[
https://issues.apache.org/jira/browse/SOLR-471?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12566144#action_12566144
]
Yonik Seeley commented on SOLR-471:
-----------------------------------
Hi Trung, have you had a look at SOLR-303 ?
It implements distributed search in Solr itself... I think that may have a
couple of advantages:
- if it's in Solr, any type of client can use it
- possible (but not easy) for custom components to be distributed
- access to schema for proper sorting
- easier multi-tier distributed search
I've been thinking about the indexing side recently too. Longer term we need
something very robust (fault tolerant on the indexing side, ability to resize
the server pool, ability to self-synchronize among shards, etc,). In the short
term I was thinking of something that simply fanned out requests to a list of
servers based on a simple hash (no need for consistent hash in this simple
scheme). I originally thought about having this simple fan-out indexer reside
outside solr, but it occured to me that if we wanted to support all of Solr's
input types (multi-doc XML, CSV, etc) that it should probably happen inside
solr after the doc had been parsed.
> Distributed Solr Client
> -----------------------
>
> Key: SOLR-471
> URL: https://issues.apache.org/jira/browse/SOLR-471
> Project: Solr
> Issue Type: New Feature
> Components: clients - java
> Affects Versions: 1.3
> Reporter: Nguyen Kien Trung
> Priority: Minor
> Attachments: distributedclient.patch
>
>
> Inspired by memcached java clients.
> The ability to update/search/delete among many solr instances
> Client parametters:
> - List of solr servers
> - Number of replicas
> Client functions:
> - Update: using consistent hashing to determine what documents are going to
> be stored in what server. Get the list of servers (equal to number of
> replicas) and issue parallel UPDATE
> - Search: parallel search all servers, aggregate distinct results
> - Delete: parallel delete in all servers
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
