Copyright © 2001-2008 Andrew Aksyonoff, <shodan(at)shodan.ru>
Table of Contents
sphinx.conf
options referenceindexer
program configuration optionssearchd
program configuration optionsSphinx is a full-text search engine, distributed under GPL version 2. Commercial licensing (eg. for embedded use) is also available upon request.
Generally, it's a standalone search engine, meant to provide fast, size-efficient and relevant full-text search functions to other applications. Sphinx was specially designed to integrate well with SQL databases and scripting languages.
Currently built-in data source drivers support fetching data either via direct connection to MySQL, or PostgreSQL, or from a pipe in a custom XML format. Adding new drivers (eg. to natively support some other DBMSes) is designed to be as easy as possible.
Search API is natively ported to PHP, Python, Perl, Ruby, Java, and also available as a pluggable MySQL storage engine. API is very lightweight so porting it to new language is known to take a few hours.
As for the name, Sphinx is an acronym which is officially decoded as SQL Phrase Index. Yes, I know about CMU's Sphinx project.
Sphinx is available through its official Web site at http://www.sphinxsearch.com/.
Currently, Sphinx distribution tarball includes the following software:
indexer
: an utility which creates fulltext indexes;search
: a simple command-line (CLI) test utility which searches through fulltext indexes;searchd
: a daemon which enables external software (eg. Web applications) to search through fulltext indexes;sphinxapi
: a set of searchd client API libraries for popular Web scripting languages (PHP, Python, Perl, Ruby).
This program is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 2 of the License, or (at your option) any later version. See COPYING file for details.
This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
You should have received a copy of the GNU General Public License along with this program; if not, write to the Free Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
If you don't want to be bound by GNU GPL terms (for instance, if you would like to embed Sphinx in your software, but would not like to disclose its source code), please contact the author to obtain a commercial license.
Sphinx initial author and current primary developer is:
<shodan(at)shodan.ru>
People who contributed to Sphinx and their contributions (in no particular order) are:
Many other people have contributed ideas, bug reports, fixes, etc. Thank you!
Sphinx development was started back in 2001, because I didn't manage to find an acceptable search solution (for a database driven Web site) which would meet my requirements. Actually, each and every important aspect was a problem:
Despite the amount of time passed and numerous improvements made in the other solutions, there's still no solution which I personally would be eager to migrate to.
Considering that and a lot of positive feedback received from Sphinx users during last years, the obvious decision is to continue developing Sphinx (and, eventually, to take over the world).
Most modern UNIX systems with a C++ compiler should be able to compile and run Sphinx without any modifications.
Currently known systems Sphinx has been successfully running on are:
CPU architectures known to work include X86, X86-64, SPARC64.
I hope Sphinx will work on other Unix platforms as well. If the platform you run Sphinx on is not in this list, please do report it.
At the moment, Windows version of Sphinx is not intended to be used in production, but rather for testing and debugging only. Two most prominent issues are missing concurrent queries support (client queries are stacked on TCP connection level instead), and missing index data rotation support. There are succesful production installations which workaround these issues. However, running high-volume search service under Windows is still not recommended.
On UNIX, you will need the following tools to build and install Sphinx:
On Windows, you will need Microsoft Visual C/C++ Studio .NET 2003 or 2005. Other compilers/environments will probably work as well, but for the time being, you will have to build makefile (or other environment specific project files) manually.
Extract everything from the distribution tarball (haven't you already?)
and go to the sphinx
subdirectory:
$ tar xzvf sphinx-0.9.7.tar.gz
$ cd sphinx
Run the configuration program:
$ ./configure
There's a number of options to configure. The complete listing may
be obtained by using --help
switch. The most important ones are:
--prefix
, which specifies where to install Sphinx;--with-mysql
, which specifies where to look for MySQL
include and library files, if auto-detection fails;--with-pgsql
, which specifies where to look for PostgreSQL
include and library files.
Build the binaries:
$ make
Install the binaries in the directory of your choice:
$ make install
If configure
fails to locate MySQL headers and/or libraries,
try checking for and installing mysql-devel
package. On some systems,
it is not installed by default.
If make
fails with a message which look like
/bin/sh: g++: command not found make[1]: *** [libsphinx_a-sphinx.o] Error 127
try checking for and installing gcc-c++
package.
If you are getting compile-time errors which look like
sphinx.cpp:67: error: invalid application of `sizeof' to incomplete type `Private::SizeError<false>'
this means that some compile-time type size check failed. The most probable reason is that off_t type is less than 64-bit on your system. As a quick hack, you can edit sphinx.h and replace off_t with DWORD in a typedef for SphOffset_t, but note that this will prohibit you from using full-text indexes larger than 2 GB. Even if the hack helps, please report such issues, providing the exact error message and compiler/OS details, so I could properly fix them in next releases.
If you keep getting any other error, or the suggestions above do not seem to help you, please don't hesitate to contact me.
All the example commands below assume that you installed Sphinx
in /usr/local/sphinx
.
To use Sphinx, you will need to:
Create a configuration file.
Default configuration file name is sphinx.conf
.
All Sphinx programs look for this file in current working directory
by default.
Sample configuration file, sphinx.conf.dist
, which has
all the options documented, is created by configure
.
Copy and edit that sample file to make your own configuration:
$ cd /usr/local/sphinx/etc
$ cp sphinx.conf.dist sphinx.conf
$ vi sphinx.conf
Sample configuration file is setup to index documents
table from MySQL database test
; so there's example.sql
sample data file to populate that table with a few documents for testing purposes:
$ mysql -u test < /usr/local/sphinx/etc/example.sql
Run the indexer to create full-text index from your data:
$ cd /usr/local/sphinx/etc
$ /usr/local/sphinx/bin/indexer
Query your newly created index!
To query the index from command line, use search
utility:
$ cd /usr/local/sphinx/etc
$ /usr/local/sphinx/bin/search test
To query the index from your PHP scripts, you need to:
Run the search daemon which your script will talk to:
$ cd /usr/local/sphinx/etc
$ /usr/local/sphinx/bin/searchd
Run the attached PHP API test script (to ensure that the daemon was succesfully started and is ready to serve the queries):
$ cd sphinx/api
$ php test.php test
Include the API (it's located in api/sphinxapi.php
)
into your own scripts and use it.
Happy searching!
The data to be indexed can generally come from very different sources: SQL databases, plain text files, HTML files, mailboxes, and so on. From Sphinx point of view, the data it indexes is a set of structured documents, each of which has the same set of fields. This is biased towards SQL, where each row correspond to a document, and each column to a field.
Depending on what source Sphinx should get the data from, different code is required to fetch the data and prepare it for indexing. This code is called data source driver (or simply driver or data source for brevity).
At the time of this writing, there are drivers for MySQL and
PostgreSQL databases, which can connect to the database using
its native C/C++ API, run queries and fetch the data. There's
also a driver called xmlpipe, which runs a specified command
and reads the data from its stdout
.
See Section 3.8, “xmlpipe data source” section for the format description.
There can be as many sources per index as necessary. They will be sequentially processed in the very same order which was specifed in index definition. All the documents coming from those sources will be merged as if they were coming from a single source.
Attributes are additional values associated with each document that can be used to perform additional filtering and sorting during search.
It is often desired to additionally process full-text search results based not only on matching document ID and its rank, but on a number of other per-document values as well. For instance, one might need to sort news search results by date and then relevance, or search through products within specified price range, or limit blog search to posts made by selected users, or group results by month. To do that efficiently, Sphinx allows to attach a number of additional attributes to each document, and store their values in the full-text index. It's then possible to use stored values to filter, sort, or group full-text matches.
A good example would be a forum posts table. Assume that only title and contentfields need to be full-text searchable - but that sometimes it is also required to limit search to a certain author or a sub-forum (ie. search only those rows that have some specific values of author_id or forum_id columns in the SQL table); or to sort matches by post_date column; or to group matching posts by month of the post_date and calculate per-group match counts.
This can be achieved by specifying all the mentioned columns (excluding title and content, that are full-text fields) as attributes, indexing them, and then using API calls to setup filtering, sorting, and grouping. Here as an example.
... sql_query = SELECT id, title, content, \ author_id, forum_id, post_date FROM my_forum_posts sql_attr_uint = author_id sql_attr_uint = forum_id sql_attr_timestamp = post_date ...
// only search posts by author whose ID is 123 $cl->SetFilter ( "author_id", array ( 123 ) ); // only search posts in sub-forums 1, 3 and 7 $cl->SetFilter ( "forum_id", array ( 1,3,7 ) ); // sort found posts by posting date in descending order $cl->SetSortMode ( SPH_SORT_ATTR_DESC, "post_date" );
Attributes are named. Attribute names are case insensitive. Attributes are not full-text indexed; they are stored in the index as is. Currently supported attribute types are:
The complete set of per-document attribute values is sometimes referred to as docinfo. Docinfos can either be
.spa
file), or.spd
file).
When using extern storage, a copy of .spa
file
(with all the attribute values for all the documents) is kept in RAM by
searchd
at all times. This is for performance reasons;
random disk I/O would be too slow. On the contrary, inline storage does not
require any additional RAM at all, but that comes at the cost of greatly
inflating the index size: remember that it copies all
attribute value every time when the document ID
is mentioned, and that is exactly as many times as there are
different keywords in the document. Inline may be the only viable
option if you have only a few attributes and need to work with big
datasets in limited RAM. However, in most cases extern storage
makes both indexing and searching much more efficient.
Search-time memory requirements for extern storage are
(1+number_of_attrs)*number_of_docs*4 bytes, ie. 10 million docs with
2 groups and 1 timestamp will take (1+2+1)*10M*4 = 160 MB of RAM.
This is PER DAEMON, not per query. searchd
will allocate 160 MB on startup, read the data and keep it shared between queries.
The children will NOT allocate any additional
copies of this data.
MVAs, or multi-valued attributes, are an important special type of per-document attributes in Sphinx. MVAs make it possible to attach lists of values to every document. They are useful for article tags, product categories, etc. Filtering and group-by (but not sorting) on MVA attributes is supported.
Currently, MVA list entries are limited to unsigned 32-bit integers.
The list length is not limited, you can have an arbitrary number of values
attached to each document as long as RAM permits (.spm
file
that contains the MVA values will be precached in RAM by searchd
).
The source data can be taken either from a separate query, or from a document field;
see source type in sql_attr_multi.
In the first case the query will have to return pairs of document ID and MVA values,
in the second one the field will be parsed for integer values.
There are absolutely no requirements as to incoming data order; the values will be
automatically grouped by document ID (and internally sorted within the same ID)
during indexing anyway.
When filtering, a document will match the filter on MVA attribute if any of the values satisfy the filtering condition. (Therefore, documents that pass through exclude filters will not contain any of the forbidden values.) When grouping by MVA attribute, a document will contribute to as many groups as there are different MVA values associated with that document. For instance, if the collection contains exactly 1 document having a 'tag' MVA with values 5, 7, and 11, grouping on 'tag' will produce 3 groups with '@count' equal to 1 and '@groupby' key values of 5, 7, and 11 respectively. Also note that grouping by MVA might lead to duplicate documents in the result set: because each document can participate in many groups, it can be chosen as the best one in in more than one group, leading to duplicate IDs. PHP API historically uses ordered hash on the document ID for the resulting rows; so you'll also need to use SetArrayResult() in order to employ group-by on MVA with PHP API.
To be able to answer full-text search queries fast, Sphinx needs to build a special data structure optimized for such queries from your text data. This structure is called index; and the process of building index from text is called indexing.
Different index types are well suited for different tasks. For example, a disk-based tree-based index would be easy to update (ie. insert new documents to existing index), but rather slow to search. Therefore, Sphinx architecture allows for different index types to be implemented easily.
The only index type which is implemented in Sphinx at the moment is designed for maximum indexing and searching speed. This comes at a cost of updates being really slow; theoretically, it might be slower to update this type of index than than to reindex it from scratch. However, this very frequently could be worked around with muiltiple indexes, see Section 3.10, “Live index updates” for details.
It is planned to implement more index types, including the type which would be updateable in real time.
There can be as many indexes per configuration file as necessary.
indexer
utility can reindex either all of them
(if --all
option is specified), or a certain explicitly
specified subset. searchd
utility will serve all
the specified indexes, and the clients can specify what indexes to
search in run time.
There are a few different restrictions imposed on the source data which is going to be indexed by Sphinx, of which the single most important one is:
ALL DOCUMENT IDS MUST BE UNIQUE UNSIGNED NON-ZERO INTEGER NUMBERS (32-BIT OR 64-BIT, DEPENDING ON BUILD TIME SETTINGS).
If this requirement is not met, different bad things can happen. For instance, Sphinx can crash with an internal assertion while indexing; or produce strange results when searching due to conflicting IDs. Also, a 1000-pound gorilla might eventually come out of your display and start throwing barrels at you. You've been warned.
When indexing some index, Sphinx fetches documents from the specified sources, splits the text into words, and does case folding so that "Abc", "ABC" and "abc" would be treated as the same word (or, to be pedantic, term).
To do that properly, Sphinx needs to know
This should be configured on a per-index basis using
charset_type
and
charset_table
options.
charset_type
specifies whether the document encoding is single-byte (SBCS) or UTF-8.
charset_table
specifies the table that maps letter characters to their case
folded versions. The characters that are not in the table are considered
to be non-letters and will be treated as word separators when indexing
or searching through this index.
Note that while default tables do not include space character (ASCII code 0x20, Unicode U+0020) as a letter, it's in fact perfectly legal to do so. This can be useful, for instance, for indexing tag clouds, so that space-separated word sets would index as a single search query term.
Default tables currently include English and Russian characters. Please do submit your tables for other languages!
With all the SQL drivers, indexing generally works as follows.
