metasync: amend GFN notifications

* always notify via metasync-post * add extra checks when _not_ to Signed-off-by: Alex Aizman <[email protected]>
NVIDIA · Jul 23, 2024 · 1a01903 · 1a01903
1 parent 162dfef
commit 1a01903
Show file tree

Hide file tree

Showing 2 changed files with 15 additions and 7 deletions.
diff --git a/ais/metasync.go b/ais/metasync.go
@@ -94,6 +94,8 @@ const (
 
 const failsync = "failing to sync"
 
+const retryConnRefused = 4
+
 type (
 	revs interface {
 		tag() string         // enum { revsSmapTag, ... }
@@ -218,7 +220,7 @@ func (y *metasyncer) Stop(err error) {
 func (y *metasyncer) notify(wait bool, pair revsPair) (failedCnt int) {
 	var (
 		failedCntAtomic = atomic.NewInt32(0)
-		req             = revsReq{pairs: []revsPair{pair}}
+		req             = revsReq{pairs: []revsPair{pair}, reqType: reqNotify}
 	)
 	if y.isPrimary() != nil {
 		return
@@ -227,7 +229,6 @@ func (y *metasyncer) notify(wait bool, pair revsPair) (failedCnt int) {
 		req.wg = &sync.WaitGroup{}
 		req.wg.Add(1)
 		req.failedCnt = failedCntAtomic
-		req.reqType = reqNotify
 	}
 	y.workCh <- req
 
@@ -369,7 +370,7 @@ func (y *metasyncer) do(pairs []revsPair, reqT int) (failedCnt int) {
 	freeBcastRes(results)
 	// step: handle connection-refused right away
 	lr := len(refused)
-	for range 4 { // retry
+	for range retryConnRefused {
 		if len(refused) == 0 {
 			if lr > 0 {
 				nlog.Infof("%s: %d node%s sync-ed", y.p, lr, cos.Plural(lr))

diff --git a/ais/prxclu.go b/ais/prxclu.go
@@ -729,7 +729,7 @@ func (p *proxy) mcastJoined(nsi *meta.Snode, msg *apc.ActMsg, flags cos.BitFlags
 		interrupted: regReq.Flags.IsSet(cos.RebalanceInterrupted),
 		restarted:   regReq.Flags.IsSet(cos.Restarted),
 	}
-	if err = p._earlyGFN(ctx, ctx.nsi); err != nil {
+	if err = p._earlyGFN(ctx, ctx.nsi, msg.Action, true /*joining*/); err != nil {
 		return
 	}
 	if err = p.owner.smap.modify(ctx); err != nil {
@@ -755,10 +755,10 @@ func (p *proxy) mcastJoined(nsi *meta.Snode, msg *apc.ActMsg, flags cos.BitFlags
 	return
 }
 
-func (p *proxy) _earlyGFN(ctx *smapModifier, si *meta.Snode /*being added or removed*/) error {
+func (p *proxy) _earlyGFN(ctx *smapModifier, si *meta.Snode, action string, joining bool) error {
 	smap := p.owner.smap.get()
 	if !smap.isPrimary(p.si) {
-		return newErrNotPrimary(p.si, smap, fmt.Sprintf("cannot add %s", si))
+		return newErrNotPrimary(p.si, smap, "cannot "+action+" "+si.StringEx())
 	}
 	if si.IsProxy() {
 		return nil
@@ -770,6 +770,13 @@ func (p *proxy) _earlyGFN(ctx *smapModifier, si *meta.Snode /*being added or rem
 		return err
 	}
 
+	if smap.CountActiveTs() == 0 {
+		return nil
+	}
+	if !joining && smap.CountActiveTs() == 1 {
+		return nil
+	}
+
 	// early-GFN notification with an empty (version-only and not yet updated) Smap and
 	// message(new target's ID)
 	msg := p.newAmsgActVal(apc.ActStartGFN, nil)
@@ -1539,7 +1546,7 @@ func (p *proxy) mcastMaint(msg *apc.ActMsg, si *meta.Snode, reb, maintPostReb bo
 		msg:     msg,
 		skipReb: !reb,
 	}
-	if err = p._earlyGFN(ctx, si); err != nil {
+	if err = p._earlyGFN(ctx, si, msg.Action, false /*joining*/); err != nil {
 		return
 	}
 	if err = p.owner.smap.modify(ctx); err != nil {