refactor: align metrics field names with current behavior (#16826)

- Rename `db_size` to `raft_log_size`, which is the size of on-disk data
  of raft-log.
- Rename `key_num` to `snapshot_key_count`.

This commit updates field names in
- the the metrics API,
- and the cluster status gRPC API.

Author: drmingdrmer

src/meta/binaries/metactl/main.rs

@@ -236,8 +236,8 @@ impl App {
         let res = client.get_cluster_status().await?;
         println!("BinaryVersion: {}", res.binary_version);
         println!("DataVersion: {}", res.data_version);
-        println!("DBSize: {}", res.db_size);
-        println!("KeyNumber: {}", res.key_num);
+        println!("RaftLogSize: {}", res.raft_log_size);
+        println!("SnapshotKeyCount: {}", res.snapshot_key_count);
         println!("Node: id={} raft={}", res.id, res.endpoint);
         println!("State: {}", res.state);
         if let Some(leader) = res.leader {

src/meta/service/src/api/grpc/grpc_service.rs

@@ -462,8 +462,8 @@ impl MetaService for MetaServiceImpl {
             binary_version: status.binary_version,
             data_version: status.data_version.to_string(),
             endpoint: status.endpoint,
-            db_size: status.db_size,
-            key_num: status.key_num as u64,
+            raft_log_size: status.raft_log_size,
+            snapshot_key_count: status.snapshot_key_count as u64,
             state: status.state,
             is_leader: status.is_leader,
             current_term: status.current_term,

src/meta/service/src/meta_service/meta_node.rs

@@ -447,8 +447,8 @@ impl MetaNode {
                 server_metrics::set_last_seq(meta_node.get_last_seq().await);
 
                 // metrics about server storage
-                server_metrics::set_db_size(meta_node.get_db_size().await);
-                server_metrics::set_snapshot_key_num(meta_node.get_key_num().await);
+                server_metrics::set_raft_log_size(meta_node.get_raft_log_size().await);
+                server_metrics::set_snapshot_key_count(meta_node.get_snapshot_key_count().await);
 
                 last_leader = mm.current_leader;
             }
@@ -817,13 +817,14 @@ impl MetaNode {
         nodes
     }
 
-    async fn get_db_size(&self) -> u64 {
+    /// Get the size in bytes of the on disk files of the raft log storage.
+    async fn get_raft_log_size(&self) -> u64 {
         self.sto.log.read().await.on_disk_size()
     }
 
-    async fn get_key_num(&self) -> u64 {
+    async fn get_snapshot_key_count(&self) -> u64 {
         self.sto
-            .try_get_snapshot_key_num()
+            .try_get_snapshot_key_count()
             .await
             .unwrap_or_default()
     }
@@ -841,8 +842,8 @@ impl MetaNode {
 
         let endpoint = self.sto.get_node_raft_endpoint(&self.sto.id).await?;
 
-        let db_size = self.get_db_size().await;
-        let key_num = self.get_key_num().await;
+        let raft_log_size = self.get_raft_log_size().await;
+        let key_count = self.get_snapshot_key_count().await;
 
         let metrics = self.raft.metrics().borrow().clone();
 
@@ -859,8 +860,8 @@ impl MetaNode {
             binary_version: METASRV_COMMIT_VERSION.as_str().to_string(),
             data_version: DATA_VERSION,
             endpoint: endpoint.to_string(),
-            db_size,
-            key_num,
+            raft_log_size,
+            snapshot_key_count: key_count,
             state: format!("{:?}", metrics.state),
             is_leader: metrics.state == openraft::ServerState::Leader,
             current_term: metrics.current_term,

src/meta/service/src/meta_service/meta_node_status.rs

@@ -32,11 +32,11 @@ pub struct MetaNodeStatus {
     /// The raft service endpoint for internal communication
     pub endpoint: String,
 
-    /// The size in bytes of the on disk data.
-    pub db_size: u64,
+    /// The size in bytes of the raft-log on disk data.
+    pub raft_log_size: u64,
 
-    /// key number of current snapshot
-    pub key_num: u64,
+    /// Total number of keys in current snapshot
+    pub snapshot_key_count: u64,
 
     /// Server state, one of "Follower", "Learner", "Candidate", "Leader".
     pub state: String,

src/meta/service/src/metrics/meta_metrics.rs

@@ -52,8 +52,8 @@ pub mod server_metrics {
         node_is_health: Gauge,
         leader_changes: Counter,
         applying_snapshot: Gauge,
-        snapshot_key_num: Gauge,
-        db_size: Gauge,
+        snapshot_key_count: Gauge,
+        raft_log_size: Gauge,
         last_log_index: Gauge,
         last_seq: Gauge,
         current_term: Gauge,
@@ -73,8 +73,8 @@ pub mod server_metrics {
                 node_is_health: Gauge::default(),
                 leader_changes: Counter::default(),
                 applying_snapshot: Gauge::default(),
-                snapshot_key_num: Gauge::default(),
-                db_size: Gauge::default(),
+                snapshot_key_count: Gauge::default(),
+                raft_log_size: Gauge::default(),
                 last_log_index: Gauge::default(),
                 last_seq: Gauge::default(),
                 current_term: Gauge::default(),
@@ -109,11 +109,15 @@ pub mod server_metrics {
                 metrics.applying_snapshot.clone(),
             );
             registry.register(
-                key!("snapshot_key_num"),
-                "snapshot key numbers",
-                metrics.snapshot_key_num.clone(),
+                key!("snapshot_key_count"),
+                "number of keys in the last snapshot",
+                metrics.snapshot_key_count.clone(),
+            );
+            registry.register(
+                key!("raft_log_size"),
+                "the size in bytes of the on disk data of raft log",
+                metrics.raft_log_size.clone(),
             );
-            registry.register(key!("db_size"), "db size", metrics.db_size.clone());
             registry.register(
                 key!("proposals_applied"),
                 "proposals applied",
@@ -174,12 +178,12 @@ pub mod server_metrics {
         SERVER_METRICS.applying_snapshot.inc_by(cnt);
     }
 
-    pub fn set_snapshot_key_num(snapshot_key_num: u64) {
-        SERVER_METRICS.snapshot_key_num.set(snapshot_key_num as i64);
+    pub fn set_snapshot_key_count(n: u64) {
+        SERVER_METRICS.snapshot_key_count.set(n as i64);
     }
 
-    pub fn set_db_size(db_size: u64) {
-        SERVER_METRICS.db_size.set(db_size as i64);
+    pub fn set_raft_log_size(raft_log_size: u64) {
+        SERVER_METRICS.raft_log_size.set(raft_log_size as i64);
     }
 
     pub fn set_proposals_applied(proposals_applied: u64) {

src/meta/service/src/store/store_inner.rs

@@ -292,7 +292,7 @@ impl StoreInner {
     /// Return snapshot id and meta of the last snapshot.
     ///
     /// It returns None if there is no snapshot or there is an error parsing snapshot meta or id.
-    pub(crate) async fn try_get_snapshot_key_num(&self) -> Option<u64> {
+    pub(crate) async fn try_get_snapshot_key_count(&self) -> Option<u64> {
         let sm = self.state_machine.read().await;
         let db = sm.levels().persisted()?;
         Some(db.stat().key_num)

src/meta/types/proto/meta.proto

@@ -167,7 +167,7 @@ message ClusterStatus {
   string binary_version = 2;
   string data_version = 3;
   string endpoint = 4;
-  uint64 db_size = 5;
+  uint64 raft_log_size = 5;
   string state = 6;
   bool is_leader = 7;
   uint64 current_term = 8;
@@ -180,7 +180,7 @@ message ClusterStatus {
   repeated string voters = 15;
   repeated string non_voters = 16;
   uint64 last_seq = 17;
-  uint64 key_num = 18;
+  uint64 snapshot_key_count = 18;
 }
 
 message ClientInfo {