From 9cb51bf106802c78b1400fba9f1d1c7e772dd5e7 Mon Sep 17 00:00:00 2001
From: tomscut <litao@bigo.sg>
Date: Mon, 1 Mar 2021 16:35:12 -0800
Subject: [PATCH] HDFS-15808. Add metrics for FSNamesystem read/write lock hold
 long time. (#2668) Contributed by tomscut.

---
 .../hdfs/server/namenode/FSNamesystem.java    | 14 ++++++++
 .../server/namenode/FSNamesystemLock.java     | 34 +++++++++++++++++++
 2 files changed, 48 insertions(+)

diff --git a/hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/namenode/FSNamesystem.java b/hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/namenode/FSNamesystem.java
index 22b4b92f44..ff03d7b052 100644
--- a/hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/namenode/FSNamesystem.java
+++ b/hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/namenode/FSNamesystem.java
@@ -4830,6 +4830,20 @@ public int getFsLockQueueLength() {
     return fsLock.getQueueLength();
   }
 
+  @Metric(value = {"ReadLockLongHoldCount", "The number of time " +
+          "the read lock has been held for longer than the threshold"},
+          type = Metric.Type.COUNTER)
+  public long getNumOfReadLockLongHold() {
+    return fsLock.getNumOfReadLockLongHold();
+  }
+
+  @Metric(value = {"WriteLockLongHoldCount", "The number of time " +
+          "the write lock has been held for longer than the threshold"},
+          type = Metric.Type.COUNTER)
+  public long getNumOfWriteLockLongHold() {
+    return fsLock.getNumOfWriteLockLongHold();
+  }
+
   int getNumberOfDatanodes(DatanodeReportType type) {
     readLock();
     try {
diff --git a/hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/namenode/FSNamesystemLock.java b/hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/namenode/FSNamesystemLock.java
index c03cfd5075..842c6b3f2d 100644
--- a/hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/namenode/FSNamesystemLock.java
+++ b/hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/namenode/FSNamesystemLock.java
@@ -109,6 +109,16 @@ public Long initialValue() {
   private final AtomicReference<LockHeldInfo> longestReadLockHeldInfo =
       new AtomicReference<>(new LockHeldInfo());
   private LockHeldInfo longestWriteLockHeldInfo = new LockHeldInfo();
+  /**
+   * The number of time the read lock
+   * has been held longer than the threshold.
+   */
+  private final AtomicLong numReadLockLongHold = new AtomicLong(0);
+  /**
+   * The number of time the write lock
+   * has been held for longer than the threshold.
+   */
+  private final AtomicLong numWriteLockLongHold = new AtomicLong(0);
 
   @VisibleForTesting
   static final String OP_NAME_OTHER = "OTHER";
@@ -182,6 +192,7 @@ public void readUnlock(String opName,
     final long readLockIntervalMs =
         TimeUnit.NANOSECONDS.toMillis(readLockIntervalNanos);
     if (needReport && readLockIntervalMs >= this.readLockReportingThresholdMs) {
+      numReadLockLongHold.incrementAndGet();
       String lockReportInfo = null;
       boolean done = false;
       while (!done) {
@@ -298,6 +309,7 @@ private void writeUnlock(String opName, boolean suppressWriteLockReport,
     LogAction logAction = LogThrottlingHelper.DO_NOT_LOG;
     if (needReport &&
         writeLockIntervalMs >= this.writeLockReportingThresholdMs) {
+      numWriteLockLongHold.incrementAndGet();
       if (longestWriteLockHeldInfo.getIntervalMs() <= writeLockIntervalMs) {
         String lockReportInfo = lockReportInfoSupplier != null ? " (" +
             lockReportInfoSupplier.get() + ")" : "";
@@ -362,6 +374,28 @@ public int getQueueLength() {
     return coarseLock.getQueueLength();
   }
 
+  /**
+   * Returns the number of time the read lock
+   * has been held longer than the threshold.
+   *
+   * @return long - Number of time the read lock
+   * has been held longer than the threshold
+   */
+  public long getNumOfReadLockLongHold() {
+    return numReadLockLongHold.get();
+  }
+
+  /**
+   * Returns the number of time the write lock
+   * has been held longer than the threshold.
+   *
+   * @return long - Number of time the write lock
+   * has been held longer than the threshold.
+   */
+  public long getNumOfWriteLockLongHold() {
+    return numWriteLockLongHold.get();
+  }
+
   /**
    * Add the lock hold time for a recent operation to the metrics.
    * @param operationName Name of the operation for which to record the time