*: keep tombstone if revision == compactAtRev

Before this patch, the tombstone can be deleted if its revision is equal
compacted revision. It causes that the watch subscriber won't get this
DELETE event. Based on Compact API[1], we should keep tombstone revision
if it's not less than the compaction revision.

> CompactionRequest compacts the key-value store up to a given revision.
> All superseded keys with a revision less than the compaction revision
> will be removed.

[1]: https://etcd.io/docs/latest/dev-guide/api_reference_v3/

Signed-off-by: Wei Fu <fuweid89@gmail.com>

server/storage/mvcc/hash.go

@@ -63,6 +63,9 @@ func (h *kvHasher) WriteKeyValue(k, v []byte) {
 	if !upper.GreaterThan(kr) {
 		return
 	}
+
+	isTombstone := BytesToBucketKey(k).tombstone
+
 	lower := Revision{Main: h.compactRevision + 1}
 	// skip revisions that are scheduled for deletion
 	// due to compacting; don't skip if there isn't one.
@@ -71,6 +74,16 @@ func (h *kvHasher) WriteKeyValue(k, v []byte) {
 			return
 		}
 	}
+
+	// NOTE: If the previous compacted revision was tombstone, existing
+	// releases already deleted that key(s). However, new releases won't
+	// delete the key(s) on compacted revision, no matter what type key it
+	// is. So, we should skip previous compacted revision to keep consistent
+	// with existing releases.
+	if kr.Main == h.compactRevision && isTombstone {
+		return
+	}
+
 	h.hash.Write(k)
 	h.hash.Write(v)
 }