Most options, such as database user/host/password, are straightforward. However, there are a few subtle things, which are discussed in more detail here.
Main query, which needs to fetch all the documents, can impose a read lock on the whole table and stall the concurrent queries (eg. INSERTs to MyISAM table), waste a lot of memory for result set, etc. To avoid this, Sphinx supports so-called ranged queries. With ranged queries, Sphinx first fetches min and max document IDs from the table, and then substitutes different ID intervals into main query text and runs the modified query to fetch another chunk of documents. Here's an example.
Example 1. Ranged query usage example
# in sphinx.conf sql_query_range = SELECT MIN(id),MAX(id) FROM documents sql_range_step = 1000 sql_query = SELECT * FROM documents WHERE id>=$start AND id<=$end
If the table contains document IDs from 1 to, say, 2345, then sql_query would be run three times:
$start
replaced with 1 and $end
replaced with 1000;$start
replaced with 1001 and $end
replaced with 2000;$start
replaced with 2000 and $end
replaced with 2345.Obviously, that's not much of a difference for 2000-row table, but when it comes to indexing 10-million-row MyISAM table, ranged queries might be of some help.
sql_post
vs. sql_post_index
The difference between post-query and post-index query is in that post-query is run immediately when Sphinx received all the documents, but further indexing may still fail for some other reason. On the contrary, by the time the post-index query gets executed, it is guaranteed that the indexing was succesful. Database connection is dropped and re-established because sorting phase can be very lengthy and would just timeout otherwise.
xmlpipe data source was designed to enable users to plug data into Sphinx without having to implement new data sources drivers themselves. It is limited to 2 fixed fields and 2 fixed attributes, and is deprecated in favor of Section 3.9, “xmlpipe2 data source” now. For new streams, use xmlpipe2.
To use xmlpipe, configure the data source in your configuration file as follows:
source example_xmlpipe_source { type = xmlpipe xmlpipe_command = perl /www/mysite.com/bin/sphinxpipe.pl }
The indexer
will run the command specified
in xmlpipe_command
,
and then read, parse and index the data it prints to stdout
.
More formally, it opens a pipe to given command and then reads
from that pipe.
indexer will expect one or more documents in custom XML format. Here's the example document stream, consisting of two documents:
Example 2. XMLpipe document stream
<document> <id>123</id> <group>45</group> <timestamp>1132223498</timestamp> <title>test title</title> <body> this is my document body </body> </document> <document> <id>124</id> <group>46</group> <timestamp>1132223498</timestamp> <title>another test</title> <body> this is another document </body> </document>
Legacy xmlpipe legacy driver uses a builtin parser
which is pretty fast but really strict and does not actually
fully support XML. It requires that all the fields must
be present, formatted exactly as in this example, and
occur exactly in the same order. The only optional
field is timestamp
; it defaults to 1.
xmlpipe2 lets you pass arbitrary full-text and attribute data to Sphinx in yet another custom XML format. It also allows to specify the schema (ie. the set of fields and attributes) either in the XML stream itself, or in the source settings.
When indexing xmlpipe2 source, indexer runs the given command, opens a pipe to its stdout, and expects well-formed XML stream. Here's sample stream data:
Example 3. xmlpipe2 document stream
<?xml version="1.0" encoding="utf-8"?> <sphinx:docset> <sphinx:schema> <sphinx:field name="subject"/> <sphinx:field name="content"/> <sphinx:attr name="published" type="timestamp"/> <sphinx:attr name="author_id" type="int" bits="16" default="1"/> </sphinx:schema> <sphinx:document id="1234"> <content>this is the main content <![CDATA[[and this <cdata> entry must be handled properly by xml parser lib]]></content> <published>1012325463</published> <subject>note how field/attr tags can be in <b class="red">randomized</b> order</subject> <misc>some undeclared element</misc> </sphinx:document> <!-- ... more documents here ... --> </sphinx:docset>
Arbitrary fields and attributes are allowed. They also can occur in the stream in arbitrary order within each document; the order is ignored. There is a restriction on maximum field length; fields longer than 2 MB will be truncated to 2 MB (this limit can be changed in the source).
The schema, ie. complete fields and attributes list, must be declared
before any document could be parsed. This can be done either in the
configuration file using xmlpipe_field
and xmlpipe_attr_XXX
settings, or right in the stream using <sphinx:schema> element.
<sphinx:schema> is optional. It is only allowed to occur as the very
first sub-element in <sphinx:docset>. If there is no in-stream
schema definition, settings from the configuration file will be used.
Otherwise, stream settings take precedence.
Unknown tags (which were not declared neither as fields nor as attributes) will be ignored with a warning. In the example above, <misc> will be ignored. All embedded tags and their attributes (such as <b> in <subject> in the example above) will be silently ignored.
Support for incoming stream encodings depends on whether iconv
is installed on the system. xmlpipe2 is parsed using libexpat
parser that understands US-ASCII, ISO-8859-1, UTF-8 and a few UTF-16 variants
natively. Sphinx configure
script will also check
for libiconv
presence, and utilize it to handle
other encodings. libexpat
also enforces the
requirement to use UTF-8 charset on Sphinx side, because the
parsed data it returns is always in UTF-8.
XML elements (tags) recognized by xmlpipe2 (and their attributes where applicable) are:
There's a frequent situation when the total dataset is too big to be reindexed from scratch often, but the amount of new records is rather small. Example: a forum with a 1,000,000 archived posts, but only 1,000 new posts per day.
In this case, "live" (almost real time) index updates could be implemented using so called "main+delta" scheme.
The idea is to set up two sources and two indexes, with one "main" index for the data which only changes rarely (if ever), and one "delta" for the new documents. In the example above, 1,000,000 archived posts would go to the main index, and newly inserted 1,000 posts/day would go to the delta index. Delta index could then be reindexed very frequently, and the documents can be made available to search in a matter of minutes.
Specifying which documents should go to what index and reindexing main index could also be made fully automatical. One option would be to make a counter table which would track the ID which would split the documents, and update it whenever the main index is reindexed.
Example 4. Fully automated live updates
# in MySQL CREATE TABLE sph_counter ( counter_id INTEGER PRIMARY KEY NOT NULL, max_doc_id INTEGER NOT NULL ); # in sphinx.conf source main { # ... sql_query_pre = SET NAMES utf8 sql_query_pre = REPLACE INTO sph_counter SELECT 1, MAX(id) FROM documents sql_query = SELECT id, title, body FROM documents \ WHERE id<=( SELECT max_doc_id FROM sph_counter WHERE counter_id=1 ) } source delta : main { sql_query_pre = SET NAMES utf8 sql_query = SELECT id, title, body FROM documents \ WHERE id>( SELECT max_doc_id FROM sph_counter WHERE counter_id=1 ) } index main { source = main path = /path/to/main # ... all the other settings } # note how all other settings are copied from main, # but source and path are overridden (they MUST be) index delta : main { source = delta path = /path/to/delta }
Note how we're overriding sql_query_pre
in the delta source.
We need to explicitly have that override. Otherwise REPLACE
query
would be run when indexing delta source too, effectively nullifying it. However,
when we issue the directive in the inherited source for the first time, it removes
all inherited values, so the encoding setup is also lost.
So sql_query_pre
in the delta can not just be empty; and we need
to issue the encoding setup query explicitly once again.
Merging two existing indexes can be more efficient that indexing the data
from scratch, and desired in some cases (such as merging 'main' and 'delta'
indexes instead of simply reindexing 'main' in 'main+delta' partitioning
scheme). So indexer
has an option to do that.
Merging the indexes is normally faster than reindexing but still
not instant on huge indexes. Basically,
it will need to read the contents of both indexes once and write
the result once. Merging 100 GB and 1 GB index, for example,
will result in 202 GB of IO (but that's still likely less than
the indexing from scratch requires).
The basic command syntax is as follows:
indexer --merge DSTINDEX SRCINDEX [--rotate]
Only the DSTINDEX index will be affected: the contents of SRCINDEX will be merged into it.
--rotate
switch will be required if DSTINDEX is already being served by searchd
.
The initially devised usage pattern is to merge a smaller update from SRCINDEX into DSTINDEX.
Thus, when merging the attributes, values from SRCINDEX will win if duplicate document IDs are encountered.
Note, however, that the "old" keywords will not be automatically removed in such cases.
For example, if there's a keyword "old" associated with document 123 in DSTINDEX, and a keyword "new" associated
with it in SRCINDEX, document 123 will be found by both keywords after the merge.
You can supply an explicit condition to remove documents from DSTINDEX to mitigate that;
the relevant switch is --merge-dst-range
:
indexer --merge main delta --merge-dst-range deleted 0 0
This switch lets you apply filters to the destination index along with merging. There can be several filters; all of their conditions must be met in order to include the document in the resulting mergid index. In the example above, the filter passes only those records where 'deleted' is 0, eliminating all records that were flagged as deleted (for instance, using UpdateAttributes() call).
There are the following matching modes available:
There also is a special "full scan" mode that will be automatically activated when the following conditions are met:
extern
.In full scan mode, all the indexed documents will be considered as matching. Such queries will still apply filters, sorting, or group by, but will not perform any real full-text searching. This can be useful to unify full-text and non-full-text searching code, or to offload SQL server (there are cases when Sphinx scans will perform better than analogous MySQL queries).
Boolean queries allow the following special operators to be used:
hello & world
hello | world
hello -world hello !world
( hello world )
Here's an example query which uses all these operators:
There always is implicit AND operator, so "hello world" query actually means "hello & world".
OR operator precedence is higher than AND, so "looking for cat | dog | mouse" means "looking for ( cat | dog | mouse )" and not "(looking for cat) | dog | mouse".
Queries like "-dog", which implicitly include all documents from the collection, can not be evaluated. This is both for technical and performance reasons. Technically, Sphinx does not always keep a list of all IDs. Performance-wise, when the collection is huge (ie. 10-100M documents), evaluating such queries could take very long.
The following special operators can be used when using the extended matching mode:
hello | world
hello -world hello !world
@title hello @body world
"hello world"
"hello world"~10
"the world is a wonderful place"/3
Here's an example query which uses most of these operators:
There always is implicit AND operator, so "hello world" means that both "hello" and "world" must be present in matching document.
OR operator precedence is higher than AND, so "looking for cat | dog | mouse" means "looking for ( cat | dog | mouse )" and not "(looking for cat) | dog | mouse".
Proximity distance is specified in words, adjusted for word count, and applies to all words within quotes. For instance, "cat dog mouse"~5 query means that there must be less than 8-word span which contains all 3 words, ie. "CAT aaa bbb ccc DOG eee fff MOUSE" document will not match this query, because this span is exactly 8 words long.
Quorum matching operator introduces a kind of fuzzy matching. It will only match those documents that pass a given threshold of given words. The example above ("the world is a wonderful place"/3) will match all documents that have at least 3 of the 6 specified words.
Nested brackets, as in queries like
aaa | ( bbb ccc | ( ddd eee ) )
are not allowed yet, but this will be fixed.
Negation (ie. operator NOT) is only allowed on top level and not within brackets (ie. groups). This isn't going to change, because supporting nested negations would make phrase ranking implementation way too complicated.
Specific weighting function (currently) depends on the search mode.
There are these major parts which are used in the weighting functions:
Phrase rank is based on a length of longest common subsequence (LCS) of search words between document body and query phrase. So if there's a perfect phrase match in some document then its phrase rank would be the highest possible, and equal to query words count.
Statistical rank is based on classic BM25 function which only takes word frequencies into account. If the word is rare in the whole database (ie. low frequency over document collection) or mentioned a lot in specific document (ie. high frequency over matching document), it receives more weight. Final BM25 weight is a floating point number between 0 and 1.
In all modes, per-field weighted phrase ranks are computed as a product of LCS multiplied by per-field weight speficifed by user. Per-field weights are integer, default to 1, and can not be set lower than 1.
In SPH_MATCH_BOOLEAN mode, no weighting is performed at all, every match weight is set to 1.
In SPH_MATCH_ALL and SPH_MATCH_PHRASE modes, final weight is a sum of weighted phrase ranks.
In SPH_MATCH_ANY mode, the idea is essentially the same, but it also adds a count of matching words in each field. Before that, weighted phrase ranks are additionally mutliplied by a value big enough to guarantee that higher phrase rank in any field will make the match ranked higher, even if it's field weight is low.
In SPH_MATCH_EXTENDED mode, final weight is a sum of weighted phrase ranks and BM25 weight, multiplied by 1000 and rounded to integer.
This is going to be changed, so that MATCH_ALL and MATCH_ANY modes use BM25 weights as well. This would improve search results in those match spans where phrase ranks are equal; this is especially useful for 1-word queries.
The key idea (in all modes, besides boolean) is that better subphrase matches are ranked higher, and perfect matches are pulled to the top. Author's experience is that this phrase proximity based ranking provides noticeably better search quality than any statistical scheme alone (such as BM25, which is commonly used in other search engines).
There are the following result sorting modes available:
SPH_SORT_RELEVANCE ignores any additional parameters and always sorts matches by relevance rank. All other modes require an additional sorting clause, with the syntax depending on specific mode. SPH_SORT_ATTR_ASC, SPH_SORT_ATTR_DESC and SPH_SORT_TIME_SEGMENTS modes require simply an attribute name. SPH_SORT_RELEVANCE is equivalent to sorting by "@weight DESC, @id ASC" in extended mode, SPH_SORT_ATTR_ASC is equivalent to "attribute ASC, @weight DESC, @id ASC", and SPH_SORT_ATTR_DESC to "attribute DESC, @weight DESC, @id ASC" respectively.
In SPH_SORT_TIME_SEGMENTS mode, attribute values are split into so-called time segments, and then sorted by time segment first, and by relevance second.
