Question

我正在使用NEO4j 3.5来存储和查询人与人之间的关系。我有带有“用户”标签的节点和带有“朋友”标签的关系。我可以结识朋友，但查询时间太长。当前返回的时间是4秒到6秒。这不是高事务性的neo4j数据库，并且服务器具有大量可用的CPU和内存。服务器上的负载不到3个，共有8个核心。它在AWS EC2实例上运行。数据库中大约有25万个节点，数据库总大小不到750mb。

这是我当前正在使用的查询：

MATCH (user:User {user_id:1145})-[:FRIENDS*3]->(fof:User)
WHERE NOT (user:User)-[:FRIENDS]->(fof:User)
RETURN count(distinct fof.user_id)

此密码查询返回的计数为69,704，这是正确的。

可以对密码查询或NEO4j数据库引擎进行哪些优化以更快地返回结果？

执行计划

+-----------------------+----------------+--------+---------+-----------+-----------------------------+------------------+--------------------------------------------+
| Operator              | Estimated Rows | Rows   | DB Hits | Cache H/M | Identifiers                 | Ordered by       | Other                                      |
+-----------------------+----------------+--------+---------+-----------+-----------------------------+------------------+--------------------------------------------+
| +ProduceResults       |              0 |      1 |       0 |       0/0 | count(distinct fof.user_id) |                  | 0.0                                        |
| |                     +----------------+--------+---------+-----------+-----------------------------+------------------+--------------------------------------------+
| +EagerAggregation     |              0 |      1 |  326421 |       0/0 | count(distinct fof.user_id) |                  | 0.0                                        |
| |                     +----------------+--------+---------+-----------+-----------------------------+------------------+--------------------------------------------+
| +AntiSemiApply        |              0 | 256717 |       0 |       0/0 | anon[33], fof, user         | user.user_id ASC | 0.0                                        |
| |\                    +----------------+--------+---------+-----------+-----------------------------+------------------+--------------------------------------------+
| | +Expand(Into)       |              0 |      0 | 8006149 |       0/0 |   REL80, fof, user          |                  | 0.0; (user)-[  REL80:FRIENDS]->(fof)       |
| | |                   +----------------+--------+---------+-----------+-----------------------------+------------------+--------------------------------------------+
| | +Filter             |              1 | 260120 |  520240 |       0/0 | fof, user                   |                  | 0.0; fof:User                              |
| | |                   +----------------+--------+---------+-----------+-----------------------------+------------------+--------------------------------------------+
| | +Argument           |              1 | 260120 |       0 |       0/0 | fof, user                   |                  | 0.0                                        |
| |                     +----------------+--------+---------+-----------+-----------------------------+------------------+--------------------------------------------+
| +Filter               |              0 | 260120 |  260120 |       0/0 | anon[33], fof, user         | user.user_id ASC | 0.0; fof:User                              |
| |                     +----------------+--------+---------+-----------+-----------------------------+------------------+--------------------------------------------+
| +VarLengthExpand(All) |              0 | 260120 |  267999 |       0/0 | anon[33], fof, user         | user.user_id ASC | 0.0; (user)-[anon[33]:FRIENDS*3..3]->(fof) |
| |                     +----------------+--------+---------+-----------+-----------------------------+------------------+--------------------------------------------+
| +NodeIndexSeek        |              1 |      1 |       3 |       0/0 | user                        | user.user_id ASC | 0.0; :User(user_id)                        |
+-----------------------+----------------+--------+---------+-----------+-----------------------------+------------------+--------------------------------------------+

Answer 1

您的WHERE子句包含一个模式，该模式要求每个fof附加数据库命中数。您可以通过在内存中保留user的所有直接好友的列表，并更改您的WHERE子句以使其仅在列表中进行搜索，来避免这些数据库命中。（根据您的个人资料数据，这可以节省8006149 + 520240，或节省超过850万个数据库匹配-这是整个查询的大部分匹配。）
在您的查询中，如果同一fof节点多次匹配，则每次都将执行相同的WHERE测试。您可以通过在进行fof测试之前在过滤掉重复的WHERE个节点来避免这种情况。这也意味着您以后不再需要删除重复项。

例如：

MATCH (user:User {user_id:1145})-[:FRIENDS]->(f:User)
WITH user, COLLECT(f) AS friends
MATCH (user)-[:FRIENDS*3]->(fof:User)
WITH DISTINCT friends, fof
WHERE NOT fof IN friends
RETURN COUNT(fof)

查询响应时间慢-简单的好友密码查询

1 个答案: