1 <?xml version="1.0" encoding="UTF-8" standalone="no"?>
2 <!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd"><html xmlns="http://www.w3.org/1999/xhtml"><head><meta http-equiv="Content-Type" content="text/html; charset=UTF-8" /><title>65.6. Hash Indexes</title><link rel="stylesheet" type="text/css" href="stylesheet.css" /><link rev="made" href="pgsql-docs@lists.postgresql.org" /><meta name="generator" content="DocBook XSL Stylesheets Vsnapshot" /><link rel="prev" href="brin.html" title="65.5. BRIN Indexes" /><link rel="next" href="storage.html" title="Chapter 66. Database Physical Storage" /></head><body id="docContent" class="container-fluid col-10"><div class="navheader"><table width="100%" summary="Navigation header"><tr><th colspan="5" align="center">65.6. Hash Indexes</th></tr><tr><td width="10%" align="left"><a accesskey="p" href="brin.html" title="65.5. BRIN Indexes">Prev</a> </td><td width="10%" align="left"><a accesskey="u" href="indextypes.html" title="Chapter 65. Built-in Index Access Methods">Up</a></td><th width="60%" align="center">Chapter 65. Built-in Index Access Methods</th><td width="10%" align="right"><a accesskey="h" href="index.html" title="PostgreSQL 18.0 Documentation">Home</a></td><td width="10%" align="right"> <a accesskey="n" href="storage.html" title="Chapter 66. Database Physical Storage">Next</a></td></tr></table><hr /></div><div class="sect1" id="HASH-INDEX"><div class="titlepage"><div><div><h2 class="title" style="clear: both">65.6. Hash Indexes <a href="#HASH-INDEX" class="id_link">#</a></h2></div></div></div><div class="toc"><dl class="toc"><dt><span class="sect2"><a href="hash-index.html#HASH-INTRO">65.6.1. Overview</a></span></dt><dt><span class="sect2"><a href="hash-index.html#HASH-IMPLEMENTATION">65.6.2. Implementation</a></span></dt></dl></div><a id="id-1.10.17.7.2" class="indexterm"></a><div class="sect2" id="HASH-INTRO"><div class="titlepage"><div><div><h3 class="title">65.6.1. Overview <a href="#HASH-INTRO" class="id_link">#</a></h3></div></div></div><p>
3 <span class="productname">PostgreSQL</span>
4 includes an implementation of persistent on-disk hash indexes,
5 which are fully crash recoverable. Any data type can be indexed by a
6 hash index, including data types that do not have a well-defined linear
7 ordering. Hash indexes store only the hash value of the data being
8 indexed, thus there are no restrictions on the size of the data column
11 Hash indexes support only single-column indexes and do not allow
14 Hash indexes support only the <code class="literal">=</code> operator,
15 so WHERE clauses that specify range operations will not be able to take
16 advantage of hash indexes.
18 Each hash index tuple stores just the 4-byte hash value, not the actual
19 column value. As a result, hash indexes may be much smaller than B-trees
20 when indexing longer data items such as UUIDs, URLs, etc. The absence of
21 the column value also makes all hash index scans lossy. Hash indexes may
22 take part in bitmap index scans and backward scans.
24 Hash indexes are best optimized for SELECT and UPDATE-heavy workloads
25 that use equality scans on larger tables. In a B-tree index, searches must
26 descend through the tree until the leaf page is found. In tables with
27 millions of rows, this descent can increase access time to data. The
28 equivalent of a leaf page in a hash index is referred to as a bucket page. In
29 contrast, a hash index allows accessing the bucket pages directly,
30 thereby potentially reducing index access time in larger tables. This
31 reduction in "logical I/O" becomes even more pronounced on indexes/data
32 larger than shared_buffers/RAM.
34 Hash indexes have been designed to cope with uneven distributions of
35 hash values. Direct access to the bucket pages works well if the hash
36 values are evenly distributed. When inserts mean that the bucket page
37 becomes full, additional overflow pages are chained to that specific
38 bucket page, locally expanding the storage for index tuples that match
39 that hash value. When scanning a hash bucket during queries, we need to
40 scan through all of the overflow pages. Thus an unbalanced hash index
41 might actually be worse than a B-tree in terms of number of block
42 accesses required, for some data.
44 As a result of the overflow cases, we can say that hash indexes are
45 most suitable for unique, nearly unique data or data with a low number
46 of rows per hash bucket.
