Merge pull request #4626 from chengtx01/add_net_metrics

feat(net): add 3 prometheus metrics for network module

Author: jwrct

common/src/main/java/org/tron/common/prometheus/MetricKeys.java

@@ -58,6 +58,9 @@ public static class Histogram {
     public static final String TCP_BYTES = "tron:tcp_bytes";
     public static final String HTTP_BYTES = "tron:http_bytes";
     public static final String INTERNAL_SERVICE_LATENCY = "tron:internal_service_latency_seconds";
+    public static final String MESSAGE_PROCESS_LATENCY = "tron:message_process_latency_seconds";
+    public static final String BLOCK_FETCH_LATENCY = "tron:block_fetch_latency_seconds";
+    public static final String BLOCK_RECEIVE_DELAY = "tron:block_receive_delay_seconds";
 
     private Histogram() {
       throw new IllegalStateException("Histogram");

common/src/main/java/org/tron/common/prometheus/MetricsHistogram.java

@@ -41,6 +41,11 @@ public class MetricsHistogram {
         "type");
     init(MetricKeys.Histogram.HTTP_BYTES, "http_bytes traffic.",
         "url", "status");
+    init(MetricKeys.Histogram.MESSAGE_PROCESS_LATENCY, "process message latency.",
+        "type");
+    init(MetricKeys.Histogram.BLOCK_FETCH_LATENCY, "fetch block latency.");
+    init(MetricKeys.Histogram.BLOCK_RECEIVE_DELAY,
+        "receive block delay time, receiveTime - blockTime.");
   }
 
   private MetricsHistogram() {

framework/src/main/java/org/tron/core/db/Manager.java

@@ -1410,7 +1410,7 @@ public BlockCapsule generateBlock(Miner miner, long blockTime, long timeout) {
     String address =  StringUtil.encode58Check(miner.getWitnessAddress().toByteArray());
     final Histogram.Timer timer = Metrics.histogramStartTimer(
         MetricKeys.Histogram.BLOCK_GENERATE_LATENCY, address);
-    Metrics.histogramObserve(MetricKeys.Histogram.MINER_LATENCY,
+    Metrics.histogramObserve(MetricKeys.Histogram.MINER_DELAY,
         (System.currentTimeMillis() - blockTime) / Metrics.MILLISECONDS_PER_SECOND, address);
     long postponedTrxCount = 0;
     logger.info("Generate block {} begin.", chainBaseManager.getHeadBlockNum() + 1);

framework/src/main/java/org/tron/core/net/TronNetService.java

@@ -5,6 +5,8 @@
 import org.springframework.stereotype.Component;
 import org.tron.common.overlay.message.Message;
 import org.tron.common.overlay.server.ChannelManager;
+import org.tron.common.prometheus.MetricKeys;
+import org.tron.common.prometheus.Metrics;
 import org.tron.core.exception.P2pException;
 import org.tron.core.exception.P2pException.TypeEnum;
 import org.tron.core.net.message.BlockMessage;
@@ -129,8 +131,12 @@ protected void onMessage(PeerConnection peer, TronMessage msg) {
       processException(peer, msg, e);
     } finally {
       long costs = System.currentTimeMillis() - startTime;
-      logger.info("Message processing costs {} ms, peer: {}, type: {}, time tag: {}",
-          costs, peer.getInetAddress(), msg.getType(), getTimeTag(costs));
+      if (costs > DURATION_STEP) {
+        logger.info("Message processing costs {} ms, peer: {}, type: {}, time tag: {}",
+            costs, peer.getInetAddress(), msg.getType(), getTimeTag(costs));
+        Metrics.histogramObserve(MetricKeys.Histogram.MESSAGE_PROCESS_LATENCY,
+            costs / Metrics.MILLISECONDS_PER_SECOND, msg.getType().name());
+      }
     }
   }
 

framework/src/main/java/org/tron/core/net/messagehandler/BlockMsgHandler.java

@@ -7,6 +7,8 @@
 import org.bouncycastle.util.encoders.Hex;
 import org.springframework.beans.factory.annotation.Autowired;
 import org.springframework.stereotype.Component;
+import org.tron.common.prometheus.MetricKeys;
+import org.tron.common.prometheus.Metrics;
 import org.tron.core.Constant;
 import org.tron.core.capsule.BlockCapsule;
 import org.tron.core.capsule.BlockCapsule.BlockId;
@@ -76,7 +78,11 @@ public void processMessage(PeerConnection peer, TronMessage msg) throws P2pExcep
       if (null != time) {
         MetricsUtil.histogramUpdateUnCheck(MetricsKey.NET_LATENCY_FETCH_BLOCK
                 + peer.getNode().getHost(), now - time);
+        Metrics.histogramObserve(MetricKeys.Histogram.BLOCK_FETCH_LATENCY,
+            (now - time) / Metrics.MILLISECONDS_PER_SECOND);
       }
+      Metrics.histogramObserve(MetricKeys.Histogram.BLOCK_RECEIVE_DELAY,
+          (now - blockMessage.getBlockCapsule().getTimeStamp()) / Metrics.MILLISECONDS_PER_SECOND);
       fetchBlockService.blockFetchSuccess(blockId);
       long interval = blockId.getNum() - tronNetDelegate.getHeadBlockId().getNum();
       processBlock(peer, blockMessage.getBlockCapsule());