If your database is very large and you are still experiencing memory issues when deleting in transactions as shown above, you could try manually batching that whole operation - something like this:

//Unwinding Batched Delete

:auto

// Step 1: Get the node count and use it to set the batch size

CALL {

MATCH (n)

RETURN COUNT(n) AS nodeCount

}

WITH nodeCount, 1000 AS batchSize

WITH nodeCount, batchSize, toInteger(ceil(toFloat(nodeCount) / batchSize)) AS numberOfBatches

// Generate the series for the UNWIND

UNWIND range(0, numberOfBatches - 1) AS i

CALL {

WITH i, numberOfBatches, batchSize

MATCH (n)

WHERE id(n) % numberOfBatches = i

WITH n LIMIT 1000

DELETE n

RETURN COUNT(n) AS count

}

// You could experiment with the number of concurrent transactions and number of rows below to achieve the best throughput

// Or just try with the default values ('IN CONCURRENT TRANSACTIONS')

IN 2 CONCURRENT TRANSACTIONS OF 50 ROWS

ON ERROR FAIL

MATCH (n) DELETE n

RETURN nodeCount, numberOfBatches, batchSize

A first step would be to try executing the DETACH DELETE operation in transactions, like this:

MATCH (n)

CALL (n) {

DETACH DELETE n

} IN TRANSACTIONS

Where you have multiple processors available, you can ou can also try that with CONCURRENT TRANSACTIONS:

MATCH (n)

CALL (n) {

DETACH DELETE n

} IN CONCURRENT TRANSACTIONS

(The level of concurrency and batch size is configurable - please see the docs here for more details: https://neo4j.com/docs/cypher-manual/current/subqueries/subqueries-in-transactions/