Home | About | Sematext search-lucene.com search-hadoop.com
 Search Hadoop and all its subprojects:

Switch to Plain View
Zookeeper, mail # dev - Quorum reads on zookeeper


+
Abhishek .E.S 2013-03-26, 05:37
+
Ted Dunning 2013-03-26, 08:21
Copy link to this message
-
Re: Quorum reads on zookeeper
Alexander Shraer 2013-03-26, 15:19
Ted, as previously discussed on this list, what you are saying is true
when the leader doesn't fail but it may not work in the general case
under some asynchrony conditions. Please see my email "sync semantics"
from Mar 1st or the thread "Ensure Leader hasn't changed" from
September 2012. In order to handle the general
case we should make the sync a quorum operation.

Alex

On Tue, Mar 26, 2013 at 1:21 AM, Ted Dunning <[EMAIL PROTECTED]> wrote:
> Because the leader is the one who leads all writes, it suffices to catch up
> with that one node.  There is no need to read from a quorum.  Another way
> to look at this is that there is no way for a quorum read to succeed
> without there also existing a leader and one read to the leader is better
> than trying to read from the entire cluster.
>
> On Tue, Mar 26, 2013 at 6:37 AM, Abhishek .E.S <[EMAIL PROTECTED]>wrote:
>
>> Hi,
>>
>> Zookeeper suggests using sync if the most up to date value is required for
>> reading.
>> Would it be possible to use Quorum reads for ensuring up to date reads ?
>>
>> There is a function called  isQuorum for indicating quorum requirement for
>> operations. Read is listed as not requiring forum in this function
>> How would changing this to true impact the behavior ?
>>
>> Any pointers would be appreciated.
>>
>> -Abhishek
>>
+
Ted Dunning 2013-03-26, 18:56