1 |
=head1 pgestraier - search Hyper Estraier indexes from PostgreSQL |
=head1 pgestraier - PostgreSQL full-text search using Hyper Estraier |
2 |
|
|
3 |
This package is essentially PostgreSQL C function which calls Hyper Estraier |
This package is essentially composed of two different parts: |
4 |
API and returns results in user defined format. |
|
5 |
|
=over 4 |
6 |
|
|
7 |
|
=item search function |
8 |
|
|
9 |
|
PostgreSQL function to search Hyper Estraier full-text index, using |
10 |
|
full-text queries and attribute filtering to return user-specified |
11 |
|
table of results. |
12 |
|
|
13 |
|
This function can mimic SQL C<LIMIT>, C<OFFSET> and C<ORDER BY> |
14 |
|
functionality much faster than using those SQL constructs on search |
15 |
|
results. |
16 |
|
|
17 |
|
=item trigger function |
18 |
|
|
19 |
|
PostgreSQL trigger function to keep Hyper Estraier in sync with PostgreSQL. |
20 |
|
It triggers after insert, update or delete and update full-text index |
21 |
|
accordingly. |
22 |
|
|
23 |
|
=back |
24 |
|
|
25 |
|
Both functions are written in C, while test framework and supporting |
26 |
|
utilities are written in perl. |
27 |
|
|
28 |
|
You can use just one of those functions. If you want just to search existing |
29 |
|
Hyper Estraier index or generate it off-line (after nightly batch jobs, for |
30 |
|
example), just use search function. |
31 |
|
|
32 |
|
On the other hand, if you want just to keep your Hyper Estraier index in |
33 |
|
sync with PostgreSQL data, you can use just trigger function to achieve that. |
34 |
|
|
35 |
=head1 Why is it written? |
=head1 Why is it written? |
36 |
|
|
37 |
Aside from providing single API to query your RDBMS and full text index |
Aside from providing single query language (SQL) to RDBMS and full text index |
38 |
(using any language that has PostgreSQL client libraries), real power is |
(using any language that has PostgreSQL client libraries), real power is |
39 |
hidden in ability to join results from full text index and structured data |
hidden in ability to join results from full text index and structured data |
40 |
in RDBMS. |
in RDBMS. |
41 |
|
|
42 |
|
For simple real-life example which address problem |
43 |
|
C<< WHERE name LIKE '%foo%' OR surname LIKE '%foo%' >> |
44 |
|
is slow see L<Tutorial> and L<pgest-index> documentation. |
45 |
|
|
46 |
=head1 How to install |
=head1 How to install |
47 |
|
|
48 |
Installation should be simple. However, you will have to have following |
Installation should be simple. However, you will have to have following |
49 |
software already installed before you try this function: |
software already installed before you try this functions: |
50 |
|
|
51 |
=over |
=over |
52 |
|
|
53 |
=item * |
=item * |
54 |
|
|
55 |
PostgreSQL (tested with versions 7.4 and 8.0) with development libraries |
PostgreSQL (tested with versions 7.4, 8.0 and 8.1) with development libraries |
56 |
|
|
57 |
|
=item * |
58 |
|
|
59 |
|
Hyper Estraier (tested with various versions, recommended 1.2.4 or newer) |
60 |
|
with development headers |
61 |
|
|
62 |
=item * |
=item * |
63 |
|
|
64 |
Hyper Estraier (tested with 0.5.0-1.0.0+, version newer than 0.9.6 are |
working C compiler (tested with gcc) |
|
recommended) |
|
65 |
|
|
66 |
=back |
=back |
67 |
|
|
68 |
To run tests you will also need: |
If you want to use helper script to create consistency triggers to keep |
69 |
|
Hyper Estraier in sync with PostgreSQL database, you will also need: |
70 |
|
|
71 |
=over |
=over |
72 |
|
|
76 |
|
|
77 |
=item * |
=item * |
78 |
|
|
79 |
perl modules C<DBI>, C<DBD::Pg>, C<Test::More> and optionally C<HyperEstraier> |
perl modules C<Search::Estraier>, C<DBI> and C<DBD::Pg> |
80 |
|
|
81 |
|
=back |
82 |
|
|
83 |
|
To run tests you will also need: |
84 |
|
|
85 |
|
=over |
86 |
|
|
87 |
|
=item * |
88 |
|
|
89 |
|
perl module C<Test::More> |
90 |
|
|
91 |
=item * |
=item * |
92 |
|
|
98 |
|
|
99 |
=item * |
=item * |
100 |
|
|
101 |
Hyper Estraier node C<trivia> with permissions for C<admin> user. |
Hyper Estraier C<estmaster> running with permissions for C<admin> user |
102 |
|
to create C<trivia> node. |
103 |
|
|
104 |
=back |
=back |
105 |
|
|
111 |
install C<pgest.so> shared library in system-wide location so that PostgreSQL |
install C<pgest.so> shared library in system-wide location so that PostgreSQL |
112 |
could access it. |
could access it. |
113 |
|
|
114 |
