Run your own transactions

When querying the database with execute_query(), the driver automatically creates a transaction. A transaction is a unit of work that is either committed in its entirety or rolled back on failure. You can include multiple Cypher statements in a single query, as for example when using MATCH and CREATE in sequence to update the database, but you cannot have multiple queries and interleave some client-logic in between them.

For these more advanced use-cases, the driver provides functions to take full control over the transaction lifecycle. These are called managed transactions, and you can think of them as a way of unwrapping the flow of execute_query() and being able to specify its desired behavior in more places.

Create a session

Before running a transaction, you need to obtain a session. Sessions act as concrete query channels between the driver and the server, and ensure causal consistency is enforced.

Sessions are created with the method Driver.session(), with the keyword argument database allowing to specify the target database. For further parameters, see Session configuration.

with driver.session(database="neo4j") as session:

Session creation is a lightweight operation, so sessions can be created and destroyed without significant cost. Always close sessions when you are done with them.

Sessions are not thread safe: you can share the main Driver object across threads, but make sure each thread creates its own sessions.

Run a managed transaction

A transaction can contain any number of queries. As Neo4j is ACID compliant, queries within a transaction will either be executed as a whole or not at all: you cannot get a part of the transaction succeeding and another failing. Use transactions to group together related queries which work together to achieve a single logical database operation.

A managed transaction is created with the methods Session.execute_read() and Session.execute_write(), depending on whether you want to retrieve data from the database or alter it. Both methods take a transaction function callback, which is responsible for actually carrying out the queries and processing the result.

Retrieve people whose name starts with Al.
def match_person_nodes(tx, name_filter): (3)
    result =""" (4)
        MATCH (p:Person) WHERE STARTS WITH $filter
        RETURN AS name ORDER BY name
        """, filter=name_filter)
    return list(result)  # return a list of Record objects (5)

with driver.session(database="neo4j") as session:  (1)
    people = session.execute_read(  (2)
    for person in people:
        print(  # obtain dict representation
1 Create a session. A single session can be the container for multiple queries. Unless created using the with construct, remember to close it when done.
2 The .execute_read() (or .execute_write()) method is the entry point into a transaction. It takes a callback to a transaction function and an arbitrary number of positional and keyword arguments which are handed down to the transaction function.
3 The transaction function callback is responsible of running queries.
4 Use the method to run queries. Each query run returns a Result object.
5 Process the result using any of the methods on Result.

Do not hardcode or concatenate parameters directly into the query. Use query parameters instead, both for performance and security reasons.

Transaction functions should never return the Result object directly. Instead, always process the result in some way; at minimum, cast it to list. Within a transaction function, a return statement results in the transaction being committed, while the transaction is automatically rolled back if an exception is raised.

The methods .execute_read() and .execute_write() have replaced .read_transaction() and .write_transaction(), which are deprecated in version 5.x and will be removed in version 6.0.
A transaction with multiple queries, client logic, and potential roll backs
from neo4j import GraphDatabase

URI = "<URI for Neo4j database>"
AUTH = ("<Username>", "<Password>")

def main():
    with GraphDatabase.driver(URI, auth=AUTH) as driver:
        with driver.session(database="neo4j") as session:
            for i in range(100):
                name = f"Thor{i}"
                org_id = session.execute_write(employ_person_tx, name)
                print(f"User {name} added to organization {org_id}")

def employ_person_tx(tx, name):
    # Create new Person node with given name, if not exists already
    result ="""
        MERGE (p:Person {name: $name})
        RETURN AS name
        """, name=name

    # Obtain most recent organization ID and the number of people linked to it
    result ="""
        MATCH (o:Organization)
        RETURN AS id, COUNT{(p:Person)-[r:WORKS_FOR]->(o)} AS employees_n
        ORDER BY o.created_date DESC
        LIMIT 1
    org = result.single()

    if org is not None and org["employees_n"] == 0:
        raise Exception("Most recent organization is empty.")
        # Transaction will roll back -> not even Person is created!

    # If org does not have too many employees, add this Person to that
    if org is not None and org.get("employees_n") < employee_threshold:
        result ="""
            MATCH (o:Organization {id: $org_id})
            MATCH (p:Person {name: $name})
            MERGE (p)-[r:WORKS_FOR]->(o)
            RETURN $org_id AS id
            """, org_id=org["id"], name=name

    # Otherwise, create a new Organization and link Person to it
        result ="""
            MATCH (p:Person {name: $name})
            CREATE (o:Organization {id: randomuuid(), created_date: datetime()})
            MERGE (p)-[r:WORKS_FOR]->(o)
            RETURN AS id
            """, name=name

    # Return the Organization ID to which the new Person ends up in
    return result.single()["id"]

if __name__ == "__main__":

Should a transaction fail for a reason that the driver deems transient, it automatically retries to run the transaction function (with an exponentially increasing delay). For this reason, transaction functions must be idempotent (i.e., they should produce the same effect when run several times), because you do not know upfront how many times they are going to be executed. In practice, this means that you should not edit nor rely on globals, for example. Note that although transactions functions might be executed multiple times, the queries inside it will always run only once.

