HDFS-17171. CONGESTION_RATIO should be configurable (#5996)

Reviewed-by: Ayush Saxena <ayushsaxena@apache.org>
Signed-off-by: Tao Li <tomscut@apache.org>

hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/DFSConfigKeys.java

@@ -1573,6 +1573,8 @@ public class DFSConfigKeys extends CommonConfigurationKeys {
   public static final boolean DFS_PIPELINE_ECN_ENABLED_DEFAULT = false;
   public static final String DFS_PIPELINE_SLOWNODE_ENABLED = "dfs.pipeline.slownode";
   public static final boolean DFS_PIPELINE_SLOWNODE_ENABLED_DEFAULT = false;
+  public static final String DFS_PIPELINE_CONGESTION_RATIO = "dfs.pipeline.congestion.ratio";
+  public static final double DFS_PIPELINE_CONGESTION_RATIO_DEFAULT = 1.5;
 
   // Key Provider Cache Expiry
   public static final String DFS_DATANODE_BLOCK_PINNING_ENABLED =
@@ -2042,5 +2044,4 @@ public class DFSConfigKeys extends CommonConfigurationKeys {
   public static final long DFS_LEASE_HARDLIMIT_DEFAULT =
       HdfsClientConfigKeys.DFS_LEASE_HARDLIMIT_DEFAULT;
 
-
 }

hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/datanode/DataNode.java

@@ -462,7 +462,7 @@ public class DataNode extends ReconfigurableBase
   private final Tracer tracer;
   private static final int NUM_CORES = Runtime.getRuntime()
       .availableProcessors();
-  private static final double CONGESTION_RATIO = 1.5;
+  private final double congestionRatio;
   private DiskBalancer diskBalancer;
   private DataSetLockManager dataSetLockManager;
 
@@ -515,6 +515,10 @@ public class DataNode extends ReconfigurableBase
     volumeChecker = new DatasetVolumeChecker(conf, new Timer());
     this.xferService =
         HadoopExecutors.newCachedThreadPool(new Daemon.DaemonFactory());
+    double congestionRationTmp = conf.getDouble(DFSConfigKeys.DFS_PIPELINE_CONGESTION_RATIO,
+        DFSConfigKeys.DFS_PIPELINE_CONGESTION_RATIO_DEFAULT);
+    this.congestionRatio = congestionRationTmp > 0 ?
+        congestionRationTmp : DFSConfigKeys.DFS_PIPELINE_CONGESTION_RATIO_DEFAULT;
   }
 
   /**
@@ -614,6 +618,10 @@ public class DataNode extends ReconfigurableBase
         new DataTransferThrottler(100, ecReconstuctReadBandwidth) : null;
     this.ecReconstuctWriteThrottler = ecReconstuctWriteBandwidth > 0 ?
         new DataTransferThrottler(100, ecReconstuctWriteBandwidth) : null;
+    double congestionRationTmp = conf.getDouble(DFSConfigKeys.DFS_PIPELINE_CONGESTION_RATIO,
+        DFSConfigKeys.DFS_PIPELINE_CONGESTION_RATIO_DEFAULT);
+    this.congestionRatio = congestionRationTmp > 0 ?
+        congestionRationTmp : DFSConfigKeys.DFS_PIPELINE_CONGESTION_RATIO_DEFAULT;
   }
 
   @Override  // ReconfigurableBase
@@ -1070,7 +1078,7 @@ public class DataNode extends ReconfigurableBase
     }
     double load = ManagementFactory.getOperatingSystemMXBean()
         .getSystemLoadAverage();
-    return load > NUM_CORES * CONGESTION_RATIO ? PipelineAck.ECN.CONGESTED :
+    return load > NUM_CORES * congestionRatio ? PipelineAck.ECN.CONGESTED :
         PipelineAck.ECN.SUPPORTED;
   }
 

hadoop-hdfs-project/hadoop-hdfs/src/main/resources/hdfs-default.xml

@@ -5619,6 +5619,14 @@
   </description>
 </property>
 
+<property>
+  <name>dfs.pipeline.congestion.ratio</name>
+  <value>1.5</value>
+  <description>
+    The ratio which is used to compute congestion load.
+  </description>
+</property>
+
 <property>
   <name>dfs.qjournal.accept-recovery.timeout.ms</name>
   <value>120000</value>