Data Search while replication

[mechravi25 <me...@yahoo.co.in>]

We would like to know what will happen to the search queries hit on a Solr
server which is a slave instance during the time of replication. Will the
searches fetch the documents from the old indexes? Can some please explain
what exactly happens to the search queries while the replication is in
progress in slave solr instance.

In our application, the master solr instances is used for indexing purpose
and the slave solr is for user search request. Hence the replication has to
happen on regular interval of time. Master solr has around 1.4 million
document(Size : 2.7 GB) and it takes more than 900 seconds for replication.
We have been reported by the users that the search request is taking longer
time during replication. 

Thanks in advance. Any pointers would be of great help.


--
View this message in context: http://lucene.472066.n3.nabble.com/Data-Search-while-replication-tp3585184p3585184.html
Sent from the Lucene - General mailing list archive at Nabble.com.

Re: Data Search while replication

[Chris Hostetter <ho...@fucit.org>]

: We would like to know what will happen to the search queries hit on a Solr
: server which is a slave instance during the time of replication. Will the
: searches fetch the documents from the old indexes? Can some please explain

yes.

: what exactly happens to the search queries while the replication is in
: progress in slave solr instance.

they continue to be served by the existing indx until the new index is 
fully loaded -- from the perspective of the slave instance, replication is 
exactly the same as any other "commit" that causes a new searcher to be 
opend (the old seracher is used until the new searcher is ready)

: document(Size : 2.7 GB) and it takes more than 900 seconds for replication.

replication time is almost entirely based on IO: the network IO available 
on both machines, and the disk IO for reading the new segments on the 
master and writting the new segments on the slave -- the only other time 
involved is instantiating the new searcher on the slave and any cache 
warming (again: not replication specific)

: We have been reported by the users that the search request is taking longer
: time during replication. 

this is usually an IO issue -- either slow network IO as the clients 
network traffic competes with the replication network traffic, or disk IO 
if not enough free memory is available to keep the old index segments in 
the filesystem cache as the new index segments are written.

: Thanks in advance. Any pointers would be of great help.

FWIW: you are likeley to get a wider breadth of help using the 
solr-user@lucene mailing list.  it has more subscribers, and discussions 
of using solr are more topical there (general@lucene is for broad 
discussion of the Lucene project as a whole)


-Hoss