database/postgres
1
0
Fork 0
mirror of https://github.com/postgres/postgres.git synced 2025-07-24 14:22:24 +03:00
Code Activity

This patch adds the following to the FTI module:

Browse Source 
* The ability to index more than one column in a table with a single
trigger.
* All uses of sprintf changed to snprintf to prevent users from crashing
Postgres.
* Error messages made more consistent
* Some changes made to bring it into line with coding requirements for
triggers specified in the docs.  (ie. check you're a trigger before casting
your context)
* The perl script that generate indices has been updated to support indexing
multiple columns in a table.
* Fairly well tested in our development environment indexing a food
database's brand and description fields.  The size of the fti index is
around 300,000 rows.
* All docs and examples upgraded.  This includes specifying more efficient
index usage that was specified before, better examples that don't produce
duplicates, etc.


Christopher Kings-Lynne & Brett
This commit is contained in:
Bruce Momjian
2001-08-04 19:41:00 +00:00
parent 16365ac75b
commit e9ea1255c0
3 changed files with 189 additions and 141 deletions
Download Patch File Download Diff File Expand all files Collapse all files

30
contrib/fulltextindex/README.fti
View File

@ -3,11 +3,13 @@ An attempt at some sort of Full Text Indexing for PostgreSQL.
The included software is an attempt to add some sort of Full Text Indexing
support to PostgreSQL. I mean by this that we can ask questions like:
Give me all rows that have 'still' and 'nash' in the 'artist' field.
Give me all rows that have 'still' and 'nash' in the 'artist' or 'title'
fields.
Ofcourse we can write this as:
select * from cds where artist ~* 'stills' and artist ~* 'nash';
select * from cds where (artist ~* 'stills' or title ~* 'stills') and
(artist ~* 'nash' or title ~* 'nash');
But this does not use any indices, and therefore, if your database
gets very large, it will not have very high performance (the above query
@ -15,8 +17,8 @@ requires at least one sequential scan, it probably takes 2 due to the
self-join).
The approach used by this add-on is to define a trigger on the table and
column you want to do this queries on. On every insert in the table, it
takes the value in the specified column, breaks the text in this column
columns you want to do this queries on. On every insert in the table, it
takes the value in the specified columns, breaks the text in these columns
up into pieces, and stores all sub-strings into another table, together
with a reference to the row in the original table that contained this
sub-string (it uses the oid of that row).
@ -24,8 +26,8 @@ sub-string (it uses the oid of that row).
By now creating an index over the 'fti-table', we can search for
substrings that occur in the original table. By making a join between
the fti-table and the orig-table, we can get the actual rows we want
(this can also be done by using subselects, and maybe there're other
ways too).
(this can also be done by using subselects - but subselects are currently
inefficient in Postgres, and maybe there're other ways too).
The trigger code also allows an array called StopWords, that prevents
certain words from being indexed.
@ -62,20 +64,22 @@ The create the function that contains the trigger::
And finally define the trigger on the 'cds' table:
create trigger cds-fti-trigger after update or insert or delete on cds
for each row execute procedure fti(cds-fti, artist);
for each row execute procedure fti(cds-fti, artist, title);
Here, the trigger will be defined on table 'cds', it will create
sub-strings from the field 'artist', and it will place those sub-strings
in the table 'cds-fti'.
sub-strings from the fields 'artist' and 'title', and it will place
those sub-strings in the table 'cds-fti'.
Now populate the table 'cds'. This will also populate the table 'cds-fti'.
It's fastest to populate the table *before* you create the indices.
It's fastest to populate the table *before* you create the indices. Use the
supplied 'fti.pl' to assist you with this.
Before you start using the system, you should at least have the following
indices:
create index cds-fti-idx on cds-fti (string, id);
create index cds-oid-idx on cds (oid);
create index cds-fti-idx on cds-fti (string); -- String matching
create index cds-fti-idx on cds-fti (id); -- For deleting a cds row
create index cds-oid-idx on cds (oid); -- For joining cds to cds-fti
To get the most performance out of this, you should have 'cds-fti'
clustered on disk, ie. all rows with the same sub-strings should be
@ -109,7 +113,7 @@ clustered : same as above, only clustered : 4.501.321 rows
A sequential scan of the artist_fti table (and thus also the clustered table)
takes around 6:16 minutes....
Unfortunately I cannot probide anybody else with this test-date, since I
Unfortunately I cannot provide anybody else with this test-data, since I
am not allowed to redistribute the data (it's a database being sold by
a couple of wholesale companies). Anyways, it's megabytes, so you probably
wouldn't want it in this distribution anyways.

