refactor(metrics): remove legacy _drops_total compatibility mirrors (security-union#690)

PR security-union#670 renamed gauges from videocall_*_drops_total to
videocall_*_drops but kept the old names as deprecated mirrors. Enough
time has passed for dashboards to migrate.

Remove DATAGRAM_DROPS_TOTAL and WEBSOCKET_DROPS_TOTAL gauge
registrations, their dual-emit sites, dual-cleanup code, and the
`or rate(...drops_total...)` fallback from the Grafana dashboard JSON.

Author: jboyd01

actix-api/src/bin/metrics_server.rs

@@ -52,7 +52,7 @@ use sec_api::metrics::{
     CLIENT_ACTIVE_SERVER_RTT_MS, CLIENT_INFO, CLIENT_LONGTASK_DURATION_MS,
     CLIENT_MEMORY_TOTAL_BYTES, CLIENT_MEMORY_USED_BYTES, CLIENT_PACKETS_RECEIVED_PER_SEC,
     CLIENT_PACKETS_SENT_PER_SEC, CLIENT_RENDER_FPS, CLIENT_SEND_QUEUE_BYTES, CLIENT_TAB_THROTTLED,
-    CLIENT_TAB_VISIBLE, DATAGRAM_DROPS, DATAGRAM_DROPS_TOTAL, DECODER_ERRORS_TOTAL,
+    CLIENT_TAB_VISIBLE, DATAGRAM_DROPS, DECODER_ERRORS_TOTAL,
     ENCODER_BITRATE_RATIO, ENCODER_FPS_RATIO, ENCODER_OUTPUT_FPS, ENCODER_P75_PEER_FPS,
     ENCODER_TARGET_BITRATE_KBPS, ENCODER_WORST_PEER_FPS, HEALTH_REPORTS_TOTAL,
     KEYFRAME_REQUESTS_SENT_TOTAL, MEETING_PARTICIPANTS, NETEQ_ACCELERATE_OPS_PER_SEC,
@@ -61,7 +61,7 @@ use sec_api::metrics::{
     PEER_AUDIO_ENABLED, PEER_CAN_LISTEN, PEER_CAN_SEE, PEER_CONNECTIONS_TOTAL, PEER_VIDEO_ENABLED,
     SCREEN_SHARING_ACTIVE, SCREEN_VIDEO_BITRATE_KBPS, SCREEN_VIDEO_FPS, SELF_AUDIO_ENABLED,
     SELF_VIDEO_ENABLED, TIER_TRANSITIONS_TOTAL, VIDEO_BITRATE_KBPS, VIDEO_FPS,
-    VIDEO_FRAMES_DROPPED, VIDEO_QUALITY_SCORE, WEBSOCKET_DROPS, WEBSOCKET_DROPS_TOTAL,
+    VIDEO_FRAMES_DROPPED, VIDEO_QUALITY_SCORE, WEBSOCKET_DROPS,
 };
 
 async fn metrics_handler(
@@ -188,9 +188,7 @@ fn remove_session_metrics(session_info: &SessionInfo) {
     let _ = ADAPTIVE_VIDEO_TIER.remove_label_values(&reporter_labels);
     let _ = ADAPTIVE_AUDIO_TIER.remove_label_values(&reporter_labels);
     let _ = DATAGRAM_DROPS.remove_label_values(&reporter_labels);
-    let _ = DATAGRAM_DROPS_TOTAL.remove_label_values(&reporter_labels);
     let _ = WEBSOCKET_DROPS.remove_label_values(&reporter_labels);
-    let _ = WEBSOCKET_DROPS_TOTAL.remove_label_values(&reporter_labels);
     let _ = KEYFRAME_REQUESTS_SENT_TOTAL.remove_label_values(&reporter_labels);
     let _ = ENCODER_FPS_RATIO.remove_label_values(&reporter_labels);
     let _ = ENCODER_WORST_PEER_FPS.remove_label_values(&reporter_labels);
@@ -604,17 +602,11 @@ fn process_health_packet_to_metrics_pb(
             DATAGRAM_DROPS
                 .with_label_values(&reporter_labels)
                 .set(drops as f64);
-            DATAGRAM_DROPS_TOTAL
-                .with_label_values(&reporter_labels)
-                .set(drops as f64);
         }
         if let Some(drops) = health_packet.websocket_drops_total {
             WEBSOCKET_DROPS
                 .with_label_values(&reporter_labels)
                 .set(drops as f64);
-            WEBSOCKET_DROPS_TOTAL
-                .with_label_values(&reporter_labels)
-                .set(drops as f64);
         }
         if let Some(kf_reqs) = health_packet.keyframe_requests_sent_total {
             KEYFRAME_REQUESTS_SENT_TOTAL

actix-api/src/metrics.rs

@@ -387,15 +387,6 @@ lazy_static! {
     )
     .expect("Failed to create datagram_drops metric");
 
-    /// Deprecated compatibility mirror for dashboards still querying the old
-    /// counter-shaped gauge name.
-    pub static ref DATAGRAM_DROPS_TOTAL: GaugeVec = register_gauge_vec!(
-        "videocall_datagram_drops_total",
-        "Cumulative datagrams dropped due to locked writable stream",
-        &["meeting_id", "session_id", "peer_id", "display_name"]
-    )
-    .expect("Failed to create datagram_drops_total metric");
-
     /// Cumulative WebSocket packet drops as of the latest client health snapshot.
     pub static ref WEBSOCKET_DROPS: GaugeVec = register_gauge_vec!(
         "videocall_websocket_drops",
@@ -404,15 +395,6 @@ lazy_static! {
     )
     .expect("Failed to create websocket_drops metric");
 
-    /// Deprecated compatibility mirror for dashboards still querying the old
-    /// counter-shaped gauge name.
-    pub static ref WEBSOCKET_DROPS_TOTAL: GaugeVec = register_gauge_vec!(
-        "videocall_websocket_drops_total",
-        "Cumulative WebSocket packets dropped due to send buffer backpressure",
-        &["meeting_id", "session_id", "peer_id", "display_name"]
-    )
-    .expect("Failed to create websocket_drops_total metric");
-
     /// Cumulative keyframe requests sent (PLI)
     pub static ref KEYFRAME_REQUESTS_SENT_TOTAL: GaugeVec = register_gauge_vec!(
         "videocall_keyframe_requests_sent_total",

helm/grafana/dashboards/meeting-investigation.json

@@ -1504,12 +1504,12 @@
       },
       "targets": [
         {
-          "expr": "rate(videocall_datagram_drops{meeting_id=~\"$meeting\"}[$__rate_interval]) or rate(videocall_datagram_drops_total{meeting_id=~\"$meeting\"}[$__rate_interval])",
+          "expr": "rate(videocall_datagram_drops{meeting_id=~\"$meeting\"}[$__rate_interval])",
           "legendFormat": "WT {{display_name}}",
           "refId": ""
         },
         {
-          "expr": "rate(videocall_websocket_drops{meeting_id=~\"$meeting\"}[$__rate_interval]) or rate(videocall_websocket_drops_total{meeting_id=~\"$meeting\"}[$__rate_interval])",
+          "expr": "rate(videocall_websocket_drops{meeting_id=~\"$meeting\"}[$__rate_interval])",
           "legendFormat": "WS {{display_name}}",
           "refId": ""
         }