(list-objects, scrub): version-changed, version-removed

* scenarios: with and without `--cached`, `--check-versions` * scrub to ask for version; drop atime * core: `lom.CheckRemoteMD` will now always return remote meta - including version and custom * x-lso: - lso entry: add head-fail flag; refactor - wanted-lso: do not override remote custom - when checking remote version: - use custom and version from lom.CheckRemoteMD (above) - but only if err nil Signed-off-by: Alex Aizman <[email protected]>
NVIDIA · Jan 15, 2025 · 939d7a6 · 939d7a6
1 parent 476399d
commit 939d7a6
Show file tree

Hide file tree

Showing 9 changed files with 83 additions and 55 deletions.
diff --git a/api/apc/lsmsg.go b/api/apc/lsmsg.go
@@ -1,6 +1,6 @@
 // Package apc: API control messages and constants
 /*
- * Copyright (c) 2018-2024, NVIDIA CORPORATION. All rights reserved.
+ * Copyright (c) 2018-2025, NVIDIA CORPORATION. All rights reserved.
  */
 package apc
 
@@ -91,7 +91,7 @@ const (
 )
 
 const (
-	// Status
+	// location _status_
 	LocOK = iota
 	LocMisplacedNode
 	LocMisplacedMountpath
@@ -105,6 +105,8 @@ const (
 	EntryIsArchive  = 1 << (EntryStatusBits + 4)
 	EntryVerChanged = 1 << (EntryStatusBits + 5) // see also: QparamLatestVer, et al.
 	EntryVerRemoved = 1 << (EntryStatusBits + 6) // ditto
+	// added v3.26
+	EntryHeadFail = 1 << (EntryStatusBits + 7)
 )
 
 // ObjEntry.Flags field

diff --git a/cmd/cli/cli/scrub.go b/cmd/cli/cli/scrub.go
@@ -254,7 +254,7 @@ func (ctx *scrCtx) ls(bck cmn.Bck) (*scrBp, error) {
 		}
 	)
 	scr.Cname = bck.Cname("")
-	propNames := []string{apc.GetPropsName, apc.GetPropsSize, apc.GetPropsAtime, apc.GetPropsCopies, apc.GetPropsLocation, apc.GetPropsCustom}
+	propNames := []string{apc.GetPropsName, apc.GetPropsSize, apc.GetPropsVersion, apc.GetPropsCopies, apc.GetPropsLocation, apc.GetPropsCustom}
 	if bck.IsRemote() {
 		lsmsg.Flags |= apc.LsVerChanged
 		lsmsg.AddProps(propNames...)

diff --git a/cmd/cli/teb/scrub.go b/cmd/cli/teb/scrub.go
@@ -23,8 +23,8 @@ const (
 	colMissingCp     = "MISSING-CP"
 	colSmallSz       = "SMALL"
 	colLargeSz       = "LARGE"
-	colVchanged      = "VERSION-CHANGED"
-	colVremoved      = "VERSION-REMOVED"
+	colVchanged      = "VER-CHANGED"
+	colVremoved      = "VER-REMOVED"
 )
 
 const (

diff --git a/cmn/objattrs.go b/cmn/objattrs.go
@@ -253,8 +253,11 @@ func (oa *ObjAttrs) FromHeader(hdr http.Header) (cksum *cos.Cksum) {
 // b) the same remote "source" and at least one matching checksum, or c) two matching checksums.
 // (See also note below.)
 //
-// Note that mismatch in any given checksum type immediately renders inequality and return
-// from the function.
+// Note that ETag, checksum, or version mismatch leads to immediate return with error
+// specifying the exact cause.
+//
+// Note version comparison may fail even when the objects are identical, content-wise:
+// same size, ETag, and checksums may still "co-exist" with different versions.
 //
 // TODO: count == 1 with matching checksum being xxhash - must be configurable :NOTE
 func (oa *ObjAttrs) CheckEq(rem cos.OAH) error {

diff --git a/cmn/objlist_utils.go b/cmn/objlist_utils.go
@@ -1,7 +1,7 @@
 // Package cmn provides common constants, types, and utilities for AIS clients
 // and AIStore.
 /*
- * Copyright (c) 2018-2024, NVIDIA CORPORATION. All rights reserved.
+ * Copyright (c) 2018-2025, NVIDIA CORPORATION. All rights reserved.
  */
 package cmn
 
