[ 
https://issues.apache.org/jira/browse/BOOKKEEPER-293?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13445014#comment-13445014
 ] 

Uma Maheswara Rao G commented on BOOKKEEPER-293:
------------------------------------------------

Ok. Yes, I have this requirement in my mind, how DNs scans the blocks 
periodically to detect the corruption of data etc.. That is why I asked for 
clarification of JIRA.
Also true that we have to decide that who should actually scan the ledgers. 
I have seen in some comment that, you have the idea of distributed Auditor. In 
that case this may be easy to give this job to auditors itself to scan the 
local bk ledgers? How about moving this to after distributed audotor idea 
implemented, if you have concrete idea ready on distributed auditor?

Another idea is that, Auditor should periodically take the BKs one by one and 
find under-Replicated ledgers for that BK alone, even though that BK not 
failed. As disk failures will be very rare than node failures, this period can 
be litle more and can track some how last scaned time per BK. After crossing 
that period only we will go and scan that BK ledgers. Very initial idea, need 
to think the scenarios.
DN block scanning time 21 days for each block.

This particular case may not be really important for Namenode's scenario, as 
ledgers will be maintained only for checkpoint period. There will be very very 
rere case for 3 disks failures with in the checkpoint period. Node failures may 
be general than disk failure scenarios.


                
> Periodic checking of ledger replication status
> ----------------------------------------------
>
>                 Key: BOOKKEEPER-293
>                 URL: https://issues.apache.org/jira/browse/BOOKKEEPER-293
>             Project: Bookkeeper
>          Issue Type: Sub-task
>          Components: bookkeeper-auto-recovery
>            Reporter: Ivan Kelly
>             Fix For: 4.2.0
>
>
> We should periodically check all ledgers to make sure that they are fully 
> replicated.

--
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

Reply via email to