[ https://issues.apache.org/jira/browse/BOOKKEEPER-464?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13505739#comment-13505739 ]
Sijie Guo commented on BOOKKEEPER-464: -------------------------------------- {quote} If a bookie starts replicating a ledger fragment because it has been added to the ensemble of ledger but it doesn't complete the replication process because, e.g., its disk becomes full, then we remove the partially replicated ledger fragment. {quote} For replication, the sequence is that 1) we picked up a bookie to replicate entries to it, and then 2) update the ledger ensemble only after those entries are replicated. if there are failures happened between replicating and updating ledger ensemble, those replicated entries will become zombies (not referenced by any ledger ensemble). I think I wrote a comment in BOOKKEEPER-249 ( https://issues.apache.org/jira/browse/BOOKKEEPER-249?focusedCommentId=13497139&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-13497139 ) to discuss all the possible cases causing zombie entries (also I gave my understanding about zombie entry in that comment). > Provide an improved GC algorithm > -------------------------------- > > Key: BOOKKEEPER-464 > URL: https://issues.apache.org/jira/browse/BOOKKEEPER-464 > Project: Bookkeeper > Issue Type: Sub-task > Components: bookkeeper-server > Reporter: Sijie Guo > Assignee: Fangmin Lv > Fix For: 4.3.0 > > Attachments: BOOKKEEPER-464.patch > > -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators For more information on JIRA, see: http://www.atlassian.com/software/jira