The segments are calculated according to the current timestamp at the time when the search is performed, so the results would change over time. The segments are as follows:
These segments are hardcoded, but it is trivial to change them if necessary.
This mode was added to support searching through blogs, news headlines, etc. When using time segments, recent records would be ranked higher because of segment, but withing the same segment, more relevant records would be ranked higher - unlike sorting by just the timestamp attribute, which would not take relevance into account at all.
In SPH_SORT_EXTENDED mode, you can specify an SQL-like sort expression with up to 5 attributes (including internal attributes), eg:
@relevance DESC, price ASC, @id DESC
Both internal attributes (that are computed by the engine on the fly)
and user attributes that were configured for this index are allowed.
Internal attribute names must start with magic @-symbol; user attribute
names can be used as is. In the example above, @relevance
and @id
are internal attributes and price
is user-specified.
Known internal attributes are:
@rank
and @relevance
are just additional
aliases to @weight
.
Expression sorting mode lets you sort the matches by an arbitrary arithmetic expression, involving attribute values, internal attributes (@id and @weight), arithmetic operations, and a number of built-in functions. Here's an example:
$cl->SetSortMode ( SPH_SORT_EXPR, "@weight + ( user_karma + ln(pageviews) )*0.1" );
The following operators and functions are supported. They are mimiced after MySQL. The functions take a number of arguments depending on the specific function.
All calculations are performed in single-precision, 32-bit IEEE 754 floating point format.
Comparison operators (eg. = or <=) return 1.0 when the condition is true and 0.0 otherwise.
For instance, (a=b)+3
will evaluate to 4 when attribute 'a' is equal to attribute 'b', and to 3 when 'a' is not.
Unlike MySQL, the equality comparisons (ie. = and <> operators) introduce a small equality threshold (1e-6 by default).
If the difference between compared values is within the threshold, they will be considered equal.
All unary and binary functions are straightforward, they behave just like their mathematical counterparts.
But IF()
behavior needs to be explained in more detail.
It takes 3 arguments, check whether the 1st argument is equal to 0.0, returns the 2nd argument if it is not zero, or the 3rd one when it is.
Note that unlike comparison operators, IF()
does not use a threshold!
Therefore, it's safe to use comparison results as its 1st argument, but arithmetic operators might produce unexpected results.
For instance, the following two calls will produce different results even though they are logically equivalent:
IF ( sqrt(3)*sqrt(3)-3<>0, a, b ) IF ( sqrt(3)*sqrt(3)-3, a, b )
In the first case, the comparison operator <> will return 0.0 (false)
because of a threshold, and IF()
will always return 'b' as a result.
In the second one, the same sqrt(3)*sqrt(3)-3
expression will be compared
with zero without threshold by the IF()
function itself.
But its value will be slightly different from zero because of limited floating point
calculations precision. Because of that, the comparison with 0.0 done by IF()
will not pass, and the second variant will return 'a' as a result.
Sometimes it could be useful to group (or in other terms, cluster) search results and/or count per-group match counts - for instance, to draw a nice graph of how much maching blog posts were there per each month; or to group Web search results by site; or to group matching forum posts by author; etc.
In theory, this could be performed by doing only the full-text search in Sphinx and then using found IDs to group on SQL server side. However, in practice doing this with a big result set (10K-10M matches) would typically kill performance.
To avoid that, Sphinx offers so-called grouping mode. It is enabled with SetGroupBy() API call. When grouping, all matches are assigned to different groups based on group-by value. This value is computed from specified attribute using one of the following built-in functions:
The final search result set then contains one best match per group. Grouping function value and per-group match count are returned along as "virtual" attributes named @group and @count respectively.
The result set is sorted by group-by sorting clause, with the syntax similar
to SPH_SORT_EXTENDED
sorting clause
syntax. In addition to @id
and @weight
,
group-by sorting clause may also include:
The default mode is to sort by groupby value in descending order,
ie. by "@group desc"
.
On completion, total_found
result parameter would
contain total amount of matching groups over he whole index.
WARNING: grouping is done in fixed memory
and thus its results are only approximate; so there might be more groups reported
in total_found
than actually present. @count
might also
be underestimated. To reduce inaccuracy, one should raise max_matches
.
If max_matches
allows to store all found groups, results will be 100% correct.
For example, if sorting by relevance and grouping by "published"
attribute with SPH_GROUPBY_DAY
function, then the result set will
contain
To scale well, Sphinx has distributed searching capabilities. Distributed searching is useful to improve query latency (ie. search time) and throughput (ie. max queries/sec) in multi-server, multi-CPU or multi-core environments. This is essential for applications which need to search through huge amounts data (ie. billions of records and terabytes of text).
The key idea is to horizontally partition (HP) searched data accross search nodes and then process it in parallel.
Partitioning is done manually. You should
indexer
and searchd
)
on different servers;searchd
instances;This index only contains references to other local and remote indexes - so it could not be directly reindexed, and you should reindex those indexes which it references instead.
When searchd
receives a query against distributed index,
it does the following:
From the application's point of view, there are no differences between usual and distributed index at all.
Any searchd
instance could serve both as a master
(which aggregates the results) and a slave (which only does local searching)
at the same time. This has a number of uses:
It is scheduled to implement better HA support which would allow to specify which agents mirror each other, do health checks, keep track of alive agents, load-balance requests, etc.
searchd
logs all succesfully executed search queries
into query log file. Here's an example:
[Fri Jun 29 21:17:58 2007] 0.004 sec [all/0/rel 35254 (0,20)] [lj] test [Fri Jun 29 21:20:34 2007] 0.024 sec [all/0/rel 19886 (0,20) @channel_id] [lj] test
This log format is as follows:
[query-date] query-time [match-mode/filters-count/sort-mode total-matches (offset,limit) @groupby-attr] [index-name] query
Match mode can take one of the following values:
Sort mode can take one of the following values:
There is a number of native searchd client API implementations for Sphinx. As of time of this writing, we officially support our own PHP, Python, and Java implementations. There also are third party free, open-source API implementations for Perl, Ruby, and C++.
The reference API implementation is in PHP, because (we believe) Sphinx is most widely used with PHP than any other language. This reference documentation is in turn based on reference PHP API, and all code samples in this section will be given in PHP.
However, all other APIs provide the same methods and implement the very same network protocol. Therefore the documentation does apply to them as well. There might be minor differences as to the method naming conventions or specific data structures used. But the provided functionality must not differ across languages.
Prototype: function GetLastError()
Returns last error message, as a string, in human readable format. If there were no errors during the previous API call, empty string is returned.
You should call it when any other function (such as Query()) fails (typically, the failing function returns false). The returned string will contain the error description.
The error message is not reset by this call; so you can safely call it several times if needed.
Prototype: function GetLastWarning ()
Returns last warning message, as a string, in human readable format. If there were no warnings during the previous API call, empty string is returned.
You should call it to verify whether your request (such as Query()) was completed but with warnings. For instance, search query against a distributed index might complete succesfully even if several remote agents timed out. In that case, a warning message would be produced.
The warning message is not reset by this call; so you can safely call it several times if needed.
Prototype: function SetServer ( $host, $port )
Sets searchd
host name and TCP port.
All subsequent requests will use the new host and port settings.
Default host and port are 'localhost' and 3312, respectively.
Prototype: function SetRetries ( $count, $delay=0 )
Sets distributed retry count and delay.
On temporary failures searchd
will attempt up to
$count
retries per agent. $delay
is the delay
between the retries, in milliseconds. Retries are disabled by default.
Note that this call will not make the API itself retry on
temporary failure; it only tells searchd
to do so.
Currently, the list of temporary failures includes all kinds of connect()
failures and maxed out (too busy) remote agents.
Prototype: function SetArrayResult ( $arrayresult )
PHP specific. Controls matches format in the search results set (whether matches should be returned as an array or a hash).
$arrayresult
argument must be boolean. If $arrayresult
is false
(the default mode), matches will returned in PHP hash format with
document IDs as keys, and other information (weight, attributes)
as values. If $arrayresult
is true, matches will be returned
as a plain array with complete per-match information including
document ID.
Introduced along with GROUP BY support on MVA attributes. Group-by-MVA result sets may contain duplicate document IDs. Thus they need to be returned as plain arrays, because hashes will only keep one entry per document ID.
Prototype: function SetLimits ( $offset, $limit, $max_matches=0, $cutoff=0 )
Sets offset into server-side result set ($offset
) and amount of matches
to return to client starting from that offset ($limit
). Can additionally
control maximum server-side result set size for current query ($max_matches
)
and the threshold amount of matches to stop searching at ($cutoff
).
All parameters must be non-negative integers.
First two parameters to SetLimits() are identical in behavior to MySQL
LIMIT clause. They instruct searchd
to return at
most $limit
matches starting from match number $offset
.
The default offset and limit settings are 0 and 20, that is, to return
first 20 matches.
max_matches
setting controls how much matches searchd
will keep in RAM while searching. All matching documents will be normally
processed, ranked, filtered, and sorted even if max_matches
is set to 1.
But only best N documents are stored in memory at any given moment for performance
and RAM usage reasons, and this setting controls that N. Note that there are
two places where max_matches
limit is enforced. Per-query
limit is controlled by this API call, but there also is per-server limit
controlled by max_matches
setting in the config file. To prevent
RAM usage abuse, server will not allow to set per-query limit
higher than the per-server limit.
You can't retrieve more than max_matches
matches to the client application.
The default limit is set to 1000. Normally, you must not have to go over
this limit. One thousand records is enough to present to the end user.
And if you're thinking about pulling the results to application
for further sorting or filtering, that would be much more efficient
if performed on Sphinx side.
$cutoff
setting is intended for advanced performance control.
It tells searchd
to forcibly stop search query
once $cutoff
matches had been found and processed.
Prototype: function SetMaxQueryTime ( $max_query_time )
Sets maximum search query time, in milliseconds. Parameter must be a non-negative integer. Default valus is 0 which means "do not limit".
Similar to $cutoff
setting from SetLimits(),
but limits elapsed query time instead of processed matches count. Local search queries
will be stopped once that much time has elapsed. Note that if you're performing
a search which queries several local indexes, this limit applies to each index
separately.
Prototype: function SetMatchMode ( $mode )
Sets full-text query matching mode, as described in Section 4.1, “Matching modes”. Parameter must be a constant specifying one of the known modes.
WARNING: (PHP specific) you must not take the matching mode constant name in quotes, that syntax specifies a string and is incorrect:
$cl->SetMatchMode ( "SPH_MATCH_ANY" ); // INCORRECT! will not work as expected $cl->SetMatchMode ( SPH_MATCH_ANY ); // correct, works OK
Prototype: function SetRankingMode ( $ranker )
Sets ranking mode. Only available in SPH_MATCH_EXTENDED2 matching mode at the time of this writing. Parameter must be a constant specifying one of the known modes.
By default, Sphinx computes two factors which contribute to the final match weight. The major part is query phrase proximity to document text. The minor part is so-called BM25 statistical function, which varies from 0 to 1 depending on the keyword frequency within document (more occurrences yield higher weight) and within the whole index (more rare keywords yield higher weight).
However, in some cases you'd want to compute weight differently - or maybe avoid computing it at all for performance reasons because you're sorting the result set by something else anyway. This can be accomplished by setting the appropriate ranking mode.
Currently implemented modes are:
Prototype: function SetSortMode ( $mode, $sortby="" )
Set matches sorting mode, as described in Section 4.5, “Sorting modes”. Parameter must be a constant specifying one of the known modes.
WARNING: (PHP specific) you must not take the matching mode constant name in quotes, that syntax specifies a string and is incorrect:
$cl->SetSortMode ( "SPH_SORT_ATTR_DESC" ); // INCORRECT! will not work as expected $cl->SetSortMode ( SPH_SORT_ATTR_ASC ); // correct, works OK
Prototype: function SetWeights ( $weights )
Binds per-field weights in the order of appearance in the index. DEPRECATED, use SetFieldWeights() instead.
Prototype: function SetFieldWeights ( $weights )
Binds per-field weights by name. Parameter must be a hash (associative array) mapping string field names to integer weights.
Match ranking can be affected by per-field weights. For instance, see Section 4.4, “Weighting” for an explanation how phrase proximity ranking is affected. This call lets you specify what non-default weights to assign to different full-text fields.
The weights must be positive 32-bit integers. The final weight will be a 32-bit integer too. Default weight value is 1. Unknown field names will be silently ignored.
There is no enforced limit on the maximum weight value at the moment. However, beware that if you set it too high you can start hitting 32-bit wraparound issues. For instance, if you set a weight of 10,000,000 and search in extended mode, then maximum possible weight will be equal to 10 million (your weight) by 1 thousand (internal BM25 scaling factor, see Section 4.4, “Weighting”) by 1 or more (phrase proximity rank). The result is at least 10 billion that does not fit in 32 bits and will be wrapped around, producing unexpected results.
Prototype: function SetIndexWeights ( $weights )
Sets per-index weights, and enables weighted summing of match weights across different indexes. Parameter must be a hash (associative array) mapping string index names to integer weights. Default is empty array that means to disable weighting summing.
When a match with the same document ID is found in several different local indexes, by default Sphinx simply chooses the match from the index specified last in the query. This is to support searching through partially overlapping index partitions.
However in some cases the indexes are not just partitions, and you
might want to sum the weights across the indexes instead of picking one.
SetIndexWeights()
lets you do that. With summing enabled,
final match weight in result set will be computed as a sum of match
weight coming from the given index multiplied by respective per-index
weight specified in this call. Ie. if the document 123 is found in
index A with the weight of 2, and also in index B with the weight of 3,
and you called SetIndexWeights ( array ( "A"=>100, "B"=>10 ) )
,
the final weight return to the client will be 2*100+3*10 = 230.
