[ https://issues.apache.org/jira/browse/HDFS-16083?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17371915#comment-17371915 ]
Jinglun commented on HDFS-16083: -------------------------------- Hi [~lei w], thanks your patch, some comments. In EditLogTailer.java: # I prefer using `shouldRollLog` instead of avoidTriggerActiveLogRoll. {code:java} if (shouldRollLog && tooLongSinceLastLoad() && lastRollTriggerTxId < lastLoadedTxnId) {{code} In TestStandbyRollEditsLogOnly.java: # The test case and setup method should not be static. # We need a License for the new file. In TestStandbyRollEditsLogOnly#testOnlyStandbyRollEditlog: # When you compare observerRollTimeMs1, could you use assertEquals instead of assertTrue. # The message of the assert should be more specific. Something like: "Standby should roll the log." and "The observer is not expected to roll the log." # I'd prefer using standbyInitialRollTime and standbyLastRollTime instead of using numbers standbyRollTimeMs1 and standbyRollTimeMs2. # The sleep time is too long, can we make it faster ? In TestStandbyRollEditsLogOnly#testTransObToStandbyThenRollLog: # It fails, could you give it a check. # The verify logic is very like testOnlyStandbyRollEditlog, can we extract the same part as a new method. # The idea of this test is good. We can transition the state and verify roll edit more times. May be do it 3 times ? There is also some checkstyle issue. Please follow jenkins suggestions. I'll re-submit v03 as v04 to trigger jenkins. > Forbid Observer NameNode trigger active namenode log roll > ---------------------------------------------------------- > > Key: HDFS-16083 > URL: https://issues.apache.org/jira/browse/HDFS-16083 > Project: Hadoop HDFS > Issue Type: Improvement > Components: namanode > Reporter: lei w > Assignee: lei w > Priority: Minor > Attachments: HDFS-16083.001.patch, HDFS-16083.002.patch, > HDFS-16083.003.patch, activeRollEdits.png > > > When the Observer NameNode is turned on in the cluster, the Active NameNode > will receive rollEditLog RPC requests from the Standby NameNode and Observer > NameNode in a short time. Observer NameNode's rollEditLog request is a > repetitive operation, so should we forbid Observer NameNode trigger active > namenode log roll ? We 'dfs.ha.log-roll.period' configured is 300( 5 > minutes) and active NameNode receives rollEditLog RPC as shown in > activeRollEdits.png -- This message was sent by Atlassian Jira (v8.3.4#803005) --------------------------------------------------------------------- To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org