Elasticsearch is one of the data stores that can be considered for quick lookup and near real time queries for searches, and can form a part of the serving layer as well. What we mean by real time here is that document is available for retrieval as soon as it is ingested. Elasticsearch supports key based document lookup in absolute real-time as well as supports near real-time searches as needed for data service and its contract. This storage layer should be ideally placed behind the data services as the data store fulfilling data requirements of the data services. It is an eventually consistent system that does not support ACID transactions but does support lightweight optimistic locking for minimum level of transactional consistency. This option is good for scenarios that require partial matches and searches and need complete document representation of data.

Other options may include NoSQL stores like HBase and Cassandra that are proven for very fast key based lookup and can also be invoked using Java based drivers. Both HBase and Cassandra provide column family based storage that can facilitate access pattern based key design for real-time lookup scenarios. The key advantage here with HBase is that it works over HDFS storage and does not require direct attached storage like Elasticsearch.