s3: route versioned PutObject finalize off the DLM (#9631)

s3: route versioned PutObject finalize off the distributed lock

A versioned write's finalize (flip the .versions pointer to the newest version,
demote the prior latest) now runs as a single RECOMPUTE_LATEST ObjectTransaction
on the object's owner filer, under its per-path lock, instead of the unserialized
updateLatestVersionInDirectory. The version file is written first; the owner
re-derives the pointer by scanning the directory.

RECOMPUTE_LATEST gains size_to_key / mtime_to_key to cache the chosen version's
size and mtime on the pointer, and demote_key / demote_value to stamp the
displaced prior latest (NoncurrentSinceNs for lifecycle) when the pointer moves.

Falls back to updateLatestVersionInDirectory when no owner is known yet.

weed/server/filer_grpc_server.go

@@ -8,6 +8,7 @@ import (
 	"math"
 	"os"
 	"path/filepath"
+	"strconv"
 	"time"
 
 	"github.com/seaweedfs/seaweedfs/weed/cluster"
@@ -411,6 +412,12 @@ func (fs *FilerServer) applyRecomputeLatest(ctx context.Context, m *filer_pb.Obj
 		pointer.Extended = make(map[string][]byte)
 	}
 
+	// Remember the prior chosen child so it can be demoted once the pointer moves.
+	var priorName string
+	if rc.NameToKey != "" {
+		priorName = string(pointer.Extended[rc.NameToKey])
+	}
+
 	// The store streams entries ascending by name. For the lowest-name pick we
 	// only need the first entry, so cap the listing at one; for the highest-name
 	// pick we must scan all and keep the last (the store has no reverse order).
@@ -427,12 +434,15 @@ func (fs *FilerServer) applyRecomputeLatest(ctx context.Context, m *filer_pb.Obj
 		return listErr
 	}
 
+	cleared := []string{rc.NameToKey, rc.SizeToKey, rc.MtimeToKey}
 	if chosen == nil {
 		for pointerKey := range rc.CopyExtended {
 			delete(pointer.Extended, pointerKey)
 		}
-		if rc.NameToKey != "" {
-			delete(pointer.Extended, rc.NameToKey)
+		for _, k := range cleared {
+			if k != "" {
+				delete(pointer.Extended, k)
+			}
 		}
 	} else {
 		for pointerKey, sourceKey := range rc.CopyExtended {
@@ -445,9 +455,39 @@ func (fs *FilerServer) applyRecomputeLatest(ctx context.Context, m *filer_pb.Obj
 		if rc.NameToKey != "" {
 			pointer.Extended[rc.NameToKey] = []byte(chosen.Name())
 		}
+		if rc.SizeToKey != "" {
+			pointer.Extended[rc.SizeToKey] = []byte(strconv.FormatUint(chosen.FileSize, 10))
+		}
+		if rc.MtimeToKey != "" {
+			pointer.Extended[rc.MtimeToKey] = []byte(strconv.FormatInt(chosen.Mtime.Unix(), 10))
+		}
 	}
 
-	return fs.filer.UpdateEntry(ctx, pointer, pointer)
+	if err := fs.filer.UpdateEntry(ctx, pointer, pointer); err != nil {
+		return err
+	}
+
+	// Stamp the displaced prior child (e.g. NoncurrentSinceNs for lifecycle).
+	newName := ""
+	if chosen != nil {
+		newName = chosen.Name()
+	}
+	if rc.DemoteKey != "" && priorName != "" && priorName != newName {
+		priorEntry, perr := fs.filer.FindEntry(ctx, util.NewFullPath(rc.ScanDir, priorName))
+		if perr == filer_pb.ErrNotFound {
+			return nil
+		}
+		if perr != nil {
+			return perr
+		}
+		if priorEntry.Extended == nil {
+			priorEntry.Extended = make(map[string][]byte)
+		}
+		priorEntry.Extended[rc.DemoteKey] = rc.DemoteValue
+		return fs.filer.UpdateEntry(ctx, priorEntry, priorEntry)
+	}
+
+	return nil
 }
 
 func (fs *FilerServer) UpdateEntry(ctx context.Context, req *filer_pb.UpdateEntryRequest) (*filer_pb.UpdateEntryResponse, error) {

weed/server/filer_grpc_server_object_txn_test.go

@@ -2,6 +2,7 @@ package weed_server
 
 import (
 	"context"
+	"strconv"
 	"testing"
 	"time"
 