290
contrib/fulltextindex/fti.c
View File

@ -6,61 +6,87 @@
#include "commands/trigger.h"
/*
* Trigger function takes 2 arguments:
1. relation in which to store the substrings
2. field to extract substrings from
* Trigger function accepts variable number of arguments:
*
* 1. relation in which to store the substrings
* 2. fields to extract substrings from
*
* The relation in which to insert *must* have the following layout:
*
* string varchar(#)
* id oid
*
* where # is the largest size of the varchar columns being indexed
*
* Example:
*
* -- Create the SQL function based on the compiled shared object
* create function fti() returns opaque as
* '/usr/local/pgsql/lib/contrib/fti.so' language 'C';
*
* -- Create the FTI table
* create table product_fti (string varchar(255), id oid);
*
* -- Create an index to assist string matches
* create index product_fti_string_idx on product_fti (string);
*
* -- Create an index to assist trigger'd deletes
* create index product_fti_id_idx on product_fti (id);
*
* -- Create an index on the product oid column to assist joins
* -- between the fti table and the product table
* create index product_oid_idx on product (oid);
*
* -- Create the trigger to perform incremental changes to the full text index.
* create trigger product_fti_trig after update or insert or delete on product
* for each row execute procedure fti(product_fti, title, artist);
* ^^^^^^^^^^^
* table where full text index is stored
* ^^^^^^^^^^^^^
* columns to index in the base table
*
* After populating 'product', try something like:
*
* SELECT DISTINCT(p.*) FROM product p, product_fti f1, product_fti f2 WHERE
* f1.string ~ '^slippery' AND f2.string ~ '^wet' AND p.oid=f1.id AND p.oid=f2.id;
*
* To check that your indicies are being used correctly, make sure you
* EXPLAIN SELECT ... your test query above.
*
* CHANGELOG
* ---------
*
* august 3 2001
* Extended fti function to accept more than one column as a
* parameter and all specified columns are indexed. Changed
* all uses of sprintf to snprintf. Made error messages more
* consistent.
*
* march 4 1998 Changed breakup() to return less substrings. Only breakup
* in word parts which are in turn shortened from the start
* of the word (ie. word, ord, rd)
* Did allocation of substring buffer outside of breakup()
*
* oct. 5 1997, fixed a bug in string breakup (where there are more nonalpha
* characters between words then 1).
*
* oct 4-5 1997 implemented the thing, at least the basic functionallity
* of it all....
*
* TODO
* ----
*
* prevent generating duplicate words for an oid in the fti table
* save a plan for deletes
* create a function that will make the index *after* we have populated
* the main table (probably first delete all contents to be sure there's
* nothing in it, then re-populate the fti-table)
*
* can we do something with operator overloading or a seperate function
* that can build the final query automatigally?
*/
The relation in which to insert *must* have the following layout:
string varchar(#)
id oid
Example:
create function fti() returns opaque as
'/home/boekhold/src/postgresql-6.2/contrib/fti/fti.so' language 'C';
create table title_fti (string varchar(25), id oid);
create index title_fti_idx on title_fti (string);
create trigger title_fti_trigger after update or insert or delete on product
for each row execute procedure fti(title_fti, title);
^^^^^^^^^
where to store index in
^^^^^
which column to index
ofcourse don't forget to create an index on title_idx, column string, else
you won't notice much speedup :)
After populating 'product', try something like:
select p.* from product p, title_fti f1, title_fti f2 where
f1.string='slippery' and f2.string='wet' and f1.id=f2.id and p.oid=f1.id;
*/
/*
march 4 1998 Changed breakup() to return less substrings. Only breakup
in word parts which are in turn shortened from the start
