Use custom loop to watch for upgrade plan changes instead of agency callback

Author: nikita-vanyasin

CHANGELOG.md

@@ -2,7 +2,7 @@
 
 ## [master](https://github.com/arangodb-helper/arangodb/tree/master) (N/A)
 - Fix context handling in WaitUntilStarterReady for tests
-- Proper leader election without using observe and TTL features
+- Proper leader election and cluster upgrade without using observe and TTL features
 
 ## [0.15.8](https://github.com/arangodb-helper/arangodb/tree/0.15.8) (2023-06-02)
 - Add passing ARANGODB_SERVER_DIR env variable when starting arangod instances

service/server.go

@@ -173,8 +173,6 @@ func (s *httpServer) Run(hostAddr, containerAddr string, tlsConfig *tls.Config,
 		mux.HandleFunc("/database-version", s.databaseVersionHandler)
 		mux.HandleFunc("/shutdown", s.shutdownHandler)
 		mux.HandleFunc("/database-auto-upgrade", s.databaseAutoUpgradeHandler)
-		// Agency callback
-		mux.HandleFunc("/cb/upgradePlanChanged", s.cbUpgradePlanChanged)
 
 		// JWT Rotation
 		s.registerJWTFunctions(mux)
@@ -744,20 +742,6 @@ func (s *httpServer) databaseAutoUpgradeHandler(w http.ResponseWriter, r *http.R
 	}
 }
 
-// cbUpgradePlanChanged is a callback called by the agency when the upgrade plan is modified.
-func (s *httpServer) cbUpgradePlanChanged(w http.ResponseWriter, r *http.Request) {
-	s.log.Debug().Msgf("Upgrade plan changed callback from %s", r.RemoteAddr)
-	if r.Method != "POST" {
-		w.WriteHeader(http.StatusMethodNotAllowed)
-		return
-	}
-
-	// Interrupt upgrade manager
-	s.context.UpgradeManager().UpgradePlanChangedCallback()
-	w.WriteHeader(http.StatusOK)
-	w.Write([]byte("OK"))
-}
-
 func handleError(w http.ResponseWriter, err error) {
 	if loc, ok := IsRedirect(err); ok {
 		header := w.Header()

service/upgrade_manager.go

@@ -27,6 +27,7 @@ import (
 	"context"
 	"fmt"
 	"net/url"
+	"reflect"
 	"strings"
 	"sync"
 	"time"
@@ -73,9 +74,6 @@ type UpgradeManager interface {
 
 	// RunWatchUpgradePlan keeps watching the upgrade plan until the given context is canceled.
 	RunWatchUpgradePlan(context.Context)
-
-	// UpgradePlanChangedCallback is an agency callback to notify about changes in the upgrade plan
-	UpgradePlanChangedCallback()
 }
 
 // UpgradeManagerContext holds methods used by the upgrade manager to control its context.
@@ -155,6 +153,11 @@ func (p *UpgradePlan) ResetFailures() {
 	}
 }
 
+// Equals returns true if other plan is the same
+func (p UpgradePlan) Equals(other UpgradePlan) bool {
+	return reflect.DeepEqual(p, other)
+}
+
 // UpgradeEntryType is a strongly typed upgrade plan item
 type UpgradeEntryType string
 
@@ -823,39 +826,76 @@ func (m *upgradeManager) removeUpgradePlan(ctx context.Context) error {
 	return nil
 }
 
+// waitForPlanChanges returns unbuffered channel on which
+func (m *upgradeManager) waitForPlanChanges(ctx context.Context) chan UpgradePlan {
+	ch := make(chan UpgradePlan)
+
+	go func() {
+		var oldPlan UpgradePlan
+		for {
+			delay := time.Second * 3
+			plan, err := m.readUpgradePlan(ctx)
+			if agency.IsKeyNotFound(err) || plan.IsEmpty() {
+				// Just try later
+			} else if err != nil {
+				// Failed to read plan
+				m.log.Info().Err(err).Msg("Failed to read upgrade plan")
+			} else if !oldPlan.Equals(plan) {
+				ch <- plan
+				oldPlan = plan
+				delay = time.Millisecond * 500
+			}
+
+			timer := time.NewTimer(delay)
+			select {
+			case <-timer.C:
+				// Continue
+			case <-ctx.Done():
+				// Context canceled
+				if !timer.Stop() {
+					<-timer.C
+				}
+				return
+			}
+		}
+	}()
+
+	return ch
+}
+
 // RunWatchUpgradePlan keeps watching the upgrade plan in the agency.
 // Once it detects that this starter has to act, it does.
 func (m *upgradeManager) RunWatchUpgradePlan(ctx context.Context) {
-	_, myPeer, mode := m.upgradeManagerContext.ClusterConfig()
-	ownURL := myPeer.CreateStarterURL("/")
+	_, _, mode := m.upgradeManagerContext.ClusterConfig()
 	if !mode.HasAgency() {
 		// Nothing to do here without an agency
 		return
 	}
-	registeredCallback := false
-	defer func() {
-		if registeredCallback {
-			m.unregisterUpgradePlanChangedCallback(ctx, ownURL)
-		}
-	}()
+
+	planChanges := m.waitForPlanChanges(ctx)
+
+	var plan UpgradePlan
+	delay := 1 * time.Second
 	for {
-		delay := time.Minute
-		if !registeredCallback {
-			m.log.Debug().Msg("Registering upgrade plan changed callback...")
-			if err := m.registerUpgradePlanChangedCallback(ctx, ownURL); err != nil {
-				m.log.Info().Err(err).Msg("Failed to register upgrade plan changed callback")
-			} else {
-				registeredCallback = true
+		timer := time.NewTimer(delay)
+		select {
+		case <-timer.C:
+			// Continue
+		case plan = <-planChanges:
+			// Continue
+			if !timer.Stop() {
+				<-timer.C
 			}
+		case <-ctx.Done():
+			// Context canceled
+			if !timer.Stop() {
+				<-timer.C
+			}
+			return
 		}
-		plan, err := m.readUpgradePlan(ctx)
-		if agency.IsKeyNotFound(err) || plan.IsEmpty() {
-			// Just try later
-		} else if err != nil {
-			// Failed to read plan
-			m.log.Info().Err(err).Msg("Failed to read upgrade plan")
-		} else if plan.IsReady() {
-			// Plan entries have aal been processes
+
+		if plan.IsReady() {
+			// Plan entries have been processed
 			if !plan.Finished {
 				// Let's show the user that we're done
 				if err := m.finishUpgradePlan(ctx, plan); err != nil {
@@ -871,62 +911,7 @@ func (m *upgradeManager) RunWatchUpgradePlan(ctx context.Context) {
 			}
 			delay = time.Second
 		}
-
-		select {
-		case <-time.After(delay):
-			// Continue
-		case <-m.cbTrigger.Done():
-			// Continue
-		case <-ctx.Done():
-			// Context canceled
-			return
-		}
-	}
-}
-
-// UpgradePlanChangedCallback is an agency callback to notify about changes in the upgrade plan
-func (m *upgradeManager) UpgradePlanChangedCallback() {
-	m.cbTrigger.Trigger()
-}
-
-// registerUpgradePlanChangedCallback registers our callback URL with the agency
-func (m *upgradeManager) registerUpgradePlanChangedCallback(ctx context.Context, ownURL string) error {
-	// Get api client
-	api, err := m.createAgencyAPI()
-	if err != nil {
-		return maskAny(err)
-	}
-	// Register callback
-	cbURL, err := getURLWithPath(ownURL, "/cb/upgradePlanChanged")
-	if err != nil {
-		return maskAny(err)
 	}
-	ctx, cancel := context.WithTimeout(ctx, time.Second*10)
-	defer cancel()
-	if err := api.RegisterChangeCallback(ctx, upgradePlanKey, cbURL); err != nil {
-		return maskAny(err)
-	}
-	return nil
-}
-
-// unregisterUpgradePlanChangedCallback removes our callback URL from the agency
-func (m *upgradeManager) unregisterUpgradePlanChangedCallback(ctx context.Context, ownURL string) error {
-	// Get api client
-	api, err := m.createAgencyAPI()
-	if err != nil {
-		return maskAny(err)
-	}
-	// Register callback
-	cbURL, err := getURLWithPath(ownURL, "/cb/upgradePlanChanged")
-	if err != nil {
-		return maskAny(err)
-	}
-	ctx, cancel := context.WithTimeout(ctx, time.Second*10)
-	defer cancel()
-	if err := api.UnregisterChangeCallback(ctx, upgradePlanKey, cbURL); err != nil {
-		return maskAny(err)
-	}
-	return nil
 }
 
 // processUpgradePlan inspects the first entry of the given plan and acts upon