HDFS-17231. HA: Safemode should exit when resources are from low to a…

…vailable. (apache#6207). Contributed by Gu Peng. Reviewed-by: Xing Lin <[email protected]> Signed-off-by: He Xiaoqiao <[email protected]>
jiajunmao · Feb 6, 2024 · 1355958 · 1355958
1 parent c6d36d0
commit 1355958
Show file tree

Hide file tree

Showing 2 changed files with 20 additions and 0 deletions.
diff --git a/...roject/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/namenode/FSNamesystem.java b/...roject/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/namenode/FSNamesystem.java
@@ -4534,6 +4534,11 @@ public void run () {
               LOG.warn(lowResourcesMsg + "Already in safe mode.");
             }
             enterSafeMode(true);
+          } else {
+            if (isNoManualAndResourceLowSafeMode()) {
+              LOG.info("Namenode has sufficient available resources, exiting safe mode.");
+              leaveSafeMode(false);
+            }
           }
           try {
             Thread.sleep(resourceRecheckInterval);
@@ -5265,6 +5270,13 @@ private synchronized boolean isInManualOrResourceLowSafeMode() {
     return manualSafeMode || resourceLowSafeMode;
   }
 
+  /**
+   * @return true if it is not in manual safe mode and resource low safe mode.
+   */
+  private synchronized boolean isNoManualAndResourceLowSafeMode() {
+    return !manualSafeMode && resourceLowSafeMode;
+  }
+
   private synchronized void setManualAndResourceLowSafeMode(boolean manual,
       boolean resourceLow) {
     this.manualSafeMode = manual;

diff --git a/...dfs/src/test/java/org/apache/hadoop/hdfs/server/namenode/TestNameNodeResourceChecker.java b/...dfs/src/test/java/org/apache/hadoop/hdfs/server/namenode/TestNameNodeResourceChecker.java
@@ -130,6 +130,14 @@ public void testCheckThatNameNodeResourceMonitorIsRunning()
 
       assertTrue("NN should be in safe mode after resources crossed threshold",
           cluster.getNameNode().isInSafeMode());
+
+      mockResourceChecker.setResourcesAvailable(true);
+      while (cluster.getNameNode().isInSafeMode() &&
+          Time.now() < startMillis + (60 * 1000)) {
+        Thread.sleep(1000);
+      }
+      assertTrue("NN should leave safe mode after resources not crossed threshold",
+          !cluster.getNameNode().isInSafeMode());
     } finally {
       if (cluster != null)
         cluster.shutdown();