Prototype: function SetIDRange ( $min, $max )
Sets an accepted range of document IDs. Parameters must be integers. Defaults are 0 and 0; that combination means to not limit by range.
After this call, only those records that have document ID
between $min
and $max
(including IDs
exactly equal to $min
or $max
)
will be matched.
Prototype: function SetFilter ( $attribute, $values, $exclude=false )
Adds new integer values set filter.
On this call, additional new filter is added to the existing
list of filters. $attribute
must be a string with
attribute name. $values
must be a plain array
containing integer values. $exclude
must be a boolean
value; it controls whether to accept the matching documents
(default mode, when $exclude
is false) or reject them.
Only those documents where $attribute
column value
stored in the index matches any of the values from $values
array will be matched (or rejected, if $exclude
is true).
Prototype: function SetFilterRange ( $attribute, $min, $max, $exclude=false )
Adds new integer range filter.
On this call, additional new filter is added to the existing
list of filters. $attribute
must be a string with
attribute name. $min
and $max
must be
integers that define the acceptable attribute values range
(including the boundaries). $exclude
must be a boolean
value; it controls whether to accept the matching documents
(default mode, when $exclude
is false) or reject them.
Only those documents where $attribute
column value
stored in the index is between $min
and $max
(including values that are exactly equal to $min
or $max
)
will be matched (or rejected, if $exclude
is true).
Prototype: function SetFilterFloatRange ( $attribute, $min, $max, $exclude=false )
Adds new float range filter.
On this call, additional new filter is added to the existing
list of filters. $attribute
must be a string with
attribute name. $min
and $max
must be
floats that define the acceptable attribute values range
(including the boundaries). $exclude
must be a boolean
value; it controls whether to accept the matching documents
(default mode, when $exclude
is false) or reject them.
Only those documents where $attribute
column value
stored in the index is between $min
and $max
(including values that are exactly equal to $min
or $max
)
will be matched (or rejected, if $exclude
is true).
Prototype: function SetGeoAnchor ( $attrlat, $attrlong, $lat, $long )
Sets anchor point for and geosphere distance (geodistance) calculations, and enable them.
$attrlat
and $attrlong
must be strings that contain the names
of latitude and longitude attributes, respectively. $lat
and $long
are floats that specify anchor point latitude and longitude, in radians.
Once an anchor point is set, you can use magic "@geodist"
attribute
name in your filters and/or sorting expressions. Sphinx will compute geosphere distance
between the given anchor point and a point specified by latitude and lognitude
attributes from each full-text match, and attach this value to the resulting match.
The latitude and longitude values both in SetGeoAnchor
and the index
attribute data are expected to be in radians. The result will be returned in meters,
so geodistance value of 1000.0 means 1 km. 1 mile is approximately 1609.344 meters.
Prototype: function SetGroupBy ( $attribute, $func, $groupsort="@group desc" )
Sets grouping attribute, function, and groups sorting mode; and enables grouping (as described in Section 4.6, “Grouping (clustering) search results ”).
$attribute
is a string that contains group-by attribute name.
$func
is a constant that chooses a function applied to the attribute value in order to compute group-by key.
$groupsort
is a clause that controls how the groups will be sorted. Its syntax is similar
to that described in Section 4.5, “SPH_SORT_EXTENDED mode”.
Grouping feature is very similar in nature to GROUP BY clause from SQL. Results produces by this function call are going to be the same as produced by the following pseudo code:
SELECT ... GROUP BY $func($attribute) ORDER BY $groupsort
Note that it's $groupsort
that affects the order of matches
in the final result set. Sorting mode (see Section 5.3.3, “SetSortMode”)
affect the ordering of matches within group, ie.
what match will be selected as the best one from the group.
So you can for instance order the groups by matches count
and select the most relevant match within each group at the same time.
Prototype: function SetGroupDistinct ( $attribute )
Sets attribute name for per-group distinct values count calculations. Only available for grouping queries.
$attribute
is a string that contains the attribute name.
For each group, all values of this attribute will be stored (as RAM limits
permit), then the amount of distinct values will be calculated and returned
to the client. This feature is similar to COUNT(DISTINCT)
clause in standard SQL; so these Sphinx calls:
$cl->SetGroupBy ( "category", SPH_GROUPBY_ATTR, "@count desc" ); $cl->SetGroupDistinct ( "vendor" );
can be expressed using the following SQL clauses:
SELECT id, weight, all-attributes, COUNT(DISTINCT vendor) AS @distinct, COUNT(*) AS @count FROM products GROUP BY category ORDER BY @count DESC
In the sample pseudo code shown just above, SetGroupDistinct()
call
corresponds to COUNT(DISINCT vendor)
clause only.
GROUP BY
, ORDER BY
, and COUNT(*)
clauses are all an equivalent of SetGroupBy()
settings. Both queries
will return one matching row for each category. In addition to indexed attributes,
matches will also contain total per-category matches count, and the count
of distinct vendor IDs within each category.
Prototype: function Query ( $query, $index="*" )
Connects to searchd
server, runs given search query
with current settings, obtains and returns the result set.
$query
is a query string. $index
is an index name (or names) string.
Returns false and sets GetLastError()
message on general error.
Returns search result set on success.
Default value for $index
is "*"
that means
to query all local indexes. Characters allowed in index names include
Latin letters (a-z), numbers (0-9), minus sign (-), and underscore (_);
everything else is considered a separator. Therefore, all of the
following samples calls are valid and will search the same
two indexes:
$cl->Query ( "test query", "main delta" ); $cl->Query ( "test query", "main;delta" ); $cl->Query ( "test query", "main, delta" );
Index specification order matters. If document with identical IDs are found in two or more indexes, weight and attribute values from the very last matching index will be used for sorting and returning to client (unless explicitly overridden with SetIndexWeights()). Therefore, in the example above, matches from "delta" index will always win over matches from "main".
On success, Query()
returns a result set that contains
some of the found matches (as requested by SetLimits())
and additional general per-query statistics. The result set is a hash
(PHP specific; other languages might utilize other structures instead
of hash) with the following keys and values:
searchd
(string, human readable). Empty if there were no errors.searchd
(string, human readable). Empty if there were no warnings.
Prototype: function AddQuery ( $query, $index="*" )
Adds additional query with current settings to multi-query batch.
$query
is a query string. $index
is an index name (or names) string.
Returns index to results array returned from RunQueries().
Batch queries (or multi-queries) enable searchd
to perform internal
optimizations if possible. They also reduce network connection overheads and search process
creation overheads in all cases. They do not result in any additional overheads compared
to simple queries. Thus, if you run several different queries from your web page,
you should always consider using multi-queries.
For instance, running the same full-text query but with different
sorting or group-by settings will enable searchd
to perform expensive full-text search and ranking operation only once,
but compute multiple group-by results from its output.
This can be a big saver when you need to display not just plain search results but also some per-category counts, such as the amount of products grouped by vendor. Without multi-query, you would have to run several queries which perform essentially the same search and retrieve the same matches, but create result sets differently. With multi-query, you simply pass all these querys in a single batch and Sphinx optimizes the redundant full-text search internally.
AddQuery()
internally saves full current settings state
along with the query, and you can safely change them afterwards for subsequent
AddQuery()
calls. Already added queries will not be affected;
there's actually no way to change them at all. Here's an example:
$cl->SetSortMode ( SPH_SORT_RELEVANCE ); $cl->AddQuery ( "hello world", "documents" ); $cl->SetSortMode ( SPH_SORT_ATTR_DESC, "price" ); $cl->AddQuery ( "ipod", "products" ); $cl->AddQuery ( "harry potter", "books" ); $results = $cl->RunQueries ();
With the code above, 1st query will search for "hello world" in "documents" index
and sort results by relevance, 2nd query will search for "ipod" in "products"
index and sort results by price, and 3rd query will search for "harry potter"
in "books" index while still sorting by price. Note that 2nd SetSortMode()
call
does not affect the first query (because it's already added) but affects both other
subsequent queries.
AddQuery()
does not modify the current state. That is,
all current sorting, filtering, and grouping settings will not be affected by
this call; so subsequent queries can easily reuse current query settings.
AddQuery()
returns an index into an array of results
that will be returned from RunQueries()
call. It is simply
a sequentially increasing 0-based integer, ie. first call will return 0,
second will return 1, and so on. Just a small helper so you won't have
to track the indexes manualy if you need then.
Prototype: function RunQueries ()
Connect to searchd, runs a batch of all queries added using AddQuery()
,
obtains and returns the result sets. Returns false and sets GetLastError()
message on general error (such as network I/O failure). Returns a plain array
of result sets on success.
Each result set in the returned array is exactly the same as
the result set returned from Query()
.
Note that the batch query request itself almost always succeds - unless there's a network error, blocking index rotation in progress, or another general failure which prevents the whole request from being processed.
However individual queries within the batch might very well fail.
In this case their respective result sets will contain non-empty "error"
message,
but no matches or query statistics. In the extreme case all queries within the batch
could fail. There still will be no general error reported, because API was able to
succesfully connect to searchd
, submit the batch, and receive
the results - but every result set will have a specific error message.
Prototype: function ResetFilters ()
Clears all currently set filters.
This call is only normally required when using multi-queries. You might want
to set different filters for different queries in the batch. To do that,
you should call ResetFilters()
and add new filters using
the respective calls.
Prototype: function ResetGroupBy ()
Clears all currently group-by settings, and disables group-by.
This call is only normally required when using multi-queries.
You can change individual group-by settings using SetGroupBy()
and SetGroupDistinct()
calls, but you can not disable
group-by using those calls. ResetGroupBy()
fully resets previous group-by settings and disables group-by mode
in the current state, so that subsequent AddQuery()
calls can perform non-grouping searches.
Prototype: function BuildExcerpts ( $docs, $index, $words, $opts=array() )
Excerpts (snippets) builder function. Connects to searchd
,
asks it to generate excerpts (snippets) from given documents, and returns the results.
$docs
is a plain array of strings that carry the documents' contents.
$index
is an index name string. Different settings (such as charset,
morphology, wordforms) from given index will be used.
$words
is a string that contains the keywords to highlight. They will
be processed with respect to index settings. For instance, if English stemming
is enabled in the index, "shoes" will be highlighted even if keyword is "shoe".
$opts
is a hash which contains additional optional highlighting parameters:
Returns false on failure. Returns a plain array of strings with excerpts (snippets) on success.
Prototype: function UpdateAttributes ( $index, $attrs, $values )
Instantly updates given attribute values in given documents. Returns number of actually updated documents (0 or more) on success, or -1 on failure.
$index
is a name of the index (or indexes) to be updated.
$attrs
is a plain array with string attribute names, listing attributes that are updated.
$values
is a hash where key is document ID, and value is a plain array of new attribute values.
$index
can be either a single index name or a list, like in Query()
.
Unlike Query()
, wildcard is not allowed and all the indexes
to update must be specified explicitly. The list of indexes can include
distributed index names. Updates on distributed indexes will be pushed
to all agents.
The updates only work with docinfo=extern
storage strategy.
They are very fast because they're working fully in RAM, but they can also
be made persistent: updates are saved on disk on clean searchd
shutdown initiated by SIGTERM signal.
Usage example:
$cl->UpdateAttributes ( "test1", array("group_id"), array(1=>array(456)) ); $cl->UpdateAttributes ( "products", array ( "price", "amount_in_stock" ), array ( 1001=>array(123,5), 1002=>array(37,11), 1003=>(25,129) ) );
The first sample statement will update document 1 in index "test1", setting "group_id" to 456. The second one will update documents 1001, 1002 and 1003 in index "products". For document 1001, the new price will be set to 123 and the new amount in stock to 5; for document 1002, the new price will be 37 and the new amount will be 11; etc.
Prototype: function BuildKeywords ( $query, $index, $hits )
Extracts keywords from query using tokenizer settings for given index, optionally with per-keyword occurrence statistics. Returns an array of hashes with per-keyword information.
$query
is a query to extract keywords from.
$index
is a name of the index to get tokenizing settings and keyword occurrence statistics from.
$hits
is a boolean flag that indicates whether keyword occurrence statistics are required.
Usage example:
$keywords = $cl->BuildKeywords ( "this.is.my query", "test1", false );
Prototype: function EscapeString ( $string )
Escapes characters that are treated as special operators by the query language parser. Returns an escaped string.
$string
is a string to escape.
This function might seem redundant because it's trivial to implement in any calling application. However, as the set of special characters might change over time, it makes sense to have an API call that is guaranteed to escape all such characters at all times.
Usage example:
$escaped = $cl->EscapeString ( "escaping-sample@query/string" );
SphinxSE is MySQL storage engine which can be compiled into MySQL server 5.x using its pluggable architecure. It is not available for MySQL 4.x series. It also requires MySQL 5.0.22 or higher in 5.0.x series, or MySQL 5.1.12 or higher in 5.1.x series.
Despite the name, SphinxSE does not
actually store any data itself. It is actually a built-in client
which allows MySQL server to talk to searchd
,
run search queries, and obtain search results. All indexing and
searching happen outside MySQL.
Obvious SphinxSE applications include:
You will need to obtain a copy of MySQL sources, prepare those, and then recompile MySQL binary. MySQL sources (mysql-5.x.yy.tar.gz) could be obtained from dev.mysql.com Web site.
For some MySQL versions, there are delta tarballs with already prepared source versions available from Sphinx Web site. After unzipping those over original sources MySQL would be ready to be configured and built with Sphinx support.
If such tarball is not available, or does not work for you for any reason, you would have to prepare sources manually. You will need to GNU Autotools framework (autoconf, automake and libtool) installed to do that.
