Debian Codegraph @

src/docsrc/html/architecture.html

raw
1 <!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Strict//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-strict.dtd">
2 <!--
3 Copyright 2004-2019 H2 Group. Multiple-Licensed under the MPL 2.0,
4 and the EPL 1.0 (http://h2database.com/html/license.html).
5 Initial Developer: H2 Group
6 -->
7 <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en">
8 <head>
9 <meta http-equiv="Content-Type" content="text/html;charset=utf-8" />
10 <meta name="viewport" content="width=device-width, initial-scale=1" />
11 <title>
12 Architecture
13 </title>
14 <link rel="stylesheet" type="text/css" href="stylesheet.css" />
15 <!-- [search] { -->
16 <script type="text/javascript" src="navigation.js"></script>
17 </head><body onload="frameMe();">
18 <table class="content"><tr class="content"><td class="content"><div class="contentDiv">
19 <!-- } -->
20
21 <h1>Architecture</h1>
22 <a href="#introduction">
23 Introduction</a><br />
24 <a href="#top_down">
25 Top-down overview</a><br />
26 <a href="#jdbc">
27 JDBC driver</a><br />
28 <a href="#connection">
29 Connection/session management</a><br />
30 <a href="#command">
31 Command execution and planning</a><br />
32 <a href="#table">
33 Table/index/constraints</a><br />
34 <a href="#transaction">
35 Undo log, redo log, and transactions layer</a><br />
36 <a href="#btree">
37 B-tree engine and page-based storage allocation</a><br />
38 <a href="#filesystem">
39 Filesystem abstraction</a><br />
40
41 <h2 id="introduction">Introduction</h2>
42 <p>
43 H2 implements an embedded and standalone ANSI-SQL89 compliant SQL engine on top of a B-tree based disk store.
44 </p>
45 <p>
46 As of October 2013, Thomas is still working on our next-generation storage engine called MVStore. This will
47 in time replace the B-tree based storage engine.
48 </p>
49
50 <h2 id="top_down">Top-down Overview</h2>
51 <p>
52 Working from the top down, the layers look like this:
53 <ul><li>JDBC driver.
54 </li><li>Connection/session management.
55 </li><li>SQL Parser.
56 </li><li>Command execution and planning.
57 </li><li>Table/Index/Constraints.
58 </li><li>Undo log, redo log, and transactions layer.
59 </li><li>B-tree engine and page-based storage allocation.
60 </li><li>Filesystem abstraction.
61 </li></ul>
62 </p>
63
64 <h2 id="jdbc">JDBC Driver</h2>
65 <p>
66 The JDBC driver implementation lives in <code>org.h2.jdbc, org.h2.jdbcx</code>
67 </p>
68
69 <h2 id="connection">Connection/session management</h2>
70 <p>
71 The primary classes of interest are:
72 <table class="main">
73 <tr><th>Package</th><th>Description</th></tr>
74 <tr><td>org.h2.engine.Database</td><td>the root/global class</td></tr>
75 <tr><td>org.h2.engine.SessionInterface</td>
76 <td>abstracts over the differences between embedded and remote sessions</td></tr>
77 <tr><td>org.h2.engine.Session</td>
78 <td>local/embedded session</td></tr>
79 <tr><td>org.h2.engine.SessionRemote</td>
80 <td>remote session</td></tr>
81 </table>
82 </p>
83
84 <h2 id="jdbc">Parser</h2>
85 <p>
86 The parser lives in <code>org.h2.command.Parser</code>. It uses a straightforward recursive-descent design.
87 </p>
88 <p>
89 See Wikipedia <a href="https://en.wikipedia.org/wiki/Recursive_descent_parser">Recursive descent parser</a> page.
90 </p>
91
92
93 <h2 id="command">Command execution and planning</h2>
94 <p>
95 Unlike other databases, we do not have an intermediate step where we generate some kind of IR (intermediate representation) of the query.
96 The parser class directly generates a command execution object.
97 Then we run some optimisation steps over the command to possibly generate a more efficient command.
98
99 The primary packages of interest are:
100 <table class="main">
101 <tr><th>Package</th><th>Description</th></tr>
102 <tr><td>org.h2.command.ddl</td><td>Commands that modify schema data structures</td></tr>
103 <tr><td>org.h2.command.dml</td><td>Commands that modify data</td></tr>
104 </table>
105 </p>
106
107 <h2 id="table">Table/Index/Constraints</h2>
108 <p>
109 One thing to note here is that indexes are simply stored as special kinds of tables.
110 </p>
111 <p>
112 The primary packages of interest are:
113 <table class="main">
114 <tr><th>Package</th><th>Description</th></tr>
115 <tr><td>org.h2.table</td><td>Implementations of different kinds of tables</td></tr>
116 <tr><td>org.h2.index</td><td>Implementations of different kinds of indices</td></tr>
117 </table>
118 </p>
119
120 <h2 id="transaction">Undo log, redo log, and transactions layer</h2>
121 <p>
122 We have a transaction log, which is shared among all sessions. See also
123 <a href="https://en.wikipedia.org/wiki/Transaction_log">https://en.wikipedia.org/wiki/Transaction_log</a>
124 <a href="http://h2database.com/html/grammar.html#set_log">http://h2database.com/html/grammar.html#set_log</a>
125 </p>
126 <p>
127 We also have an undo log, which is per session, to undo an operation (an update that fails for example)
128 and to rollback a transaction.
129 Theoretically, the transaction log could be used, but for simplicity, H2 currently uses it's
130 own "list of operations" (usually in-memory).
131 </p>
132 <p>
133 With the MVStore, this is no longer needed (just the transaction log).
134 </p>
135
136 <h2 id="btree">B-tree engine and page-based storage allocation.</h2>
137 <p>
138 The primary package of interest is <code>org.h2.store</code>.
139 </p>
140 <p>
141 This implements a storage mechanism which allocates pages of storage (typically 2k in size)
142 and also implements a b-tree over those pages to allow fast retrieval and update.
143 </p>
144
145 <h2 id="filesystem">Filesystem abstraction.</h2>
146 <p>
147 The primary class of interest is <code>org.h2.store.FileStore</code>.
148 </p>
149 <p>
150 This implements an abstraction of a random-access file.
151 This allows the higher layers to treat in-memory vs. on-disk vs. zip-file databases the same.
152 </p>
153
154 <!-- [close] { --></div></td></tr></table><!-- } --><!-- analytics --></body></html>