HDocDB - HBase as a JSON Document Database

Build Status Maven Javadoc

HDocDB is a client layer for using HBase as a store for JSON documents. It implements many of the interfaces in the OJAI framework.


Releases of HDocDB are deployed to Maven Central.



You can also choose to build HDocDB manually. Prerequisites for building:

git clone https://github.com/rayokota/hdocdb.git
cd hdocdb
mvn clean package -DskipTests


Currently HDocDB does not make use of coprocessors. However, HDocDB does make use of server-side filters. To deploy HDocDB:


To initialize HDocDB, an HBase connection is required. For example,

Configuration config = HBaseConfiguration.create();
Connection conn = ConnectionFactory.createConnection(config);
HDocumentDB hdocdb = new HDocumentDB(conn);

Next is to obtain a document collection.

HDocumentCollection coll = hdocdb.getCollection("mycollection");

Each document collection is backed by an HBase table.

Creating Documents

Once a document collection is in hand, creating documents is straightforward.

Document doc = new HDocument()
    .set("firstName", "John")
    .set("lastName", "Doe")
    .set("dateOfBirth", ODate.parse("1970-10-10"));

You can also use the insertOrReplace() method, which will replace the document with the same ID if it already exists.


Retrieving Documents

To retrieve all documents in a collection, use the find() method.

DocumentStream docs = coll.find();

To retrieve a single document by ID, use the findById() method.

Document doc = coll.findById("jdoe");

You can also pass a condition to the find() method.

QueryCondition condition = new HQueryCondition()
    .is("lastName", QueryCondition.Op.EQUAL, "Doe")
    .is("dateOfBirth", QueryCondition.Op.LESS, ODate.parse("1981-01-01"))
DocumentStream docs = coll.find(condition);

Updating Documents

To update a document, first create a document mutation.

DocumentMutation mutation = new HDocumentMutation()
    .setOrReplace("firstName", "Jim")
    .setOrReplace("dateOfBirth", ODate.parse("1970-10-09"));
coll.update("jdoe", mutation);

Here are the different types of methods supported with HDocumentMutation.

All of the methods other than the setOrReplace() method perform a read-modify-write at the client side.

Deleting Documents

To delete a document:


Saving and Retrieving Objects

Since OJAI has Jackson integration, HDocDB can treat HBase as an object store. Assuming your Java class is annotated as follows:

public class User {

    private String id;
    private String firstName;
    private String lastName;

    public User(@JsonProperty("_id")       String id,
                @JsonProperty("firstName") String firstName,
                @JsonProperty("lastName")  String lastName) {
        this.id = id;
        this.firstName = firstName;
        this.lastName = lastName;

    public String getId() { return id; }

    public String getFirstName() { return firstName; }

    public String getLastName() { return lastName; }

Then instances of your class can be saved and retrieved using HDocDB.

User user = new User("jsmith", "John", "Smith");
Document doc = Json.newDocument(user);
user = coll.findById("jsmith").toJavaBean(User.class);

Global Secondary Indexes

HDocDB also has basic support for global secondary indexes. For more sophisticated indexing support, an engine that can perform full text searches, such as ElasticSearch or Solr, is recommended.

Index management is performed mostly on the client-side, so it is not as performant as a coprocessor-based solution such as that provided by Apache Phoenix. Also, covered indexes are not supported, so each index lookup requires a join. However, the currrent index implementation should still help speed up some reads (at the cost of slightly slower writes).

To create a secondary index on the lastName field:

coll.createIndex("myindex" "lastName", Value.Type.STRING);

If the index is created after documents have already been added to the database, then the index will be populated in the background asynchronously. Since the indexing is performed on the client, this may take some time for a large collection.

Now, when performing a query such as the following, the index above will be used.

QueryCondition condition = new HQueryCondition()
    .is("lastName", QueryCondition.Op.EQUAL, "Doe")
    .is("dateOfBirth", QueryCondition.Op.LESS, ODate.parse("1981-01-01"))
DocumentStream docs = coll.find(condition);

A query will use at most one index. We can verify which index was used as follows.


which should print the following.

    "plan": "index scan",
    "indexName": "myindex",
    "indexBounds": {"lastName": "[Doe‥Doe]"},
    "staleIndexesRunningCount": 0

We can also specify which index to use.

DocumentStream docs = coll.findWithIndex("myindex", condition);

Or that no index should be used.

DocumentStream docs = coll.findWithIndex(Index.NONE, condition);

You can also create compound indexes.

IndexBuilder builder = coll.newIndexBuilder("myindex2")
    .add("lastName", Value.Type.STRING)
    .add("firstName", Value.Type.STRING)

HDocDB Shell with Nashorn Integration

The HDocDB shell is a command-line shell with Nashorn integration, so that MongoDB-like queries can be specified interactively or in a Nashorn script.

To start the HDocDB shell you need to use jrunscript that comes with Java (typically found in $JAVA_HOME/bin).

$ jrunscript -cp <hbase-conf-dir>:target/hdocdb-1.0.0.jar -f target/classes/shell/hdocdb.js -f - 

Here is a sample run.

nashorn> db.mycoll.insert( { _id: "jdoe", first_name: "John", last_name: "Doe" } )
nashorn> var doc = db.mycoll.find( { last_name: "Doe" } )[0]
nashorn> print(doc)
nashorn> db.mycoll.update( { last_name: "Doe" }, { $set: { first_name: "Jim" } } )
nashorn> var doc = db.mycoll.find( { last_name: "Doe" } )[0]
nashorn> print(doc)
nashorn> db.mycoll.delete( "jdoe" )

To run a script:

$ jrunscript -cp <hbase-conf-dir>:target/hdocdb-1.0.0.jar -f target/classes/shell/hdocdb.js -f <script>

Implementation Notes

Each document is stored as a separate row in HBase. This allows multiple operations on a document to be performed together atomically. The document is essentially "shredded" using a technique called key-flattening, as described in the Argo paper. That technique was developed for use with a relational database, but in HDocDB it has been adapted for HBase.

The implementation of global secondary indexes is based on blogs by Hofhansl and Yates.