Babayaga 0906b299f4 Maintenance Love :)

2026-03-21 20:18:25 +01:00

33 KiB

Raw Blame History

Full-Text Search and VFS Indexing Architecture

This document covers the end-to-end architecture for Full-Text Search (FTS) regarding the Virtual File System (VFS). It details the extraction of textual content from uploaded files all the way to performing concurrent WebSearch language processing algorithms native to PostgreSQL.

1. Indexing & Text Extraction Pipeline

When an administrator forces a VFS reindex with the fullText option, the server systematically walks the specified mount, performing strict checks to extract the file content safely into the database.

File Indexing Sequence

sequenceDiagram
    participant Admin as VFS Admin (UI)
    participant UI as [StorageManager.tsx](../src/components/admin/StorageManager.tsx)
    participant Route as [vfs-routes.ts](../server/src/products/storage/api/vfs-routes.ts)
    participant VFS as [vfs.ts](../server/src/products/storage/api/vfs.ts)
    participant FS as Local File System
    participant DB as Supabase DB

    Admin->>UI: Select "Index All" + Full Text
    UI->>Route: POST /api/vfs/index/{mount}?fullText=true
    Route->>VFS: handleVfsIndex()
    
    loop Every File Found in Walk
        VFS->>VFS: Check Type and Size (Limit: VFS_INDEX_MAX_FILE_SIZE_KB)
        VFS->>VFS: Validate Extension against allowFullTextExtensions
        
        opt Passes Filters
            VFS->>FS: Stream start readingz
            FS-->>VFS: Output Stream chunks
            VFS->>VFS: Verify NOT Binary (No Null Bytes)
        end
        
        VFS->>DB: Upsert Node Data to vfs_index
        Note right of DB: fts tsvector auto-updated<br/>by trigger/stored procedure
    end

    VFS-->>Route: Return Batched OK
    Route-->>UI: OK

Core Components

StorageManager.tsx: UI providing the checkbox to trigger fullText parameter for reindexing.
vfs.json: Externalized configuration supporting allowFullTextExtensions defining valid target patterns (e.g. md, cpp).
vfs.ts: Contains the central mechanism inside handleVfsIndex that implements extraction via async pipeline strategies (filters and transformers) wrapped in race limits to block hanging operations.
vfs_index.sql: Contains the table structure. Specifically, you can see fts tsvector GENERATED ALWAYS AS (...) mapped tightly to combine the name, path, and content.

2. Searching & WebSearch Concurrency

The main unified search bar queries /api/search which dispatches nested parallel logic depending on the internal components. For files, we natively blend strict regex string checks (ilike) and Deep WebSearch functionality (textSearch).

File Search Sequence

sequenceDiagram
    participant User
    participant Router as [db-search.ts](../server/src/products/serving/db/db-search.ts)
    participant VFS as [vfs.ts](../server/src/products/storage/api/vfs.ts)
    participant Supabase

    User->>Router: GET /api/search?q=Anycubic+Chiron+Marlin&type=files
    Router->>VFS: searchAllSearchableVfs(query)
    
    par Path Substring Match
        VFS->>Supabase: ilike('path', '%Anycubic%') & ilike('path', '%Marlin%')
    and FTS Content Match
        VFS->>Supabase: .textSearch('fts', query, { type: 'websearch' })
        Note right of Supabase: Translates unquoted words to "&" organically
    end

    Supabase-->>VFS: return [pathRes, ftsRes]
    
    VFS->>VFS: Map.set(row.id, deduplicated)
    VFS-->>Router: Combined INode[] (with deep metadata)
    Router->>Router: Enrich & Filter Visibility (ACL)
    Router-->>User: Final Feed Items (meta.url mapped)

Core Components

db-search.ts: Coordinates overarching feed blending globally. It processes hit arrays from various pipelines in parallel (Pages, Posts, Files etc.).
vfs.ts: Inside searchAllSearchableVfs, we issue the actual .textSearch querying function to Supabase and cleanly map any internal SQL errors safely to our standard error logging stream.
search-fts.e2e.test.ts: Validates exactly this layer ensuring end-to-end integration and data propagation accurately hit native database FTS queries safely.

