptionID
 }
 
+func (e *Error) HasRestartDisabled() bool {
+	if e.Context == nil {
+		return false
+	}
+	contextCode, ok := e.Context["code"].(string)
+	return ok && contextCode == ContextCodeRestartDisabled
+}
+
 // StatusCode returns HTTP status code.
 func (e *Error) StatusCode() int {
 	return e.response.StatusCode
diff --git a/internal/pkg/service/appsproxy/dataapps/api/wakeup.go b/internal/pkg/service/appsproxy/dataapps/api/wakeup.go
@@ -21,7 +21,7 @@ func (a *API) WakeupApp(appID AppID) request.APIRequest[request.NoResult] {
 				attribute.Int(attrSandboxesServiceStatusCode, response.StatusCode()),
 			}
 			span.SetAttributes(attrs...)
-			return nil
+			return err
 		}).
 		WithPatch("apps/{appId}").
 		AndPathParam("appId", appID.String()).
diff --git a/internal/pkg/service/appsproxy/dataapps/wakeup/wakeup.go b/internal/pkg/service/appsproxy/dataapps/wakeup/wakeup.go
@@ -12,6 +12,7 @@ import (
 	"github.com/keboola/keboola-as-code/internal/pkg/log"
 	"github.com/keboola/keboola-as-code/internal/pkg/service/appsproxy/dataapps/api"
 	"github.com/keboola/keboola-as-code/internal/pkg/service/appsproxy/syncmap"
+	"github.com/keboola/keboola-as-code/internal/pkg/utils/errors"
 )
 
 // Interval sets how often the proxy sends wakeup request to sandboxes service.
@@ -72,6 +73,15 @@ func (l *Manager) Wakeup(ctx context.Context, appID api.AppID) error {
 
 	// Send the notification
 	_, err := l.api.WakeupApp(appID).Send(ctx)
+
+	// Check if it's a restart disabled error via context code
+	var apiErr *api.Error
+	if errors.As(err, &apiErr) && apiErr.HasRestartDisabled() {
+		// This is expected for apps with restart disabled, don't log as error
+		l.logger.Infof(ctx, `app "%s" has restart disabled`, appID)
+		return err // Still return the error so it can be handled by the proxy
+	}
+
 	// If it does not succeed but app is currently stopping do not log it as error, log only other errors
 	// Instead of implementing state machine as in sandboxes service, we want to skip valid state that the
 	// pod is deallocating, and we want to wait till pod is `stopped` and we can `start` the pod again.
diff --git a/internal/pkg/service/appsproxy/proxy/apphandler/authproxy/oauthproxy/pagewriter.go b/internal/pkg/service/appsproxy/proxy/apphandler/authproxy/oauthproxy/pagewriter.go
@@ -99,7 +99,7 @@ func (pw *pageWriter) WriteErrorPage(w http.ResponseWriter, req *http.Request, o
 }
 
 func (pw *pageWriter) ProxyErrorHandler(w http.ResponseWriter, req *http.Request, err error) {
-	pw.pageWriter.ProxyErrorHandler(w, req, pw.app, err)
+	pw.pageWriter.ProxyErrorHandler(w, req, pw.app, nil, err)
 }
 
 func (pw *pageWriter) WriteRobotsTxt(w http.ResponseWriter, req *http.Request) {
diff --git a/internal/pkg/service/appsproxy/proxy/apphandler/upstream/upstream.go b/internal/pkg/service/appsproxy/proxy/apphandler/upstream/upstream.go
@@ -49,13 +49,14 @@ type Manager struct {
 }
 
 type AppUpstream struct {
-	manager       *Manager
-	app           api.AppConfig
-	target        *url.URL
-	handler       *chain.Chain
-	wsHandler     *chain.Chain
-	cancelWs      context.CancelCauseFunc
-	activeWsCount atomic.Int64
+	manager         *Manager
+	app             api.AppConfig
+	target          *url.URL
+	handler         *chain.Chain
+	wsHandler       *chain.Chain
+	cancelWs        context.CancelCauseFunc
+	activeWsCount   atomic.Int64
+	restartDisabled atomic.Bool
 }
 
 type dependencies interface {
@@ -141,7 +142,7 @@ func (u *AppUpstream) ServeHTTPOrError(rw http.ResponseWriter, req *http.Request
 func (u *AppUpstream) newProxy(timeout time.Duration) *chain.Chain {
 	proxy := httputil.NewSingleHostReverseProxy(u.target)
 	proxy.Transport = u.manager.transport
-	proxy.ErrorHandler = u.manager.pageWriter.ProxyErrorHandlerFor(u.app)
+	proxy.ErrorHandler = u.manager.pageWriter.ProxyErrorHandlerFor(u.app, &u.restartDisabled)
 
 	return chain.
 		New(chain.HandlerFunc(func(w http.ResponseWriter, req *http.Request) error {
@@ -160,7 +161,7 @@ func (u *AppUpstream) newProxy(timeout time.Duration) *chain.Chain {
 func (u *AppUpstream) newWebsocketProxy(timeout time.Duration) *chain.Chain {
 	proxy := httputil.NewSingleHostReverseProxy(u.target)
 	proxy.Transport = u.manager.transport
-	proxy.ErrorHandler = u.manager.pageWriter.ProxyErrorHandlerFor(u.app)
+	proxy.ErrorHandler = u.manager.pageWriter.ProxyErrorHandlerFor(u.app, &u.restartDisabled)
 
 	return chain.
 		New(chain.HandlerFunc(func(w http.ResponseWriter, req *http.Request) error {
@@ -232,6 +233,13 @@ func (u *AppUpstream) wakeup(ctx context.Context, err error) {
 
