> On 2011-11-09 17:48:34, fpj wrote:
> > Good job, Sijie. I only have a few minor comments. One important general 
> > comment I have is that we need documentation for this feature. Ideally we 
> > commit a patch for BOOKKEEPER-82 that includes doc changes. Otherwise, I'd 
> > like to have a blocker jira to make sure that we don't have a release 
> > without the doc changes.

I would add doc for this feature.


> On 2011-11-09 17:48:34, fpj wrote:
> > http://svn.apache.org/repos/asf/zookeeper/bookkeeper/trunk/bookkeeper-server/src/main/java/org/apache/bookkeeper/bookie/Bookie.java,
> >  line 592
> > <https://reviews.apache.org/r/2745/diff/1/?file=56625#file56625line592>
> >
> >     I think we can remove this todo, right?

Yes.


> On 2011-11-09 17:48:34, fpj wrote:
> > http://svn.apache.org/repos/asf/zookeeper/bookkeeper/trunk/bookkeeper-server/src/main/java/org/apache/bookkeeper/bookie/Bookie.java,
> >  line 598
> > <https://reviews.apache.org/r/2745/diff/1/?file=56625#file56625line598>
> >
> >     Typo on journal.

I will fix it.


> On 2011-11-09 17:48:34, fpj wrote:
> > http://svn.apache.org/repos/asf/zookeeper/bookkeeper/trunk/bookkeeper-server/src/main/java/org/apache/bookkeeper/bookie/EntryLogger.java,
> >  line 460
> > <https://reviews.apache.org/r/2745/diff/1/?file=56626#file56626line460>
> >
> >     If there are corrupted entries, it is true that we can fetch copies 
> > from other bookies, but I don't think we have a mechanism for doing it yet, 
> > so it sounds safer to keep throwing an exception here for now, no?


I changed throwing an exception to skipping it, because I found bookie servers 
can't be started due to entry logger corruption after restartBookieServers in 
test cases.
This corruption is because that entry logger uses a BufferedChannel but it 
doesn't do flush when shutting down. So we read partial entry after those 
servers are restarted.

I will add code in EntryLogger#shutdown to ensure buffered data persisted to 
disk, so it can pass unit tests.

Since corruption is a complex issue, I agreed keeping throwing an exception 
here for now.

I created a jira BOOKKEEPER-62 before. Maybe we can handle data corruption in 
that jira.


- Sijie


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/2745/#review3130
-----------------------------------------------------------


On 2011-11-07 17:46:11, Sijie Guo wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/2745/
> -----------------------------------------------------------
> 
> (Updated 2011-11-07 17:46:11)
> 
> 
> Review request for bookkeeper.
> 
> 
> Summary
> -------
> 
> implement journal rolling
> 
> 1) remove journals whose id are less than marked journal id. since they 
> journals' data and index has been persisted to disk.
> 2) replay entires started from marked position in marked journal to speed up 
> bookie start up.
> 
> 
> This addresses bug BOOKKEEPER-82.
>     http://issues.apache.org/jira/browse/BOOKKEEPER-82
> 
> 
> Diffs
> -----
> 
>   
> http://svn.apache.org/repos/asf/zookeeper/bookkeeper/trunk/bookkeeper-server/src/main/java/org/apache/bookkeeper/bookie/Bookie.java
>  1198776 
>   
> http://svn.apache.org/repos/asf/zookeeper/bookkeeper/trunk/bookkeeper-server/src/main/java/org/apache/bookkeeper/bookie/EntryLogger.java
>  1198776 
>   
> http://svn.apache.org/repos/asf/zookeeper/bookkeeper/trunk/bookkeeper-server/src/test/java/org/apache/bookkeeper/test/BookieJournalRollingTest.java
>  PRE-CREATION 
> 
> Diff: https://reviews.apache.org/r/2745/diff
> 
> 
> Testing
> -------
> 
> 
> Thanks,
> 
> Sijie
> 
>

Reply via email to