postgresql/doc/src/sgml/bloom.sgml

<!-- doc/src/sgml/bloom.sgml -->

<sect1 id="bloom" xreflabel="bloom">
 <title>bloom &mdash; bloom filter index access method</title>

 <indexterm zone="bloom">
  <primary>bloom</primary>
 </indexterm>

 <para>
  <literal>bloom</literal> provides an index access method based on
  <ulink url="https://en.wikipedia.org/wiki/Bloom_filter">Bloom filters</ulink>.
 </para>

 <para>
  A Bloom filter is a space-efficient data structure that is used to test
  whether an element is a member of a set.  In the case of an index access
  method, it allows fast exclusion of non-matching tuples via signatures
  whose size is determined at index creation.
 </para>

 <para>
  A signature is a lossy representation of the indexed attribute(s), and as
  such is prone to reporting false positives; that is, it may be reported
  that an element is in the set, when it is not.  So index search results
  must always be rechecked using the actual attribute values from the heap
  entry.  Larger signatures reduce the odds of a false positive and thus
  reduce the number of useless heap visits, but of course also make the index
  larger and hence slower to scan.
 </para>

 <para>
  This type of index is most useful when a table has many attributes and
  queries test arbitrary combinations of them.  A traditional btree index is
  faster than a bloom index, but it can require many btree indexes to support
  all possible queries where one needs only a single bloom index.  Note
  however that bloom indexes only support equality queries, whereas btree
  indexes can also perform inequality and range searches.
 </para>

 <sect2 id="bloom-parameters">
  <title>Parameters</title>

  <para>
   A <literal>bloom</literal> index accepts the following parameters in its
   <literal>WITH</literal> clause:
  </para>

   <variablelist>
   <varlistentry>
    <term><literal>length</literal></term>
    <listitem>
     <para>
      Length of each signature (index entry) in bits. It is rounded up to the
      nearest multiple of <literal>16</literal>. The default is
      <literal>80</literal> bits and the maximum is <literal>4096</literal>.
     </para>
    </listitem>
   </varlistentry>
   </variablelist>
   <variablelist>
   <varlistentry>
    <term><literal>col1 &mdash; col32</literal></term>
    <listitem>
     <para>
      Number of bits generated for each index column. Each parameter's name
      refers to the number of the index column that it controls.  The default
      is <literal>2</literal> bits and the maximum is <literal>4095</literal>.
      Parameters for index columns not actually used are ignored.
     </para>
    </listitem>
   </varlistentry>
   </variablelist>
 </sect2>

 <sect2 id="bloom-examples">
  <title>Examples</title>

  <para>
   This is an example of creating a bloom index:
  </para>

<programlisting>
CREATE INDEX bloomidx ON tbloom USING bloom (i1,i2,i3)
       WITH (length=80, col1=2, col2=2, col3=4);
</programlisting>

  <para>
   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 <literal>length</literal>, <literal>col1</literal>,
   and <literal>col2</literal> specifications since those have the default values.
  </para>

  <para>
   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.
  </para>

<programlisting>
=# CREATE TABLE tbloom AS
   SELECT
     (random() * 1000000)::int as i1,
     (random() * 1000000)::int as i2,
     (random() * 1000000)::int as i3,
     (random() * 1000000)::int as i4,
     (random() * 1000000)::int as i5,
     (random() * 1000000)::int as i6
   FROM
  generate_series(1,10000000);
SELECT 10000000
</programlisting>

  <para>
   A sequential scan over this large table takes a long time:
<programlisting>
=# EXPLAIN ANALYZE SELECT * FROM tbloom WHERE i2 = 898732 AND i5 = 123451;
                                              QUERY PLAN
-------------------------------------------------------------------&zwsp;-----------------------------------
 Seq Scan on tbloom  (cost=0.00..213744.00 rows=250 width=24) (actual time=357.059..357.059 rows=0.00 loops=1)
   Filter: ((i2 = 898732) AND (i5 = 123451))
   Rows Removed by Filter: 10000000
   Buffers: shared hit=63744
 Planning Time: 0.346 ms
 Execution Time: 357.076 ms
(6 rows)
</programlisting>
  </para>

  <para>
   Even with the btree index defined the result will still be a
   sequential scan:
<programlisting>
=# CREATE INDEX btreeidx ON tbloom (i1, i2, i3, i4, i5, i6);
CREATE INDEX
=# SELECT pg_size_pretty(pg_relation_size('btreeidx'));
 pg_size_pretty
----------------
 386 MB
(1 row)
=# EXPLAIN ANALYZE SELECT * FROM tbloom WHERE i2 = 898732 AND i5 = 123451;
                                              QUERY PLAN
-------------------------------------------------------------------&zwsp;-----------------------------------
 Seq Scan on tbloom  (cost=0.00..213744.00 rows=2 width=24) (actual time=351.016..351.017 rows=0.00 loops=1)
   Filter: ((i2 = 898732) AND (i5 = 123451))
   Rows Removed by Filter: 10000000
   Buffers: shared hit=63744
 Planning Time: 0.138 ms
 Execution Time: 351.035 ms
(6 rows)
</programlisting>
  </para>

