Query Interface

Custom API

Cypher is a declarative, SQL-inspired language for describing patterns in graphs visually using an ascii-art syntax. One of the most powerful ability of Cypher is the notion to use -> operator to define relationships which are important in Graph Databases.

Views

Not Supported

Neo4j does not support the concept of views like traditional RDMS. However, one can create a special VIEW Node. Querying a view now implies querying the created view node, traversing the edges to find the results.

Concurrency Control

Two-Phase Locking (Deadlock Detection)

Neo4j uses locks for transactions which may lead to deadlock. Deadlocks are detected and the transaction is marked for rollback. The transaction may be retried if needed by the user. The retry logic is left to the user.

Logging

Logical Logging

Logical transaction logs in Neo4j are used in scenarios when the database needs to be recovered after a unclean shutdown. They are also used for online backup operations, especially for incremental backups. These transaction log files are rotated after surpassing a certain size (25 Mb in size). The amount of log files or the used space can be configured.

Joins

Hash Join

NodeHashJoin API provided in the documentation joins the input coming from the left with the input coming from the right. The support for ValueHashJoins was also merged with the github Neo4j 3.0 repository.

Storage Model

Custom

Each node and its properties are stored on the disk using a doubly linked list. Layout : byte 0 : 4 high bits of previous, 4 high bits of next pointers, bytes 1-4 : previous property record, bytes 5-8 : next property record, bytes 9-40 : payload

Isolation Levels

Read Committed Serializable

Neo4j supports ACID Transactions and the default isolation provided is READ_COMMITTED. The user can however acquire locks on nodes and relationships to achieve the SERIALIZABLE isolation. Deadlock Detection is inbuilt in the system.

Query Execution

Tuple-at-a-Time Model

Stored Procedures

Supported

Stored procedures in Neo4j are called using Cypher calls. Arguments can be supplied directly within the query or taken from the associated parameter set. Procedures can be written in Java, compiled as jar files and put into the database as plugins.

Storage Architecture

Disk-oriented

Neo4j has a disk-based, native storage manager optimised (optimisations typically revolve around using the cache effectively to avoid trips to the disk) for storing graph structures with maximum performance and scalability.

Checkpoints

Non-Blocking

Neo4j can be backed up while it continues to serve user traffic (called “online” backup). Neo4j offers two backup options: full or incremental. These strategies can be combined to provide the best mix of safety and efficiency. Depending on the risk profile of the system, a typical strategy might be to have daily full backups and hourly incremental backups, or weekly full backups with daily incremental backups.

Foreign Keys

Supported

The traditional notion of foreign keys in RDBMS is used in Neo4j as relationships.The simple abstraction of nodes and relationships in connected structures helps enable the concept of foreign keys in Neo4j.

System Architecture

Shared-Nothing

Neo4j offers a shared-nothing architecture with a single write master and multiple read replicas thus consumes significant instance based storage.

Query Compilation

Code Generation JIT Compilation

Neo4j uses (byte) code generation in addition to JIT compilation. The JIT compilation is provided by the JVM.