我不明白为什么在我的情况下没有使用定义的哈希索引：
FOR a in products_en
FOR i in a.infos
// This one works and uses the index:
LET test = "6GK1411-5AB10"
// This one does not use the index:
// LET test = i.name
LET catalogs = (FOR c IN MallCatalog
FILTER test IN c.accessoires.mlfbs
RETURN c.defaultURL
)
RETURN LENGTH(catalogs)
在i.name中，始终只有一个字符串。 我也试过像LET test = TO_STRING（i.name）和其他愚蠢的东西这样的技巧，但没办法。因此，即使每个集合中有10.000个元素，此查询的运行时也会变得恐怖： - （
以下是解释信息，第一个是固定字符串示例：
Execution plan:
Id NodeType Est. Comment
1 SingletonNode 1 * ROOT
2 EnumerateCollectionNode 5809 - FOR a IN products_en /* full collection scan */
3 CalculationNode 5809 - LET #7 = a.`infos` /* attribute expression */ /* collections used: a : products_en */
4 EnumerateListNode 580900 - FOR i IN #7 /* list iteration */
12 SubqueryNode 580900 - LET catalogs = ... /* const subquery */
6 SingletonNode 1 * ROOT
15 IndexNode 5 - FOR c IN MallCatalog /* hash index scan */
10 CalculationNode 5 - LET #11 = c.`defaultURL` /* attribute expression */ /* collections used: c : MallCatalog */
11 ReturnNode 5 - RETURN #11
13 CalculationNode 580900 - LET #13 = LENGTH(catalogs) /* simple expression */
14 ReturnNode 580900 - RETURN #13
Indexes used:
By Type Collection Unique Sparse Selectivity Fields Ranges
15 hash MallCatalog false true 17.29 % [ `accessoires.mlfbs[*]` ] ("6GK1411-5AB10" in c.`accessoires`.`mlfbs`)
Optimization rules applied:
Id RuleName
1 move-calculations-up
2 move-filters-up
3 remove-unnecessary-calculations
4 move-calculations-up-2
5 move-filters-up-2
6 use-indexes
7 remove-filter-covered-by-index
8 remove-unnecessary-calculations-2
这里有一个非固定替代方案：
Execution plan:
Id NodeType Est. Comment
1 SingletonNode 1 * ROOT
2 EnumerateCollectionNode 5809 - FOR a IN products_en /* full collection scan */
3 CalculationNode 5809 - LET #7 = a.`infos` /* attribute expression */ /* collections used: a : products_en */
4 EnumerateListNode 580900 - FOR i IN #7 /* list iteration */
5 CalculationNode 580900 - LET test = i.`name` /* attribute expression */
12 SubqueryNode 580900 - LET catalogs = ... /* subquery */
6 SingletonNode 1 * ROOT
7 EnumerateCollectionNode 10837 - FOR c IN MallCatalog /* full collection scan */
8 CalculationNode 10837 - LET #9 = (test in c.`accessoires`.`mlfbs`) /* simple expression */ /* collections used: c : MallCatalog */
9 FilterNode 10837 - FILTER #9
10 CalculationNode 10837 - LET #11 = c.`defaultURL` /* attribute expression */ /* collections used: c : MallCatalog */
11 ReturnNode 10837 - RETURN #11
13 CalculationNode 580900 - LET #13 = LENGTH(catalogs) /* simple expression */
14 ReturnNode 580900 - RETURN #13
Indexes used:
none
Optimization rules applied:
Id RuleName
1 move-calculations-up
2 move-filters-up
3 move-calculations-up-2
4 move-filters-up-2
有人可以帮帮我吗？ 提前致谢
答案 0 :(得分：2)
报告的行为不适用于devel和3.3。两个版本都优化了查询，如下所示：
arangod> gut()
Query string:
FOR i in infos
LET test = "6GK1411-5AB10"
LET catalogs = ( FOR c IN MallCatalog
FILTER test IN c.accessoires.mlfbs
RETURN c.defaultURL
)
RETURN LENGTH(catalogs)
Execution plan:
Id NodeType Est. Comment
1 SingletonNode 1 * ROOT
2 EnumerateCollectionNode 0 - FOR i IN infos /* full collection scan */
10 SubqueryNode 0 - LET catalogs = ... /* const subquery */
4 SingletonNode 1 * ROOT
13 IndexNode 1 - FOR c IN MallCatalog /* hash index scan */
8 CalculationNode 1 - LET #8 = c.`defaultURL` /* attribute expression */ /* collections used: c : MallCatalog */
9 ReturnNode 1 - RETURN #8
11 CalculationNode 0 - LET #10 = LENGTH(catalogs) /* simple expression */
12 ReturnNode 0 - RETURN #10
Indexes used:
By Type Collection Unique Sparse Selectivity Fields Ranges
13 hash MallCatalog false false 100.00 % [ `accessoires.mlfbs[*]` ] ("6GK1411-5AB10" in c.`accessoires`.`mlfbs`)
Optimization rules applied:
Id RuleName
1 move-calculations-up
2 move-filters-up
3 remove-unnecessary-calculations
4 move-calculations-up-2
5 move-filters-up-2
6 use-indexes
7 remove-filter-covered-by-index
8 remove-unnecessary-calculations-2
arangod> gugu()
Query string:
FOR i in infos
LET test = i.name
LET catalogs = ( FOR c IN MallCatalog
FILTER test IN c.accessoires.mlfbs
RETURN c.defaultURL
)
RETURN LENGTH(catalogs)
Execution plan:
Id NodeType Est. Comment
1 SingletonNode 1 * ROOT
2 EnumerateCollectionNode 0 - FOR i IN infos /* full collection scan */
10 SubqueryNode 0 - LET catalogs = ... /* subquery */
4 SingletonNode 1 * ROOT
13 IndexNode 1 - FOR c IN MallCatalog /* hash index scan */
8 CalculationNode 1 - LET #8 = c.`defaultURL` /* attribute expression */ /* collections used: c : MallCatalog */
9 ReturnNode 1 - RETURN #8
11 CalculationNode 0 - LET #10 = LENGTH(catalogs) /* simple expression */
3 CalculationNode 0 - LET test = i.`name` /* attribute expression */ /* collections used: i : infos */
12 ReturnNode 0 - RETURN #10
Indexes used:
By Type Collection Unique Sparse Selectivity Fields Ranges
13 hash MallCatalog false false 100.00 % [ `accessoires.mlfbs[*]` ] (i.`name` in c.`accessoires`.`mlfbs`)
Optimization rules applied:
Id RuleName
1 move-calculations-up
2 move-filters-up
3 move-calculations-up-2
4 move-filters-up-2
5 use-indexes
6 remove-filter-covered-by-index
7 remove-unnecessary-calculations-2
8 move-calculations-down
编辑：
此外，对于稀疏索引，预期会出现所描述的行为，因为测试的动态值无法证明不 null。这就是为什么稀疏索引只能与修复值一起使用。