47 One possible way to avoid problems is to exclude highly non-unique
48 values from the index using a partial index condition, but this may
49 not be suitable in many cases.
51 Like B-Trees, hash indexes perform simple index tuple deletion. This
52 is a deferred maintenance operation that deletes index tuples that are
53 known to be safe to delete (those whose item identifier's LP_DEAD bit
54 is already set). If an insert finds no space is available on a page we
55 try to avoid creating a new overflow page by attempting to remove dead
56 index tuples. Removal cannot occur if the page is pinned at that time.
57 Deletion of dead index pointers also occurs during VACUUM.
59 If it can, VACUUM will also try to squeeze the index tuples onto as
60 few overflow pages as possible, minimizing the overflow chain. If an
61 overflow page becomes empty, overflow pages can be recycled for reuse
62 in other buckets, though we never return them to the operating system.
63 There is currently no provision to shrink a hash index, other than by
64 rebuilding it with REINDEX.
65 There is no provision for reducing the number of buckets, either.
67 Hash indexes may expand the number of bucket pages as the number of
68 rows indexed grows. The hash key-to-bucket-number mapping is chosen so that
69 the index can be incrementally expanded. When a new bucket is to be added to
70 the index, exactly one existing bucket will need to be "split", with some of
71 its tuples being transferred to the new bucket according to the updated
72 key-to-bucket-number mapping.
74 The expansion occurs in the foreground, which could increase execution
75 time for user inserts. Thus, hash indexes may not be suitable for tables
76 with rapidly increasing number of rows.
77 </p></div><div class="sect2" id="HASH-IMPLEMENTATION"><div class="titlepage"><div><div><h3 class="title">65.6.2. Implementation <a href="#HASH-IMPLEMENTATION" class="id_link">#</a></h3></div></div></div><p>
78 There are four kinds of pages in a hash index: the meta page (page zero),
79 which contains statically allocated control information; primary bucket
80 pages; overflow pages; and bitmap pages, which keep track of overflow
81 pages that have been freed and are available for re-use. For addressing
82 purposes, bitmap pages are regarded as a subset of the overflow pages.
84 Both scanning the index and inserting tuples require locating the bucket
85 where a given tuple ought to be located. To do this, we need the bucket
86 count, highmask, and lowmask from the metapage; however, it's undesirable
87 for performance reasons to have to have to lock and pin the metapage for
88 every such operation. Instead, we retain a cached copy of the metapage
89 in each backend's relcache entry. This will produce the correct bucket
90 mapping as long as the target bucket hasn't been split since the last
93 Primary bucket pages and overflow pages are allocated independently since
94 any given index might need more or fewer overflow pages relative to its
95 number of buckets. The hash code uses an interesting set of addressing
96 rules to support a variable number of overflow pages while not having to
97 move primary bucket pages around after they are created.
99 Each row in the table indexed is represented by a single index tuple in
100 the hash index. Hash index tuples are stored in bucket pages, and if
101 they exist, overflow pages. We speed up searches by keeping the index entries
102 in any one index page sorted by hash code, thus allowing binary search to be
103 used within an index page. Note however that there is *no* assumption about
104 the relative ordering of hash codes across different index pages of a bucket.
106 The bucket splitting algorithms to expand the hash index are too complex to
107 be worthy of mention here, though are described in more detail in
108 <code class="filename">src/backend/access/hash/README</code>.
109 The split algorithm is crash safe and can be restarted if not completed
111 </p></div></div><div class="navfooter"><hr /><table width="100%" summary="Navigation footer"><tr><td width="40%" align="left"><a accesskey="p" href="brin.html" title="65.5. BRIN Indexes">Prev</a> </td><td width="20%" align="center"><a accesskey="u" href="indextypes.html" title="Chapter 65. Built-in Index Access Methods">Up</a></td><td width="40%" align="right"> <a accesskey="n" href="storage.html" title="Chapter 66. Database Physical Storage">Next</a></td></tr><tr><td width="40%" align="left" valign="top">65.5. BRIN Indexes </td><td width="20%" align="center"><a accesskey="h" href="index.html" title="PostgreSQL 18.0 Documentation">Home</a></td><td width="40%" align="right" valign="top"> Chapter 66. Database Physical Storage</td></tr></table></div></body></html>