@@ -67,11 +67,8 @@ func appSorted(entries LsoEntries, ne *LsoEnt) LsoEntries {
 func (be *LsoEnt) IsPresent() bool { return be.Flags&apc.EntryIsCached != 0 }
 func (be *LsoEnt) SetPresent()     { be.Flags |= apc.EntryIsCached }
 
-// see also: "latest-ver", QparamLatestVer, et al.
-func (be *LsoEnt) SetVerChanged()     { be.Flags |= apc.EntryVerChanged }
-func (be *LsoEnt) IsVerChanged() bool { return be.Flags&apc.EntryVerChanged != 0 }
-func (be *LsoEnt) SetVerRemoved()     { be.Flags |= apc.EntryVerRemoved }
-func (be *LsoEnt) IsVerRemoved() bool { return be.Flags&apc.EntryVerRemoved != 0 }
+func (be *LsoEnt) SetFlag(fl uint16)           { be.Flags |= fl }
+func (be *LsoEnt) IsAnyFlagSet(fl uint16) bool { return be.Flags&fl != 0 }
 
 func (be *LsoEnt) IsStatusOK() bool   { return be.Status() == 0 }
 func (be *LsoEnt) Status() uint16     { return be.Flags & apc.EntryStatusMask }

diff --git a/core/ldp.go b/core/ldp.go
@@ -1,6 +1,6 @@
 // Package core provides core metadata and in-cluster API
 /*
- * Copyright (c) 2018-2024, NVIDIA CORPORATION. All rights reserved.
+ * Copyright (c) 2018-2025, NVIDIA CORPORATION. All rights reserved.
  */
 package core
 
@@ -143,7 +143,7 @@ func (lom *LOM) CheckRemoteMD(locked, sync bool, origReq *http.Request) (res CRM
 
 	if !locked {
 		// return info (neq and, possibly, not-found), and be done
-		return CRMD{ErrCode: ecode, Err: err}
+		return CRMD{ObjAttrs: oa, ErrCode: ecode, Err: err}
 	}
 
 	// rm remotely-deleted
@@ -160,7 +160,7 @@ func (lom *LOM) CheckRemoteMD(locked, sync bool, origReq *http.Request) (res CRM
 	}
 
 	lom.Uncache()
-	return CRMD{ErrCode: ecode, Err: err}
+	return CRMD{ObjAttrs: oa, ErrCode: ecode, Err: err}
 }
 
 // NOTE: Sync is false (ie., not deleting)

diff --git a/xact/xs/nextpage.go b/xact/xs/nextpage.go
@@ -1,7 +1,7 @@
 // Package xs is a collection of eXtended actions (xactions), including multi-object
 // operations, list-objects, (cluster) rebalance and (target) resilver, ETL, and more.
 /*
- * Copyright (c) 2018-2024, NVIDIA CORPORATION. All rights reserved.
+ * Copyright (c) 2018-2025, NVIDIA CORPORATION. All rights reserved.
  */
 package xs
 
@@ -112,19 +112,19 @@ func (npg *npgCtx) nextPageR(nentries cmn.LsoEntries, inclStatusLocalMD bool) (l
 
 func (npg *npgCtx) populate(lst *cmn.LsoRes) error {
 	post := npg.wi.lomVisitedCb
-	for _, obj := range lst.Entries {
-		if obj.IsDir() {
+	for _, en := range lst.Entries {
+		if en.IsDir() {
 			// collecting virtual dir-s when apc.LsNoRecursion is on - skipping here
 			continue
 		}
-		si, err := npg.wi.smap.HrwName2T(npg.bck.MakeUname(obj.Name))
+		si, err := npg.wi.smap.HrwName2T(npg.bck.MakeUname(en.Name))
 		if err != nil {
 			return err
 		}
 		if si.ID() != core.T.SID() {
 			continue
 		}
-		lom := core.AllocLOM(obj.Name)
+		lom := core.AllocLOM(en.Name)
 		if err := lom.InitBck(npg.bck.Bucket()); err != nil {
 			core.FreeLOM(lom)
 			if cmn.IsErrBucketNought(err) {
@@ -137,8 +137,8 @@ func (npg *npgCtx) populate(lst *cmn.LsoRes) error {
 			continue
 		}
 
