expire lockers if lockers are offline (#10749)

lockers currently might leave stale lockers,
in unknown ways waiting for downed lockers.

locker check interval is high enough to safely
cleanup stale locks.

cmd/admin-handlers.go

@@ -348,7 +348,7 @@ func (a adminAPIHandlers) DataUsageInfoHandler(w http.ResponseWriter, r *http.Re
 	writeSuccessResponseJSON(w, dataUsageInfoJSON)
 }
 
-func lriToLockEntry(l lockRequesterInfo, resource, server string, rquorum, wquorum int) *madmin.LockEntry {
+func lriToLockEntry(l lockRequesterInfo, resource, server string) *madmin.LockEntry {
 	entry := &madmin.LockEntry{
 		Timestamp:  l.Timestamp,
 		Resource:   resource,
@@ -356,18 +356,17 @@ func lriToLockEntry(l lockRequesterInfo, resource, server string, rquorum, wquor
 		Source:     l.Source,
 		Owner:      l.Owner,
 		ID:         l.UID,
+		Quorum:     l.Quorum,
 	}
 	if l.Writer {
 		entry.Type = "WRITE"
-		entry.Quorum = wquorum
 	} else {
 		entry.Type = "READ"
-		entry.Quorum = rquorum
 	}
 	return entry
 }
 
-func topLockEntries(peerLocks []*PeerLocks, rquorum, wquorum int, stale bool) madmin.LockEntries {
+func topLockEntries(peerLocks []*PeerLocks, stale bool) madmin.LockEntries {
 	entryMap := make(map[string]*madmin.LockEntry)
 	for _, peerLock := range peerLocks {
 		if peerLock == nil {
@@ -379,7 +378,7 @@ func topLockEntries(peerLocks []*PeerLocks, rquorum, wquorum int, stale bool) ma
 					if val, ok := entryMap[lockReqInfo.UID]; ok {
 						val.ServerList = append(val.ServerList, peerLock.Addr)
 					} else {
-						entryMap[lockReqInfo.UID] = lriToLockEntry(lockReqInfo, k, peerLock.Addr, rquorum, wquorum)
+						entryMap[lockReqInfo.UID] = lriToLockEntry(lockReqInfo, k, peerLock.Addr)
 					}
 				}
 			}
@@ -429,10 +428,7 @@ func (a adminAPIHandlers) TopLocksHandler(w http.ResponseWriter, r *http.Request
 
 	peerLocks := globalNotificationSys.GetLocks(ctx, r)
 
-	rquorum := getReadQuorum(objectAPI.SetDriveCount())
-	wquorum := getWriteQuorum(objectAPI.SetDriveCount())
-
-	topLocks := topLockEntries(peerLocks, rquorum, wquorum, stale)
+	topLocks := topLockEntries(peerLocks, stale)
 
 	// Marshal API response upto requested count.
 	if len(topLocks) > count && count > 0 {

cmd/erasure.go

@@ -178,7 +178,9 @@ func getDisksInfo(disks []StorageAPI, endpoints []string) (disksInfo []madmin.Di
 		}
 	}
 
-	if len(disksInfo) == rootDiskCount {
+	// Count offline disks as well to ensure consistent
+	// reportability of offline drives on local setups.
+	if len(disksInfo) == (rootDiskCount + offlineDisks.Sum()) {
 		// Success.
 		return disksInfo, errs, onlineDisks, offlineDisks
 	}

cmd/local-locker.go

@@ -35,6 +35,8 @@ type lockRequesterInfo struct {
 	// Owner represents the UUID of the owner who originally requested the lock
 	// useful in expiry.
 	Owner string
+	// Quorum represents the quorum required for this lock to be active.
+	Quorum int
 }
 
 // isWriteLock returns whether the lock is a write or read lock.
@@ -96,6 +98,7 @@ func (l *localLocker) Lock(ctx context.Context, args dsync.LockArgs) (reply bool
 				UID:           args.UID,
 				Timestamp:     UTCNow(),
 				TimeLastCheck: UTCNow(),
+				Quorum:        args.Quorum,
 			},
 		}
 	}
@@ -153,6 +156,7 @@ func (l *localLocker) RLock(ctx context.Context, args dsync.LockArgs) (reply boo
 		UID:           args.UID,
 		Timestamp:     UTCNow(),
 		TimeLastCheck: UTCNow(),
+		Quorum:        args.Quorum,
 	}
 	resource := args.Resources[0]
 	if lri, ok := l.lockMap[resource]; ok {

cmd/lock-rest-client.go

@@ -23,6 +23,7 @@ import (
 	"errors"
 	"io"
 	"net/url"
+	"strconv"
 
