[ https://issues.apache.org/jira/browse/HDFS-14158?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16744104#comment-16744104 ]
Kihwal Lee commented on HDFS-14158: ----------------------------------- Are you running BackupNode? I has received little to no attention since people have not used it much for a long time. The standard way of checkpointing until the HA feature was to use secondary namenode, which has its own checks. "periodMsec is always 5 minutes or lower" might have been intentional and reasonable long time ago when BackupNode was first created. This, however, was fixed in HDFS-1572 by [~jghoman] in January 2011, only to be accidentally reverted by the merge of HDFS-1052 (HDFS Federation) three months later. If the same thing was done on SecondaryNamenode, people would have noticed it very soon. > Checkpointer ignores configured time period > 5 minutes > ------------------------------------------------------- > > Key: HDFS-14158 > URL: https://issues.apache.org/jira/browse/HDFS-14158 > Project: Hadoop HDFS > Issue Type: Bug > Components: namenode > Affects Versions: 2.8.1 > Reporter: Timo Walter > Priority: Minor > Labels: checkpoint, hdfs, namenode > > The checkpointer always triggers a checkpoint every 5 minutes and ignores the > flag "*dfs.namenode.checkpoint.period*", if its greater than 5 minutes. > See the code below (in Checkpointer.java): > {code:java} > //Main work loop of the Checkpointer > public void run() { > // Check the size of the edit log once every 5 minutes. > long periodMSec = 5 * 60; // 5 minutes > if(checkpointConf.getPeriod() < periodMSec) { > periodMSec = checkpointConf.getPeriod(); > } > {code} > If the configured period ("*dfs.namenode.checkpoint.period*") is lower than 5 > minutes, you choose use the configured one. But it always ignores it, if it's > greater than 5 minutes. > > In my opinion, the if-expression should be: > {code:java} > if(checkpointConf.getPeriod() > periodMSec) { > periodMSec = checkpointConf.getPeriod(); > } > {code} > > Then "*dfs.namenode.checkpoint.period*" won't get ignored. -- This message was sent by Atlassian JIRA (v7.6.3#76005) --------------------------------------------------------------------- To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org