How to load data?

There are several different ways of writing data into the system, optimized for different use cases.

Inline queries

For ad-hoc experimenting, it is usually enough to create individual nodes and edges directly with Cypher queries.

  (lilysr: Person { name: "Lily Potter", gender: "female", birth_year: 1960 }),
  (jamessr:Person { name: "James Potter", gender: "male", birth_year: 1960 }),
  (molly:Person { name: "Molly Weasley", gender: "female", birth_year: 1949 }),
  (arthur:Person { name: "Arthur Weasley", gender: "male", birth_year: 1950 }),
  (harry:Person { name: "Harry Potter", gender: "male", birth_year: 1980 }),
  (ginny:Person { name: "Ginny Weasley", gender: "female", birth_year: 1981 }),
  (ron:Person { name: "Ron Weasley", gender: "male", birth_year: 1980 }),
  (hermione:Person { name: "Hermione Granger", gender: "female", birth_year: 1979 }),
  (jamesjr:Person { name: "James Sirius Potter", gender: "male", birth_year: 2003 }),
  (albus:Person { name: "Albus Severus Potter", gender: "male", birth_year: 2005 }),
  (lilyjr:Person { name: "Lily Luna", gender: "female",  birth_year: 2007 }),
  (rose:Person { name: "Rose Weasley", gender: "female",  birth_year: 2005 }),
  (hugo:Person { name: "Hugo Weasley", gender: "male",  birth_year: 2008 }),

This sort of query can be entered in the Exploration UI, through cypher-shell, or directly via the REST API (see the “Cypher query language” section). Entering the above graph in the Exploration UI and then querying MATCH (n) RETURN n produces the following graph.

harry potter graph

Queries that read from files

For larger static datasets, it isn’t always feasible or convenient to be constructing large Cypher queries. If these datasets are CSVs or line-based JSON files that are publicly available on the web, it is is possible to write Cypher queries that will iterate through the records in the file, executing some query action for each entry.

For instance, consider the same Harry Potter dataset in a JSON file. Using the custom loadJsonLines procedure to load data from either a file or web URL, we can iterate over each record and create a node for it along with edges to its children.

CALL loadJsonLines("") YIELD value AS person
MATCH (p) WHERE id(p) = idFrom('name',
SET p = { name:, gender: person.gender, birth_year: person.birth_year }
SET p: Person
WITH person.children AS childrenNames, p
UNWIND childrenNames AS childName
MATCH (c) WHERE id(c) = idFrom('name', childName)
CREATE (c)-[:has_parent]->(p)

If the data is in a CSV format, you can use the LOAD CSV clause.


idFrom is a function that hashes its arguments into a valid ID. It takes an arbitrary number of arguments, so that multiple bits of data can factor into the deterministic ID. By convention, the first of these arguments is a string describing a namespace for the IDs being generated. This is important to avoid accidentally producing collisions in IDs that exist in different namespaces: idFrom('year', 2000) is different from idFrom('part number', 2000).

Streaming Data Ingest

Connect is engineered first and foremost as a stream processing system. Data ingest pipelines are almost always streams, and batch processing is something done for want of streaming capabilities. Batch processing is often used to work around other limitations of an ingest system (eg. slow query times and inability to properly trigger computation on new data). These are problems which we believe can be avoided entirely in Connect through judicious use of standing queries.

Connect supports defining ingest streams that connect to existing industry streaming systems such as Kafka and Kinesis. Since it is expected that ingest streams will run for long-periods of time, the REST API is designed to make it easy to

  • list or lookup the configuration of currently running streams as well as their ingest progress
  • create fresh new ingest streams
  • halt currently running ingest streams

See the “Ingest streams” section of the REST API for more details.