 	"github.com/minio/minio/cmd/http"
 	xhttp "github.com/minio/minio/cmd/http"
@@ -93,6 +94,7 @@ func (client *lockRESTClient) restCall(ctx context.Context, call string, args ds
 	values.Set(lockRESTUID, args.UID)
 	values.Set(lockRESTOwner, args.Owner)
 	values.Set(lockRESTSource, args.Source)
+	values.Set(lockRESTQuorum, strconv.Itoa(args.Quorum))
 	var buffer bytes.Buffer
 	for _, resource := range args.Resources {
 		buffer.WriteString(resource)

cmd/lock-rest-server-common.go

@@ -21,7 +21,7 @@ import (
 )
 
 const (
-	lockRESTVersion       = "v3"
+	lockRESTVersion       = "v4" // Add Quorum query param
 	lockRESTVersionPrefix = SlashSeparator + lockRESTVersion
 	lockRESTPrefix        = minioReservedBucketPath + "/lock"
 )
@@ -43,6 +43,10 @@ const (
 	// Source contains the line number, function and file name of the code
 	// on the client node that requested the lock.
 	lockRESTSource = "source"
+
+	// Quroum value to be saved along lock requester info, useful
+	// in verifying stale locks
+	lockRESTQuorum = "quorum"
 )
 
 var (

cmd/lock-rest-server.go

@@ -24,6 +24,7 @@ import (
 	"net/http"
 	"path"
 	"sort"
+	"strconv"
 	"time"
 
 	"github.com/gorilla/mux"
@@ -32,10 +33,10 @@ import (
 
 const (
 	// Lock maintenance interval.
-	lockMaintenanceInterval = 30 * time.Second
+	lockMaintenanceInterval = 15 * time.Second
 
 	// Lock validity check interval.
-	lockValidityCheckInterval = 30 * time.Second
+	lockValidityCheckInterval = 5 * time.Second
 )
 
 // To abstract a node over network.
@@ -63,10 +64,16 @@ func (l *lockRESTServer) IsValid(w http.ResponseWriter, r *http.Request) bool {
 }
 
 func getLockArgs(r *http.Request) (args dsync.LockArgs, err error) {
+	quorum, err := strconv.Atoi(r.URL.Query().Get(lockRESTQuorum))
+	if err != nil {
+		return args, err
+	}
+
 	args = dsync.LockArgs{
 		Owner:  r.URL.Query().Get(lockRESTOwner),
 		UID:    r.URL.Query().Get(lockRESTUID),
 		Source: r.URL.Query().Get(lockRESTSource),
+		Quorum: quorum,
 	}
 
 	var resources []string
@@ -277,12 +284,8 @@ func lockMaintenance(ctx context.Context, interval time.Duration) error {
 	for lendpoint, nlrips := range getLongLivedLocks(interval) {
 		nlripsMap := make(map[string]nlock, len(nlrips))
 		for _, nlrip := range nlrips {
-			// Locks are only held on first zone, make sure that
-			// we only look for ownership of locks from endpoints
-			// on first zone.
 			for _, c := range allLockersFn() {
 				if !c.IsOnline() || c == nil {
-					updateNlocks(nlripsMap, nlrip.name, nlrip.lri.Writer)
 					continue
 				}
 
@@ -306,14 +309,8 @@ func lockMaintenance(ctx context.Context, interval time.Duration) error {
 				}
 			}
 
-			// Read locks we assume quorum for be N/2 success
-			quorum := getReadQuorum(objAPI.SetDriveCount())
-			if nlrip.lri.Writer {
-				quorum = getWriteQuorum(objAPI.SetDriveCount())
-			}
-
 			// less than the quorum, we have locks expired.
-			if nlripsMap[nlrip.name].locks < quorum {
+			if nlripsMap[nlrip.name].locks < nlrip.lri.Quorum {
 				// The lock is no longer active at server that originated
 				// the lock, attempt to remove the lock.
 				globalLockServers[lendpoint].mutex.Lock()