Skips steps 1-3 if using already prepared delta tarball.
copy sphinx.5.0.yy.diff
patch file
into MySQL sources directory and run
patch -p1 < sphinx.5.0.yy.diff
If there's no .diff file exactly for the specific version you need to build, try applying .diff with closest version numbers. It is important that the patch should apply with no rejects.
sh BUILD/autorun.sh
sql/sphinx
directory in and copy all files in mysqlse
directory
from Sphinx sources there. Example:
cp -R /root/builds/sphinx-0.9.7/mysqlse /root/builds/mysql-5.0.24/sql/sphinx
./configure --with-sphinx-storage-engine
make make install
Skip steps 1-2 if using already prepared delta tarball.
storage/sphinx
directory in and copy all files in mysqlse
directory
from Sphinx sources there. Example:
cp -R /root/builds/sphinx-0.9.7/mysqlse /root/builds/mysql-5.1.14/storage/sphinx
sh BUILD/autorun.sh
./configure --with-plugins=sphinx
make make install
SHOW ENGINES
query. You should see a list
of all available engines. Sphinx should be present and "Support"
column should contain "YES":
mysql> show engines; +------------+----------+----------------------------------------------------------------+ | Engine | Support | Comment | +------------+----------+----------------------------------------------------------------+ | MyISAM | DEFAULT | Default engine as of MySQL 3.23 with great performance | ... | SPHINX | YES | Sphinx storage engine | ... +------------+----------+----------------------------------------------------------------+ 13 rows in set (0.00 sec)
To search via SphinxSE, you would need to create special ENGINE=SPHINX "search table", and then SELECT from it with full text query put into WHERE clause for query column.
Let's begin with an example create statement and search query:
CREATE TABLE t1 ( id INTEGER NOT NULL, weight INTEGER NOT NULL, query VARCHAR(3072) NOT NULL, group_id INTEGER, INDEX(query) ) ENGINE=SPHINX CONNECTION="sphinx://localhost:3312/test"; SELECT * FROM t1 WHERE query='test it;mode=any';
First 3 columns of search table must be INTEGER
,
INTEGER
and VARCHAR
which will be mapped to document ID,
match weight and search query accordingly. Query column must be indexed;
all the others must be kept unindexed. Columns' names are ignored so you
can use arbitrary ones.
Additional columns must be either INTEGER
or TIMESTAMP
.
They will be bound to attributes provided in Sphinx result set by name, so their
names must match attribute names specified in sphinx.conf
.
If there's no such attribute name in Sphinx search results, column will have
NULL
values.
Special "virtual" attributes names can also be bound to SphinxSE columns.
_sph_
needs to be used instead of @
for that.
For instance, to obtain @group
and @count
virtual attributes, use _sph_group
and _sph_count
column names.
CONNECTION
string parameter can be used to specify default
searchd host, port and indexes for queries issued using this table.
If no connection string is specified in CREATE TABLE
,
index name "*" (ie. search all indexes) and localhost:3312 are assumed.
Connection string syntax is as follows:
CONNECTION="sphinx://HOST:PORT/INDEXNAME"
You can change the default connection string later:
ALTER TABLE t1 CONNECTION="sphinx://NEWHOST:NEWPORT/NEWINDEXNAME";
You can also override all these parameters per-query.
As seen in example, both query text and search options should be put into WHERE clause on search query column (ie. 3rd column); the options are separated by semicolons; and their names from values by equality sign. Any number of options can be specified. Available options are:
... WHERE query='test;sort=attr_asc:group_id'; ... WHERE query='test;sort=extended:@weight desc, group_id asc';
... WHERE query='test;index=test1;'; ... WHERE query='test;index=test1,test2,test3;';
... WHERE query='test;weights=1,2,3;';
# only include groups 1, 5 and 19 ... WHERE query='test;filter=group_id,1,5,19;'; # exclude groups 3 and 11 ... WHERE query='test;!filter=group_id,3,11;';
# include groups from 3 to 7, inclusive ... WHERE query='test;range=group_id,3,7;'; # exclude groups from 5 to 25 ... WHERE query='test;!range=group_id,5,25;';
... WHERE query='test;maxmatches=2000;';
... WHERE query='test;groupby=day:published_ts;'; ... WHERE query='test;groupby=attr:group_id;';
... WHERE query='test;groupsort=@count desc;';
... WHERE query='test;indexweights=idx_exact,2,idx_stemmed,1;';
One very important note that it is much more efficient to allow Sphinx to perform sorting, filtering and slicing the result set than to raise max matches count and use WHERE, ORDER BY and LIMIT clauses on MySQL side. This is for two reasons. First, Sphinx does a number of optimizations and performs better than MySQL on these tasks. Second, less data would need to be packed by searchd, transferred and unpacked by SphinxSE.
Additional query info besides result set could be
retrieved with SHOW ENGINE SPHINX STATUS
statement:
mysql> SHOW ENGINE SPHINX STATUS; +--------+-------+-------------------------------------------------+ | Type | Name | Status | +--------+-------+-------------------------------------------------+ | SPHINX | stats | total: 25, total found: 25, time: 126, words: 2 | | SPHINX | words | sphinx:591:1256 soft:11076:15945 | +--------+-------+-------------------------------------------------+ 2 rows in set (0.00 sec)
You could perform JOINs on SphinxSE search table and tables using other engines. Here's an example with "documents" from example.sql:
mysql> SELECT content, date_added FROM test.documents docs -> JOIN t1 ON (docs.id=t1.id) -> WHERE query="one document;mode=any"; +-------------------------------------+---------------------+ | content | docdate | +-------------------------------------+---------------------+ | this is my test document number two | 2006-06-17 14:04:28 | | this is my test document number one | 2006-06-17 14:04:28 | +-------------------------------------+---------------------+ 2 rows in set (0.00 sec) mysql> SHOW ENGINE SPHINX STATUS; +--------+-------+---------------------------------------------+ | Type | Name | Status | +--------+-------+---------------------------------------------+ | SPHINX | stats | total: 2, total found: 2, time: 0, words: 2 | | SPHINX | words | one:1:2 document:2:2 | +--------+-------+---------------------------------------------+ 2 rows in set (0.00 sec)
Unfortunately, Sphinx is not yet 100% bug free (even though I'm working hard towards that), so you might occasionally run into some issues.
Reporting as much as possible about each bug is very important - because to fix it, I need to be able either to reproduce and debug the bug, or to deduce what's causing it from the information that you provide. So here are some instructions on how to do that.
If Sphinx fails to build for some reason, please do the following:
mysql-devel
package is present);
mysql_config gcc --version uname -a
configure
or gcc
(it should be to include error message itself only,
not the whole build log).
If Sphinx builds and runs, but there are any problems running it, please do the following:
mysql --version gcc --version uname -a
make distclean ./configure --with-debug make install killall -TERM searchd
searchd
, include
relevant entries from searchd.log
and
query.log
in your bug report;
searchd
, try
running it in console mode and check if it dies with an assertion:
./searchd --console
If any program dies with an assertion, crashes without an assertion or hangs up, you would additionally need to generate a core dump and examine it.
ulimit
:
ulimit -c 32768
kill -SEGV
from another console to force it to exit and dump core:
kill -SEGV HANGED-PROCESS-ID
gdb
to examine the core file
and obtain a backtrace:
gdb ./CRASHED-PROGRAM-FILE-NAME CORE-DUMP-FILE-NAME (gdb) bt (gdb) quit
Note that HANGED-PROCESS-ID, CRASHED-PROGRAM-FILE-NAME and CORE-DUMP-FILE-NAME must all be replaced with specific numbers and file names. For example, hanged searchd debugging session would look like:
# kill -SEGV 12345 # ls *core* core.12345 # gdb ./searchd core.12345 (gdb) bt ... (gdb) quit
Note that ulimit
is not server-wide
and only affects current shell session. This means that you will not
have to restore any server-wide limits - but if you relogin,
you will have to set ulimit
again.
Core dumps should be placed in current working directory (and Sphinx programs do not change it), so this is where you would look for them.
Please do not immediately remove the core file because there could be additional helpful information which could be retrieved from it. You do not need to send me this file (as the debug info there is closely tied to your system) but I might need to ask you a few additional questions about it.
Data source type.
Mandatory, no default value.
Known types are mysql
, pgsql
, xmlpipe
and xmlpipe2
.
All other per-source options depend on source type selected by this option. Names of the options used for SQL sources (ie. MySQL and PostgreSQL) start with "sql_"; names of the ones used for xmlpipe and xmlpipe2 start with "xmlpipe_".
type = mysql
SQL server host to connect to.
Mandatory, no default value.
Applies to SQL source types (mysql
and pgsql
) only.
In the simplest case when Sphinx resides on the same host with your MySQL or PostgreSQL installation, you would simply specify "localhost". Note that MySQL client library chooses whether to connect over TCP/IP or over UNIX socket based on the host name. Generally speaking, "localhost" will force it to use UNIX socket (this is the default and generally recommended mode) and "127.0.0.1" will force TCP/IP usage. Refer to MySQL manual for more details.
sql_host = localhost
SQL server IP port to connect to.
Optional, default is 3306 for mysql
source type and 5432 for pgsql
type.
Applies to SQL source types (mysql
and pgsql
) only.
Note that it depends on sql_host setting whether this value will actually be used.
sql_port = 3306
SQL user to use when connecting to sql_host.
Mandatory, no default value.
Applies to SQL source types (mysql
and pgsql
) only.
sql_user = test
SQL user password to use when connecting to sql_host.
Mandatory, no default value.
Applies to SQL source types (mysql
and pgsql
) only.
sql_pass = mysecretpassword
SQL database (in MySQL terms) to use after the connection and perform further queries within.
Mandatory, no default value.
Applies to SQL source types (mysql
and pgsql
) only.
sql_db = test
UNIX socket name to connect to for local SQL servers.
Optional, default value is empty (use client library default settings).
Applies to SQL source types (mysql
and pgsql
) only.
On Linux, it would typically be /var/lib/mysql/mysql.sock
.
On FreeBSD, it would typically be /tmp/mysql.sock
.
Note that it depends on sql_host setting whether this value will actually be used.
sql_sock = /tmp/mysql.sock
MySQL client connection flags.
Optional, default value is 0 (do not set any flags).
Applies to mysql
source type only.
This option must contain an integer value with the sum of the flags. The value will be passed to mysql_real_connect() verbatim. The flags are enumerated in mysql_com.h include file. Flags that are especially interesting in regard to indexing, with their respective values, are as follows:
For instance, you can specify 2080 (2048+32) to use both compression and SSL,
or 32768 to use new authentication only. Initially, this option was introduced
to be able to use compression when the indexer
and mysqld
are on different hosts. Compression on 1 Gbps
links is most likely to hurt indexing time though it reduces network traffic,
both in theory and in practice. However, enabling compression on 100 Mbps links
may improve indexing time significantly (upto 20-30% of the total indexing time
improvement was reported). Your mileage may vary.
mysql_connect_flags = 32 # enable compression
Pre-fetch query, or pre-query.
Multi-value, optional, default is empty list of queries.
Applies to SQL source types (mysql
and pgsql
) only.
Multi-value means that you can specify several pre-queries. They are executed before the main fetch query, and they will be exectued exactly in order of appeareance in the configuration file. Pre-query results are ignored.
Pre-queries are useful in a lot of ways. They are used to setup encoding, mark records that are going to be indexed, update internal counters, set various per-connection SQL server options and variables, and so on.
Perhaps the most frequent pre-query usage is to specify the encoding that the server will use for the rows it returnes. It must match the encoding that Sphinx expects (as specified by charset_type and charset_table options). Two MySQL specific examples of setting the encoding are:
sql_query_pre = SET CHARACTER_SET_RESULTS=cp1251 sql_query_pre = SET NAMES utf8
Also specific to MySQL sources, it is useful to disable query cache (for indexer connection only) in pre-query, because indexing queries are not going to be re-run frequently anyway, and there's no sense in caching their results. That could be achieved with:
sql_query_pre = SET SESSION query_cache_type=OFF
sql_query_pre = SET NAMES utf8 sql_query_pre = SET SESSION query_cache_type=OFF
Main document fetch query.
Mandatory, no default value.
Applies to SQL source types (mysql
and pgsql
) only.
There can be only one main query. This is the query which is used to retrieve documents from SQL server. You can specify up to 32 full-text fields (formally, upto SPH_MAX_FIELDS from sphinx.h), and an arbitrary amount of attributes. All of the columns that are neither document ID (the first one) nor attributes will be full-text indexed.
Document ID MUST be the very first field,
and it MUST BE UNIQUE UNSIGNED POSITIVE (NON-ZERO, NON-NEGATIVE) INTEGER NUMBER.
It can be either 32-bit or 64-bit, depending on how you built Sphinx;
by default it builds with 32-bit IDs support but --enable-id64
option
to configure
allows to build with 64-bit document and word IDs support.
sql_query = \ SELECT id, group_id, UNIX_TIMESTAMP(date_added) AS date_added, \ title, content \ FROM documents
Range query setup.
Optional, default is empty.
Applies to SQL source types (mysql
and pgsql
) only.
Setting this option enables ranged document fetch queries (see Section 3.7, “Ranged queries”). Ranged queries are useful to avoid notorious MyISAM table locks when indexing lots of data. (They also help with other less notorious issues, such as reduced performance caused by big result sets, or additional resources consumed by InnoDB to serialize big read transactions.)
The query specified in this option must fetch min and max document IDs that will be used as range boundaries. It must return exactly two integer fields, min ID first and max ID second; the field names are ignored.
When ranged queries are enabled, sql_query
will be required to contain $start
and $end
macros
(because it obviously would be a mistake to index the whole table many times over).
Note that the intervals specified by $start
..$end
will not overlap, so you should not remove document IDs that are
exactly equal to $start
or $end
from your query.
The example in Section 3.7, “Ranged queries”) illustrates that; note how it
uses greater-or-equal and less-or-equal comparisons.
sql_query_range = SELECT MIN(id),MAX(id) FROM documents
Range query step.
Optional, default is 1024.
Applies to SQL source types (mysql
and pgsql
) only.
Only used when ranged queries are enabled. The full document IDs interval fetched by sql_query_range will be walked in this big steps. For example, if min and max IDs fetched are 12 and 3456 respectively, and the step is 1000, indexer will call sql_query several times with the following substitutions:
sql_range_step = 1000
Unsigned integer attribute declaration.
Multi-value (there might be multiple attributes declared), optional.
Applies to SQL source types (mysql
and pgsql
) only.
The column value should fit into 32-bit unsigned integer range. Values outside this range will be accepted but wrapped around. For instance, -1 will be wrapped around to 2^32-1 or 4,294,967,295.
You can specify bit count for integer attributes by appending
':BITCOUNT' to attribute name (see example below). Attributes with
less than default 32-bit size, or bitfields, perform slower.
But they require less RAM when using extern storage:
such bitfields are packed together in 32-bit chunks in .spa
attribute data file. Bit size settings are ignored if using
inline storage.
sql_attr_uint = group_id sql_attr_uint = forum_id:9 # 9 bits for forum_id
Boolean attribute declaration.
Multi-value (there might be multiple attributes declared), optional.
Applies to SQL source types (mysql
and pgsql
) only.
Equivalent to sql_attr_uint declaration with a bit count of 1.
sql_attr_bool = is_deleted # will be packed to 1 bit
UNIX timestamp attribute declaration.
Multi-value (there might be multiple attributes declared), optional.
Applies to SQL source types (mysql
and pgsql
) only.
The column value should be a timestamp in UNIX format, ie. 32-bit unsigned integer number of seconds elapsed since midnight, January 01, 1970, GMT. Timestamps are internally stored and handled as integers everywhere. But in addition to working with timestamps as integers, it's also legal to use them along with different date-based functions - such as time segments sorting mode, or day/week/month/year extraction for GROUP BY. Note that DATE or DATETIME column types in MySQL can not be directly used as timestamps; you need to explicitly convert such columns using UNIX_TIMESTAMP function.
sql_attr_timestamp = UNIX_TIMESTAMP(added_datetime) AS added_ts
Ordinal string number attribute declaration.
Multi-value (there might be multiple attributes declared), optional.
Applies to SQL source types (mysql
and pgsql
) only.
This attribute type (so-called ordinal, for brevity) is intended to allow sorting by string values, but without storing the strings themselves. When indexing ordinals, string values are fetched from database, temporarily stored, sorted, and then replaced by their respective ordinal numbers in the array of sorted strings. So, the ordinal number is an integer such that sorting by it produces the same result as if lexicographically sorting by original strings. by string values lexicographically.
Earlier versions could consume a lot of RAM for indexing ordinals. Starting with revision r1112, ordinals accumulation and sorting also runs in fixed memory (at the cost of using additional temporary disk space), and honors mem_limit settings.
Ideally the strings should be sorted differently, depending on the encoding and locale. For instance, if the strings are known to be Russian text in KOI8R encoding, sorting the bytes 0xE0, 0xE1, and 0xE2 should produce 0xE1, 0xE2 and 0xE0, because in KOI8R value 0xE0 encodes a character that is (noticeably) after characters encoded by 0xE1 and 0xE2. Unfortunately, Sphinx does not support that at the moment and will simply sort the strings bytewise.
Note that the ordinals are by construction local to each index, and it's therefore impossible to merge ordinals while retaining the proper order. The processed strings are replaced by their sequential number in the index they occurred in, but different indexes have different sets of strings. For instance, if 'main' index contains strings "aaa", "bbb", "ccc", and so on up to "zzz", they'll be assigned numbers 1, 2, 3, and so on up to 26, respectively. But then if 'delta' only contains "zzz" the assigned number will be 1. And after the merge, the order will be broken. Unfortunately, this is impossible to workaround without storing the original strings (and once Sphinx supports storing the original strings, ordinals will not be necessary any more).
sql_attr_str2ordinal = author_name
Floating point attribute declaration.
Multi-value (there might be multiple attributes declared), optional.
Applies to SQL source types (mysql
and pgsql
) only.
The values will be stored in single precision, 32-bit IEEE 754 format. Represented range is approximately from 1e-38 to 1e+38. The amount of decimal digits that can be stored precisely is approximately 7. One important usage of the float attributes is storing latitude and longitude values (in radians), for further usage in query-time geosphere distance calculations.
sql_attr_float = lat_radians sql_attr_float = long_radians
Multi-valued attribute (MVA) declaration.
Multi-value (ie. there may be more than one such attribute declared), optional.
Applies to SQL source types (mysql
and pgsql
) only.
Plain attributes only allow to attach 1 value per each document. However, there are cases (such as tags or categories) when it is desired to attach multiple values of the same attribute and be able to apply filtering or grouping to value lists.
The declaration format is as follows (backslashes are for clarity only; everything can be declared in a single line as well):
sql_attr_multi = ATTR-TYPE ATTR-NAME 'from' SOURCE-TYPE \ [;QUERY] \ [;RANGE-QUERY]
where
sql_attr_multi = uint tag from query; SELECT id, tag FROM tags sql_attr_multi = uint tag from ranged-query; \ SELECT id, tag FROM tags WHERE id>=$start AND id<=$end; \ SELECT MIN(id), MAX(id) FROM tags
Post-fetch query.
Optional, default value is empty.
Applies to SQL source types (mysql
and pgsql
) only.
This query is executed immediately after sql_query completes successfully. When post-fetch query produces errors, they are reported as warnings, but indexing is not terminated. It's result set is ignored. Note that indexing is not yet completed at the point when this query gets executed, and further indexing still may fail. Therefore, any permanent updates should not be done from here. For instance, updates on helper table that permanently change the last successfully indexed ID should not be run from post-fetch query; they should be run from post-index query instead.
sql_query_post = DROP TABLE my_tmp_table
Post-index query.
Optional, default value is empty.
Applies to SQL source types (mysql
and pgsql
) only.
This query is executed when indexing is fully and succesfully completed.
If this query produces errors, they are reported as warnings,
but indexing is not terminated. It's result set is ignored.
$maxid
macro can be used in its text; it will be
expanded to maximum document ID which was actually fetched
from the database during indexing.
sql_query_post_index = REPLACE INTO counters ( id, val ) \ VALUES ( 'max_indexed_id', $maxid )
Ranged query throttling period, in milliseconds.
Optional, default is 0 (no throttling).
Applies to SQL source types (mysql
and pgsql
) only.
Throttling can be useful when indexer imposes too much load on the database server. It causes the indexer to sleep for given amount of milliseconds once per each ranged query step. This sleep is unconditional, and is performed before the fetch query.
sql_ranged_throttle = 1000 # sleep for 1 sec before each query step
Document info query.
Optional, default is empty.
Applies to mysql
source type only.
Only used by CLI search to fetch and display document information,
only works with MySQL at the moment, and only intended for debugging purposes.
This query fetches the row that will be displayed by CLI search utility
for each document ID. It is required to contain $id
macro
that expands to the queried document ID.
sql_query_info = SELECT * FROM documents WHERE id=$id
Shell command that invokes xmlpipe stream producer.
Mandatory.
Applies to xmlpipe
and xmlpipe2
source types only.
Specifies a command that will be executed and which output will be parsed for documents. Refer to Section 3.8, “xmlpipe data source” or Section 3.9, “xmlpipe2 data source” for specific format description.
xmlpipe_command = cat /home/sphinx/test.xml
xmlpipe field declaration.
Multi-value, optional.
Applies to xmlpipe2
source type only. Refer to Section 3.9, “xmlpipe2 data source”.
xmlpipe_field = subject xmlpipe_field = content
xmlpipe integer attribute declaration.
Multi-value, optional.
Applies to xmlpipe2
source type only.
Syntax fully matches that of sql_attr_uint.
xmlpipe_attr_uint = author
xmlpipe boolean attribute declaration.
Multi-value, optional.
Applies to xmlpipe2
source type only.
Syntax fully matches that of sql_attr_bool.
xmlpipe_attr_bool = is_deleted # will be packed to 1 bit
xmlpipe UNIX timestamp attribute declaration.
Multi-value, optional.
Applies to xmlpipe2
source type only.
Syntax fully matches that of sql_attr_timestamp.
xmlpipe_attr_timestamp = published
xmlpipe string ordinal attribute declaration.
Multi-value, optional.
Applies to xmlpipe2
source type only.
Syntax fully matches that of sql_attr_str2ordinal.
xmlpipe_attr_str2ordinal = author_sort
xmlpipe floating point attribute declaration.
Multi-value, optional.
Applies to xmlpipe2
source type only.
Syntax fully matches that of sql_attr_float.
xmlpipe_attr_float = lat_radians xmlpipe_attr_float = long_radians
xmlpipe MVA attribute declaration.
Multi-value, optional.
Applies to xmlpipe2
source type only.
This setting declares an MVA attribute tag in xmlpipe2 stream. The contents of the specified tag will be parsed and a list of integers that will constitute the MVA will be extracted, similar to how sql_attr_multi parses SQL column contents when 'field' MVA source type is specified.
xmlpipe_attr_multi = taglist
Index type. Optional, default is empty (index is plain local index). Known values are empty string or 'distributed'.
Sphinx supports two different types of indexes: local, that are stored
and processed on the local machine; and distributed, that involve not only
local searching but querying remote searchd
instances
over the network as well. Index type settings lets you choose this type.
By default, indexes are local. Specifying 'distributed' for type enables
distributed searching, see Section 4.7, “Distributed searching”.
type = distributed
Adds document source to local index. Multi-value, mandatory.
Specifies document source to get documents from when the current index is indexed. There must be at least one source. There may be multiple sources, without any restrictions on the source types: ie. you can pull part of the data from MySQL server, part from PostgreSQL, part from the filesystem using xmlpipe2 wrapper.
However, there are some restrictions on the source data. First, document IDs must be globally unique across all sources. If that condition is not met, you might get unexpected search results. Second, source schemas must be the same in order to be stored within the same index.
No source ID is stored automatically. Therefore, in order to be able to tell what source the matched document came from, you will need to store some additional information yourself. Two typical approaches include:
source src1 { sql_query = SELECT id*10+1, ... FROM table1 ... } source src2 { sql_query = SELECT id*10+2, ... FROM table2 ... }
source src1 { sql_query = SELECT id, 1 AS source_id FROM table1 sql_attr_uint = source_id ... } source src2 { sql_query = SELECT id, 2 AS source_id FROM table2 sql_attr_uint = source_id ... }
source = srcpart1 source = srcpart2 source = srcpart3
Index files path and file name (without extension). Mandatory.
Path specifies both directory and file name, but without extension.
indexer
will append different extensions
to this path when generating final names for both permanent and
temporary index files. Permanent data files have several different
extensions starting with '.sp'; temporary files' extensions
start with '.tmp'. It's safe to remove .tmp*
files is if indexer fails to remove them automatically.
For reference, different index files store the following data:
.spa
stores document attributes (used in extern docinfo storage mode only);.spd
stores matching document ID lists for each word ID;.sph
stores index header information;.spi
stores word lists (word IDs and pointers to .spd
file);.spm
stores MVA data;.spp
stores hit (aka posting, aka word occurence) lists for each word ID.
path = /var/data/test1
Document attribute values (docinfo) storage mode. Optional, default is 'extern'. Known values are 'none', 'extern' and 'inline'.
Docinfo storage mode defines how exactly docinfo will be
physically stored on disk and RAM. "none" means that there will be
no docinfo at all (ie. no attributes). Normally you need not to set
"none" explicitly because Sphinx will automatically select "none"
when there are no attributes configured. "inline" means that the
docinfo will be stored in the .spd
file,
along with the document ID lists. "extern" means that the docinfo
will be stored separately (externally) from document ID lists,
in a special .spa
file.
Basically, externally stored docinfo must be kept in RAM when querying. for performance reasons. So in some cases "inline" might be the only option. However, such cases are infrequent, and docinfo defaults to "extern". Refer to Section 3.2, “Attributes” for in-depth discussion and RAM usage estimates.
docinfo = inline
Memory locking for cached data. Optional, default is 0 (do not call mlock()).
For search performance, searchd
preloads
a copy of .spa
and .spi
files in RAM, and keeps that copy in RAM at all times. But if there
are no searches on the index for some time, there are no accesses
to that cached copy, and OS might decide to swap it out to disk.
First queries to such "cooled down" index will cause swap-in
and their latency will suffer.
Setting mlock option to 1 makes Sphinx lock physical RAM used
for that cached data using mlock(2) system call, and that prevents
swapping (see man 2 mlock for details). mlock(2) is a privileged call,
so it will require searchd
to be either run
from root account, or be granted enough privileges otherwise.
If mlock() fails, a warning is emitted, but index continues
working.
mlock = 1
A list of morphology preprocessors to apply. Optional, default is empty (do not apply any preprocessor).
Morphology preprocessors can be applied to the words being indexed to replace different forms of the same word with the base, normalized form. For instance, English stemmer will normalize both "dogs" and "dog" to "dog", making search results for both searches the same.
Built-in preprocessors include English stemmer, Russian stemmer
(that supports UTF-8 and Windows-1251 encodings), Soundex,
and Metaphone. The latter two replace the words with special
phonetic codes that are equal is words are phonetically close.
Additional stemmers provided by Snowball
project libstemmer library
can be enabled at compile time using --with-libstemmer
configure
option.
Built-in English and Russian stemmers should be faster than their
libstemmer counterparts, but can produce slightly different results,
because they are based on an older version. Metaphone implementation
is based on Double Metaphone algorithm and indexes the primary code.
