[Building Sakai] . Re: Elastic Search (SRCH-111) (John Bush)

[Ian Boston]

On 27 January 2013 16:13, Colin Hebert <colin.hebert at it.ox.ac.uk> wrote:
> I'm not sure why you would _need_ to implement a WAL.

To avoid having to commit segments to disk on every update ?

If you soft commit you risk loosing data. If you commit on every
update you cause a reader reload on every update which flushes all
term vectors and cached information. Having a WAL you can replay in
the event of failure means you can do a soft commit to memory avoiding
having to perform a hard commit. At least, thats the way ES avoids
hard commits on every update.  Unfortunately, Solr master/slave
replication requires hard commits to replicate. If you buy a
commercial Solr distribution from Lucid then IIUC you get replication
of soft commit. ES replicates by replicating the write operation
avoiding the hard commit issue completely.

HTH
Ian

>
> On 27 January 2013 16:34, Ian Boston <ieb at tfd.co.uk> wrote:
>>  elasticsearch.index.number_of_replicas=1
>>>
>>>
>>
>>
>> Make sure you have enough replicas to make the index resilient. Iiuc that
>> means >= 3.
>>
>> Also, if using sol4 you will need to manage replicas yourself, and might
>> also need to implement your own write ahead log. I did some blg posts on
>> this a while back.
>> Hth
>> Ian
>>
>> _______________________________________________
>> sakai-dev mailing list
>> sakai-dev at collab.sakaiproject.org
>> http://collab.sakaiproject.org/mailman/listinfo/sakai-dev
>>
>> TO UNSUBSCRIBE: send email to sakai-dev-unsubscribe at collab.sakaiproject.org
>> with a subject of "unsubscribe"