[ https://issues.apache.org/jira/browse/HDFS-17438?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17833352#comment-17833352 ]
ASF GitHub Bot commented on HDFS-17438: --------------------------------------- KeeProMise commented on code in PR #6655: URL: https://github.com/apache/hadoop/pull/6655#discussion_r1548803547 ########## hadoop-hdfs-project/hadoop-hdfs-rbf/src/test/java/org/apache/hadoop/hdfs/server/federation/resolver/TestNamenodeResolver.java: ########## @@ -359,11 +359,12 @@ public void testRegistrationNamenodeSelection() FederationNamenodeServiceState.ACTIVE); // 1) ns0:nn0 - Standby (oldest) - // 2) ns0:nn1 - Standby (newest) - // 3) ns0:nn2 - Standby - // Verify the selected entry is the newest standby entry + // 2) ns0:nn1 - Standby + // 3) ns0:nn2 - Standby (newest) + // Verify the selected entry is the oldest standby entry assertTrue(namenodeResolver.registerNamenode(createNamenodeReport( NAMESERVICES[0], NAMENODES[0], HAServiceState.STANDBY))); + Thread.sleep(1500); Review Comment: This suggestion sounds good, I added GenericTestUtils#atLeastWaitFor method, please help to take a look. > RBF: The newest STANDBY and UNAVAILABLE nn should be the lowest priority. > ------------------------------------------------------------------------- > > Key: HDFS-17438 > URL: https://issues.apache.org/jira/browse/HDFS-17438 > Project: Hadoop HDFS > Issue Type: Improvement > Reporter: Jian Zhang > Assignee: Jian Zhang > Priority: Major > Labels: pull-request-available > Attachments: HDFS-17438.001.patch > > > At present, when the status of all namenodes in an ns in the router is the > same, the namenode which is the newest reported will be placed at the top of > the cache. when the client accesses the ns through the router, it will first > access the namenode. > If multiple namenodes in this route are in an active state, or if there are > namenodes with multiple observer states, the existing logic is not a problem, > because the newest reported active or observer state namenode have a higher > probability of being true active or observer compared to the namenode that > reported active or observer state a long time ago. > Similarly, the newest reported namenode with a status of standby or > unavailable has a higher probability of being a standby or unavailable > namenode compared to the namenode reported with a status of standby or > unavailable a long time ago. Therefore, the newest nn reported as standby or > unavailable status should have a lower priority for access, the oldest nn > reported as standby or unavailable status should have a higher priority for > access. -- This message was sent by Atlassian Jira (v8.20.10#820010) --------------------------------------------------------------------- To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org