volume: skip directory fsync on Windows, report a failed makeupDiff (#10572)

* volume: skip directory fsync on Windows

* ci: run the windows jobs for the whole vacuum path

Both windows jobs start the same weed mini cluster, so both exercise the
volume server's vacuum path, but only one of them watched a single file
in it. Cover the compact, reconcile and load files in both.

* volume: report a failed makeupDiff instead of discarding it

The cleanup removes assigned to the same err the makeupDiff failure was
held in, so an aborted compaction returned nil once both removes
succeeded. The master then recorded the vacuum as committed and the
volume reloaded against the discarded generation.

* volume: correct the fsyncDir comments after the windows skip

Both comments described the old shape, where windows fell through to a
sync whose error was swallowed.

* volume: keep the makeupDiff failure ahead of its cleanup errors

A failed remove of .cpd/.cpx outranked the failure that abandoned the
compaction, so the caller saw the cleanup error instead of the cause.
Log it and return the original, matching the Rust do_commit_compact. A
leftover temp file is rolled back by reconcile on the next start.

.github/workflows/mount-windows-conformance.yml

@@ -6,6 +6,9 @@ on:
     paths:
       - 'weed/mount/**'
       - 'weed/command/mount*.go'
+      - 'weed/storage/volume_vacuum*.go'
+      - 'weed/storage/volume_loading.go'
+      - 'weed/storage/disk_location.go'
       - 'test/winfsp-conformance/**'
       - '.github/workflows/mount-windows-conformance.yml'
   # No base branch filter: this is the only thing that runs the Windows mount,
@@ -14,6 +17,9 @@ on:
     paths:
       - 'weed/mount/**'
       - 'weed/command/mount*.go'
+      - 'weed/storage/volume_vacuum*.go'
+      - 'weed/storage/volume_loading.go'
+      - 'weed/storage/disk_location.go'
       - 'test/winfsp-conformance/**'
       - '.github/workflows/mount-windows-conformance.yml'
 

.github/workflows/mount-windows.yml

@@ -6,6 +6,9 @@ on:
     paths:
       - 'weed/mount/**'
       - 'weed/command/mount*.go'
+      - 'weed/storage/volume_vacuum*.go'
+      - 'weed/storage/volume_loading.go'
+      - 'weed/storage/disk_location.go'
       - 'test/winfsp/**'
       - '.github/workflows/mount-windows.yml'
   # No base branch filter: this is the only thing that runs the Windows mount,
@@ -14,6 +17,9 @@ on:
     paths:
       - 'weed/mount/**'
       - 'weed/command/mount*.go'
+      - 'weed/storage/volume_vacuum*.go'
+      - 'weed/storage/volume_loading.go'
+      - 'weed/storage/disk_location.go'
       - 'test/winfsp/**'
       - '.github/workflows/mount-windows.yml'
 

seaweed-volume/src/storage/volume.rs

@@ -3627,7 +3627,7 @@ fn get_append_at_ns(last: u64) -> u64 {
 /// durable, propagating a sync failure so the commit path can abort rather than
 /// proceed with an undurable rename or marker. A path with no openable parent is
 /// tolerated; directory fsync is unsupported on Windows, so it is a no-op there
-/// (matching the Go fsyncDir helper, which ignores that error).
+/// (matching the Go fsyncDir helper, which skips it too).
 pub(crate) fn fsync_dir(path: &str) -> io::Result<()> {
     #[cfg(windows)]
     {

weed/storage/volume_vacuum.go

@@ -187,36 +187,36 @@ func (v *Volume) CommitCompact() error {
 	v.DataBackend = nil
 	stats.VolumeServerVolumeGauge.WithLabelValues(v.Collection, "volume").Dec()
 
-	var e error
-	if e = v.makeupDiff(v.FileName(".cpd"), v.FileName(".cpx"), v.FileName(".dat"), v.FileName(".idx")); e != nil {
-		glog.V(0).Infof("makeupDiff in CommitCompact volume %d failed %v", v.Id, e)
-		e = os.Remove(v.FileName(".cpd"))
-		if e != nil {
-			return e
+	if compactErr := v.makeupDiff(v.FileName(".cpd"), v.FileName(".cpx"), v.FileName(".dat"), v.FileName(".idx")); compactErr != nil {
+		glog.V(0).Infof("makeupDiff in CommitCompact volume %d failed %v", v.Id, compactErr)
+		if e := os.Remove(v.FileName(".cpd")); e != nil && !os.IsNotExist(e) {
+			glog.V(0).Infof("remove %s: %v", v.FileName(".cpd"), e)
 		}
-		e = os.Remove(v.FileName(".cpx"))
-		if e != nil {
-			return e
-		}
-	} else {
-		// makeupDiff has fsynced the .cpd/.cpx contents. Persist a durable .cpc
-		// commit marker BEFORE renaming so the two renames are atomic across a
-		// crash: a marker on disk means the swap is decided and reconcile rolls
-		// forward; no marker means roll back. Without it, a crash between the
-		// two renames leaves a stale .idx that a later vacuum compacts to empty.
-		if e = v.writeCompactCommitMarker(); e != nil {
-			return e
-		}
-		if e = v.applyCompactSwap(); e != nil {
-			return e
+		if e := os.Remove(v.FileName(".cpx")); e != nil && !os.IsNotExist(e) {
+			glog.V(0).Infof("remove %s: %v", v.FileName(".cpx"), e)
 		}
+		// Report the abandoned compaction rather than a cleanup failure that
+		// reconcile rolls back anyway, and never fall through to the reload.
+		return compactErr
+	}
+
+	// makeupDiff has fsynced the .cpd/.cpx contents. Persist a durable .cpc
+	// commit marker BEFORE renaming so the two renames are atomic across a
+	// crash: a marker on disk means the swap is decided and reconcile rolls
+	// forward; no marker means roll back. Without it, a crash between the
+	// two renames leaves a stale .idx that a later vacuum compacts to empty.
+	if e := v.writeCompactCommitMarker(); e != nil {
+		return e
+	}
+	if e := v.applyCompactSwap(); e != nil {
+		return e
 	}
 
 	//glog.V(3).Infof("Pretending to be vacuuming...")
 	//time.Sleep(20 * time.Second)
 
 	glog.V(3).Infof("Loading volume %d commit file...", v.Id)
-	if e = v.load(true, false, v.needleMapKind, 0, v.Version()); e != nil {
+	if e := v.load(true, false, v.needleMapKind, 0, v.Version()); e != nil {
 		return e
 	}
 	glog.V(3).Infof("Finish committing volume %d", v.Id)
@@ -377,9 +377,13 @@ func (v *Volume) cleanupCompact() error {
 }
 
 // fsyncDir fsyncs a directory so a rename/create/unlink inside it is durable.
-// A failure to open the directory for sync is non-fatal on platforms that do
-// not support it.
+// Windows has no directory fsync, so the .cpc protocol leans on NTFS metadata
+// ordering there. An unopenable directory is tolerated, unlike util.FsyncDir,
+// because the caller has already closed the needle map.
 func fsyncDir(dir string) error {
+	if runtime.GOOS == "windows" {
+		return nil
+	}
 	d, err := os.Open(dir)
 	if err != nil {
 		return nil