filesystem: adjust live daemon binary check (#699)

Commit b5bceed introduced a regression where nydusd binary is
required even for proxy mode (used by Kata Containers for guest
pulling) where no nydusd daemon runs on the host.

The hot upgrade logic now unconditionally calls GetDaemonGitCommit()
which fails when nydusd binary doesn't exist. This breaks proxy mode
deployments where only the snapshotter runs on the host.

Fix by only checking for the nydusd binary when nydusd is already
running.

Fixes regression from: b5bceed

Signed-off-by: Fabiano Fidêncio <[email protected]>
Signed-off-by: Zephyr <[email protected]>

Author: fidencio

pkg/filesystem/fs.go

@@ -157,44 +157,47 @@ func NewFileSystem(ctx context.Context, opt ...NewFSOpt) (*Filesystem, error) {
 		return nil, err
 	}
 
-	newNydusImageBinaryCommit, err := daemon.GetDaemonGitCommit(fs.nydusdBinaryPath)
-	if err != nil {
-		return nil, errors.Wrapf(err, "failed to get git commit from nydusd binary at path: %s", fs.nydusdBinaryPath)
-	}
+	// Only check nydusd binary commit when there are live daemons that may need hot upgrade.
+	if len(liveDaemons) > 0 {
+		newNydusImageBinaryCommit, err := daemon.GetDaemonGitCommit(fs.nydusdBinaryPath)
+		if err != nil {
+			return nil, errors.Wrapf(err, "failed to get git commit from nydusd binary at path: %s", fs.nydusdBinaryPath)
+		}
 
-	egLive, _ := errgroup.WithContext(context.Background())
-	for _, d := range liveDaemons {
-		d := d
-		egLive.Go(func() error {
-			if d.Supervisor == nil {
-				log.L.Warnf("Daemon %s is skipped for hot upgrade because recover policy is not set to 'failover'", d.ID())
-				return nil
-			}
-			daemonInfo, err := d.GetDaemonInfo()
-			if err != nil {
-				log.L.Warnf("Failed to get daemon info from daemon %s, skipping: %v", d.ID(), err)
-				return nil
-			}
-			if newNydusImageBinaryCommit != daemonInfo.DaemonVersion().GitCommit {
-				fsManager, err := fs.getManager(d.States.FsDriver)
-				if err != nil {
-					log.L.Warnf("Failed to get filesystem manager for daemon %s, skipping: %v", d.ID(), err)
+		egLive, _ := errgroup.WithContext(context.Background())
+		for _, d := range liveDaemons {
+			d := d
+			egLive.Go(func() error {
+				if d.Supervisor == nil {
+					log.L.Warnf("Daemon %s is skipped for hot upgrade because recover policy is not set to 'failover'", d.ID())
 					return nil
 				}
-				newDaemon, upgradeErr := fsManager.DoDaemonUpgrade(d, fs.nydusdBinaryPath, fsManager)
-				if upgradeErr != nil {
-					log.L.Warnf("Daemon %s hot upgrade failed, skipping: %v", d.ID(), upgradeErr)
+				daemonInfo, err := d.GetDaemonInfo()
+				if err != nil {
+					log.L.Warnf("Failed to get daemon info from daemon %s, skipping: %v", d.ID(), err)
 					return nil
 				}
-				fs.TryRetainSharedDaemon(newDaemon)
-			} else {
-				fs.TryRetainSharedDaemon(d)
-			}
-			return nil
-		})
-	}
-	if err := egLive.Wait(); err != nil {
-		return nil, err
+				if newNydusImageBinaryCommit != daemonInfo.DaemonVersion().GitCommit {
+					fsManager, err := fs.getManager(d.States.FsDriver)
+					if err != nil {
+						log.L.Warnf("Failed to get filesystem manager for daemon %s, skipping: %v", d.ID(), err)
+						return nil
+					}
+					newDaemon, upgradeErr := fsManager.DoDaemonUpgrade(d, fs.nydusdBinaryPath, fsManager)
+					if upgradeErr != nil {
+						log.L.Warnf("Daemon %s hot upgrade failed, skipping: %v", d.ID(), upgradeErr)
+						return nil
+					}
+					fs.TryRetainSharedDaemon(newDaemon)
+				} else {
+					fs.TryRetainSharedDaemon(d)
+				}
+				return nil
+			})
+		}
+		if err := egLive.Wait(); err != nil {
+			return nil, err
+		}
 	}
 	return &fs, nil
 }