[ https://issues.apache.org/jira/browse/HDFS-384?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13188361#comment-13188361 ]
Harsh J commented on HDFS-384: ------------------------------ Shouldn't this be done via the offline image/edits viewers, by an experienced user? Automating this would not help in all cases. (For instance once the fsimage was only barely written before the storage went down. Recovery would've been possible, but multiple sections of the fsimage that tail after the file list were missing altogether and can't be faked.) I guess we could provide a manual recovery tool for edits corruption, usable at the user's jurisdiction, but if fsimage is corrupted we should not interfere in any automatic way. > optionally ignore a bad entry in namenode state when starting up > ---------------------------------------------------------------- > > Key: HDFS-384 > URL: https://issues.apache.org/jira/browse/HDFS-384 > Project: Hadoop HDFS > Issue Type: Improvement > Reporter: Yoram Arnon > Assignee: Sameer Paranjpye > > if the namenode state (fsimage, edits) contains a bad entry, the namenode > refused to start. > Normally that's a good thing, alerting the administrator that something's > corrupted. > An option to ignore those entries is useful for recovering from such a > condition. Anyone but a hard core developer would be helpless in the face of > a corruption like that, and would prefer to lose a couple of records and be > able to run than to be down, or remove the entire state of dfs. -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa For more information on JIRA, see: http://www.atlassian.com/software/jira