如何在Neo4J中通过单个查询执行多LOAD CSV子查询并利用单个文件生成默认图数据库

Let's break down your problem and walk through a solution that mirrors the "single script" workflow you're used to with MySQL.

Why Your Original Query Failed #

Your initial approach of stacking multiple LOAD CSV statements in one query caused an OutOfMemoryError because:

• Neo4j loads all rows from every CSV into memory at once when they're in the same query, then attempts to create all nodes in a single transaction. Even if each CSV is small alone, combining them overwhelms memory.
• USING PERIODIC COMMIT doesn't work here because it only supports implicit transactions (queries centered on a single LOAD CSV). Stacking multiple LOAD CSV turns it into an explicit transaction, hence the semantic error.

Step-by-Step Solution for a Single Initialization Script #

1. Clear Existing Data (Equivalent to DROP DATABASE) #

First, wipe the graph to start fresh (skip this if you don't need a full reset):

// Delete all nodes and their associated relationships
MATCH (n)
DETACH DELETE n;

DETACH DELETE is critical here—Neo4j requires you to remove relationships before deleting nodes, and this command handles both automatically.

2. Load Data in Batched, Separate Queries #

Split each CSV import into its own query block with batch processing to avoid memory overload. For Neo4j 3.x, use USING PERIODIC COMMIT; for Neo4j 4.0+, prefer CALL { ... } IN TRANSACTIONS (more flexible for complex logic).

Example: Load Ingredients

// Batch-import ingredient nodes
USING PERIODIC COMMIT 250
LOAD CSV WITH HEADERS FROM 'file:///C:/Users/Enes/CSV_import/ingredients.csv' AS row
CREATE (ing:Ingredient {
    name: row.ingredientName,
    ingredientName: row.ingredientName
});

Example: Load Users (with Data Type Conversion)

// Batch-import user nodes (convert string boolean to actual boolean type)
USING PERIODIC COMMIT 250
LOAD CSV WITH HEADERS FROM 'file:///C:/Users/Enes/CSV_import/users.csv' AS row FIELDTERMINATOR ';'
CREATE (user:User {
    name: row.userName,
    userName: row.userName,
    userEmail: row.userEmail,
    userPassword: row.userPassword,
    enabled: toBoolean(row.enabled)
});

Example: Load Recipes (with Numeric Type Conversion)

// Batch-import recipe nodes (convert string numbers to integers)
USING PERIODIC COMMIT 250
LOAD CSV WITH HEADERS FROM 'file:///C:/Users/Enes/CSV_import/recipes.csv' AS row FIELDTERMINATOR ';'
CREATE (recipe:Recipe {
    name: row.recipeName,
    recipeName: row.recipeName,
    prepTimeInMin: toInteger(row.prepTimeInMin),
    restTimeInMinutes: toInteger(row.restTimeInMinutes),
    prepText: row.prepText,
    people: toInteger(row.people),
    viewCount: toInteger(row.viewCount),
    difficultyName: row.difficultyName,
    mealTypeName: row.mealTimeName,
    createdByUser: row.createdByUser
});

3. Add Constraints & Indexes (For Integrity & Performance) #

Add these after importing nodes to enforce data uniqueness and speed up queries:

// Enforce unique user emails
CREATE CONSTRAINT user_email_unique FOR (user:User) REQUIRE user.userEmail IS UNIQUE;

// Index for fast ingredient lookups
CREATE INDEX ingredient_name_index FOR (ing:Ingredient) ON (ing.ingredientName);

4. Create Relationships (If Needed) #

If your data has connections (e.g., recipes using ingredients), add a separate query block:

// Batch-create relationships between recipes and ingredients
USING PERIODIC COMMIT 250
LOAD CSV WITH HEADERS FROM 'file:///C:/Users/Enes/CSV_import/recipe_ingredients.csv' AS row
MATCH (recipe:Recipe {recipeName: row.recipeName})
MATCH (ing:Ingredient {ingredientName: row.ingredientName})
CREATE (recipe)-[:CONTAINS_INGREDIENT]->(ing);

5. Create Views & Procedures #

For views (Neo4j Enterprise Edition only) and stored procedures, add these final blocks:

// Create a view for popular recipes
CREATE VIEW popularRecipes AS
MATCH (r:Recipe)
WHERE r.viewCount > 1000
RETURN r.recipeName, r.viewCount, r.difficultyName;

// Example stored procedure (requires deployment of custom logic)
CREATE PROCEDURE getRecipesByDifficulty(difficulty: String)
YIELD recipeName, viewCount
MATCH (r:Recipe {difficultyName: difficulty})
RETURN r.recipeName AS recipeName, r.viewCount AS viewCount;

How to Execute the Script #

• Neo4j Desktop: Paste the entire script into the query editor and click "Run"—Neo4j will execute each query block in separate transactions automatically.
• Command Line: Use cypher-shell to run the file directly:

  cypher-shell -u neo4j -p your_password -f /path/to/your/init_script.cypher
  

Key Takeaways #

• Never stack multiple LOAD CSV statements in one query—this floods memory with all row data at once.
• Use batch processing (USING PERIODIC COMMIT or CALL { ... } IN TRANSACTIONS) to split large imports into manageable chunks.
• A single .cypher file can contain all your initialization logic: data cleanup, imports, constraints, relationships, views, and procedures.

内容的提问来源于stack exchange，提问作者Enes Sabi