  <para>
   Having the bloom index defined on the table is better than btree in
   handling this type of search:
<programlisting>
=# CREATE INDEX bloomidx ON tbloom USING bloom (i1, i2, i3, i4, i5, i6);
CREATE INDEX
=# SELECT pg_size_pretty(pg_relation_size('bloomidx'));
 pg_size_pretty
----------------
 153 MB
(1 row)
=# EXPLAIN ANALYZE SELECT * FROM tbloom WHERE i2 = 898732 AND i5 = 123451;
                                                     QUERY PLAN
-------------------------------------------------------------------&zwsp;--------------------------------------------------
 Bitmap Heap Scan on tbloom  (cost=1792.00..1799.69 rows=2 width=24) (actual time=22.605..22.606 rows=0.00 loops=1)
   Recheck Cond: ((i2 = 898732) AND (i5 = 123451))
   Rows Removed by Index Recheck: 2300
   Heap Blocks: exact=2256
   Buffers: shared hit=21864
   -&gt;  Bitmap Index Scan on bloomidx  (cost=0.00..178436.00 rows=1 width=0) (actual time=20.005..20.005 rows=2300.00 loops=1)
         Index Cond: ((i2 = 898732) AND (i5 = 123451))
         Index Searches: 1
         Buffers: shared hit=19608
 Planning Time: 0.099 ms
 Execution Time: 22.632 ms
(11 rows)
</programlisting>
  </para>

  <para>
   Now, the main problem with the btree search is that btree is inefficient
   when the search conditions do not constrain the leading index column(s).
   A better strategy for btree is to create a separate index on each column.
   Then the planner will choose something like this:
<programlisting>
=# CREATE INDEX btreeidx1 ON tbloom (i1);
CREATE INDEX
=# CREATE INDEX btreeidx2 ON tbloom (i2);
CREATE INDEX
=# CREATE INDEX btreeidx3 ON tbloom (i3);
CREATE INDEX
=# CREATE INDEX btreeidx4 ON tbloom (i4);
CREATE INDEX
=# CREATE INDEX btreeidx5 ON tbloom (i5);
CREATE INDEX
=# CREATE INDEX btreeidx6 ON tbloom (i6);
CREATE INDEX
=# EXPLAIN ANALYZE SELECT * FROM tbloom WHERE i2 = 898732 AND i5 = 123451;
                                                        QUERY PLAN
-------------------------------------------------------------------&zwsp;--------------------------------------------------------
 Bitmap Heap Scan on tbloom  (cost=9.29..13.30 rows=1 width=24) (actual time=0.032..0.033 rows=0.00 loops=1)
   Recheck Cond: ((i5 = 123451) AND (i2 = 898732))
   Buffers: shared read=6
   -&gt;  BitmapAnd  (cost=9.29..9.29 rows=1 width=0) (actual time=0.047..0.047 rows=0.00 loops=1)
         Buffers: shared hit=6
         -&gt;  Bitmap Index Scan on btreeidx5  (cost=0.00..4.52 rows=11 width=0) (actual time=0.026..0.026 rows=7.00 loops=1)
               Index Cond: (i5 = 123451)
               Index Searches: 1
               Buffers: shared hit=3
         -&gt;  Bitmap Index Scan on btreeidx2  (cost=0.00..4.52 rows=11 width=0) (actual time=0.007..0.007 rows=8.00 loops=1)
               Index Cond: (i2 = 898732)
               Index Searches: 1
               Buffers: shared hit=3
 Planning Time: 0.264 ms
 Execution Time: 0.047 ms
(15 rows)
</programlisting>
   Although this query runs much faster than with either of the single
   indexes, we pay a penalty in index size.  Each of the single-column
   btree indexes occupies 88.5 MB, so the total space needed is 531 MB,
   over three times the space used by the bloom index.
  </para>
 </sect2>

 <sect2 id="bloom-operator-class-interface">
  <title>Operator Class Interface</title>

  <para>
   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 <type>text</type> data type:
  </para>

<programlisting>
CREATE OPERATOR CLASS text_ops
DEFAULT FOR TYPE text USING bloom AS
    OPERATOR    1   =(text, text),
    FUNCTION    1   hashtext(text);
</programlisting>
 </sect2>

 <sect2 id="bloom-limitations">
  <title>Limitations</title>
  <para>
   <itemizedlist>
    <listitem>
     <para>
      Only operator classes for <type>int4</type> and <type>text</type> are
      included with the module.
     </para>
    </listitem>

    <listitem>
     <para>
      Only the <literal>=</literal> operator is supported for search.  But
      it is possible to add support for arrays with union and intersection
      operations in the future.
     </para>
    </listitem>

    <listitem>
     <para>
       <literal>bloom</literal> access method doesn't support
       <literal>UNIQUE</literal> indexes.
     </para>
    </listitem>

    <listitem>
     <para>
       <literal>bloom</literal> access method doesn't support searching for
       <literal>NULL</literal> values.
     </para>
    </listitem>
   </itemizedlist>
  </para>
 </sect2>

 <sect2 id="bloom-authors">
  <title>Authors</title>

  <para>
   Teodor Sigaev <email>teodor@postgrespro.ru</email>,
   Postgres Professional, Moscow, Russia
  </para>

  <para>
   Alexander Korotkov <email>a.korotkov@postgrespro.ru</email>,
   Postgres Professional, Moscow, Russia
  </para>

  <para>
   Oleg Bartunov <email>obartunov@postgrespro.ru</email>,
   Postgres Professional, Moscow, Russia
  </para>
 </sect2>

</sect1>