Question

我的图表由 Thing ， Link 和 Tag 节点组成。 Thing 可以：CONNECT 编辑到链接和标记。该图大小约为45GB，具有4个节点和84个关系。节点仅包含一个字段，其中包含用于外部系统引用的键。

我在我的图表中寻找 Thing s（t2），它通过 Link 与另一个 Thing （t1）紧密相连。

Cypher查询例如：

MATCH (t1:Thing {key: "a80b3828-6fec-4031-b552-d3397d1737b7"}) 
MATCH p=(t1)-[:CONNECT]->(l:Link)<-[:CONNECT]-(t2:Thing) 
WHERE t1<>t2 WITH t2, count(DISTINCT p) AS links 
RETURN t2, links 
ORDER BY links DESC 
LIMIT 25

结果：

+------------------------------------------------------------------+
| t2                                                       | links |
+------------------------------------------------------------------+
| Node[166590]{key:"4b06471e-0849-4e56-b5c9-6bc04730c899"} | 854   |
| Node[190480]{key:"8934c635-17de-449a-9437-a24857b8b1c6"} | 698   |
...
| Node[754437]{key:"925b4f3a-0c69-46b9-8f35-294b9539a98b"} | 345   |
| Node[656436]{key:"50424c32-8ce8-495b-a8d3-f4864b1c3adc"} | 342   |
+------------------------------------------------------------------+
25 rows
18232 ms

表现不是很令人惊讶。但这对我来说不是主要问题。

我现在想在我的图表中查找通过链接 Thing （t1）的 Thing s（t2） >并连接到特定的标记。

MATCH (tag:`Tag` {key: "da8115ff-95fb-46d7-bd14-24bdd16f0a04"}) 
MATCH (t1:Thing {key: "a80b3828-6fec-4031-b552-d3397d1737b7"}) 
MATCH p=(t1)-[:CONNECT]->(l:Link)<-[:CONNECT]-(t2:Thing)-[:CONNECT]->(tag) 
WHERE t1<>t2 
RETURN t2, count(DISTINCT p) AS links 
ORDER BY links DESC 
LIMIT 25

问题是，此查询将永远运行。在我看来，这应该比第一个Cpyher查询运行得更快，因为我感兴趣的图形部分要小得多？

我正在使用运行在80GB SDD上的Neo4j 2.1.2，用于Java的16GB RAM，用于具有4核的OS的16GB RAM。

Answer 1

我认为有两种不同的关系类型会有所帮助。

您在:Tag(key)上有索引或约束吗？

当您不限制/订购时，会返回多少路径？如果你只使用没有订单的限制，性能差异是什么？

MATCH (t1:Thing {key: "a80b3828-6fec-4031-b552-d3397d1737b7"})-[:CONNECT]->
      (l:Link)<-[:CONNECT]-(t2:Thing) 
RETURN t2, count(DISTINCT p) AS links 
ORDER BY links DESC LIMIT 25

您可以在查询中尝试新的实验性查询工具吗？

CYPHER 2.1.experimental
MATCH (t1:Thing {key: "a80b3828-6fec-4031-b552-d3397d1737b7"})-[:CONNECT]->
      (l:Link)<-[:CONNECT]-(t2:Thing) 
RETURN t2, count(DISTINCT p) AS links 
ORDER BY links DESC LIMIT 25

<强>更新

试试这个：

MATCH (t1:Thing {key: "a80b3828-6fec-4031-b552-d3397d1737b7"})-[:CONNECT]->(l)
WITH distinct l,t1
MATCH (l)<-[:CONNECT]-(t2)  
WHERE t1<>t2  
RETURN t2, count(distinct l) AS links  
ORDER BY links DESC 
LIMIT 25

Answer 2

观看http://vimeo.com/84900121后我最快的版本是：

MATCH (t1:Thing {key: "a80b3828-6fec-4031-b552-d3397d1737b7"}) 
MATCH (t1)-[:CONNECT]->(l:Link) MATCH (l)<-[:CONNECT]-(t2:Thing) 
WHERE t1<>t2 WITH t2, count(l) AS links 
ORDER BY links DESC LIMIT 25 RETURN t2, links

