F.5. bloom
Bloom filter 是一種節省空間的資料結構，用於測試元素是否為集合的成員。在使用索引方法時，它透過 signatures（其大小是在建立索引時確定的）快速排除不相符的內容。
Signature 是索引屬性的失真表示，因此很容易誤報。也就是說，可能會回報某個元素在集合中（實際上沒有）。因此，必須使用資料中的實際屬性值來重新檢查索引並搜尋結果。較大的 signature 會減少了誤報的機率，也減少了無效的存取次數，但當然也會使索引變大，造成掃描速度變慢。
當資料表具有許多屬性並且查詢測試它們的任意組合時，這種類型的索引最有用。傳統的 btree 索引會比 Bloom 索引快，但是它可能需要許多 btree 索引來支援所有可能的查詢，而其中一個查詢只需要一個 Bloom 索引。但是請注意，bloom 索引僅支援相等查詢，而 btree 索引也可以用於不相等和範圍查詢。
F.5.1. 參數
Bloom 索引的 WITH 子句接受以下參數：
length每個 signature（索引項目）的長度（以位元為單位）。四捨五入到最接近的 16 的倍數。預設值為 80 位元，最大值為 4096。
col1 — col32每個索引欄位產成的位元數。每個參數的名稱指的是它控制的索引欄位的編號。預設值為 2 位元，最大值為 4095。實際未使用的索引欄位的參數將被忽略。
F.5.2. Examples
This is an example of creating a bloom index:
1
CREATE INDEX bloomidx ON tbloom USING bloom (i1,i2,i3)
2
WITH (length=80, col1=2, col2=2, col3=4);
Copied!
The index is created with a signature length of 80 bits, with attributes i1 and i2 mapped to 2 bits, and attribute i3 mapped to 4 bits. We could have omitted the
length, col1, and col2 specifications since those have the default values.Here is a more complete example of bloom index definition and usage, as well as a comparison with equivalent btree indexes. The bloom index is considerably smaller than the btree index, and can perform better.
1
=# CREATE TABLE tbloom AS
2
SELECT
3
(random() * 1000000)::int as i1,
4
(random() * 1000000)::int as i2,
5
(random() * 1000000)::int as i3,
6
(random() * 1000000)::int as i4,
7
(random() * 1000000)::int as i5,
8
(random() * 1000000)::int as i6
9
FROM
10
generate_series(1,10000000);
11
SELECT 10000000
12
=# CREATE INDEX bloomidx ON tbloom USING bloom (i1, i2, i3, i4, i5, i6);
13
CREATE INDEX
14
=# SELECT pg_size_pretty(pg_relation_size('bloomidx'));
15
pg_size_pretty
16
----------------
17
153 MB
18
(1 row)
19
=# CREATE index btreeidx ON tbloom (i1, i2, i3, i4, i5, i6);
20
CREATE INDEX
21
=# SELECT pg_size_pretty(pg_relation_size('btreeidx'));
22
pg_size_pretty
23
----------------
24
387 MB
25
(1 row)
Copied!
A sequential scan over this large table takes a long time:
1
=# EXPLAIN ANALYZE SELECT * FROM tbloom WHERE i2 = 898732 AND i5 = 123451;
2
QUERY PLAN
3
------------------------------------------------------------------------------------------------------------
4
Seq Scan on tbloom (cost=0.00..213694.08 rows=1 width=24) (actual time=1445.438..1445.438 rows=0 loops=1)
5
Filter: ((i2 = 898732) AND (i5 = 123451))
6
Rows Removed by Filter: 10000000
7
Planning time: 0.177 ms
8
Execution time: 1445.473 ms
9
(5 rows)
Copied!
So the planner will usually select an index scan if possible. With a btree index, we get results like this:
1
=# EXPLAIN ANALYZE SELECT * FROM tbloom WHERE i2 = 898732 AND i5 = 123451;
2
QUERY PLAN
3
--------------------------------------------------------------------------------------------------------------------------------
4
Index Only Scan using btreeidx on tbloom (cost=0.56..298311.96 rows=1 width=24) (actual time=445.709..445.709 rows=0 loops=1)
5
Index Cond: ((i2 = 898732) AND (i5 = 123451))
6
Heap Fetches: 0
7
Planning time: 0.193 ms
8
Execution time: 445.770 ms
9
(5 rows)
Copied!