Built-in values that be used in morphology
option are:
'none', 'stem_en', 'stem_ru', 'stem_enru', 'soundex', and 'metaphone'.
Additional values provided by libstemmer are in 'libstemmer_XXX' format,
where XXX is libstemmer algorithm codename (refer to
libstemmer_c/libstemmer/modules.txt
for a complete list).
Several stemmers can be specified (comma-separated). They will be applied to incoming words in the order they are listed, and the processing will stop once one of the stemmers actually modifies the word. Also when wordforms feature is enabled the word will be looked up in word forms dictionary first, and if there is a matching entry in the dictionary, stemmers will not be applied at all. Or in other words, wordforms can be used to implement stemming exceptions.
morphology = stem_en, libstemmer_sv
Stopword files list (space separated). Optional, default is empty.
Stopwords are the words that will not be indexed. Typically you'd put most frequent words in the stopwords list because they do not add much value to search results but consume a lot of resources to process.
You can specify several file names, separated by spaces. All the files will be loaded. Stopwords file format is simple plain text. The encoding must match index encoding specified in charset_type. File data will be tokenized with respect to charset_table settings, so you can use the same separators as in the indexed data. The stemmers will also be applied when parsing stopwords file.
While stopwords are not indexed, they still do affect the keyword positions. For instance, assume that "the" is a stopword, that document 1 contains the line "in office", and that document 2 contains "in the office". Searching for "in office" as for exact phrase will only return the first document, as expected, even though "the" in the second one is stopped.
stopwords = /usr/local/sphinx/data/stopwords.txt stopwords = stopwords-ru.txt stopwords-en.txt
Word forms dictionary. Optional, default is empty.
Word forms are applied after tokenizing the incoming text by charset_table rules. They essentialy let you replace one word with another. Normally, that would be used to bring different word forms to a single normal form (eg. to normalize all the variants such as "walks", "walked", "walking" to the normal form "walk"). It can also be used to implement stemming exceptions, because stemming is not applied to words found in the forms list.
Dictionaries are used to normalize incoming words both during indexing
and searching. Therefore, to pick up changes in wordforms file
it's required to reindex and restart searchd
.
Word forms support in Sphinx is designed to support big dictionaries well.
They moderately affect indexing speed: for instance, a dictionary with 1 million
entries slows down indexing about 1.5 times. Searching speed is not affected at all.
Additional RAM impact is roughly equal to the dictionary file size,
and dictionaries are shared across indexes: ie. if the very same 50 MB wordforms
file is specified for 10 different indexes, additional searchd
RAM usage will be about 50 MB.
Dictionary file should be in a simple plain text format. Each line should contain source and destination word forms, in exactly the same encoding as specified in charset_type, separated by "greater" sign. Rules from the charset_table will be applied when the file is loaded. So basically it's as case sensitive as your other full-text indexed data, ie. typically case insensitive. Here's the file contents sample:
walks > walk walked > walk walking > walk
There is bundled spelldump
utility that
helps you create a dictionary file in the format Sphinx can read
from source .dict
and .aff
dictionary files in ispell
format.
wordforms = /usr/local/sphinx/data/wordforms.txt
Tokenizing exceptions file. Optional, default is empty.
Exceptions allow to map one or more tokens (including tokens with characters that would normally be excluded) to a single keyword. They are similar to wordforms in that they also perform mapping, but have a number of important differences.
Short summary of the differences is as follows:
The expected file format is also plain text, with one line per exception, and the line format is as follows:
map-from-tokens => map-to-token
Example file:
AT & T => AT&T AT&T => AT&T Standarten Fuehrer => standartenfuhrer Standarten Fuhrer => standartenfuhrer MS Windows => ms windows Microsoft Windows => ms windows C++ => cplusplus c++ => cplusplus C plus plus => cplusplus
All tokens here are case sensitive: they will not be processed by charset_table rules. Thus, with the example exceptions file above, "At&t" text will be tokenized as two keywords "at" and "t", because of lowercase letters. On the other hand, "AT&T" will match exactly and produce single "AT&T" keyword.
Note that this map-to keyword is a) always interpereted as a single word, and b) is both case and space sensitive! In our sample, "ms windows" query will not match the document with "MS Windows" text. The query will be interpreted as a query for two keywords, "ms" and "windows". And what "MS Windows" gets mapped to is a single keyword "ms windows", with a space in the middle. On the other hand, "standartenfuhrer" will retrieve documents with "Standarten Fuhrer" or "Standarten Fuehrer" contents (capitalized exactly like this), or any capitalization variant of the keyword itself, eg. "staNdarTenfUhreR". (It won't catch "standarten fuhrer", however: this text does not match any of the listed exceptions because of case sensitivity, and gets indexed as two separate keywords.)
Whitespace in the map-from tokens list matters, but its amount does not. Any amount of the whitespace in the map-form list will match any other amount of whitespace in the indexed document or query. For instance, "AT & T" map-from token will match "AT & T" text, whatever the amount of space in both map-from part and the indexed text. Such text will therefore be indexed as a special "AT&T" keyword, thanks to the very first entry from the sample.
Exceptions also allow to capture special characters (that are exceptions from general charset_table rules; hence the name). Assume that you generally do not want to treat '+' as a valid character, but still want to be able search for some exceptions from this rule such as 'C++'. The sample above will do just that, totally independent of what characters are in the table and what are not.
Exceptions are applied to raw incoming document and query data
during indexing and searching respectively. Therefore, to pick up
changes in the file it's required to reindex and restart
searchd
.
exceptions = /usr/local/sphinx/data/exceptions.txt
Minimum indexed word length. Optional, default is 1 (index everything).
Only those words that are not shorter than this minimum will be indexed. For instance, if min_word_len is 4, then 'the' won't be indexed, but 'they' will be.
min_word_len = 4
Character set encoding type. Optional, default is 'sbcs'. Known values are 'sbcs' and 'utf-8'.
Different encodings have different methods for mapping their internal characters codes into specific byte sequences. Two most common methods in use today are single-byte encoding and UTF-8. Their corresponding charset_type values are 'sbcs' (stands for Single Byte Character Set) and 'utf-8'. The selected encoding type will be used everywhere where the index is used: when indexing the data, when parsing the query against this index, when generating snippets, etc.
Note that while 'utf-8' implies that the decoded values must be treated as Unicode codepoint numbers, there's a family of 'sbcs' encodings that may in turn treat different byte values differently, and that should be properly reflected in your charset_table settings. For example, the same byte value of 224 (0xE0 hex) maps to different Russian letters depending on whether koi-8r or windows-1251 encoding is used.
charset_type = utf-8
Accepted characters table, with case folding rules. Optional, default value depends on charset_type value.
charset_table is the main workhorse of Sphinx tokenizing process, ie. the process of extracting keywords from document text or query txet. It controls what characters are accepted as valid and what are not, and how the accepted characters should be transformed (eg. should the case be removed or not).
You can think of charset_table as of a big table that has a mapping for each and every of 100K+ characters in Unicode (or as of a small 256-character table if you're using SBCS). By default, every character maps to 0, which means that it does not occur within keywords and should be treated as a separator. Once mentioned in the table, character is mapped to some other character (most frequently, either to itself or to a lowercase letter), and is treated as a valid keyword part.
The expected value format is a commas-separated list of mappings. Two simplest mappings simply declare a character as valid, and map a single character to another single character, respectively. But specifying the whole table in such form would result in bloated and barely manageable specifications. So there are several syntax shortcuts that let you map ranges of characters at once. The complete list is as follows:
Control characters with codes from 0 to 31 are always treated as separators. Characters with codes 32 to 127, ie. 7-bit ASCII characters, can be used in the mappings as is. To avoid configuration file encoding issues, 8-bit ASCII characters and Unicode characters must be specified in U+xxx form, where 'xxx' is hexadecimal codepoint number. This form can also be used for 7-bit ASCII characters to encode special ones: eg. use U+20 to encode space, U+2E to encode dot, U+2C to encode comma.
# 'sbcs' defaults for English and Russian charset_table = 0..9, A..Z->a..z, _, a..z, \ U+A8->U+B8, U+B8, U+C0..U+DF->U+E0..U+FF, U+E0..U+FF # 'utf-8' defaults for English and Russian charset_table = 0..9, A..Z->a..z, _, a..z, \ U+410..U+42F->U+430..U+44F, U+430..U+44F
Ignored characters list. Optional, default is empty.
Useful in the cases when some characters, such as soft hyphenation mark (U+00AD), should be not just treated as separators but rather fully ignored. For example, if '-' is simply not in the charset_table, "abc-def" text will be indexed as "abc" and "def" keywords. On the contrary, if '-' is added to ignore_chars list, the same text will be indexed as a single "abcdef" keyword.
The syntax is the same as for charset_table, but it's only allowed to declare characters, and not allowed to map them. Also, the ignored characters must not be present in charset_table.
ignore_chars = U+AD
Minimum word prefix length to index. Optional, default is 0 (do not index prefixes).
Prefix indexing allows to implement wildcard searching by 'wordstart*' wildcards (refer to enable_star option for details on wildcard syntax). When mininum prefix length is set to a positive number, indexer will index all the possible keyword prefixes (ie. word beginnings) in addition to the keywords themselves. Too short prefixes (below the minimum allowed length) will not be indexed.
For instance, indexing a keyword "example" with min_prefix_len=3 will result in indexing "exa", "exam", "examp", "exampl" prefixes along with the word itself. Searches against such index for "exam" will match documents that contain "example" word, even if they do not contain "exam" on itself. However, indexing prefixes will make the index grow significantly (because of many more indexed keywords), and will degrade both indexing and searching times.
There's no automatic way to rank perfect word matches higher in a prefix index, but there's a number of tricks to achieve that. First, you can setup two indexes, one with prefix indexing and one without it, search through both, and use SetIndexWeights() call to combine weights. Second, you can enable star-syntax and rewrite your extended-mode queries:
# in sphinx.conf enable_star = 1 // in query $cl->Query ( "( keyword | keyword* ) other keywords" );
min_prefix_len = 3
Minimum infix prefix length to index. Optional, default is 0 (do not index infixes).
Infix indexing allows to implement wildcard searching by 'start*', '*end', and '*middle*' wildcards (refer to enable_star option for details on wildcard syntax). When mininum infix length is set to a positive number, indexer will index all the possible keyword infixes (ie. substrings) in addition to the keywords themselves. Too short infixes (below the minimum allowed length) will not be indexed.For instance, indexing a keyword "test" with min_infix_len=2 will result in indexing "te", "es", "st", "tes", "est" infixes along with the word itself. Searches against such index for "es" will match documents that contain "test" word, even if they do not contain "es" on itself. However, indexing infixes will make the index grow significantly (because of many more indexed keywords), and will degrade both indexing and searching times.
There's no automatic way to rank perfect word matches higher in an infix index, but the same tricks as with prefix indexes can be applied.
min_infix_len = 3
The list of full-text fields to limit prefix indexing to. Optional, default is empty (index all fields in prefix mode).
Because prefix indexing impacts both indexing and searching performance, it might be desired to limit it to specific full-text fields only: for instance, to provide prefix searching through URLs, but not through page contents. prefix_fields specifies what fields will be prefix-indexed; all other fields will be indexed in normal mode. The value format is a comma-separated list of field names.
prefix_fields = url, domain
The list of full-text fields to limit infix indexing to. Optional, default is empty (index all fields in infix mode).
Similar to prefix_fields, but lets you limit infix-indexing to given fields.
infix_fields = url, domain
Enables star-syntax (or wildcard syntax) when searching through prefix/infix indexes. Optional, default is is 0 (do not use wildcard syntax), for compatibility with 0.9.7. Known values are 0 and 1.
This feature enables "star-syntax", or wildcard syntax, when searching
through indexes which were created with prefix or infix indexing enabled.
It only affects searching; so it can be changed without reindexing
by simply restarting searchd
.
The default value is 0, that means to disable star-syntax and treat all keywords as prefixes or infixes respectively, depending on indexing-time min_prefix_len and min_infix_len settings. The value of 1 means that star ('*') can be used at the start and/or the end of the keyword. The star will match zero or more characters.
For example, assume that the index was built with infixes and that enable_star is 1. Searching should work as follows:
enable_star = 1
N-gram lengths for N-gram indexing. Optional, default is 0 (disable n-gram indexing). Known values are 0 and 1 (other lengths to be implemented).
N-grams provide basic CJK (Chinese, Japanse, Koreasn) support for unsegmented texts. The issue with CJK searching is that there could be no clear separators between the words. Ideally, the texts would be filtered through a special program called segmenter that would insert separators in proper locations. However, segmenters are slow and error prone, and it's common to index contiguous groups of N characters, or n-grams, instead.
When this feature is enabled, streams of CJK characters are indexed as N-grams. For example, if incoming text is "ABCDEF" (where A to F represent some CJK characters) and length is 1, in will be indexed as if it was "A B C D E F". (With length equal to 2, it would produce "AB BC CD DE EF"; but only 1 is supported at the moment.) Only those characters that are listed in ngram_chars table will be split this way; other ones will not be affected.
Note that if search query is segmented, ie. there are separators between individual words, then wrapping the words in quotes and using extended mode will resut in proper matches being found even if the text was not segmented. For instance, assume that the original query is BC DEF. After wrapping in quotes on the application side, it should look like "BC" "DEF" (with quotes). This query will be passed to Sphinx and internally split into 1-grams too, resulting in "B C" "D E F" query, still with quotes that are the phrase matching operator. And it will match the text even though there were no separators in the text.
Even if the search query is not segmented, Sphinx should still produce good results, thanks to phrase based ranking: it will pull closer phrase matches (which in case of N-gram CJK words can mean closer multi-character word matches) to the top.
ngram_len = 1
N-gram characters list. Optional, default is empty.
