From ed5ed7e490577457c927156f311de8d8ffa3cbec Mon Sep 17 00:00:00 2001
From: Klaus Post <klauspost@gmail.com>
Date: Tue, 22 Oct 2024 14:10:34 -0700
Subject: [PATCH] Trace ILM errors (#20576)

Some paths would attempt transitions but in case of failures
no traces would be emitted.

Add traces (with errors) when transition operations fail.
---
 cmd/bucket-lifecycle.go | 15 ++++++++++-----
 cmd/data-scanner.go     |  9 ++++++---
 cmd/erasure-object.go   |  2 ++
 3 files changed, 18 insertions(+), 8 deletions(-)

diff --git a/cmd/bucket-lifecycle.go b/cmd/bucket-lifecycle.go
index 29c88c529..eef2ee6e7 100644
--- a/cmd/bucket-lifecycle.go
+++ b/cmd/bucket-lifecycle.go
@@ -71,7 +71,7 @@ func NewLifecycleSys() *LifecycleSys {
 	return &LifecycleSys{}
 }
 
-func ilmTrace(startTime time.Time, duration time.Duration, oi ObjectInfo, event string, metadata map[string]string) madmin.TraceInfo {
+func ilmTrace(startTime time.Time, duration time.Duration, oi ObjectInfo, event string, metadata map[string]string, err string) madmin.TraceInfo {
 	sz, _ := oi.GetActualSize()
 	return madmin.TraceInfo{
 		TraceType: madmin.TraceILM,
@@ -81,18 +81,22 @@ func ilmTrace(startTime time.Time, duration time.Duration, oi ObjectInfo, event
 		Duration:  duration,
 		Path:      pathJoin(oi.Bucket, oi.Name),
 		Bytes:     sz,
-		Error:     "",
+		Error:     err,
 		Message:   getSource(4),
 		Custom:    metadata,
 	}
 }
 
-func (sys *LifecycleSys) trace(oi ObjectInfo) func(event string, metadata map[string]string) {
+func (sys *LifecycleSys) trace(oi ObjectInfo) func(event string, metadata map[string]string, err error) {
 	startTime := time.Now()
-	return func(event string, metadata map[string]string) {
+	return func(event string, metadata map[string]string, err error) {
 		duration := time.Since(startTime)
 		if globalTrace.NumSubscribers(madmin.TraceILM) > 0 {
-			globalTrace.Publish(ilmTrace(startTime, duration, oi, event, metadata))
+			e := ""
+			if err != nil {
+				e = err.Error()
+			}
+			globalTrace.Publish(ilmTrace(startTime, duration, oi, event, metadata, e))
 		}
 	}
 }
@@ -362,6 +366,7 @@ func (es *expiryState) Worker(input <-chan expiryOp) {
 				err := deleteObjectFromRemoteTier(es.ctx, oi.TransitionedObject.Name, oi.TransitionedObject.VersionID, oi.TransitionedObject.Tier)
 				if ignoreNotFoundErr(err) != nil {
 					transitionLogIf(es.ctx, err)
+					traceFn(ILMFreeVersionDelete, nil, err)
 					return
 				}
 
diff --git a/cmd/data-scanner.go b/cmd/data-scanner.go
index 3604d2281..1b9acee09 100644
--- a/cmd/data-scanner.go
+++ b/cmd/data-scanner.go
@@ -1325,10 +1325,13 @@ func applyExpiryOnNonTransitionedObjects(ctx context.Context, objLayer ObjectLay
 	dobj, err = objLayer.DeleteObject(ctx, obj.Bucket, encodeDirObject(obj.Name), opts)
 	if err != nil {
 		if isErrObjectNotFound(err) || isErrVersionNotFound(err) {
+			traceFn(ILMExpiry, nil, nil)
 			return false
 		}
 		// Assume it is still there.
-		ilmLogOnceIf(ctx, fmt.Errorf("DeleteObject(%s, %s): %w", obj.Bucket, obj.Name, err), "non-transition-expiry"+obj.Name)
+		err := fmt.Errorf("DeleteObject(%s, %s): %w", obj.Bucket, obj.Name, err)
+		ilmLogOnceIf(ctx, err, "non-transition-expiry"+obj.Name)
+		traceFn(ILMExpiry, nil, err)
 		return false
 	}
 	if dobj.Name == "" {
@@ -1549,7 +1552,7 @@ const (
 	ILMTransition = " ilm:transition"
 )
 
-func auditLogLifecycle(ctx context.Context, oi ObjectInfo, event string, tags map[string]string, traceFn func(event string, metadata map[string]string)) {
+func auditLogLifecycle(ctx context.Context, oi ObjectInfo, event string, tags map[string]string, traceFn func(event string, metadata map[string]string, err error)) {
 	var apiName string
 	switch event {
 	case ILMExpiry:
@@ -1567,5 +1570,5 @@ func auditLogLifecycle(ctx context.Context, oi ObjectInfo, event string, tags ma
 		VersionID: oi.VersionID,
 		Tags:      tags,
 	})
-	traceFn(event, tags)
+	traceFn(event, tags, nil)
 }
diff --git a/cmd/erasure-object.go b/cmd/erasure-object.go
index 367f489e9..8ee2ae6dd 100644
--- a/cmd/erasure-object.go
+++ b/cmd/erasure-object.go
@@ -2365,6 +2365,7 @@ func (er erasureObjects) TransitionObject(ctx context.Context, bucket, object st
 
 	destObj, err := genTransitionObjName(bucket)
 	if err != nil {
+		traceFn(ILMTransition, nil, err)
 		return err
 	}
 
@@ -2378,6 +2379,7 @@ func (er erasureObjects) TransitionObject(ctx context.Context, bucket, object st
 	rv, err = tgtClient.Put(ctx, destObj, pr, fi.Size)
 	pr.CloseWithError(err)
 	if err != nil {
+		traceFn(ILMTransition, nil, err)
 		return err
 	}
 	fi.TransitionStatus = lifecycle.TransitionComplete