Next, you will have to create test index. You have two options: |
=head2 Create sample index using Hyper Estraier perl bindings |
115 |
|
|
116 |
=head2 Create index using estcmd |
Perl bindings for Hyper Estraier are available at CPAN: |
117 |
|
|
118 |
This will create temporary files on disk and index them using estcmd gather |
L<http://search.cpan.org/~dpavlin/Search-Estraier/> |
119 |
|
|
|
cd data |
|
|
make index |
|
|
cd .. |
|
120 |
|
|
121 |
B<Warning:> this method is incomplete and won't create node index needed |
After installing C<Search::Estraier> you can create index using following commands: |
|
to run last examples in C<test.sql> correctly. Solution is simple: either |
|
|
symlink your newly created index to Hyper Estraier C<_node> directory or |
|
|
create node and fill re-create index using C<estcall>. |
|
|
|
|
|
=head2 Create index using Hyper Estraier perl bindings |
|
|
|
|
|
Perl bindings for Hyper Estraier are available at |
|
|
|
|
|
L<http://hyperestraier.sourceforge.net/binding/> |
|
|
|
|
|
However, they don't support node API (yet), so you will have to use |
|
|
my modified version which is available at |
|
|
L<http://svn.rot13.org/> in C<hyperestraier_wrappers> repository. |
|
|
|
|
|
If you installed bindings as documented in README file, you can use |
|
|
perl binding to create index about three times faster than using C<estcmd> |
|
|
(to be fair, I must say that creation of intermediate files take most time, |
|
|
not indexing). |
|
|
|
|
|
However, you will first need to create node I<trivia> using Hyper Estraier's |
|
|
administration interface at L<http://localhost:1978/masterui>. You will also |
|
|
need user C<admin> with password C<admin> because those values are |
|
|
hard-coded in C<indexer.pl>. If you want to use different user on index |
|
|
name, feel free to change script. |
|
122 |
|
|
123 |
cd data |
cd data |
124 |
make perl |
make index |
125 |
cd .. |
cd .. |
126 |
|
|
127 |
To run tests (which require that you have estcmd in your $PATH) issue |
To run tests (which require that you have estcmd in your $PATH) issue |
128 |
|
|
129 |
make test |
make test |
130 |
|
|
131 |
See also included file test.sql for more examples of usage. |
See also included file C<test.sql> for more examples of usage. |
132 |
|
|
133 |
=head1 Usage of pgest from SQL |
=head1 Usage of search function pgest from SQL |
134 |
|
|
135 |
C<pgest> PostgreSQL function has two different prototypes (number of arguments) depending on usage. |
C<pgest> PostgreSQL function tries to mimic usage of normal database tables (with support for attribute filtering, limit and offset) in following way: |
136 |
|
|
137 |
SELECT |
SELECT |
138 |
-- columns to return (defined later) |
-- columns to return (defined later) |
139 |
id,title,size |
id,title,size |
140 |
FROM pgest( |
FROM pgest( |
141 |
-- path to index OR URL to node, user-name and password |
-- node URI, login, password and depth of search |
|
-- you will need JUST ONE of following two lines, depending |
|
|
-- on your usage described below, for direct access |
|
|
'/full/path/to/casket', |
|
|
-- or for node API specify node URI, login, password |
|
|
-- and depth of search |
|
142 |
'http://localhost:1978/node/trivia', 'admin', 'admin', 42, |
'http://localhost:1978/node/trivia', 'admin', 'admin', 42, |
143 |
-- query |
-- query |
144 |
'blade runner', |
'blade runner', |
158 |
id text, title text, size text |
id text, title text, size text |
159 |
); |
); |
160 |
|
|
161 |
|
You should note that Hyper Estraier uses UTF-8 encoding, while your |
162 |
|
PostgreSQL installation might use different encoding. To fix that, use |
163 |
|
C<convert> function in PostgreSQL to convert encodings. |
164 |
|
|
165 |
|
=head2 Using index via C<estmaster> server process |
166 |
|
|
167 |
|
This is default and recommended way to use C<pgest> functionality. In this |
168 |
|
case, C<pgest> will use node API and access index through C<estmaster> |
169 |
|
process which should be running on (local or remote) machine. |
170 |
|
|
171 |
|
This will remove database opening overhead, at a cost of (small) additional network |
172 |
|
traffic. However, you can have Hyper Estraier C<estmaster> process running on |
173 |
|
different machine or update index while doing searches, so benefits of this |
174 |
|
approach are obvious. |
175 |
|
|
176 |
=head2 Accessing database directly |
