Neo4j是先MATCH再过滤还是先过滤再MATCH？查询逻辑与自查问询

Great question—let’s break this down so you can see exactly what’s happening under the hood.

执行逻辑：先过滤节点，再搜索路径（通常情况） #

Neo4j’s Cypher query planner is cost-based, meaning it picks the lowest-cost execution path possible. For your use case, it will almost always filter the eligible Left and Right nodes first (using your left_ids and right_ids sets) before trying to find paths between them.

Here’s why: generating a Cartesian product of all Left and Right nodes and then filtering would be astronomically expensive (especially with large datasets), and the planner is designed to avoid that. Instead, it’ll first narrow down the nodes to only those in your ID sets, then look for paths connecting those pre-filtered nodes.

That said, the exact behavior depends on how you structure your query. For example, if you write a query that doesn’t leverage labels or indexes for your nodes, the planner might have to do more work—but even then, it’s unlikely to default to a Cartesian product first.

直观自查方式：用执行计划工具 #

You don’t have to guess—Neo4j gives you tools to see exactly what the planner is doing:

• EXPLAIN command: Prepending EXPLAIN to your query will show you the planned execution steps without actually running the query. Look for steps like NodeByIdLookup (if you’re using node IDs) or NodeByLabelScan followed by a filter on the ID sets. If these steps come before any path-matching operations (like Expand), that confirms the planner is filtering nodes first.
• PROFILE command: This runs the query and shows you the actual execution stats (like how many rows each step processed, time taken, etc.). It’s even more concrete than EXPLAIN because it’s based on real execution data. Compare the row counts for the node-filtering steps vs. path-matching steps—you’ll see the filtered node sets are much smaller.

The official Neo4j documentation also has deep dives into the Cypher query optimizer, including how it prioritizes low-cost operations like node filtering before expensive path traversals.

内容的提问来源于stack exchange，提问作者danodonovan