为什么这三个非常相似的neo4j cypher查询的性能差别很大？

时间：2014-11-25 05:13:10

标签： neo4j cypher database-performance

为什么第一个密码查询比第二个和第三个密码查询速度快得多？

第一次查询： 

MATCH (source:Product { product_id:14603 }), 
      (destination:Product{product_id:286502}), 
      p = (source-[r]-()-[*0..3]-destination) 
RETURN p, length(p) as pathLength LIMIT 50

==＆GT; 14秒

==> 50 rows
==> 
==> |             Operator | Rows | DbHits |                                      Identifiers |                                 Other |
==> +----------------------+------+--------+--------------------------------------------------+---------------------------------------+
==> |         ColumnFilter |   50 |      0 |                                                  |            keep columns p, pathLength |
==> |                Slice |   50 |      0 |                                                  |                          {  AUTOINT2} |
==> |              Extract |   50 |      0 |                                                  |                            pathLength |
==> |          ExtractPath |   50 |      0 |                                                p |                                       |
==> | SimplePatternMatcher |   50 |      0 | source,   UNNAMED90, destination, r,   UNNAMED89 |                                       |
==> |          SchemaIndex |   99 |    198 |                                   source, source |            {  AUTOINT0}; :Product(product_id) |
==> |     TraversalMatcher |   99 | 419669 |                                                  |   UNNAMED89,   UNNAMED90,   UNNAMED89 |
==> +----------------------+------+--------+--------------------------------------------------+---------------------------------------+
==> 
==> Total database accesses: 419867

第二次查询： 

MATCH (source:Product { product_id:14603 }), 
      (destination:Product{product_id:286502}), 
      p = (source-[r]-()-[*0..2]-destination) 
RETURN p, length(p) as pathLength LIMIT 50

==＆GT; 140秒

==> 13 rows
==>  
==> +----------------------+---------+---------+--------------------------------------------------+---------------------------------------+
==> |             Operator |    Rows |  DbHits |                                      Identifiers |                                 Other |
==> +----------------------+---------+---------+--------------------------------------------------+---------------------------------------+
==> |         ColumnFilter |      13 |       0 |                                                  |            keep columns p, pathLength |
==> |                Slice |      13 |       0 |                                                  |                          {  AUTOINT2} |
==> |              Extract |      13 |       0 |                                                  |                            pathLength |
==> |          ExtractPath |      13 |       0 |                                                p |                                       |
==> | SimplePatternMatcher |      13 |       0 | source,   UNNAMED90, destination, r,   UNNAMED89 |                                       |
==> |          SchemaIndex | 2266560 | 4533120 |                                   source, source |            {  AUTOINT0}; :Product(product_id) |
==> |     TraversalMatcher | 2266560 | 2266605 |                                                  |   UNNAMED89,   UNNAMED90,   UNNAMED89 |
==> +----------------------+---------+---------+--------------------------------------------------+---------------------------------------+
==> 
==> Total database accesses: 6799725

messages.log：

2014-11-26 03:41:18.282+0000 WARN  [o.n.k.EmbeddedGraphDatabase]: GC Monitor: Application threads blocked for an additional 139ms [total block time: 14.49s]
2014-11-26 03:41:20.086+0000 WARN  [o.n.k.EmbeddedGraphDatabase]: GC Monitor: Application threads blocked for an additional 101ms [total block time: 14.591s]
2014-11-26 03:41:29.022+0000 WARN  [o.n.k.EmbeddedGraphDatabase]: GC Monitor: Application threads blocked for an additional 144ms [total block time: 14.735s]
2014-11-26 03:41:43.230+0000 WARN  [o.n.k.EmbeddedGraphDatabase]: GC Monitor: Application threads blocked for an additional 151ms [total block time: 14.886s]
2014-11-26 03:41:45.160+0000 WARN  [o.n.k.EmbeddedGraphDatabase]: GC Monitor: Application threads blocked for an additional 127ms [total block time: 15.013s]
2014-11-26 03:41:48.830+0000 WARN  [o.n.k.EmbeddedGraphDatabase]: GC Monitor: Application threads blocked for an additional 135ms [total block time: 15.148s]
2014-11-26 03:42:01.956+0000 WARN  [o.n.k.EmbeddedGraphDatabase]: GC Monitor: Application threads blocked for an additional 134ms [total block time: 15.282s]
2014-11-26 03:42:05.731+0000 WARN  [o.n.k.EmbeddedGraphDatabase]: GC Monitor: Application threads blocked for an additional 171ms [total block time: 15.453s]
2014-11-26 03:42:07.555+0000 WARN  [o.n.k.EmbeddedGraphDatabase]: GC Monitor: Application threads blocked for an additional 122ms [total block time: 15.575s]
2014-11-26 03:42:13.271+0000 WARN  [o.n.k.EmbeddedGraphDatabase]: GC Monitor: Application threads blocked for an additional 144ms [total block time: 15.719s]
2014-11-26 03:42:36.465+0000 WARN  [o.n.k.EmbeddedGraphDatabase]: GC Monitor: Application threads blocked for an additional 115ms [total block time: 15.834s]

第三次查询： 

MATCH (source:Product { product_id:14603 }), 
      (destination:Product{product_id:286502}), 
      p = (source-[*0..3]-destination)
RETURN p, length(p) as pathLength LIMIT 50

==＆GT;无限运行时间（10分钟后最大化CPU，我重新启动neo4j服务器）

服务器数据：

  • Neo4J版本：community-2.1.1
  • 节点：650,000
  • 产品节点：550,000
  • 属性：8,000,000
  • 关系：6,000,000
  • 关系类型：9
  • 32GB RAM
  • 768GB SSD
  • 12核心CPU

索引：

ON：产品（product_id）ONLINE（用于唯一性约束）
[...]其他一些不是这些查询的相关索引

neo4j.properties

  • neostore.nodestore.db.mapped_memory = 50M
  • neostore.relationshipstore.db.mapped_memory = 400M
  • neostore.propertystore.db.mapped_memory = 400M
  • neostore.propertystore.db.strings.mapped_memory = 400M
  • neostore.propertystore.db.arrays.mapped_memory = 400M
  • node_cache_size = 5G
  • relationship_cache_size = 5G
  • 所有其他属性均为默认

非常感谢提前！

1 个答案:

答案 0 :(得分：0)

如果您尝试将每个模式作为单独的MATCH而不是逗号分隔，它会有什么不同吗？ e.g。

MATCH (source:Product { product_id:14603 })
MATCH  (destination:Product{product_id:286502})
MATCH p = (source-[r]-()-[*0..3]-destination) 
RETURN p, length(p) as pathLength LIMIT 50

此外，如果您在/ webadmin运行查询，那么您可以尝试在其前面添加单词＆＃39; PROFILE＆＃39;并将其输出粘贴到此处，我将查看它。

相关问题
最新问题