A session can chain multiple transactions, but only one single transaction can be active within a session at any given time. To maintain multiple concurrent transactions, use multiple concurrent sessions.

Transaction function configuration

The decorator unit_of_work() allows to exert further control on transaction functions. It allows to specify:

  • a transaction timeout (in seconds). Transactions that run longer will be terminated by the server. The default value is set on the server side. The minimum value is one millisecond (0.001).

  • a dictionary of metadata that gets attached to the transaction. These metadata get logged in the server query.log, and are visible in the output of the SHOW TRANSACTIONS Cypher command. Use this to tag transactions.

from neo4j import unit_of_work

@unit_of_work(timeout=5, metadata={"app_name": "people_tracker"})
def count_people(tx):
    result ="MATCH (a:Person) RETURN count(a) AS people")
    record = result.single()
    return record["people"]

with driver.session(database="neo4j") as session:
    people_n = session.execute_read(count_people)

Run an explicit transaction

You can achieve full control over transactions by manually beginning one with the method Session.begin_transaction(). You may then run queries inside an explicit transaction with the method

with driver.session(database="neo4j") as session:
    with session.begin_transaction() as tx:
        # use to run queries and tx.commit() when done"<QUERY 1>")"<QUERY 2>")


An explicit transaction can be committed with Transaction.commit() or rolled back with Transaction.rollback(). If no explicit action is taken, the driver will automatically roll back the transaction at the end of its lifetime.

Explicit transactions are most useful for applications that need to distribute Cypher execution across multiple functions for the same transaction, or for applications that need to run multiple queries within a single transaction but without the automatic retries provided by managed transactions.

An explicit transaction example involving an external API
import neo4j

URI = "<URI for Neo4j database>"
AUTH = ("<Username>", "<Password>")

def main():
    with neo4j.GraphDatabase.driver(URI, auth=AUTH) as driver:
        customer_id = create_customer(driver)
        other_bank_id = 42
        transfer_to_other_bank(driver, customer_id, other_bank_id, 999)

def create_customer(driver):
    result, _, _ = driver.execute_query("""
        MERGE (c:Customer {id: rand()})
        RETURN AS id
    """, database_ = "neo4j")
    return result[0]["id"]

def transfer_to_other_bank(driver, customer_id, other_bank_id, amount):
    with driver.session(database="neo4j") as session:
        with session.begin_transaction() as tx:
            if not customer_balance_check(tx, customer_id, amount):
                # give up

            other_bank_transfer_api(customer_id, other_bank_id, amount)
            # Now the money has been transferred => can't rollback anymore
            # (cannot rollback external services interactions)

                decrease_customer_balance(tx, customer_id, amount)
            except Exception as e:
                request_inspection(customer_id, other_bank_id, amount, e)
                raise  # roll back

def customer_balance_check(tx, customer_id, amount):
    query = ("""
        MATCH (c:Customer {id: $id})
        RETURN c.balance >= $amount AS sufficient
    result =, id=customer_id, amount=amount)
    record = result.single(strict=True)
    return record["sufficient"]

def other_bank_transfer_api(customer_id, other_bank_id, amount):
    # make some API call to other bank

def decrease_customer_balance(tx, customer_id, amount):
    query = ("""
        MATCH (c:Customer {id: $id})
        SET c.balance = c.balance - $amount
    result =, id=customer_id, amount=amount)

def request_inspection(customer_id, other_bank_id, amount, e):
    # manual cleanup required; log this or similar
    print("WARNING: transaction rolled back due to exception:", repr(e))
    print("customer_id:", customer_id, "other_bank_id:", other_bank_id,
          "amount:", amount)

if __name__ == "__main__":

Process query results

The driver’s output of a query is a Result object, which encapsulates the Cypher result in a rich data structure that requires some parsing on the client side. There are two main points to be aware of:

  • The result records are not immediately and entirely fetched and returned by the server. Instead, results come as a lazy stream. In particular, when the driver receives some records from the server, they are initially buffered in a background queue. Records stay in the buffer until they are consumed by the application, at which point they are removed from the buffer. When no more records are available, the result is exhausted.

  • The result acts as a cursor. This means that there is no way to retrieve a previous record from the stream, unless you saved it in an auxiliary data structure.

The animation below follows the path of a single query: it shows how the driver works with result records and how the application should handle results.

The easiest way of processing a result is by casting it to list, which yields a list of Record objects. Otherwise, a Result object implements a number of methods for processing records. The most commonly needed ones are listed below.

Name Description

value(key=0, default=None)

Return the remainder of the result as a list. If key is specified, only the given property is included, while default allows to specify a value for nodes lacking that property.


Return up to n records from the result.


Return the next and only remaining record, or None. Calling this method always exhausts the result.

If more (or less) than one record is available,

  • strict==False — a warning is generated and the first of these is returned (if any);

  • strict==True — a ResultNotSingleError is raised.


Return the next record from the result without consuming it. This leaves the record in the buffer for further processing.


Return a JSON-like dump of the raw result. Only use it for debugging/prototyping purposes.


Return the query result summary. It exhausts the result, so should only be called when data processing is over.

For a complete list of Result methods, see API documentation — Result.

Session configuration

Database selection

It is recommended to always specify the database explicitly with the database parameter, even on single-database instances. This allows the driver to work more efficiently, as it saves a network round-trip to the server to resolve the home database. If no database is given, the default database set in the Neo4j instance settings is used.

with driver.session(
    database="neo4j"  # mark-line
) as session:
Specifying the database through the configuration method is preferred over the USE Cypher clause. If the server runs on a cluster, queries with USE require server-side routing to be enabled. Queries may also take longer to execute as they may not reach the right cluster member at the first attempt, and need to be routed to one containing the requested database.

Request routing

In a cluster environment, all sessions are opened in write mode, routing them to the leader. You can change this by explicitly setting the default_access_mode parameter to either neo4j.READ_ACCESS or neo4j.WRITE_ACCESS. Note that .execute_read() and .execute_write() automatically override the session’s default access mode.

import neo4j

with driver.session(
    default_access_mode=neo4j.READ_ACCESS  # mark-line
) as session:

Although executing a write query in read mode likely results in a runtime error, you should not rely on this for access control. The difference between the two modes is that read transactions will be routed to any node of a cluster, whereas write ones will be directed to the leader. In other words, there is no guarantee that a write query submitted in read mode will be rejected.

Similar remarks hold for the .executeRead() and .executeWrite() methods.

Run queries as a different user (impersonation)

You can execute a query under the security context of a different user with the parameter impersonated_user, specifying the name of the user to impersonate. For this to work, the user under which the Driver was created needs to have the appropriate permissions. Impersonating a user is cheaper than creating a new Driver object.

with driver.session(
    impersonated_user="somebody_else"  # mark-line
) as session:

When impersonating a user, the query is run within the complete security context of the impersonated user and not the authenticated user (i.e., home database, permissions, etc.).

Close sessions

Each connection pool has a finite number of sessions, so if you open sessions without ever closing them, your application could run out of them. It is thus recommended to create sessions using the with statement, which automatically closes them when the application is done with them. When a session is closed, it is returned to the connection pool to be later reused.

If you do not use with, remember to call the .close() method when you have finished using a session.

session = driver.session(database="neo4j")

# session usage




A Long Term Support release is one guaranteed to be supported for a number of years. Neo4j 4.4 is LTS, and Neo4j 5 will also have an LTS version.


Aura is Neo4j’s fully managed cloud service. It comes with both free and paid plans.


Cypher is Neo4j’s graph query language that lets you retrieve data from the database. It is like SQL, but for graphs.


Awesome Procedures On Cypher (APOC) is a library of (many) functions that can not be easily expressed in Cypher itself.


Bolt is the protocol used for interaction between Neo4j instances and drivers. It listens on port 7687 by default.


Atomicity, Consistency, Isolation, Durability (ACID) are properties guaranteeing that database transactions are processed reliably. An ACID-compliant DBMS ensures that the data in the database remains accurate and consistent despite failures.

eventual consistency

A database is eventually consistent if it provides the guarantee that all cluster members will, at some point in time, store the latest version of the data.

causal consistency

A database is causally consistent if read and write queries are seen by every member of the cluster in the same order. This is stronger than eventual consistency.


The null marker is not a type but a placeholder for absence of value. For more information, see Cypher → Working with null.


A transaction is a unit of work that is either committed in its entirety or rolled back on failure. An example is a bank transfer: it involves multiple steps, but they must all succeed or be reverted, to avoid money being subtracted from one account but not added to the other.


Backpressure is a force opposing the flow of data. It ensures that the client is not being overwhelmed by data faster than it can handle.

transaction function

A transaction function is a callback executed by an execute_read or execute_write call. The driver automatically re-executes the callback in case of server failure.


A Driver object holds the details required to establish connections with a Neo4j database.