3. TODOs & Future Enhancements

Google Search Bot Indexability: Investigate and implement strategies for how web crawlers (like Googlebot) can discover and index our dynamic search results. Potential solutions include:
- Server-Side Rendering (SSR) of Search Pages: Ensuring initial page loads return fully populated HTML for search queries if accessed directly via URL parameters.
- Dynamic Sitemaps (sitemap.xml): Generating sitemap endpoints that list popular or pre-computed search queries to guide crawlers.
- Internal Linking Structure: Exposing curated search links or tags throughout the application surface area (e.g., related topics, tag clouds) that crawlers can follow.
- Structured Data: Injecting JSON-LD or schema.org metadata representing the search capability and search result pages.

Full Text Search

How to use full text search in PostgreSQL.

Postgres has built-in functions to handle Full Text Search queries. This is like a "search engine" within Postgres.

Preparation

For this guide we'll use the following example data:

id	title	author	description
1	The Poky Little Puppy	Janette Sebring Lowrey	Puppy is slower than other, bigger animals.
2	The Tale of Peter Rabbit	Beatrix Potter	Rabbit eats some vegetables.
3	Tootle	Gertrude Crampton	Little toy train has big dreams.
4	Green Eggs and Ham	Dr. Seuss	Sam has changing food preferences and eats unusually colored food.
5	Harry Potter and the Goblet of Fire	J.K. Rowling	Fourth year of school starts, big drama ensues.

create table books (
  id serial primary key,
  title text,
  author text,
  description text
);

insert into books
  (title, author, description)
values
  (
    'The Poky Little Puppy',
    'Janette Sebring Lowrey',
    'Puppy is slower than other, bigger animals.'
  ),
  ('The Tale of Peter Rabbit', 'Beatrix Potter', 'Rabbit eats some vegetables.'),
  ('Tootle', 'Gertrude Crampton', 'Little toy train has big dreams.'),
  (
    'Green Eggs and Ham',
    'Dr. Seuss',
    'Sam has changing food preferences and eats unusually colored food.'
  ),
  (
    'Harry Potter and the Goblet of Fire',
    'J.K. Rowling',
    'Fourth year of school starts, big drama ensues.'
  );

Usage

The functions we'll cover in this guide are:

`to_tsvector()` [#to-tsvector]

Converts your data into searchable tokens. to_tsvector() stands for "to text search vector." For example:

select to_tsvector('green eggs and ham');
-- Returns 'egg':2 'green':1 'ham':4

Collectively these tokens are called a "document" which Postgres can use for comparisons.

`to_tsquery()` [#to-tsquery]

Converts a query string into tokens to match. to_tsquery() stands for "to text search query."

This conversion step is important because we will want to "fuzzy match" on keywords. For example if a user searches for eggs, and a column has the value egg, we probably still want to return a match.

Postgres provides several functions to create tsquery objects:

to_tsquery() - Requires manual specification of operators (&, |, !)
plainto_tsquery() - Converts plain text to an AND query: plainto_tsquery('english', 'fat rats') → 'fat' & 'rat'
phraseto_tsquery() - Creates phrase queries: phraseto_tsquery('english', 'fat rats') → 'fat' <-> 'rat'
websearch_to_tsquery() - Supports web search syntax with quotes, "or", and negation

Match: `@@` [#match]

The @@ symbol is the "match" symbol for Full Text Search. It returns any matches between a to_tsvector result and a to_tsquery result.

Take the following example:

select *
from books
where title = 'Harry';

const { data, error } = await supabase.from('books').select().eq('title', 'Harry')

final result = await client
  .from('books')
  .select()
  .eq('title', 'Harry');

let response = try await supabase.from("books")
  .select()
  .eq("title", value: "Harry")
  .execute()

val data = supabase.from("books").select {
    filter {
        eq("title", "Harry")
    }
}

data = supabase.from_('books').select().eq('title', 'Harry').execute()

The equality symbol above (=) is very "strict" on what it matches. In a full text search context, we might want to find all "Harry Potter" books and so we can rewrite the example above:

select *
from books
where to_tsvector(title) @@ to_tsquery('Harry');

const { data, error } = await supabase.from('books').select().textSearch('title', `'Harry'`)

final result = await client
  .from('books')
  .select()
  .textSearch('title', "'Harry'");

let response = try await supabase.from("books")
  .select()
  .textSearch("title", value: "'Harry'")

val data = supabase.from("books").select {
    filter {
        textSearch("title", "'Harry'", TextSearchType.NONE)
    }
}

