[ https://issues.apache.org/jira/browse/HDFS-11448?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15995387#comment-15995387 ]
Jing Zhao commented on HDFS-11448: ---------------------------------- Thanks for continuing working on this, [~hanishakoneru]. And thanks for the review, [~arpitagarwal]. I have 2 further minor comments: # I'm not very sure if we still need to check if the current directory exists when starting a sync iteration. We have already checked if current directory exists while initializing the journal. Then during the upgrade/rollback, both the {{moveTmpSegmentToCurrent}} and {{doRollback}} hold the Journal object's monitor, {{moveTmpSegmentToCurrent}} also checks the {{committedTxnId}} before moving. Thus to me it is not necessary to have this current directory check in {{canJournalSync}}. {code} public boolean canJournalSync() { // JN should not sync if there is no current directory (during upgrade or // rollback). return storage.getCurrentDir().exists(); } {code} # About the name of the temporary directory: maybe we can have a more specific name like "edits.tmp" or "edits.sync"? Otherwise the patch looks good to me. > JN log segment syncing should support HA upgrade > ------------------------------------------------ > > Key: HDFS-11448 > URL: https://issues.apache.org/jira/browse/HDFS-11448 > Project: Hadoop HDFS > Issue Type: Bug > Components: hdfs > Reporter: Hanisha Koneru > Assignee: Hanisha Koneru > Attachments: HDFS-11448.001.patch > > > HDFS-4025 adds support for sychronizing past log segments to JNs that missed > them. But, as pointed out by [~jingzhao], if the segment download happens > when an admin tries to rollback, it might fail ([see > comment|https://issues.apache.org/jira/browse/HDFS-4025?focusedCommentId=15850633&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-15850633]). -- This message was sent by Atlassian JIRA (v6.3.15#6346) --------------------------------------------------------------------- To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org