-		npg.wi.setWanted(obj, lom)
-		obj.SetPresent()
+		npg.wi.setWanted(en, lom)
+		en.SetPresent()
 
 		if post != nil {
 			post(lom)

diff --git a/xact/xs/wanted_lso.go b/xact/xs/wanted_lso.go
@@ -1,7 +1,7 @@
 // Package xs contains most of the supported eXtended actions (xactions) with some
 // exceptions that include certain storage services (mirror, EC) and extensions (downloader, lru).
 /*
- * Copyright (c) 2022-2024, NVIDIA CORPORATION. All rights reserved.
+ * Copyright (c) 2022-2025, NVIDIA CORPORATION. All rights reserved.
  */
 package xs
 
@@ -35,7 +35,10 @@ func wanted(msg *apc.LsoMsg) (flags cos.BitFlags) {
 	return
 }
 
-func (wi *walkInfo) setWanted(e *cmn.LsoEnt, lom *core.LOM) {
+func (wi *walkInfo) setWanted(en *cmn.LsoEnt, lom *core.LOM) {
+	var (
+		checkVchanged = wi.msg.IsFlagSet(apc.LsVerChanged)
+	)
 	for name, fl := range allmap {
 		if !wi.wanted.IsSet(fl) {
 			continue
@@ -46,41 +49,51 @@ func (wi *walkInfo) setWanted(e *cmn.LsoEnt, lom *core.LOM) {
 		case apc.GetPropsCached: // via obj.SetPresent()
 
 		case apc.GetPropsSize:
-			if e.Size > 0 && lom.Lsize() != e.Size {
-				e.SetVerChanged()
+			if en.Size > 0 && lom.Lsize() != en.Size {
+				en.SetFlag(apc.EntryVerChanged)
 			}
-			e.Size = lom.Lsize()
+			en.Size = lom.Lsize()
 		case apc.GetPropsVersion:
-			e.Version = lom.Version()
+			// remote VersionObjMD takes precedence over ais incremental numbering
+			if en.Version == "" {
+				en.Version = lom.Version()
+			}
 		case apc.GetPropsChecksum:
-			e.Checksum = lom.Checksum().Value()
+			en.Checksum = lom.Checksum().Value()
 		case apc.GetPropsAtime:
-			e.Atime = cos.FormatNanoTime(lom.AtimeUnix(), wi.msg.TimeFormat)
+			// atime vs remote LastModified
+			en.Atime = cos.FormatNanoTime(lom.AtimeUnix(), wi.msg.TimeFormat)
 		case apc.GetPropsLocation:
-			e.Location = lom.Location()
+			en.Location = lom.Location()
 		case apc.GetPropsCopies:
-			e.Copies = int16(lom.NumCopies())
+			en.Copies = int16(lom.NumCopies())
 
 		case apc.GetPropsEC:
-			// TODO?: risk of significant slow-down loading EC metafiles
+			// TODO at the risk of significant slow-down
 
 		case apc.GetPropsCustom:
-			if md := lom.GetCustomMD(); len(md) > 0 {
-				e.Custom = cmn.CustomMD2S(md)
+			// en.Custom is set via one of the two alternative flows:
+			// - checkRemoteMD => HEAD(obj)
+			// - backend.List* api call
+			if en.Custom == "" {
+				if md := lom.GetCustomMD(); len(md) > 0 {
+					en.Custom = cmn.CustomMD2S(md)
+					checkVchanged = false
+				}
 			}
 		default:
 			debug.Assert(false, name)
 		}
 	}
-	if wi.msg.IsFlagSet(apc.LsVerChanged) && !e.IsVerChanged() {
-		//
-		// slow path: extensive 'version-changed' check
-		//
-		cmn.S2CustomMD(wi.custom, e.Custom, e.Version)
+
+	// slow path: extensive 'version-changed' check
+	if checkVchanged && !en.IsAnyFlagSet(apc.EntryVerChanged|apc.EntryVerRemoved) {
+		cmn.S2CustomMD(wi.custom, en.Custom, en.Version)
 		if len(wi.custom) > 0 {
-			oa := cmn.ObjAttrs{Size: e.Size, CustomMD: wi.custom}
+			oa := cmn.ObjAttrs{Size: en.Size, CustomMD: wi.custom}
 			if lom.CheckEq(&oa) != nil {
-				e.SetVerChanged()
+				// lom.CheckEq returned err contains the cause
+				en.SetFlag(apc.EntryVerChanged)
 			}
 			clear(wi.custom)
 		}