Basic full text queries

Search a single column

To find all books where the description contain the word big:

select
  *
from
  books
where
  to_tsvector(description)
  @@ to_tsquery('big');

const { data, error } = await supabase.from('books').select().textSearch('description', `'big'`)

final result = await client
  .from('books')
  .select()
  .textSearch('description', "'big'");

let response = await client.from("books")
  .select()
  .textSearch("description", value: "'big'")
  .execute()

val data = supabase.from("books").select {
    filter {
        textSearch("description", "'big'", TextSearchType.NONE)
    }
}

data = supabase.from_('books').select().text_search('description', "'big'").execute()

id	title	author	description
3	Tootle	Gertrude Crampton	Little toy train has big dreams.
5	Harry Potter and the Goblet of Fire	J.K. Rowling	Fourth year of school starts, big drama ensues.

Search multiple columns

Right now there is no direct way to use JavaScript or Dart to search through multiple columns but you can do it by creating computed columns on the database.

To find all books where description or title contain the word little:

select
  *
from
  books
where
  to_tsvector(description || ' ' || title) -- concat columns, but be sure to include a space to separate them!
  @@ to_tsquery('little');

create function title_description(books) returns text as $$
  select $1.title || ' ' || $1.description;
$$ language sql immutable;

const { data, error } = await supabase
  .from('books')
  .select()
  .textSearch('title_description', `little`)

create function title_description(books) returns text as $$
  select $1.title || ' ' || $1.description;
$$ language sql immutable;

final result = await client
  .from('books')
  .select()
  .textSearch('title_description', "little")

create function title_description(books) returns text as $$
  select $1.title || ' ' || $1.description;
$$ language sql immutable;

let response = try await client
  .from("books")
  .select()
  .textSearch("title_description", value: "little")
  .execute()

create function title_description(books) returns text as $$
  select $1.title || ' ' || $1.description;
$$ language sql immutable;

val data = supabase.from("books").select {
    filter {
        textSearch("title_description", "title", TextSearchType.NONE)
    }
}

create function title_description(books) returns text as $$
  select $1.title || ' ' || $1.description;
$$ language sql immutable;

data = supabase.from_('books').select().text_search('title_description', "little").execute()

id	title	author	description
1	The Poky Little Puppy	Janette Sebring Lowrey	Puppy is slower than other, bigger animals.
3	Tootle	Gertrude Crampton	Little toy train has big dreams.

Match all search words

To find all books where description contains BOTH of the words little and big, we can use the & symbol:

select
  *
from
  books
where
  to_tsvector(description)
  @@ to_tsquery('little & big'); -- use & for AND in the search query

const { data, error } = await supabase
  .from('books')
  .select()
  .textSearch('description', `'little' & 'big'`)

final result = await client
  .from('books')
  .select()
  .textSearch('description', "'little' & 'big'");

let response = try await client
  .from("books")
  .select()
  .textSearch("description", value: "'little' & 'big'");
  .execute()

val data = supabase.from("books").select {
    filter {
        textSearch("description", "'title' & 'big'", TextSearchType.NONE)
    }
}

data = supabase.from_('books').select().text_search('description', "'little' & 'big'").execute()

id	title	author	description
3	Tootle	Gertrude Crampton	Little toy train has big dreams.

Match any search words

To find all books where description contain ANY of the words little or big, use the | symbol:

select
  *
from
  books
where
  to_tsvector(description)
  @@ to_tsquery('little | big'); -- use | for OR in the search query

const { data, error } = await supabase
  .from('books')
  .select()
  .textSearch('description', `'little' | 'big'`)

final result = await client
  .from('books')
  .select()
  .textSearch('description', "'little' | 'big'");

let response = try await client
  .from("books")
  .select()
  .textSearch("description", value: "'little' | 'big'")
  .execute()

val data = supabase.from("books").select {
    filter {
        textSearch("description", "'title' | 'big'", TextSearchType.NONE)
    }
}

response = client.from_('books').select().text_search('description', "'little' | 'big'").execute()

id	title	author	description
1	The Poky Little Puppy	Janette Sebring Lowrey	Puppy is slower than other, bigger animals.
3	Tootle	Gertrude Crampton	Little toy train has big dreams.

Notice how searching for big includes results with the word bigger (or biggest, etc).

Partial search

Partial search is particularly useful when you want to find matches on substrings within your data.