To be used in conjunction with in ngram_len, this list defines characters, sequences of which are subject to N-gram extraction. Words comprised of other characters will not be affected by N-gram indexing feature. The value format is identical to charset_table.
ngram_chars = U+3000..U+2FA1F
Phrase boundary characters list. Optional, default is empty.
This list controls what characters will be treated as phrase boundaries, in order to adjust word positions and enable phrase-level search emulation through proximity search. The syntax is similar to charset_table. Mappings are not allowed and the boundary characters must not overlap with anything else.
On phrase boundary, additional word position increment (specified by phrase_boundary_step) will be added to current word position. This enables phrase-level searching through proximity queries: words in different phrases will be guaranteed to be more than phrase_boundary_step distance away from each other; so proximity search within that distance will be equivalent to phrase-level search.
Phrase boundary condition will be raised if and only if such character is followed by a separator; this is to avoid abbreviations such as S.T.A.L.K.E.R or URLs being treated as several phrases.
phrase_boundary = ., ?, !, U+2026 # horizontal ellipsis
Phrase boundary word position increment. Optional, default is 0.
On phrase boundary, current word position will be additionally incremented by this number. See phrase_boundary for details.
phrase_boundary_step = 100
Whether to strip HTML markup from incoming full-text data. Optional, default is 0. Known values are 0 (disable stripping) and 1 (enable stripping).
Stripping does not work with xmlpipe
source type
(it's suggested to upgrade to xmlpipe2 anyway). It should work with
properly formed HTML and XHTML, but, just as most browsers, may produce
unexpected results on malformed input (such as HTML with stray <'s
or unclosed >'s).
Only the tags themselves, and also HTML comments, are stripped. To strip the contents of the tags too (eg. to strip embedded scripts), see html_remove_elements option. There are no restrictions on tag names; ie. everything that looks like a valid tag start, or end, or a comment will be stripped.
html_strip = 1
A list of markup attributes to index when stripping HTML. Optional, default is emptu (do not index markup attributes).
Specifies HTML markup attributes whose contents should be retained and indexed even though other HTML markup is stripped. The format is per-tag enumeration of indexable attributes, as shown in the example below.
html_index_attrs = img=alt,title; a=title;
A list of HTML elements for which to strip contents along with the elements themselves. Optional, default is empty string (do not strip contents of any elements).
This feature allows to strip element contents, ie. everything that is between the opening and the closing tags. It is useful to remove embedded scripts, CSS, etc. Short tag form for empty elements (ie. <br />) is properly supported; ie. the text that follows such tag will not be removed.
The value is a comma-separated list of element (tag) names whose contents should be removed. Tag names are case insensitive.
html_remove_elements = style, script
Local index declaration in the distributed index. Multi-value, optional, default is empty.
This setting is used to declare local indexes that will be searched when
given distributed index is searched. All local indexes will be searched
sequentially, utilizing only 1 CPU or core; to parallelize processing,
you can configure searchd
to query itself (refer to
Section 8.2.27, “agent” for the details). There might be several local
indexes declared per each distributed index. Any local index can be mentioned
several times in other distributed indexes.
local = chunk1 local = chunk2
Remote agents and indexes declaration in the distributed index. Multi-value, optional, default is empty.
This setting is used to declare remote agents that will be searched
when given distributed index is searched. The agents can be thought of
as network pointers that specify host, port, and index names. In the basic
case agents would correspond to remote physical machines. More formally,
that is not always correct: you can point several agents to the
same remote machine; or you can even point agents to the very same
single instance of searchd
(in order to utilize
many CPUs or cores).
The value format is as follows:
agent = hostname:port:remote-indexes-list
where 'hostname' is remote host name; 'port' is remote TCP port; and 'remote-indexes-list' is a comma-separated list of remote index names.
All agents will be searched in parallel. However, all indexes specified for a given agent will be searched sequentially in this agent. This lets you fine-tune the configuration to the hardware. For instance, if two remote indexes are stored on the same physical HDD, it's better to configure one agent with several sequentially searched indexes to avoid HDD steping. If they are stored on different HDDs, having two agents will be advantageous, because the work will be fully parallelized. The same applies to CPUs; though CPU performance impact caused by two processes stepping on each other is somewhat smaller and frequently can be ignored at all.
On machines with many CPUs and/or HDDs, agents can be pointed
to the same machine to utilize all of the hardware in parallel
and reduce query latency. There is no need to setup several
searchd
instances for that; it's legal
to configure the instance to contact itself. Here's an example
setup, intended for a 4-CPU machine, that will use up to
4 CPUs in parallel to process each query:
index dist { type = distributed local = chunk1 agent = localhost:3312:chunk2 agent = localhost:3312:chunk3 agent = localhost:3312:chunk4 }
Note how one of the chunks is searched locally and the same instance of searchd queries itself to launch searches through three other ones in parallel.
agent = localhost:3312:chunk2 # contact itself agent = searchbox2:3312:chunk3,chunk4 # search remote indexes
Remote agent connection timeout, in milliseconds. Optional, default is 1000 (ie. 1 second).
When connecting to remote agents, searchd
will wait at most this much time for connect() call to complete
succesfully. If the timeout is reached but connect() does not complete,
and retries are enabled,
retry will be initiated.
agent_connect_timeout = 300
Remote agent query timeout, in milliseconds. Optional, default is 3000 (ie. 3 seconds).
After connection, searchd
will wait at most this
much time for remote queries to complete. This timeout is fully separate
from connection timeout; so the maximum possible delay caused by
a remote agent equals to the sum of agent_connection_timeout
and
agent_query_timeout
. Queries will not be retried
if this timeout is reached; a warning will be produced instead.
agent_query_timeout = 10000 # our query can be long, allow up to 10 sec
Whether to pre-open all index files, or open them per each query. Optional, default is 0 (do not preopen).
This option tells searchd
that it should pre-open
all index files on startup (or rotation) and keep them open while it runs.
Currently, the default mode is not to pre-open the files (this may
change in the future). Preopened indexes take a few (currently 2) file
descriptors per index. However, they save on per-query open()
calls;
and also they are invulnerable to subtle race conditions that may happen during
index rotation under high load. On the other hand, when serving many indexes
(100s to 1000s), it still might be desired to open the on per-query basis
in order to save file descriptors.
preopen = 1
Indexing RAM usage limit. Optional, default is 32M.
Enforced memory usage limit that the indexer
will not go above. Can be specified in bytes, or kilobytes
(using K postfix), or megabytes (using M postfix); see the example.
This limit will be automatically raised if set to extremely low value
causing I/O buffers to be less than 8 KB; the exact lower bound
for that depends on the indexed data size. If the buffers are
less than 256 KB, a warning will be produced.
Maximum possible limit is 2047M. Too low values can hurt
indexing speed, but 256M to 1024M should be enough for most
if not all datasets. Setting this value too high can cause
SQL server timeouts. During the document collection phase,
there will be periods when the memory buffer is partially
sorted and no communication with the database is performed;
and the database server can timeout. You can resolve that
either by raising timeouts on SQL server side or by lowering
mem_limit
.
mem_limit = 256M # mem_limit = 262144K # same, but in KB # mem_limit = 268435456 # same, but in bytes
Maximum I/O operations per second, for I/O throttling. Optional, default is 0 (unlimited).
I/O throttling related option. It limits maximum count of I/O operations (reads or writes) per any given second. A value of 0 means that no limit is imposed.
indexer
can cause bursts of intensive disk I/O during
indexing, and it might desired to limit its disk activity (and keep something
for other programs running on the same machine, such as searchd
).
I/O throttling helps to do that. It works by enforcing a minimum guaranteed
delay between subsequent disk I/O operations performed by indexer
.
Modern SATA HDDs are able to perform up to 70-100+ I/O operations per second
(that's mostly limited by disk heads seek time). Limiting indexing I/O
to a fraction of that can help reduce search performance dedgradation
caused by indexing.
max_iops = 40
Maximum allowed I/O operation size, in bytes, for I/O throttling. Optional, default is 0 (unlimited).
I/O throttling related option. It limits maximum file I/O operation
(read or write) size for all operations performed by indexer
.
A value of 0 means that no limit is imposed.
Reads or writes that are bigger than the limit
will be split in several smaller operations, and counted as several operation
by max_iops setting. At the time of this
writing, all I/O calls should be under 256 KB (default internal buffer size)
anyway, so max_iosize
values higher than 256 KB must not affect anything.
max_iosize = 1048576
Interface IP address to bind on. Optional, default is 0.0.0.0 (ie. listen on all interfaces).
address
setting lets you specify which network interface
searchd
will bind to, listen on, and accept incoming
network connections on. The default value is 0.0.0.0 which means to listen
on all interfaces. At the time, you can not specify multiple interfaces.
address = 192.168.0.1
Log file name.
Optional, default is 'searchd.log'.
All searchd
run time events will be logged in this file.
log = /var/log/searchd.log
Query log file name.
Optional, default is empty (do not log queries).
All search queries will be logged in this file. The format is described in Section 4.8, “searchd
query log format”.
query_log = /var/log/query.log
Network client request read timeout, in seconds.
Optional, default is 5 seconds.
searchd
will forcibly close the client connections which fail to send a query within this timeout.
read_timeout = 1
Maximum amount of children to fork (or in other words, concurrent searches to run in parallel). Optional, default is 0 (unlimited).
Useful to control server load. There will be no more than this much concurrent searches running, at all times. When the limit is reached, additional incoming clients are dismissed with temporarily failure (SEARCHD_RETRY) status code and a message stating that the server is maxed out.
max_children = 10
searchd
process ID file name.
Mandatory.
PID file will be re-created (and locked) on startup. It will contain
head daemon process ID while the daemon is running, and it will be unlinked
on daemon shutdown. It's mandatory because Sphinx uses it internally
for a number of things: to check whether there already is a running instance
of searchd
; to stop searchd
;
to notify it that it should rotate the indexes. Can also be used for
different external automation scripts.
pid_file = /var/run/searchd.pid
Maximum amount of matches that the daemon keeps in RAM for each index and can return to the client. Optional, default is 1000.
Introduced in order to control and limit RAM usage, max_matches
setting defines how much matches will be kept in RAM while searching each index.
Every match found will still be processed; but only
best N of them will be kept in memory and return to the client in the end.
Assume that the index contains 2,000,000 matches for the query. You rarely
(if ever) need to retrieve all of them. Rather, you need
to scan all of them, but only choose "best" at most, say, 500 by some criteria
(ie. sorted by relevance, or price, or anything else), and display those
500 matches to the end user in pages of 20 to 100 matches. And tracking
only the best 500 matches is much more RAM and CPU efficient than keeping
all 2,000,000 matches, sorting them, and then discarding everything but
the first 20 needed to display the search results page. max_matches
controls N in that "best N" amount.
This parameter noticeably affects per-query RAM and CPU usage.
Values of 1,000 to 10,000 are generally fine, but higher limits must be
used with care. Recklessly raising max_matches
to 1,000,000
means that searchd
will have to allocate and
initialize 1-million-entry matches buffer for every
query. That will obviously increase per-query RAM usage, and in some cases
can also noticeably impact performance.
CAVEAT EMPTOR! Note that there also is another place where this limit
is enforced. max_matches
can be decreased on the fly
through the corresponding API call,
and the default value in the API is also set to 1,000. So in order
to retrieve more than 1,000 matches to your application, you will have
to change the configuration file, restart searchd, and set proper limit
in SetLimits() call.
Also note that you can not set the value in the API higher than the value
in the .conf file. This is prohibited in order to have some protection
against malicious and/or malformed requests.
max_matches = 10000
Prevents searchd
stalls while rotating indexes with huge amounts of data to precache.
Optional, default is 1 (enable seamless rotation).
Indexes may contain some data that needs to be precached in RAM.
At the moment, .spa
, .spi
and
.spm
files are fully precached (they contain attribute data,
MVA data, and keyword index, respectively.)
Without seamless rotate, rotating an index tries to use as little RAM
as possible and works as follows:
searchd
waits for all currently running queries to finish;searchd
resumes serving queries from new index.
However, if there's a lot of attribute or dictionary data, then preloading step could take noticeble time - up to several minutes in case of preloading 1-5+ GB files.
With seamless rotate enabled, rotation works as follows:
Seamless rotate comes at the cost of higher peak
memory usage during the rotation (because both old and new copies of
.spa/.spi/.spm
data need to be in RAM while
preloading new copy). Average usage stays the same.
seamless_rotate = 1
Whether to forcibly preopen all indexes on startup. Optional, default is 0 (do not preopen). Enforces enabled preopen on all served indexes, to avoid manually specifying it in every index.
preopen_indexes = 1
sql_str2ordinal_column
max_prefix_len
, max_infix_len
)@*
syntax to reset current field to query languagedocinfo=none
casemmap()
limits for attributes and wordlists (now able to map over 4 GB on x64 and over 2 GB on x32 where possible)malloc()
pressure in head daemon (search time should not degrade with time any more)test.php
command line options.spl
files getting unlinkedpgsql
source typemmap()ing
for attributes and wordlist (improves search time, speeds up fork()
greatly)-g
, removed -fomit-frame-pointer
)unlink()ed
on bind()
failure--with-mysql-includes/libs
(they conflicted with well-known paths)max_matches
per-query--with-debug
option to configure to compile in debug mode-DNDEBUG
when compiling in default modemin_word_len
) words prepended to next fieldconfigure
scriptmin_word_len
option to indexmax_matches
option to searchd, removed hardcoded MAX_MATCHES limitexample.sql
--stdin
command-line option to search utility--noprogress
option to indexer--index
option to searchtime(NULL)
calls in time-segments mode