remove serializing WalkDir() across all buckets/prefixes on SSDs (#17707)

slower drives get knocked off because they are too slow via 
active monitoring, we do not need to block calls arbitrarily.

Serializing adds latencies for already slow calls, remove
it for SSDs/NVMEs

Also, add a selection with context when writing to `out <-`
channel, to avoid any potential blocks.

cmd/xl-storage.go

@@ -115,8 +115,8 @@ type xlStorage struct {
 	formatData []byte
 
 	// mutex to prevent concurrent read operations overloading walks.
-	walkMu     sync.Mutex
-	walkReadMu sync.Mutex
+	walkMu     *sync.Mutex
+	walkReadMu *sync.Mutex
 }
 
 // checkPathLength - returns error if given path name length more than 255
@@ -216,18 +216,17 @@ func newXLStorage(ep Endpoint, cleanUp bool) (s *xlStorage, err error) {
 		return nil, err
 	}
 
+	info, err := disk.GetInfo(path)
+	if err != nil {
+		return nil, err
+	}
+
 	var rootDisk bool
 	if !globalIsCICD && !globalIsErasureSD {
 		if globalRootDiskThreshold > 0 {
 			// Use MINIO_ROOTDISK_THRESHOLD_SIZE to figure out if
-			// this disk is a root disk.
-			info, err := disk.GetInfo(path)
-			if err != nil {
-				return nil, err
-			}
-
-			// treat those disks with size less than or equal to the
-			// threshold as rootDisks.
+			// this disk is a root disk. treat those disks with
+			// size less than or equal to the threshold as rootDisks.
 			rootDisk = info.Total <= globalRootDiskThreshold
 		} else {
 			rootDisk, err = disk.IsRootDisk(path, SlashSeparator)
@@ -247,6 +246,12 @@ func newXLStorage(ep Endpoint, cleanUp bool) (s *xlStorage, err error) {
 		diskIndex:  -1,
 	}
 
+	// We stagger listings only on HDDs.
+	if info.Rotational == nil || *info.Rotational {
+		s.walkMu = &sync.Mutex{}
+		s.walkReadMu = &sync.Mutex{}
+	}
+
 	if cleanUp {
 		bgFormatErasureCleanupTmp(s.diskPath) // cleanup any old data.
 	}