diff --git a/xact/xs/wi_lso.go b/xact/xs/wi_lso.go
@@ -88,53 +88,66 @@ func (wi *walkInfo) match(objName string) bool {
 }
 
 // new entry to be added to the listed page (note: slow path)
-func (wi *walkInfo) ls(lom *core.LOM, status uint16) (e *cmn.LsoEnt) {
-	e = &cmn.LsoEnt{Name: lom.ObjName, Flags: status | apc.EntryIsCached}
+func (wi *walkInfo) ls(lom *core.LOM, status uint16) (en *cmn.LsoEnt) {
+	en = &cmn.LsoEnt{Name: lom.ObjName, Flags: status | apc.EntryIsCached}
 
 	if lom.IsFntl() {
 		orig := lom.OrigFntl()
 		if orig != nil {
 			saved := lom.PushFntl(orig)
 			if wi.msg.IsFlagSet(apc.LsVerChanged) {
-				checkRemoteMD(lom, e)
+				checkRemoteMD(lom, en)
 			}
 			lom.PopFntl(saved)
-			e.Name = orig[1]
+			en.Name = orig[1]
 		}
 	} else if wi.msg.IsFlagSet(apc.LsVerChanged) {
-		checkRemoteMD(lom, e)
+		// may set en.custom and en.version
+		checkRemoteMD(lom, en)
 	}
 	if wi.msg.IsFlagSet(apc.LsNameOnly) {
 		return
 	}
-	wi.setWanted(e, lom)
+
+	// fill out even more of `en`
+	wi.setWanted(en, lom)
+
 	wi.lomVisitedCb(lom)
 	return
 }
 
 // NOTE: slow path if lom.Bck is remote
-func checkRemoteMD(lom *core.LOM, e *cmn.LsoEnt) {
+func checkRemoteMD(lom *core.LOM, en *cmn.LsoEnt) {
 	res := lom.CheckRemoteMD(false /*locked*/, false /*sync*/, nil /*origReq*/)
+
 	switch {
 	case res.Eq:
 		debug.AssertNoErr(res.Err)
 	case cos.IsNotExist(res.Err, res.ErrCode):
-		e.SetVerRemoved()
+		en.SetFlag(apc.EntryVerRemoved)
+	case res.Err == nil:
+		en.SetFlag(apc.EntryVerChanged)
+
+		// expecting custom and version set
+		debug.Assert(len(res.ObjAttrs.CustomMD) > 0)
+		en.Custom = cmn.CustomMD2S(res.ObjAttrs.CustomMD)
+		debug.Assert(res.ObjAttrs.Ver != nil)
+		en.Version = *res.ObjAttrs.Ver
 	default:
-		e.SetVerChanged()
+		en.SetFlag(apc.EntryHeadFail)
 	}
 }
 
 // Performs a number of syscalls to load object metadata.
-func (wi *walkInfo) callback(fqn string, de fs.DirEntry) (entry *cmn.LsoEnt, err error) {
+func (wi *walkInfo) callback(fqn string, de fs.DirEntry) (en *cmn.LsoEnt, err error) {
 	if de.IsDir() {
 		return
 	}
 
 	lom := core.AllocLOM("")
-	entry, err = wi._cb(lom, fqn)
+	en, err = wi._cb(lom, fqn)
 	core.FreeLOM(lom)
-	return entry, err
+	return en, err
 }
 
 func (wi *walkInfo) _cb(lom *core.LOM, fqn string) (*cmn.LsoEnt, error) {