Improving Consistency in SolrCloud

Darsh Shah and Yifan Li
Carnegie Mellon University

ABSTRACT 

SolrCloud is the distributed search engine from Apache Solr
project. There is a known bug in the SolrCloud community that results
in lost writes in the SolrCloud system. In this report, we studied the
architecture of the SolrCloud system and used a simulation system to
address the consistency problems around this bug. We have proposed a
quorum based writing strategy to handle the writes. In this case, we
have favored consistency over write availability. In the evaluation
part, we have shown that this new quorum based writing strategy is
worth the price of availability, as it will improve the query success
rate in a large-scale real world deployment under heavy write
workloads.