=head2 Accessing database directly |
177 |
|
|
178 |
If you want to access database directly (without running C<estmaster> process), first argument is full path to database file. |
B<Please note that direct access to database is depreciated.> As such, it's |
179 |
|
not stated in example, and it's kept just for backward compatibility, but it |
180 |
|
will probably be removed in future versions of C<pgest>. |
181 |
|
|
182 |
|
If you want to access database directly (without running C<estmaster> process), you |
183 |
|
have to replace node URI, login, password and depth with full path to database file. |
184 |
|
|
185 |
Have in mind that C<postgres> user under which PostgreSQL is running must |
Have in mind that C<postgres> user under which PostgreSQL is running must |
186 |
have read permission on Hyper Estraier database files. |
have read permission on Hyper Estraier database files. |
189 |
index grows bigger, you might consider using node API to remove overhead of |
index grows bigger, you might consider using node API to remove overhead of |
190 |
database opening on each query. |
database opening on each query. |
191 |
|
|
192 |
B<Please note that direct access to database is depriciated.> |
=head1 Usage of trigger function pgest_trigger from SQL |
193 |
|
|
194 |
=head2 Using index via C<estmaster> server process |
Let's first say that I really suggest that you use C<dbi-index.pl> helper script to |
195 |
|
create triggers because it already supports following steps automatically: |
196 |
|
|
197 |
If first argument is URL to node (like C<http://localhost:1978/node/trivia>) |
=over |
|
and there are two additional parameters (user-name and password) after it, |
|
|
C<pgest> will use node API and access index through C<estmaster> process which should be running on (local or remote) machine. |
|
198 |
|
|
199 |
This will remove database opening overhead, at a cost of additional network |
=item begin transaction |
200 |
traffic. However, you can have Hyper Estraier C<estmaster> process running on |
|
201 |
different machine or update index while doing searches, so benefits of this |
Transaction is needed to catch updates which might happen while creation |
202 |
approach are obvious. |
of full-text index is in progress (and on huge collections this can take a while, |
203 |
|
just like normal index creation in PostgreSQL). |
204 |
|
|
205 |
|
=item insert all existing data in full-text index |
206 |
|
|
207 |
|
This will be done directly from PostgreSQL database to Hyper Estraier index. |
208 |
|
This is somewhat faster than waiting for trigger to fire for each existing |
209 |
|
row. |
210 |
|
|
211 |
|
=item create insert, update and delete triggers |
212 |
|
|
213 |
|
Which will keep data in sync later |
214 |
|
|
215 |
|
=item commit transaction |
216 |
|
|
217 |
|
=back |
218 |
|
|
219 |
|
If you still want to do that manually, you will need to know format of |
220 |
|
C<pgest_trigger> function: |
221 |
|
|
222 |
|
CREATE TRIGGER pgest_trigger_insert AFTER INSERT |
223 |
|
ON table FOR EACH ROW |
224 |
|
EXECUTE PROCEDURE pgest_trigger( |
225 |
|
-- node URI, login and password |
226 |
|
'http://localhost:1978/node/trivia', 'admin', 'admin', |
227 |
|
-- name of primary key column |
228 |
|
'id', |
229 |
|
-- names of all other columns to index (one or more) |
230 |
|
'column', 'another_one', 'and_another' |
231 |
|
) |
232 |
|
|
233 |
|
You have to create triggers for C<UPDATE> and C<DELETE> in similar way. |
234 |
|
|
235 |
=head1 Who wrote this? |
=head1 Who wrote this? |
236 |
|
|
237 |
Hyper Estraier is written by Mikio Hirabayashi. |
Hyper Estraier is written by Mikio Hirabayashi. |
238 |
|
|
|
Perl bindings for Hyper Estraier are written by MATSUNO Tokuhiro. |
|
|
|
|
239 |
PostgreSQL is written by hackers calling themselves PostgreSQL Global |
PostgreSQL is written by hackers calling themselves PostgreSQL Global |
240 |
Development Group. |
Development Group. |
241 |
|
|
242 |
This small C function is written by Dobrica Pavlinusic, dpavlin@rot13.org. |
This small C functions are written by L<Dobrica Pavlinusic|http://www.rot13.org/~dpavlin/>, dpavlin@rot13.org. |
243 |
|
|
244 |
|
=head1 See also |
245 |
|
|
246 |
|
L<Tutorial> - how to create first full-text index in under 10 minutes! |
247 |
|
|
248 |
|
L<ChangeLog> - what has changed since last version |
249 |
|
|
250 |
|
L<pgest-index> - helper script to create index and triggers |