diff --git a/backend/hdfs/fs.go b/backend/hdfs/fs.go
index 7d48ffed6..8ba9d2366 100644
--- a/backend/hdfs/fs.go
+++ b/backend/hdfs/fs.go
@@ -21,6 +21,7 @@ import (
 	"github.com/rclone/rclone/fs/config/configmap"
 	"github.com/rclone/rclone/fs/config/configstruct"
 	"github.com/rclone/rclone/fs/hash"
+	"github.com/rclone/rclone/lib/pacer"
 )
 
 // Fs represents a HDFS server
@@ -31,8 +32,15 @@ type Fs struct {
 	opt      Options        // options for this backend
 	ci       *fs.ConfigInfo // global config
 	client   *hdfs.Client
+	pacer    *fs.Pacer // pacer for API calls
 }
 
+const (
+	minSleep      = 20 * time.Millisecond
+	maxSleep      = 10 * time.Second
+	decayConstant = 2 // bigger for slower decay, exponential
+)
+
 // copy-paste from https://github.com/colinmarc/hdfs/blob/master/cmd/hdfs/kerberos.go
 func getKerberosClient() (*krb.Client, error) {
 	configPath := os.Getenv("KRB5_CONFIG")
@@ -114,6 +122,7 @@ func NewFs(ctx context.Context, name, root string, m configmap.Mapper) (fs.Fs, e
 		opt:    *opt,
 		ci:     fs.GetConfig(ctx),
 		client: client,
+		pacer:  fs.NewPacer(ctx, pacer.NewDefault(pacer.MinSleep(minSleep), pacer.MaxSleep(maxSleep), pacer.DecayConstant(decayConstant))),
 	}
 
 	f.features = (&fs.Features{
diff --git a/backend/hdfs/object.go b/backend/hdfs/object.go
index 31c8282f8..a2c2c1087 100644
--- a/backend/hdfs/object.go
+++ b/backend/hdfs/object.go
@@ -5,10 +5,12 @@ package hdfs
 
 import (
 	"context"
+	"errors"
 	"io"
 	"path"
 	"time"
 
+	"github.com/colinmarc/hdfs/v2"
 	"github.com/rclone/rclone/fs"
 	"github.com/rclone/rclone/fs/hash"
 	"github.com/rclone/rclone/lib/readers"
@@ -141,7 +143,23 @@ func (o *Object) Update(ctx context.Context, in io.Reader, src fs.ObjectInfo, op
 		return err
 	}
 
-	err = out.Close()
+	// If the datanodes have acknowledged all writes but not yet
+	// to the namenode, FileWriter.Close can return ErrReplicating
+	// (wrapped in an os.PathError). This indicates that all data
+	// has been written, but the lease is still open for the file.
+	//
+	// It is safe in this case to either ignore the error (and let
+	// the lease expire on its own) or to call Close multiple
+	// times until it completes without an error. The Java client,
+	// for context, always chooses to retry, with exponential
+	// backoff.
+	err = o.fs.pacer.Call(func() (bool, error) {
+		err := out.Close()
+		if err == nil {
+			return false, nil
+		}
+		return errors.Is(err, hdfs.ErrReplicating), err
+	})
 	if err != nil {
 		cleanup()
 		return err