我有下一个图表模型：
(:PaveView {Number:int, Page:string}), (:Page {Name:string})
(:PageView)-[:At]->(:Page)
(:PageView)-[:Next]->(:PageView)
架构：
Indexes
ON :Page(Name) ONLINE (for uniqueness constraint)
ON :PageView(Page) ONLINE
ON :PageView(Revision) ONLINE (for uniqueness constraint)
Constraints
ON (pageview:PageView) ASSERT pageview.Number IS UNIQUE
ON (page:Page) ASSERT page.Name IS UNIQUE
我想做类似于post
的事情我试图找到这种结构的没有循环的流行路径：
(:PageView)-[:Next*2]->(:PageView)
我的尝试：
1。来自post
MATCH p = (:PageView)-[:Next*2]->(:PageView)
WITH p, EXTRACT(v IN NODES(p) | v.Page) AS pages
UNWIND pages AS views
WITH p, COUNT(DISTINCT views) AS distinct_views
WHERE distinct_views = LENGTH(NODES(p))
RETURN EXTRACT(v in NODES(p) | v.Page), count(p)
ORDER BY count(p) DESC
LIMIT 10;
个人资料输出：
10 rows
177270 ms
Compiler CYPHER 2.2-rule
ColumnFilter(0)
|
+Extract(0)
|
+ColumnFilter(1)
|
+Top
|
+EagerAggregation(0)
|
+Extract(1)
|
+ColumnFilter(2)
|
+Filter(0)
|
+Extract(2)
|
+ColumnFilter(3)
|
+EagerAggregation(1)
|
+UNWIND
|
+ColumnFilter(4)
|
+Extract(3)
|
+ExtractPath
|
+Filter(1)
|
+TraversalMatcher
+---------------------+---------+----------+------------------------------------------------------------------------+------------------------------------------------------------------------------------------------+
| Operator | Rows | DbHits | Identifiers | Other |
+---------------------+---------+----------+------------------------------------------------------------------------+------------------------------------------------------------------------------------------------+
| ColumnFilter(0) | 10 | 0 | EXTRACT(v in NODES(p) | v.Page), count(p) | keep columns EXTRACT(v in NODES(p) | v.Page), count(p) |
| Extract(0) | 10 | 0 | FRESHID225, FRESHID258, EXTRACT(v in NODES(p) | v.Page), count(p) | EXTRACT(v in NODES(p) | v.Page), count(p) |
| ColumnFilter(1) | 10 | 0 | FRESHID225, FRESHID258 | keep columns , |
| Top | 10 | 0 | FRESHID225, INTERNAL_AGGREGATEf7fa022b-cdb5-4ef2-bec5-a2f4f10706b6 | { AUTOINT0}; Cached( INTERNAL_AGGREGATEf7fa022b-cdb5-4ef2-bec5-a2f4f10706b6 of type Integer) |
| EagerAggregation(0) | 212828 | 0 | FRESHID225, INTERNAL_AGGREGATEf7fa022b-cdb5-4ef2-bec5-a2f4f10706b6 | |
| Extract(1) | 1749120 | 10494720 | FRESHID225, distinct_views, p | |
| ColumnFilter(2) | 1749120 | 0 | distinct_views, p | keep columns distinct_views, p |
| Filter(0) | 1749120 | 0 | FRESHID196, distinct_views, p | CoercedPredicate(anon[196]) |
| Extract(2) | 2115766 | 0 | FRESHID196, distinct_views, p | |
| ColumnFilter(3) | 2115766 | 0 | distinct_views, p | keep columns p, distinct_views |
| EagerAggregation(1) | 2115766 | 0 | INTERNAL_AGGREGATEb0939c81-a40c-4012-afd6-4852b17cf2e4, p | p |
| UNWIND | 6347298 | 0 | p, pages, views | |
| ColumnFilter(4) | 2115766 | 0 | p, pages | keep columns p, pages |
| Extract(3) | 2115766 | 12694596 | p, pages | pages |
| ExtractPath | 2115766 | 0 | p | |
| Filter(1) | 2115766 | 2115766 | | hasLabel(anon[34]:PageView(0)) |
| TraversalMatcher | 2115766 | 16926150 | | , , , |
+---------------------+---------+----------+------------------------------------------------------------------------+------------------------------------------------------------------------------------------------+
Total database accesses: 42231232
2。
match (p1:PageView)-[:Next]->(p2:PageView)-[:Next]->(p3:PageView)
where p1.Page<>p2.Page and p1.Page<>p3.Page and p2.Page<>p3.Page
RETURN [p1.Page,p2.Page,p3.Page], count(*) as count
ORDER BY count DESC
LIMIT 10;
个人资料输出：
10 rows
28660 ms
Compiler CYPHER 2.2-cost
Projection(0)
|
+Top
|
+EagerAggregation
|
+Projection(1)
|
+Filter(0)
|
+Expand(0)
|
+Filter(1)
|
+Expand(1)
|
+NodeByLabelScan
+------------------+---------------+---------+----------+------------------------------------------------+--------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| Operator | EstimatedRows | Rows | DbHits | Identifiers | Other |
