Introduce install TTL markers

.mockery.yaml

@@ -18,6 +18,7 @@ packages:
     interfaces:
       WatcherHelper: {}
       watcherGrappler: {}
+      availableRollbacksSource: {}
   github.com/elastic/elastic-agent/internal/pkg/agent/cmd:
     interfaces:
       agentWatcher: {}

internal/pkg/agent/application/application.go

@@ -64,7 +64,7 @@ func New(
 	fleetInitTimeout time.Duration,
 	disableMonitoring bool,
 	override CfgOverrider,
-	initialUpgradeDetails *details.Details,
+	initialUpdateMarker *upgrade.UpdateMarker,
 	modifiers ...component.PlatformModifier,
 ) (*coordinator.Coordinator, coordinator.ConfigManager, composable.Controller, error) {
 
@@ -129,7 +129,9 @@ func New(
 
 	// monitoring is not supported in bootstrap mode https://github.com/elastic/elastic-agent/issues/1761
 	isMonitoringSupported := !disableMonitoring && cfg.Settings.V1MonitoringEnabled
-	upgrader, err := upgrade.NewUpgrader(log, cfg.Settings.DownloadConfig, cfg.Settings.Upgrade, agentInfo, new(upgrade.AgentWatcherHelper))
+
+	availableRollbacksSource := upgrade.NewTTLMarkerRegistry(log, paths.Top())
+	upgrader, err := upgrade.NewUpgrader(log, cfg.Settings.DownloadConfig, cfg.Settings.Upgrade, agentInfo, new(upgrade.AgentWatcherHelper), availableRollbacksSource)
 	if err != nil {
 		return nil, nil, nil, fmt.Errorf("failed to create upgrader: %w", err)
 	}
@@ -153,6 +155,12 @@ func New(
 		return nil, nil, nil, fmt.Errorf("failed to initialize runtime manager: %w", err)
 	}
 
+	// prepare initialUpgradeDetails for injecting it in coordinator later on
+	var initialUpgradeDetails *details.Details
+	if initialUpdateMarker != nil && initialUpdateMarker.Details != nil {
+		initialUpgradeDetails = initialUpdateMarker.Details
+	}
+
 	var configMgr coordinator.ConfigManager
 	var managed *managedConfigManager
 	var compModifiers = []coordinator.ComponentsModifier{InjectAPMConfig}

internal/pkg/agent/application/coordinator/coordinator_unit_test.go

@@ -467,7 +467,8 @@ func TestCoordinatorReportsInvalidPolicy(t *testing.T) {
 		}
 	}()
 
-	upgradeMgr, err := upgrade.NewUpgrader(log, &artifact.Config{}, nil, &info.AgentInfo{}, new(upgrade.AgentWatcherHelper))
+	tmpDir := t.TempDir()
+	upgradeMgr, err := upgrade.NewUpgrader(log, &artifact.Config{}, nil, &info.AgentInfo{}, new(upgrade.AgentWatcherHelper), upgrade.NewTTLMarkerRegistry(nil, tmpDir))
 	require.NoError(t, err, "errored when creating a new upgrader")
 
 	// Channels have buffer length 1, so we don't have to run on multiple

internal/pkg/agent/application/upgrade/rollback_test.go

@@ -769,6 +769,6 @@ func createUpdateMarker(t *testing.T, log *logger.Logger, topDir, newAgentVersio
 		time.Now(),
 		newAgentInstall,
 		oldAgentInstall,
-		nil, nil, disableRollbackWindow)
+		nil, nil, nil)
 	require.NoError(t, err, "error writing fake update marker")
 }

internal/pkg/agent/application/upgrade/step_mark.go

@@ -23,10 +23,9 @@ import (
 const markerFilename = ".update-marker"
 const disableRollbackWindow = time.Duration(0)
 
-// RollbackAvailable identifies an elastic-agent install available for rollback
-type RollbackAvailable struct {
+// TTLMarker marks an elastic-agent install available for rollback
+type TTLMarker struct {
 	Version    string    `json:"version" yaml:"version"`
-	Home       string    `json:"home" yaml:"home"`
 	ValidUntil time.Time `json:"valid_until" yaml:"valid_until"`
 }
 
@@ -55,7 +54,7 @@ type UpdateMarker struct {
 
 	Details *details.Details `json:"details,omitempty" yaml:"details,omitempty"`
 
-	RollbacksAvailable []RollbackAvailable `json:"rollbacks_available,omitempty" yaml:"rollbacks_available,omitempty"`
+	RollbacksAvailable map[string]TTLMarker `json:"rollbacks_available,omitempty" yaml:"rollbacks_available,omitempty"`
 }
 
 // GetActionID returns the Fleet Action ID associated with the
@@ -112,7 +111,7 @@ type updateMarkerSerializer struct {
 	Acked              bool                 `yaml:"acked"`
 	Action             *MarkerActionUpgrade `yaml:"action"`
 	Details            *details.Details     `yaml:"details"`
-	RollbacksAvailable []RollbackAvailable  `yaml:"rollbacks_available,omitempty"`
+	RollbacksAvailable map[string]TTLMarker `yaml:"rollbacks_available,omitempty"`
 }
 
 func newMarkerSerializer(m *UpdateMarker) *updateMarkerSerializer {
@@ -142,35 +141,23 @@ type updateActiveCommitFunc func(log *logger.Logger, topDirPath, hash string, wr
 
 // markUpgrade marks update happened so we can handle grace period
 func markUpgradeProvider(updateActiveCommit updateActiveCommitFunc, writeFile writeFileFunc) markUpgradeFunc {
-	return func(log *logger.Logger, dataDirPath string, updatedOn time.Time, agent, previousAgent agentInstall, action *fleetapi.ActionUpgrade, upgradeDetails *details.Details, rollbackWindow time.Duration) error {
+	return func(log *logger.Logger, dataDirPath string, updatedOn time.Time, agent, previousAgent agentInstall, action *fleetapi.ActionUpgrade, upgradeDetails *details.Details, availableRollbacks map[string]TTLMarker) error {
 
 		if len(previousAgent.hash) > hashLen {
 			previousAgent.hash = previousAgent.hash[:hashLen]
 		}
 
 		marker := &UpdateMarker{
-			Version:           agent.version,
-			Hash:              agent.hash,
-			VersionedHome:     agent.versionedHome,
-			UpdatedOn:         updatedOn,
-			PrevVersion:       previousAgent.version,
-			PrevHash:          previousAgent.hash,
-			PrevVersionedHome: previousAgent.versionedHome,
-			Action:            action,
-			Details:           upgradeDetails,
-		}
-
-		if rollbackWindow > disableRollbackWindow && agent.parsedVersion != nil && !agent.parsedVersion.Less(*Version_9_2_0_SNAPSHOT) {
-			// if we have a not empty rollback window, write the prev version in the rollbacks_available field
-			// we also need to check the destination version because the manual rollback and delayed cleanup will be
-			// handled by that version of agent, so it needs to be recent enough
-			marker.RollbacksAvailable = []RollbackAvailable{
-				{
-					Version:    previousAgent.version,
-					Home:       previousAgent.versionedHome,
-					ValidUntil: updatedOn.Add(rollbackWindow),
-				},
-			}
+			Version:            agent.version,
+			Hash:               agent.hash,
+			VersionedHome:      agent.versionedHome,
+			UpdatedOn:          updatedOn,
+			PrevVersion:        previousAgent.version,
+			PrevHash:           previousAgent.hash,
+			PrevVersionedHome:  previousAgent.versionedHome,
+			Action:             action,
+			Details:            upgradeDetails,
+			RollbacksAvailable: availableRollbacks,
 		}
 
 		markerBytes, err := yaml.Marshal(newMarkerSerializer(marker))