WIP - add rollbacks available to upgrade marker

Author: pchila

internal/pkg/agent/application/application.go

@@ -120,7 +120,7 @@ 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, agentInfo)
+	upgrader, err := upgrade.NewUpgrader(log, cfg.Settings.DownloadConfig, cfg.Settings.Upgrade, agentInfo)
 	if err != nil {
 		return nil, nil, nil, fmt.Errorf("failed to create upgrader: %w", err)
 	}

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

@@ -459,11 +459,7 @@ func TestCoordinatorReportsInvalidPolicy(t *testing.T) {
 		}
 	}()
 
-	upgradeMgr, err := upgrade.NewUpgrader(
-		log,
-		&artifact.Config{},
-		&info.AgentInfo{},
-	)
+	upgradeMgr, err := upgrade.NewUpgrader(log, &artifact.Config{}, nil, &info.AgentInfo{})
 	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/details/details.go

@@ -59,6 +59,19 @@ type Metadata struct {
 	// Reason is a string that may give out more information about transitioning to the current state. It has been
 	// introduced initially to distinguish between manual and automatic rollbacks
 	Reason string `json:"reason,omitempty" yaml:"reason,omitempty"`
+
+	RollbacksAvailable []RollbackAvailable `json:"rollbacks_available,omitempty" yaml:"rollbacks_available,omitempty"`
+}
+
+//rollback_available:
+//- version: 8.16.0-SNAPSHOT,
+//home: data/elastic-agent-8.16.0-SNAPSHOT-b65953
+//valid_until: "2024-11-21T14:42:21Z"
+
+type RollbackAvailable struct {
+	Version    string    `json:"version" yaml:"version"`
+	Home       string    `json:"home" yaml:"home"`
+	ValidUntil time.Time `json:"valid_until" yaml:"valid_until"`
 }
 
 func NewDetails(targetVersion string, initialState State, actionID string) *Details {

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

@@ -504,10 +504,6 @@ func createUpdateMarker(t *testing.T, log *logger.Logger, topDir, newAgentVersio
 		hash:          oldAgentHash,
 		versionedHome: oldAgentVersionedHome,
 	}
-	err := markUpgrade(log,
-		paths.DataFrom(topDir),
-		newAgentInstall,
-		oldAgentInstall,
-		nil, nil, OUTCOME_UPGRADE)
+	err := markUpgrade(log, paths.DataFrom(topDir), newAgentInstall, oldAgentInstall, nil, nil, OUTCOME_UPGRADE, 0)
 	require.NoError(t, err, "error writing fake update marker")
 }

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

@@ -91,7 +91,7 @@ func TestDownloadWithRetries(t *testing.T) {
 			return &mockDownloader{expectedDownloadPath, nil}, nil
 		}
 
-		u, err := NewUpgrader(testLogger, &settings, &info.AgentInfo{})
+		u, err := NewUpgrader(testLogger, &settings, nil, &info.AgentInfo{})
 		require.NoError(t, err)
 
 		parsedVersion, err := agtversion.ParseVersion("8.9.0")
@@ -141,7 +141,7 @@ func TestDownloadWithRetries(t *testing.T) {
 			return nil, nil
 		}
 
-		u, err := NewUpgrader(testLogger, &settings, &info.AgentInfo{})
+		u, err := NewUpgrader(testLogger, &settings, nil, &info.AgentInfo{})
 		require.NoError(t, err)
 
 		parsedVersion, err := agtversion.ParseVersion("8.9.0")
@@ -196,7 +196,7 @@ func TestDownloadWithRetries(t *testing.T) {
 			return nil, nil
 		}
 
-		u, err := NewUpgrader(testLogger, &settings, &info.AgentInfo{})
+		u, err := NewUpgrader(testLogger, &settings, nil, &info.AgentInfo{})
 		require.NoError(t, err)
 
 		parsedVersion, err := agtversion.ParseVersion("8.9.0")
@@ -241,7 +241,7 @@ func TestDownloadWithRetries(t *testing.T) {
 			return &mockDownloader{"", errors.New("download failed")}, nil
 		}
 
-		u, err := NewUpgrader(testLogger, &settings, &info.AgentInfo{})
+		u, err := NewUpgrader(testLogger, &settings, nil, &info.AgentInfo{})
 		require.NoError(t, err)
 
 		parsedVersion, err := agtversion.ParseVersion("8.9.0")

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

@@ -197,7 +197,7 @@ type agentInstall struct {
 }
 
 // markUpgrade marks update happened so we can handle grace period
-func markUpgrade(log *logger.Logger, dataDirPath string, agent, previousAgent agentInstall, action *fleetapi.ActionUpgrade, upgradeDetails *details.Details, desiredOutcome UpgradeOutcome) error {
+func markUpgrade(log *logger.Logger, dataDirPath string, agent, previousAgent agentInstall, action *fleetapi.ActionUpgrade, upgradeDetails *details.Details, desiredOutcome UpgradeOutcome, rollbackWindow time.Duration) error {
 
 	if len(previousAgent.hash) > hashLen {
 		previousAgent.hash = previousAgent.hash[:hashLen]
@@ -216,13 +216,22 @@ func markUpgrade(log *logger.Logger, dataDirPath string, agent, previousAgent ag
 		DesiredOutcome:    desiredOutcome,
 	}
 
+	if rollbackWindow > 0 {
+		// if we have a not empty rollback window, write the prev version in the rollbacks_available field
+		upgradeDetails.Metadata.RollbacksAvailable = []details.RollbackAvailable{details.RollbackAvailable{
+			Version:    previousAgent.version,
+			Home:       previousAgent.versionedHome,
+			ValidUntil: time.Now().Add(rollbackWindow),
+		}}
+	}
+
 	markerBytes, err := yaml.Marshal(newMarkerSerializer(marker))
 	if err != nil {
 		return errors.New(err, errors.TypeConfig, "failed to parse marker file")
 	}
 
 	markerPath := markerFilePath(dataDirPath)
-	log.Infow("Writing upgrade marker file", "file.path", markerPath, "hash", marker.Hash, "prev_hash", marker.PrevHash)
+	log.Infow("Writing upgrade marker file", "file.path", markerPath, "hash", marker.Hash, "prev_hash", marker.PrevHash, "content", string(markerBytes))
 	if err := os.WriteFile(markerPath, markerBytes, 0600); err != nil {
 		return errors.New(err, errors.TypeFilesystem, "failed to create update marker file", errors.M(errors.MetaKeyPath, markerPath))
 	}

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

@@ -73,12 +73,13 @@ func init() {
 
 // Upgrader performs an upgrade
 type Upgrader struct {
-	log            *logger.Logger
-	settings       *artifact.Config
-	agentInfo      info.Agent
-	upgradeable    bool
-	fleetServerURI string
-	markerWatcher  MarkerWatcher
+	log             *logger.Logger
+	settings        *artifact.Config
+	upgradeSettings *configuration.UpgradeConfig
+	agentInfo       info.Agent
+	upgradeable     bool
+	fleetServerURI  string
+	markerWatcher   MarkerWatcher
 }
 
 // IsUpgradeable when agent is installed and running as a service or flag was provided.
@@ -89,13 +90,14 @@ func IsUpgradeable() bool {
 }
 
 // NewUpgrader creates an upgrader which is capable of performing upgrade operation
-func NewUpgrader(log *logger.Logger, settings *artifact.Config, agentInfo info.Agent) (*Upgrader, error) {
+func NewUpgrader(log *logger.Logger, settings *artifact.Config, upgradeConfig *configuration.UpgradeConfig, agentInfo info.Agent) (*Upgrader, error) {
 	return &Upgrader{
-		log:           log,
-		settings:      settings,
-		agentInfo:     agentInfo,
-		upgradeable:   IsUpgradeable(),
-		markerWatcher: newMarkerFileWatcher(markerFilePath(paths.Data()), log),
+		log:             log,
+		settings:        settings,
+		upgradeSettings: upgradeConfig,
+		agentInfo:       agentInfo,
+		upgradeable:     IsUpgradeable(),
+		markerWatcher:   newMarkerFileWatcher(markerFilePath(paths.Data()), log),
 	}, nil
 }
 
@@ -147,6 +149,8 @@ func (u *Upgrader) Reload(rawConfig *config.Config) error {
 	}
 
 	u.settings = cfg.Settings.DownloadConfig
+	u.upgradeSettings = cfg.Settings.Upgrade
+
 	return nil
 }
 
@@ -346,12 +350,11 @@ func (u *Upgrader) Upgrade(ctx context.Context, version string, rollback bool, s
 		hash:          release.Commit(),
 		versionedHome: currentVersionedHome,
 	}
-
-	if err := markUpgrade(u.log,
-		paths.Data(), // data dir to place the marker in
-		current,      // new agent version data
-		previous,     // old agent version data
-		action, det, OUTCOME_UPGRADE); err != nil {
+	rollbackWindow := time.Duration(0)
+	if u.upgradeSettings != nil && u.upgradeSettings.Rollback != nil { // TODO && target version supports manual rollback and deferred cleanup
+		rollbackWindow = u.upgradeSettings.Rollback.Window
+	}
+	if err := markUpgrade(u.log, paths.Data(), current, previous, action, det, OUTCOME_UPGRADE, rollbackWindow); err != nil {
 		u.log.Errorw("Rolling back: marking upgrade failed", "error.message", err)
 		rollbackErr := rollbackInstall(ctx, u.log, paths.Top(), hashedDir, currentVersionedHome)
 		return nil, goerrors.Join(err, rollbackErr)

internal/pkg/agent/cmd/watch.go

@@ -308,7 +308,11 @@ func getConfig(streams *cli.IOStreams) *configuration.Configuration {
 }
 
 func initUpgradeDetails(marker *upgrade.UpdateMarker, saveMarker func(*upgrade.UpdateMarker, bool) error, log *logp.Logger) *details.Details {
+	// FIXME this should edit details not rewrite them
 	upgradeDetails := details.NewDetails(version.GetAgentPackageVersion(), details.StateWatching, marker.GetActionID())
+	if marker.Details != nil {
+		upgradeDetails.Metadata.RollbacksAvailable = marker.Details.Metadata.RollbacksAvailable
+	}
 	upgradeDetails.RegisterObserver(func(details *details.Details) {
 		marker.Details = details
 		if err := saveMarker(marker, true); err != nil {