+------------------+---------------+---------+----------+------------------------------------------------+--------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| Projection(0) | 1241 | 10 | 0 | FRESHID146, [p1.Page,p2.Page,p3.Page], count | [p1.Page,p2.Page,p3.Page], count |
| Top | 1241 | 10 | 0 | FRESHID146, count | { AUTOINT0}; count |
| EagerAggregation | 1241 | 212828 | 0 | FRESHID146, count | |
| Projection(1) | 1542393 | 1749120 | 10494720 | FRESHID146, p1, p2, p3 | |
| Filter(0) | 1542393 | 1749120 | 17872173 | p1, p2, p3 | (((hasLabel(p3:PageView(0)) AND NOT(Property(p1,Page(3)) == Property(p3,Page(3)))) AND NOT(anon[20] == anon[43])) AND NOT(Property(p2,Page(3)) == Property(p3,Page(3)))) |
| Expand(0) | 1904189 | 1985797 | 3971596 | p1, p2, p3 | (p2)-[:Next]->(p3) |
| Filter(1) | 1904191 | 1985799 | 10578840 | p1, p2 | (NOT(Property(p1,Page(3)) == Property(p2,Page(3))) AND hasLabel(p2:PageView(0))) |
| Expand(1) | 2115767 | 2115768 | 4231538 | p1, p2 | (p1)-[:Next]->(p2) |
| NodeByLabelScan | 2115770 | 2115770 | 2115771 | p1 | :PageView |
+------------------+---------------+---------+----------+------------------------------------------------+--------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
3。 （有循环！？我不知道为什么！我建议如果标识符不同，那么节点就不同了）
match (pv1:PageView)-[:Next]->(pv2:PageView)-[:Next]->(pv3:PageView),
(pv1)-[:At]->(p1),(pv2)-[:At]->(p2),(pv3)-[:At]->(p3)
RETURN [p1.Name,p2.Name,p3.Name], count(*) as count
ORDER BY count DESC
LIMIT 10;
个人资料输出：
10 rows
27678 ms
Compiler CYPHER 2.2-cost
Projection(0)
|
+Top
|
+EagerAggregation
|
+Projection(1)
|
+Filter(0)
|
+Expand(0)
|
+Filter(1)
|
+Expand(1)
|
+Filter(2)
|
+Expand(2)
|
+Filter(3)
|
+Expand(3)
|
+Expand(4)
|
+NodeByLabelScan
+------------------+---------------+---------+----------+------------------------------------------------+------------------------------------------------------------+
| Operator | EstimatedRows | Rows | DbHits | Identifiers | Other |
+------------------+---------------+---------+----------+------------------------------------------------+------------------------------------------------------------+
| Projection(0) | 1454 | 10 | 0 | FRESHID139, [p1.Name,p2.Name,p3.Name], count | [p1.Name,p2.Name,p3.Name], count |
| Top | 1454 | 10 | 0 | FRESHID139, count | { AUTOINT0}; count |
| EagerAggregation | 1454 | 223557 | 0 | FRESHID139, count | |
| Projection(1) | 2115760 | 2115764 | 12694584 | FRESHID139, p1, p2, p3, pv1, pv2, pv3 | |
| Filter(0) | 2115760 | 2115764 | 0 | p1, p2, p3, pv1, pv2, pv3 | (NOT(anon[116] == anon[80]) AND NOT(anon[80] == anon[98])) |
| Expand(0) | 2115760 | 2115764 | 4231530 | p1, p2, p3, pv1, pv2, pv3 | (pv1)-[:At]->(p1) |
| Filter(1) | 2115762 | 2115766 | 2115766 | p2, p3, pv1, pv2, pv3 | (hasLabel(pv1:PageView(0)) AND NOT(anon[21] == anon[45])) |
| Expand(1) | 2115762 | 2115766 | 4231532 | p2, p3, pv1, pv2, pv3 | (pv2)<-[:Next]-(pv1) |
| Filter(2) | 2115764 | 2115766 | 0 | p2, p3, pv2, pv3 | NOT(anon[116] == anon[98]) |
| Expand(2) | 2115764 | 2115766 | 4231534 | p2, p3, pv2, pv3 | (pv2)-[:At]->(p2) |
| Filter(3) | 2115766 | 2115768 | 2115768 | p3, pv2, pv3 | hasLabel(pv2:PageView(0)) |
| Expand(3) | 2115765 | 2115768 | 4231536 | p3, pv2, pv3 | (pv3)<-[:Next]-(pv2) |
| Expand(4) | 2115767 | 2115768 | 4231538 | p3, pv3 | (pv3)-[:At]->(p3) |
| NodeByLabelScan | 2115770 | 2115770 | 2115771 | pv3 | :PageView |
+------------------+---------------+---------+----------+------------------------------------------------+------------------------------------------------------------+
系统信息：
窗口8.1
250G ssd
neo4j enterprise 2.2.0-M02
缓存：hpc
ram：8G
jvm堆大小：4G
记忆映射：50％
149（：Page）节点
2115770（：PageView）节点
为什么即使这三种方法中最快的也是如此之慢？ （我想我的所有数据都在RAM中）
用循环过滤路径的最佳方法是什么？
答案 0 :(得分：0)
通过为所有标识符指定标签，可以强制Cypher打开节点标题并过滤其中的所有标签。
这是您的关系名称很重要的地方。建立关系是为了引导您进入图形，为了提高性能，不需要指定标签，因此如果您确定路径中的所有节点都有Pageview标签，则只需省略它，除了查询的开头：
match (p1:PageView)-[:Next]->(p2)-[:Next]->(p3)
where p1.Page<>p2.Page and p1.Page<>p3.Page and p2.Page<>p3.Page
RETURN [p1.Page,p2.Page,p3.Page], count(*) as count
ORDER BY count DESC
LIMIT 10;
我在与您的问题相关的答案中发布了一些查询计划结果：Neo4j: label vs. indexed property?