 		// Error is already logged by the Wakeup method itself.
 		err := u.manager.wakeup.Wakeup(wakeupCtx, u.app.ID) //nolint:contextcheck
+
+		// Check for restart disabled error
+		var apiErr *api.Error
+		if errors.As(err, &apiErr) && apiErr.HasRestartDisabled() {
+			u.restartDisabled.Store(true)
+		}
+
 		span.End(&err)
 	})
 }
diff --git a/internal/pkg/service/appsproxy/proxy/pagewriter/error.go b/internal/pkg/service/appsproxy/proxy/pagewriter/error.go
@@ -4,6 +4,7 @@ import (
 	"net"
 	"net/http"
 	"strings"
+	"sync/atomic"
 
 	"go.opentelemetry.io/otel/attribute"
 	semconv "go.opentelemetry.io/otel/semconv/v1.17.0"
@@ -14,7 +15,9 @@ import (
 	"github.com/keboola/keboola-as-code/internal/pkg/utils/errors"
 )
 
-const ExceptionIDPrefix = "keboola-appsproxy-"
+const (
+	ExceptionIDPrefix = "keboola-appsproxy-"
+)
 
 type errorPageData struct {
 	App         *AppData
@@ -24,13 +27,20 @@ type errorPageData struct {
 	ExceptionID string
 }
 
-func (pw *Writer) ProxyErrorHandlerFor(app api.AppConfig) func(w http.ResponseWriter, req *http.Request, err error) {
+func (pw *Writer) ProxyErrorHandlerFor(app api.AppConfig, restartDisabled *atomic.Bool) func(w http.ResponseWriter, req *http.Request, err error) {
 	return func(w http.ResponseWriter, req *http.Request, err error) {
-		pw.ProxyErrorHandler(w, req, app, err)
+		pw.ProxyErrorHandler(w, req, app, restartDisabled, err)
 	}
 }
 
-func (pw *Writer) ProxyErrorHandler(w http.ResponseWriter, req *http.Request, app api.AppConfig, err error) {
+func (pw *Writer) ProxyErrorHandler(w http.ResponseWriter, req *http.Request, app api.AppConfig, restartDisabled *atomic.Bool, err error) {
+	// Check for restart disabled error
+	if restartDisabled.Load() {
+		pw.logger.Info(req.Context(), "app has restart disabled, rendering restart disabled page")
+		pw.WriteRestartDisabledPage(w, req, app)
+		return
+	}
+
 	var dnsError *net.DNSError
 	if errors.As(err, &dnsError) {
 		pw.logger.Info(req.Context(), "app is not running, rendering spinner page")
diff --git a/internal/pkg/service/appsproxy/proxy/pagewriter/restart_disabled.go b/internal/pkg/service/appsproxy/proxy/pagewriter/restart_disabled.go
@@ -0,0 +1,19 @@
+package pagewriter
+
+import (
+	"net/http"
+
+	"github.com/keboola/keboola-as-code/internal/pkg/service/appsproxy/dataapps/api"
+)
+
+type RestartDisabledPageData struct {
+	App AppData
+}
+
+func (pw *Writer) WriteRestartDisabledPage(w http.ResponseWriter, req *http.Request, app api.AppConfig) {
+	w.Header().Set("Cache-Control", "no-cache, no-store, must-revalidate;")
+	w.Header().Set("pragma", "no-cache")
+	pw.writePage(w, req, "restart_disabled.gohtml", http.StatusNotFound, RestartDisabledPageData{
+		App: NewAppData(&app),
+	})
+}
diff --git a/internal/pkg/service/appsproxy/proxy/pagewriter/template/restart_disabled.gohtml b/internal/pkg/service/appsproxy/proxy/pagewriter/template/restart_disabled.gohtml
diff --git a/internal/pkg/service/appsproxy/proxy/proxy_test.go b/internal/pkg/service/appsproxy/proxy/proxy_test.go
@@ -1806,6 +1806,55 @@ func TestAppProxyRouter(t *testing.T) {
 				"123": 1,
 			},
 		},
+		{
+			name: "restart-disabled",
+			run: func(t *testing.T, client *http.Client, m []*mockoidc.MockOIDC, appServer *testutil.AppServer, service *testutil.DataAppsAPI, dnsServer *dnsmock.Server) {
+				dnsServer.RemoveARecords(dns.Fqdn("app.local"))
+
+				service.WakeUpOverrides["123"] = func(w http.ResponseWriter, req *http.Request) {
+					w.Header().Set("Content-Type", "application/json")
+					w.WriteHeader(http.StatusBadRequest)
+					_, _ = fmt.Fprintln(w, `{
+	"code": 0,
+	"context": {
+		"code": "apps.restartDisabled"
+	},
+	"error": "App restart is disabled. Contact app maintainer.",
+	"exceptionId": "exception-208db995c92ed365d47bcc701ae4d802",
+	"status": "error"
+}`)
+				}
+				// Request to public app - fails because the app doesn't have a DNS record
+				request, err := http.NewRequestWithContext(t.Context(), http.MethodGet, "https://public-123.hub.keboola.local/", nil)
+				require.NoError(t, err)
+				response, err := client.Do(request)
+				require.NoError(t, err)
+				// First request returns 503 Service Unavailable (Spinner) because wakeup is async
+				require.Equal(t, http.StatusServiceUnavailable, response.StatusCode)
+				body, err := io.ReadAll(response.Body)
+				require.NoError(t, err)
+				assert.Contains(t, string(body), "Starting your application...")
+
+				// Wait for async wakeup to complete
+				assert.Eventually(t, func() bool {
+					// Second request should return 404 Not Found (Restart Disabled)
+					request, err := http.NewRequestWithContext(t.Context(), http.MethodGet, "https://public-123.hub.keboola.local/", nil)
+					require.NoError(t, err)
+					response, err := client.Do(request)
+					require.NoError(t, err)
+					if response.StatusCode == http.StatusNotFound {
+						body, err := io.ReadAll(response.Body)
+						require.NoError(t, err)
+						return strings.Contains(string(body), "Application Disabled")
+					}
+					return false
+				}, 5*time.Second, 100*time.Millisecond)
+
+				// Expect wakeup but no notification since there was an authorized request to the app but not while it was running.
+			},
+			expectedNotifications: map[string]int{},
+			expectedWakeUps:       map[string]int{},
+		},
 		{
 			name: "private-one-provider-selector",
 			run: func(t *testing.T, client *http.Client, m []*mockoidc.MockOIDC, appServer *testutil.AppServer, service *testutil.DataAppsAPI, dnsServer *dnsmock.Server) {
diff --git a/internal/pkg/service/appsproxy/proxy/testutil/api.go b/internal/pkg/service/appsproxy/proxy/testutil/api.go
@@ -21,18 +21,20 @@ import (
 