of the word (ie. word, ord, rd)
Did allocation of substring buffer outside of breakup()
oct. 5 1997, fixed a bug in string breakup (where there are more nonalpha
characters between words then 1).
oct 4-5 1997 implemented the thing, at least the basic functionallity
of it all....
*/
/* IMPROVEMENTS:
save a plan for deletes
create a function that will make the index *after* we have populated
the main table (probably first delete all contents to be sure there's
nothing in it, then re-populate the fti-table)
can we do something with operator overloading or a seperate function
that can build the final query automatigally?
*/
#define MAX_FTI_QUERY_LENGTH 8192
extern Datum fti(PG_FUNCTION_ARGS);
static char *breakup(char *, char *);
@ -81,10 +107,10 @@ char *StopWords[] = { /* list of words to skip in indexing */
/* stuff for caching query-plans, stolen from contrib/spi/\*.c */
typedef struct
{
char *ident;
int nplans;
void **splan;
} EPlan;
char *ident;
int nplans;
void **splan;
} EPlan;
static EPlan *InsertPlans = NULL;
static EPlan *DeletePlans = NULL;
@ -99,7 +125,7 @@ PG_FUNCTION_INFO_V1(fti);
Datum
fti(PG_FUNCTION_ARGS)
{
TriggerData *trigdata = (TriggerData *) fcinfo->context;
TriggerData *trigdata;
Trigger *trigger; /* to get trigger name */
int nargs; /* # of arguments */
char **args; /* arguments */
@ -111,7 +137,7 @@ fti(PG_FUNCTION_ARGS)
bool isinsert = false;
bool isdelete = false;
int ret;
char query[8192];
char query[MAX_FTI_QUERY_LENGTH];
Oid oid;
/*
@ -124,11 +150,15 @@ fti(PG_FUNCTION_ARGS)
*/
if (!CALLED_AS_TRIGGER(fcinfo))
elog(ERROR, "Full Text Indexing: not fired by trigger manager");
elog(ERROR, "Full Text Indexing: Not fired by trigger manager");
/* It's safe to cast now that we've checked */
trigdata = (TriggerData *) fcinfo->context;
if (TRIGGER_FIRED_FOR_STATEMENT(trigdata->tg_event))
elog(ERROR, "Full Text Indexing: can't process STATEMENT events");
elog(ERROR, "Full Text Indexing: Can't process STATEMENT events");
if (TRIGGER_FIRED_BEFORE(trigdata->tg_event))
elog(ERROR, "Full Text Indexing: must be fired AFTER event");
elog(ERROR, "Full Text Indexing: Must be fired AFTER event");
if (TRIGGER_FIRED_BY_INSERT(trigdata->tg_event))
isinsert = true;
@ -148,11 +178,11 @@ fti(PG_FUNCTION_ARGS)
rettuple = trigdata->tg_newtuple;
if ((ret = SPI_connect()) < 0)
elog(ERROR, "Full Text Indexing: SPI_connect failed, returned %d\n", ret);
elog(ERROR, "Full Text Indexing: SPI_connect: Failed, returned %d\n", ret);
nargs = trigger->tgnargs;
if (nargs != 2)
elog(ERROR, "Full Text Indexing: trigger can only have 2 arguments");
if (nargs < 2)
elog(ERROR, "Full Text Indexing: Trigger must have at least 2 arguments\n");
args = trigger->tgargs;
indexname = args[0];
@ -161,7 +191,7 @@ fti(PG_FUNCTION_ARGS)
/* get oid of current tuple, needed by all, so place here */
oid = rettuple->t_data->t_oid;
if (!OidIsValid(oid))
elog(ERROR, "Full Text Indexing: oid of current tuple is NULL");
elog(ERROR, "Full Text Indexing: Oid of current tuple is invalid");
if (isdelete)
{
@ -169,8 +199,14 @@ fti(PG_FUNCTION_ARGS)
Oid *argtypes;
Datum values[1];
EPlan *plan;
int i;
snprintf(query, MAX_FTI_QUERY_LENGTH, "D%s", indexname);
for (i = 1; i < nargs; i++)
{
snprintf(query, MAX_FTI_QUERY_LENGTH, "%s$%s", query, args[i]);
}
sprintf(query, "D%s$%s", args[0], args[1]);
plan = find_plan(query, &DeletePlans, &nDeletePlans);
if (plan->nplans <= 0)
{
@ -178,15 +214,13 @@ fti(PG_FUNCTION_ARGS)
argtypes[0] = OIDOID;
sprintf(query, "DELETE FROM %s WHERE id = $1", indexname);
snprintf(query, MAX_FTI_QUERY_LENGTH, "DELETE FROM %s WHERE id = $1", indexname);
pplan = SPI_prepare(query, 1, argtypes);
if (!pplan)
elog(ERROR, "Full Text Indexing: SPI_prepare returned NULL "
"in delete");
elog(ERROR, "Full Text Indexing: SPI_prepare: Returned NULL in delete");
pplan = SPI_saveplan(pplan);
if (pplan == NULL)