@@ -431,3 +432,61 @@ func TestObjectTransactionIdempotentNoops(t *testing.T) {
 		t.Fatalf("no-op mutations should not error: %q", resp.Error)
 	}
 }
+
+// RECOMPUTE_LATEST copies the chosen child's size/mtime to the pointer and
+// stamps the demote key on the prior latest when the pointer moves.
+func TestObjectTransactionRecomputeDemoteAndAttrs(t *testing.T) {
+	t0 := time.Unix(1700000000, 0)
+	t1 := time.Unix(1700000100, 0)
+	mk := func(inode uint64, mt time.Time, size uint64, id string) *filer.Entry {
+		return &filer.Entry{
+			Attr:     filer.Attr{Inode: inode, Mtime: mt, Crtime: mt, Mode: 0644, FileSize: size},
+			Extended: map[string][]byte{"vid": []byte(id)},
+		}
+	}
+	fs, store := newTxnTestServer(map[string]*filer.Entry{
+		"/buckets/b/obj/.versions": {
+			Attr:     filer.Attr{Inode: 2, Mtime: t0, Crtime: t0, Mode: 0755 | (1 << 31)},
+			Extended: map[string][]byte{"latestName": []byte("v1.ver"), "latestVid": []byte("v1")},
+		},
+		"/buckets/b/obj/.versions/v1.ver": mk(10, t0, 100, "v1"),
+		"/buckets/b/obj/.versions/v2.ver": mk(11, t1, 250, "v2"),
+	})
+
+	resp, err := fs.ObjectTransaction(context.Background(), &filer_pb.ObjectTransactionRequest{
+		LockKey: "/buckets/b/obj",
+		Mutations: []*filer_pb.ObjectMutation{{
+			Type: filer_pb.ObjectMutation_RECOMPUTE_LATEST, Directory: "/buckets/b/obj", Name: ".versions",
+			Recompute: &filer_pb.Recompute{
+				ScanDir:      "/buckets/b/obj/.versions",
+				Descending:   true,
+				CopyExtended: map[string]string{"latestVid": "vid"},
+				NameToKey:    "latestName",
+				SizeToKey:    "latestSize",
+				MtimeToKey:   "latestMtime",
+				DemoteKey:    "noncurrentSince",
+				DemoteValue:  []byte("999"),
+			},
+		}},
+	})
+	if err != nil || resp.Error != "" {
+		t.Fatalf("txn failed: err=%v resp=%q", err, resp.Error)
+	}
+
+	ptr := store.entries["/buckets/b/obj/.versions"].Extended
+	if string(ptr["latestName"]) != "v2.ver" || string(ptr["latestVid"]) != "v2" {
+		t.Fatalf("pointer not moved to v2: name=%s vid=%s", ptr["latestName"], ptr["latestVid"])
+	}
+	if string(ptr["latestSize"]) != "250" {
+		t.Errorf("latestSize = %s, want 250", ptr["latestSize"])
+	}
+	if want := strconv.FormatInt(t1.Unix(), 10); string(ptr["latestMtime"]) != want {
+		t.Errorf("latestMtime = %s, want %s", ptr["latestMtime"], want)
+	}
+	if got := store.entries["/buckets/b/obj/.versions/v1.ver"].Extended["noncurrentSince"]; string(got) != "999" {
+		t.Errorf("prior latest v1.ver noncurrentSince = %q, want 999", got)
+	}
+	if _, ok := store.entries["/buckets/b/obj/.versions/v2.ver"].Extended["noncurrentSince"]; ok {
+		t.Errorf("new latest v2.ver should not be demoted")
+	}
+}