 type DataAppsAPI struct {
 	*httptest.Server
-	Apps          map[api.AppID]api.AppConfig
-	Notifications map[string]int
-	WakeUps       map[string]int
+	Apps            map[api.AppID]api.AppConfig
+	Notifications   map[string]int
+	WakeUps         map[string]int
+	WakeUpOverrides map[string]http.HandlerFunc
 }
 
 func StartDataAppsAPI(t *testing.T, pm server.PortManager) *DataAppsAPI {
 	t.Helper()
 
 	service := &DataAppsAPI{
-		Apps:          make(map[api.AppID]api.AppConfig),
-		Notifications: make(map[string]int),
-		WakeUps:       make(map[string]int),
+		Apps:            make(map[api.AppID]api.AppConfig),
+		Notifications:   make(map[string]int),
+		WakeUps:         make(map[string]int),
+		WakeUpOverrides: make(map[string]http.HandlerFunc),
 	}
 
 	mux := http.NewServeMux()
@@ -78,6 +80,10 @@ func StartDataAppsAPI(t *testing.T, pm server.PortManager) *DataAppsAPI {
 			service.Notifications[appID] += 1
 		}
 		if _, ok := data["desiredState"]; ok {
+			if override, ok := service.WakeUpOverrides[appID]; ok {
+				override(w, req)
+				return
+			}
 			service.WakeUps[appID] += 1
 		}
 	})

Original file line number	Diff line number	Diff line change
`@@ -21,7 +21,7 @@ func (a *API) WakeupApp(appID AppID) request.APIRequest[request.NoResult] {`
`21`	`21`	`attribute.Int(attrSandboxesServiceStatusCode, response.StatusCode()),`
`22`	`22`	`}`
`23`	`23`	`span.SetAttributes(attrs...)`
`24`		`- return nil`
	`24`	`+ return err`
`25`	`25`	`}).`
`26`	`26`	`WithPatch("apps/{appId}").`
`27`	`27`	`AndPathParam("appId", appID.String()).`
Original file line number	Diff line number	Diff line change
`@@ -99,7 +99,7 @@ func (pw pageWriter) WriteErrorPage(w http.ResponseWriter, req http.Request, o`
`99`	`99`	`}`
`100`	`100`
`101`	`101`	`func (pw pageWriter) ProxyErrorHandler(w http.ResponseWriter, req http.Request, err error) {`
`102`		`- pw.pageWriter.ProxyErrorHandler(w, req, pw.app, err)`
	`102`	`+ pw.pageWriter.ProxyErrorHandler(w, req, pw.app, nil, err)`
`103`	`103`	`}`
`104`	`104`
`105`	`105`	`func (pw pageWriter) WriteRobotsTxt(w http.ResponseWriter, req http.Request) {`