并需要7520毫秒。

使用新的实验性查询器：

CYPHER 2.1.experimental 
MATCH (t1:Thing {key: "a80b3828-6fec-4031-b552-d3397d1737b7"})  
MATCH (t1)-[:CONNECT]->(l:Link)  
MATCH (l)<-[:CONNECT]-(t2:Thing)  
WHERE t1<>t2  
WITH t2, count(l) AS links  
ORDER BY links DESC LIMIT 25  
RETURN t2, links

需要11942毫秒

您的查询

MATCH p=(t1:Thing {key: "a80b3828-6fec-4031-b552-d3397d1737b7"})-[:CONNECT]->(l:Link)<-[:CONNECT]-(t2:Thing)  
RETURN t2, count(DISTINCT p) AS links  
ORDER BY links DESC LIMIT 25

需要13315毫秒删除ORDER BY没有效果。

使用新的实验性查询器：

CYPHER 2.1.experimental  
MATCH p=(t1:Thing {key: "a80b3828-6fec-4031-b552-d3397d1737b7"})-[:CONNECT]->(l:Link)<-[:CONNECT]-(t2:Thing)  
RETURN t2, count(DISTINCT p) AS links  
ORDER BY links DESC LIMIT 25

需要21200毫秒。

典型结果如下：

+------------------------------------------------------------------+
| t2                                                       | links |
+------------------------------------------------------------------+
| Node[166590]{key:"4b06471e-0849-4e56-b5c9-6bc04730c899"} | 854   |
| Node[190480]{key:"8934c635-17de-449a-9437-a24857b8b1c6"} | 698   |
| Node[153785]{key:"476f9756-e829-495c-82a2-6d05efa24a5a"} | 665   |
| Node[217705]{key:"4e69f054-3828-4c9c-829d-bc1b94499d26"} | 595   |
| Node[434466]{key:"372832f9-338b-4818-b7df-f60ec68e0dbc"} | 586   |
| Node[188948]{key:"ed023053-9c4d-4978-b884-cd7a77f9d610"} | 580   |
| Node[129093]{key:"49dfa255-8455-4f90-962d-63e9244b4337"} | 576   |
| Node[249272]{key:"fbdcbba8-c47d-4e4d-bae1-0ebadf6e7692"} | 553   |
| Node[253605]{key:"4c7848aa-eff7-4cd8-ab1e-79cb9a17fe59"} | 533   |
| Node[149177]{key:"9e5e830e-8ebd-456d-9e0c-094b5fc6964b"} | 519   |
| Node[155016]{key:"0d1148ff-3040-4c45-8198-a5918dd96721"} | 503   |
| Node[346978]{key:"640812b0-996e-4a1b-9a55-80417b010403"} | 478   |
| Node[256312]{key:"a16b233f-e363-4883-bc4c-9ccb1e4d1738"} | 463   |
| Node[203843]{key:"e739384f-0b1d-4f47-9abb-debfdc438796"} | 421   |
| Node[632642]{key:"8788d644-11ff-4a49-805e-32268183a712"} | 420   |
| Node[222103]{key:"eb4cdb73-67b7-46aa-a26f-1612a855a7d4"} | 419   |
| Node[156930]{key:"1ee362fb-0308-45e4-9dbc-1286e4571cb9"} | 414   |
| Node[445506]{key:"c1d50c68-748c-4551-ab4f-7313bf38ec70"} | 407   |
| Node[205489]{key:"fcf2f12a-b6cf-4d74-afbc-b75f1e57dd40"} | 384   |
| Node[134227]{key:"d44a2c49-35dd-44c1-82cb-c1386bb1beec"} | 368   |
| Node[360284]{key:"8fbed168-9183-455e-8869-350fbf9310f4"} | 365   |
| Node[223032]{key:"4424b8e8-d4aa-43dc-8b21-d6c119645092"} | 359   |
| Node[769577]{key:"8b538f97-8152-4552-aa6a-eadd78cfbe9c"} | 358   |
| Node[754437]{key:"925b4f3a-0c69-46b9-8f35-294b9539a98b"} | 345   |
| Node[656436]{key:"50424c32-8ce8-495b-a8d3-f4864b1c3adc"} | 342   |
+------------------------------------------------------------------+

更改关系类型需要一些时间:)所以我稍后会回来查看数字。

如果图形的一部分变小，Cypher查询会变慢

2 个答案: