[SOLR-1524] proposal on distributed search

• Type: Sub-task
• Status: Closed
• Priority: Major
• Resolution: Duplicate
• Component/s: clients - java
• Assignee: Unassigned
• Reporter: johnson.hong
• Created: 27/Oct/09 07:19
• Updated: 23/Mar/13 00:34
• Resolved: 23/Mar/13 00:34

Description

Hi,all.
some days ago,I put a question that "why it gets slower while keep on increasing the start value accross distributed search ?".
And one reply by Shalin Shekhar Mangar was "distributed search fetches start+rows documents from each shard in order to correctly merge the results".
After this I read the source code,and I found query across distributed search would be separated into nshards' query.
Each shard query will be excute as follows:
1.get ids of matched documents into DocList //take little time
2.get all documents by id which result in step 1. //take little time
3.write all the documents found to binary string
4.parse the binary string back to SolrDocumentList // step 4 take almost all the time used
From above,I propose :It is not necessary to execute step 3 and step 4 even step 2,the ids found is enough to merge results,isn't it?

Issue Links

• duplicates: SOLR-1726 Deep Paging and Large Results Improvements