[ 
https://issues.apache.org/jira/browse/BOOKKEEPER-753?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16046967#comment-16046967
 ] 

ASF GitHub Bot commented on BOOKKEEPER-753:
-------------------------------------------

Github user reddycharan commented on the issue:

    https://github.com/apache/bookkeeper/pull/81
  
    How we handle when ledgerdirs are full when Bookie is started is little 
different
    - if isReadOnlyModeEnabled is true, then we try to bring up bookie in 
ReadOnlyMode and provide ability to create EntryLog file or Index file for 
replaying Journal or for doing compaction, though storage has reached threshold.
    
    In your change, you try to do force GC in this case and if still it hasn't 
come below threshold then you fail, if it has come below threshold then you 
start Bookie in RW mode.
    
    So I believe, theoretically speaking both the features can coexist. Like 
following
    - it tries to do force GC if storage is full during start
    - if still it cannt find space, then start Bookie in RO mode and provide 
ability to create entrylog/index file for replaying Journal or for compaction.
    
    But for pushing my code on top of yours I've to deal with bunch of 
conflicts since we are touching the same code components. 
    
    Also, I dont think your commits are in sync with current repo code. You 
need to incorporate changes for handling LedgerDirsMonitor/LedgerDirsManager 
decoupling which we did with this change - 
https://github.com/apache/bookkeeper/commit/efd8ec26926d2e75f51d6d576543a0dcc116b83f
 .
    
    @sijie @eolivelli @jvrao 


> Bookie should run garbage collection before startup when all directories 
> became full
> ------------------------------------------------------------------------------------
>
>                 Key: BOOKKEEPER-753
>                 URL: https://issues.apache.org/jira/browse/BOOKKEEPER-753
>             Project: Bookkeeper
>          Issue Type: Bug
>          Components: bookkeeper-server
>            Reporter: Sijie Guo
>            Assignee: Sijie Guo
>             Fix For: 4.5.0
>
>
> otherwise, a bookie that ran-out-of disk would never be started up again.



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)

Reply via email to