elog(ERROR, "Full Text Indexing: SPI_saveplan returned NULL "
"in delete");
elog(ERROR, "Full Text Indexing: SPI_saveplan: Returned NULL in delete");
plan->splan = (void **) malloc(sizeof(void *));
*(plan->splan) = pplan;
@ -197,21 +231,29 @@ fti(PG_FUNCTION_ARGS)
ret = SPI_execp(*(plan->splan), values, NULL, 0);
if (ret != SPI_OK_DELETE)
elog(ERROR, "Full Text Indexing: error executing plan in delete");
elog(ERROR, "Full Text Indexing: SPI_execp: Error executing plan in delete");
}
if (isinsert)
{
char *substring,
*column;
void *pplan;
Oid *argtypes;
char *substring;
char *column;
void *pplan;
Oid *argtypes;
Datum values[2];
int colnum;
struct varlena *data;
EPlan *plan;
struct varlena *data;
EPlan *plan;
int i;
char *buff;
char *string;
snprintf(query, MAX_FTI_QUERY_LENGTH, "I%s", indexname);
for (i = 1; i < nargs; i++)
{
snprintf(query, MAX_FTI_QUERY_LENGTH, "%s$%s", query, args[i]);
}
sprintf(query, "I%s$%s", args[0], args[1]);
plan = find_plan(query, &InsertPlans, &nInsertPlans);
/* no plan yet, so allocate mem for argtypes */
@ -224,67 +266,65 @@ fti(PG_FUNCTION_ARGS)
argtypes[1] = OIDOID; /* id oid); */
/* prepare plan to gain speed */
sprintf(query, "INSERT INTO %s (string, id) VALUES ($1, $2)",
snprintf(query, MAX_FTI_QUERY_LENGTH, "INSERT INTO %s (string, id) VALUES ($1, $2)",
indexname);
pplan = SPI_prepare(query, 2, argtypes);
if (!pplan)
elog(ERROR, "Full Text Indexing: SPI_prepare returned NULL "
"in insert");
elog(ERROR, "Full Text Indexing: SPI_prepare: Returned NULL in insert");
pplan = SPI_saveplan(pplan);
if (pplan == NULL)
elog(ERROR, "Full Text Indexing: SPI_saveplan returned NULL"
" in insert");
elog(ERROR, "Full Text Indexing: SPI_saveplan: Returned NULL in insert");
plan->splan = (void **) malloc(sizeof(void *));
*(plan->splan) = pplan;
plan->nplans = 1;
}
/* prepare plan for query */
colnum = SPI_fnumber(tupdesc, args[1]);
if (colnum == SPI_ERROR_NOATTRIBUTE)
elog(ERROR, "Full Text Indexing: column '%s' of '%s' not found",
args[1], args[0]);
for (i = 0; i < nargs - 1; i++)
{
colnum = SPI_fnumber(tupdesc, args[i + 1]);
if (colnum == SPI_ERROR_NOATTRIBUTE)
elog(ERROR, "Full Text Indexing: SPI_fnumber: Column '%s' of '%s' not found", args[i + 1], indexname);
/* Get the char* representation of the column with name args[1] */
column = SPI_getvalue(rettuple, tupdesc, colnum);
/* Get the char* representation of the column */
column = SPI_getvalue(rettuple, tupdesc, colnum);
if (column)
{ /* make sure we don't try to index NULL's */
char *buff;
char *string = column;
while (*string != '\0')
/* make sure we don't try to index NULL's */
if (column)
{
*string = tolower((unsigned char) *string);
string++;
string = column;
while (*string != '\0')
{
*string = tolower((unsigned char) *string);
string++;
}
data = (struct varlena *) palloc(sizeof(int32) + strlen(column) + 1);
buff = palloc(strlen(column) + 1);
/* saves lots of calls in while-loop and in breakup() */
new_tuple = true;
while ((substring = breakup(column, buff)))
{
int l;
l = strlen(substring);
data->vl_len = l + sizeof(int32);
memcpy(VARDATA(data), substring, l);
values[0] = PointerGetDatum(data);
values[1] = oid;
ret = SPI_execp(*(plan->splan), values, NULL, 0);
if (ret != SPI_OK_INSERT)
elog(ERROR, "Full Text Indexing: SPI_execp: Error executing plan in insert");
}
pfree(buff);
pfree(data);
}
data = (struct varlena *) palloc(sizeof(int32) + strlen(column) +1);
buff = palloc(strlen(column) + 1);
/* saves lots of calls in while-loop and in breakup() */
new_tuple = true;
while ((substring = breakup(column, buff)))
{
int l;
l = strlen(substring);
data->vl_len = l + sizeof(int32);
memcpy(VARDATA(data), substring, l);
values[0] = PointerGetDatum(data);
values[1] = oid;
ret = SPI_execp(*(plan->splan), values, NULL, 0);
if (ret != SPI_OK_INSERT)
elog(ERROR, "Full Text Indexing: error executing plan "
"in insert");
}
pfree(buff);
pfree(data);
}
}