Bloom is better than btree in handling this type of search:
1
=# EXPLAIN ANALYZE SELECT * FROM tbloom WHERE i2 = 898732 AND i5 = 123451;
2
QUERY PLAN
3
---------------------------------------------------------------------------------------------------------------------------
4
Bitmap Heap Scan on tbloom (cost=178435.39..178439.41 rows=1 width=24) (actual time=76.698..76.698 rows=0 loops=1)
5
Recheck Cond: ((i2 = 898732) AND (i5 = 123451))
6
Rows Removed by Index Recheck: 2439
7
Heap Blocks: exact=2408
8
-> Bitmap Index Scan on bloomidx (cost=0.00..178435.39 rows=1 width=0) (actual time=72.455..72.455 rows=2439 loops=1)
9
Index Cond: ((i2 = 898732) AND (i5 = 123451))
10
Planning time: 0.475 ms
11
Execution time: 76.778 ms
12
(8 rows)
Copied!
請注意，這裡的誤報的數量相對較多：被選擇 2439 筆資料要進行確認，但實際上沒有與查詢相符的資料。我們可以透過指定更大的 signature 長度來減少這種情況。在此範例中，建立長度為 200 的索引可將誤報數量減少到 55；但它也使索引大小增加了一倍（至 306 MB），並最終使該查詢的速度變慢（總計 125 毫秒）。
現在，btree 搜搜的主要問題在於，當搜搜條件不限制前導索引欄時，btree 的效率低下。btree 的更好策略是在每欄位上建立一個單獨的索引。然後計劃程序將會選擇規劃以下內容：
1
=# EXPLAIN ANALYZE SELECT * FROM tbloom WHERE i2 = 898732 AND i5 = 123451;
2
QUERY PLAN
3
------------------------------------------------------------------------------------------------------------------------------
4
Bitmap Heap Scan on tbloom (cost=9.29..13.30 rows=1 width=24) (actual time=0.148..0.148 rows=0 loops=1)
5
Recheck Cond: ((i5 = 123451) AND (i2 = 898732))
6
-> BitmapAnd (cost=9.29..9.29 rows=1 width=0) (actual time=0.145..0.145 rows=0 loops=1)
7
-> Bitmap Index Scan on tbloom_i5_idx (cost=0.00..4.52 rows=11 width=0) (actual time=0.089..0.089 rows=10 loops=1)
8
Index Cond: (i5 = 123451)
9
-> Bitmap Index Scan on tbloom_i2_idx (cost=0.00..4.52 rows=11 width=0) (actual time=0.048..0.048 rows=8 loops=1)
10
Index Cond: (i2 = 898732)
11
Planning time: 2.049 ms
12
Execution time: 0.280 ms
13
(9 rows)
Copied!
儘管此查詢的執行速度比使用單個索引的查詢快得多，但我們在索引大小上付出了很大的代價。每個單欄位 btree 索引佔用 214 MB，因此所需的總空間超過 1.2GB，是 Bloom 索引使用的空間 8 倍以上。
F.5.3. Operator Class Interface
An operator class for bloom indexes requires only a hash function for the indexed data type and an equality operator for searching. This example shows the operator class definition for the
text data type:1
CREATE OPERATOR CLASS text_ops
2
DEFAULT FOR TYPE text USING bloom AS
3
OPERATOR 1 =(text, text),
4
FUNCTION 1 hashtext(text);
Copied!
F.5.4. Limitations
- Only operator classes for
int4andtextare included with the module. - Only the
=operator is supported for search. But it is possible to add support for arrays with union and intersection operations in the future. bloomaccess method doesn't supportUNIQUEindexes.bloomaccess method doesn't support searching forNULLvalues.
F.5.5. Authors
Last modified 7mo ago
Export as PDF
Copy link
Edit on GitHub