HDFS-16186 Datanode kicks out hard disk logic optimization

apache · Sep 7, 2021 · 1e1dc86 · 1e1dc86
1 parent 852af87
commit 1e1dc86
Show file tree

Hide file tree

Showing 2 changed files with 33 additions and 31 deletions.
diff --git a/...t/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/datanode/VolumeExCountPair.java b/...t/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/datanode/VolumeExCountPair.java
@@ -18,43 +18,43 @@
 package org.apache.hadoop.hdfs.server.datanode;
 
 /**
- * Record volume's IOException total counts and update timestamp
+ * Record volume's IOException total counts and update timestamp.
  */
 public class VolumeExCountPair {
 
-    private long prevTs;
-    private long IoExceptionCnt;
+  private long prevTs;
+  private long ioExceptionCnt;
 
-    public VolumeExCountPair() {
-    }
+  public VolumeExCountPair() {
+  }
 
-    public VolumeExCountPair(long prevTimeStamp, long IoExceptionCnt) {
-        this.prevTs = prevTimeStamp;
-        this.IoExceptionCnt = IoExceptionCnt;
-    }
+  public VolumeExCountPair(long prevTimeStamp, long ioExceptionCnt) {
+    this.prevTs = prevTimeStamp;
+    this.ioExceptionCnt = ioExceptionCnt;
+  }
 
-    public void setNewPair(long now, long curExCnt) {
-        setPrevTs(now);
-        setIoExceptionCnt(curExCnt);
-    }
+  public void setNewPair(long now, long curExCnt) {
+    setPrevTs(now);
+    setIoExceptionCnt(curExCnt);
+  }
 
-    public VolumeExCountPair getPair() {
-        return new VolumeExCountPair(prevTs, IoExceptionCnt);
-    }
+  public VolumeExCountPair getPair() {
+    return new VolumeExCountPair(prevTs, ioExceptionCnt);
+  }
 
-    public void setPrevTs(long prevTs) {
-        this.prevTs = prevTs;
-    }
+  public void setPrevTs(long prevTs) {
+    this.prevTs = prevTs;
+  }
 
-    public void setIoExceptionCnt(long ioExceptionCnt) {
-        IoExceptionCnt = ioExceptionCnt;
-    }
+  public void setIoExceptionCnt(long ioExceptionCnt) {
+    this.ioExceptionCnt = ioExceptionCnt;
+  }
 
-    public long getPrevTs() {
-        return prevTs;
-    }
+  public long getPrevTs() {
+    return prevTs;
+  }
 
-    public long getIoExceptionCnt() {
-        return IoExceptionCnt;
-    }
+  public long getIoExceptionCnt() {
+    return ioExceptionCnt;
+  }
 }
diff --git a/...fs/src/main/java/org/apache/hadoop/hdfs/server/datanode/checker/DatasetVolumeChecker.java b/...fs/src/main/java/org/apache/hadoop/hdfs/server/datanode/checker/DatasetVolumeChecker.java
@@ -38,6 +38,7 @@
 import org.apache.hadoop.io.IOUtils;
 import org.apache.hadoop.util.DiskChecker.DiskErrorException;
 import org.apache.hadoop.util.Sets;
+import org.apache.hadoop.util.Time;
 import org.apache.hadoop.util.Timer;
 import org.slf4j.Logger;
 import org.slf4j.LoggerFactory;
@@ -400,11 +401,12 @@ private void markHealthy() {
         LOG.warn("Volume {} Exception Count is {}", reference.getVolume().toString(),
                 reference.getVolume().getExCountPair().getIoExceptionCnt());
         if (volumeExCountPair.getIoExceptionCnt() == 0) {
-          volumeExCountPair.setNewPair(System.currentTimeMillis(), totalFileIoErrors);
+          volumeExCountPair.setNewPair(Time.monotonicNow(), totalFileIoErrors);
         }else{
-          if((System.currentTimeMillis() - reference.getVolume().getExCountPair().getPrevTs())/1000/60/60 > 2){
+          long lastTime = reference.getVolume().getExCountPair().getPrevTs();
+          if((Time.monotonicNow() - lastTime)/1000/60/60 > 2){
             long cnt = reference.getVolume().getExCountPair().getPair().getIoExceptionCnt();
-            if(cnt > 100){
+            if(cnt > 30){
               failedVolumes.add(reference.getVolume());
               return;
             }