HDFS-17285. RBF: Add a safe mode check period configuration (apache#6347

) Contributed by LiuGuH. Reviewed-by: Inigo Goiri <[email protected]> Reviewed-by: Ayush Saxena <[email protected]> Signed-off-by: Shilun Fan <[email protected]>
slfan1989 · Dec 21, 2023 · b4fed58 · b4fed58
1 parent 5dd1977
commit b4fed58
Show file tree

Hide file tree

Showing 4 changed files with 22 additions and 3 deletions.
diff --git a/...hdfs-rbf/src/main/java/org/apache/hadoop/hdfs/server/federation/router/RBFConfigKeys.java b/...hdfs-rbf/src/main/java/org/apache/hadoop/hdfs/server/federation/router/RBFConfigKeys.java
@@ -279,6 +279,10 @@ public class RBFConfigKeys extends CommonConfigurationKeysPublic {
       FEDERATION_ROUTER_PREFIX + "safemode.expiration";
   public static final long DFS_ROUTER_SAFEMODE_EXPIRATION_DEFAULT =
       3 * DFS_ROUTER_CACHE_TIME_TO_LIVE_MS_DEFAULT;
+  public static final String DFS_ROUTER_SAFEMODE_CHECKPERIOD_MS =
+      FEDERATION_ROUTER_PREFIX + "safemode.checkperiod";
+  public static final long DFS_ROUTER_SAFEMODE_CHECKPERIOD_MS_DEFAULT =
+      TimeUnit.SECONDS.toMillis(5);
 
   // HDFS Router-based federation mount table entries
   /** Maximum number of cache entries to have. */

diff --git a/.../src/main/java/org/apache/hadoop/hdfs/server/federation/router/RouterSafemodeService.java b/.../src/main/java/org/apache/hadoop/hdfs/server/federation/router/RouterSafemodeService.java
@@ -133,8 +133,8 @@ protected void serviceInit(Configuration conf) throws Exception {
 
     // Use same interval as cache update service
     this.setIntervalMs(conf.getTimeDuration(
-        RBFConfigKeys.DFS_ROUTER_CACHE_TIME_TO_LIVE_MS,
-        RBFConfigKeys.DFS_ROUTER_CACHE_TIME_TO_LIVE_MS_DEFAULT,
+        RBFConfigKeys.DFS_ROUTER_SAFEMODE_CHECKPERIOD_MS,
+        RBFConfigKeys.DFS_ROUTER_SAFEMODE_CHECKPERIOD_MS_DEFAULT,
         TimeUnit.MILLISECONDS));
 
     this.startupInterval = conf.getTimeDuration(

diff --git a/hadoop-hdfs-project/hadoop-hdfs-rbf/src/main/resources/hdfs-rbf-default.xml b/hadoop-hdfs-project/hadoop-hdfs-rbf/src/main/resources/hdfs-rbf-default.xml
@@ -530,6 +530,17 @@
     </description>
   </property>
 
+  <property>
+    <name>dfs.federation.router.safemode.checkperiod</name>
+    <value>5s</value>
+    <description>
+      How often the Router should check safe mode. This
+      setting supports multiple time unit suffixes as described in
+      dfs.heartbeat.interval. If no suffix is specified then milliseconds is
+      assumed.
+    </description>
+  </property>
+
   <property>
     <name>dfs.federation.router.monitor.namenode</name>
     <value></value>

diff --git a/...rbf/src/test/java/org/apache/hadoop/hdfs/server/federation/router/TestRouterSafemode.java b/...rbf/src/test/java/org/apache/hadoop/hdfs/server/federation/router/TestRouterSafemode.java
@@ -18,6 +18,7 @@
 package org.apache.hadoop.hdfs.server.federation.router;
 
 import static org.apache.hadoop.hdfs.server.federation.router.RBFConfigKeys.DFS_ROUTER_CACHE_TIME_TO_LIVE_MS;
+import static org.apache.hadoop.hdfs.server.federation.router.RBFConfigKeys.DFS_ROUTER_SAFEMODE_CHECKPERIOD_MS;
 import static org.apache.hadoop.hdfs.server.federation.router.RBFConfigKeys.DFS_ROUTER_SAFEMODE_EXPIRATION;
 import static org.apache.hadoop.hdfs.server.federation.router.RBFConfigKeys.DFS_ROUTER_SAFEMODE_EXTENSION;
 import static org.apache.hadoop.hdfs.server.federation.store.FederationStateStoreTestUtils.deleteStateStore;
@@ -70,6 +71,9 @@ public static void create() throws IOException {
     // 200 ms cache refresh
     conf.setTimeDuration(DFS_ROUTER_CACHE_TIME_TO_LIVE_MS,
         200, TimeUnit.MILLISECONDS);
+    // 100 ms safemode checkperiod
+    conf.setTimeDuration(DFS_ROUTER_SAFEMODE_CHECKPERIOD_MS,
+        100, TimeUnit.MILLISECONDS);
     // 1 sec post cache update before entering safemode (2 intervals)
     conf.setTimeDuration(DFS_ROUTER_SAFEMODE_EXPIRATION,
         TimeUnit.SECONDS.toMillis(1), TimeUnit.MILLISECONDS);
@@ -133,7 +137,7 @@ public void testRouterExitSafemode()
     long interval =
         conf.getTimeDuration(DFS_ROUTER_SAFEMODE_EXTENSION,
             TimeUnit.SECONDS.toMillis(2), TimeUnit.MILLISECONDS) +
-        conf.getTimeDuration(DFS_ROUTER_CACHE_TIME_TO_LIVE_MS,
+        conf.getTimeDuration(DFS_ROUTER_SAFEMODE_CHECKPERIOD_MS,
             TimeUnit.SECONDS.toMillis(1), TimeUnit.MILLISECONDS);
     Thread.sleep(interval);