10
contrib/fulltextindex/fti.pl
View File

@ -17,7 +17,7 @@
#
# Example:
#
# fti.pl -u -d mydb -t mytable -c mycolumn -f myfile
# fti.pl -u -d mydb -t mytable -c mycolumn,mycolumn2 -f myfile
# sort -o myoutfile myfile
# uniq myoutfile sorted-file
#
@ -140,11 +140,13 @@ sub main {
getopts('d:t:c:f:u');
if (!$opt_d || !$opt_t || !$opt_c || !$opt_f) {
print STDERR "usage: $0 [-u] -d database -t table -c column ".
print STDERR "usage: $0 [-u] -d database -t table -c column[,column...] ".
"-f output-file\n";
return 1;
}
@cols = split(/,/, $opt_c);
if (defined($opt_u)) {
$uname = get_username();
$pwd = get_password();
@ -166,7 +168,9 @@ sub main {
PQexec($PG_CONN, "begin");
$query = "declare C cursor for select $opt_c, oid from $opt_t";
$query = "declare C cursor for select (\"";
$query .= join("\" || ' ' || \"", @cols);
$query .= "\") as string, oid from $opt_t";
$res = PQexec($PG_CONN, $query);
if (!$res || (PQresultStatus($res) != $PGRES_COMMAND_OK)) {
print STDERR "Error declaring cursor!\n";