Implementing partial search

You can use the :* syntax with to_tsquery(). Here's an example that searches for any book titles beginning with "Lit":

select title from books where to_tsvector(title) @@ to_tsquery('Lit:*');

Extending functionality with RPC

To make the partial search functionality accessible through the API, you can wrap the search logic in a stored procedure.

After creating this function, you can invoke it from your application using the SDK for your platform. Here's an example:

create or replace function search_books_by_title_prefix(prefix text)
returns setof books AS $$
begin
  return query
  select * from books where to_tsvector('english', title) @@ to_tsquery(prefix || ':*');
end;
$$ language plpgsql;

const { data, error } = await supabase.rpc('search_books_by_title_prefix', { prefix: 'Lit' })

final data = await supabase.rpc('search_books_by_title_prefix', params: { 'prefix': 'Lit' });

let response = try await supabase.rpc(
  "search_books_by_title_prefix",
  params: ["prefix": "Lit"]
)
.execute()

val rpcParams = mapOf("prefix" to "Lit")
val result = supabase.postgrest.rpc("search_books_by_title_prefix", rpcParams)

data = client.rpc('search_books_by_title_prefix', { 'prefix': 'Lit' }).execute()

This function takes a prefix parameter and returns all books where the title contains a word starting with that prefix. The :* operator is used to denote a prefix match in the to_tsquery() function.

Handling spaces in queries

When you want the search term to include a phrase or multiple words, you can concatenate words using a + as a placeholder for space:

select * from search_books_by_title_prefix('Little+Puppy');

Web search syntax with `websearch_to_tsquery()` [#websearch-to-tsquery]

The websearch_to_tsquery() function provides an intuitive search syntax similar to popular web search engines, making it ideal for user-facing search interfaces.

Basic usage

select *
from books
where to_tsvector(description) @@ websearch_to_tsquery('english', 'green eggs');

const { data, error } = await supabase
  .from('books')
  .select()
  .textSearch('description', 'green eggs', { type: 'websearch' })

Quoted phrases

Use quotes to search for exact phrases:

select * from books
where to_tsvector(description || ' ' || title) @@ websearch_to_tsquery('english', '"Green Eggs"');
-- Matches documents containing "Green" immediately followed by "Eggs"

OR searches

Use "or" (case-insensitive) to search for multiple terms:

select * from books
where to_tsvector(description) @@ websearch_to_tsquery('english', 'puppy or rabbit');
-- Matches documents containing either "puppy" OR "rabbit"

Negation

Use a dash (-) to exclude terms:

select * from books
where to_tsvector(description) @@ websearch_to_tsquery('english', 'animal -rabbit');
-- Matches documents containing "animal" but NOT "rabbit"

Complex queries

Combine multiple operators for sophisticated searches:

select * from books
where to_tsvector(description || ' ' || title) @@
  websearch_to_tsquery('english', '"Harry Potter" or "Dr. Seuss" -vegetables');
-- Matches books by "Harry Potter" or "Dr. Seuss" but excludes those mentioning vegetables

Creating indexes

Now that you have Full Text Search working, create an index. This allows Postgres to "build" the documents preemptively so that they don't need to be created at the time we execute the query. This will make our queries much faster.

Searchable columns

Let's create a new column fts inside the books table to store the searchable index of the title and description columns.

We can use a special feature of Postgres called Generated Columns to ensure that the index is updated any time the values in the title and description columns change.

alter table
  books
add column
  fts tsvector generated always as (to_tsvector('english', description || ' ' || title)) stored;

create index books_fts on books using gin (fts); -- generate the index

select id, fts
from books;

| id  | fts                                                                                                             |
| --- | --------------------------------------------------------------------------------------------------------------- |
| 1   | 'anim':7 'bigger':6 'littl':10 'poki':9 'puppi':1,11 'slower':3                                                 |
| 2   | 'eat':2 'peter':8 'rabbit':1,9 'tale':6 'veget':4                                                               |
| 3   | 'big':5 'dream':6 'littl':1 'tootl':7 'toy':2 'train':3                                                         |
| 4   | 'chang':3 'color':9 'eat':7 'egg':12 'food':4,10 'green':11 'ham':14 'prefer':5 'sam':1 'unus':8                |
| 5   | 'big':6 'drama':7 'ensu':8 'fire':15 'fourth':1 'goblet':13 'harri':9 'potter':10 'school':4 'start':5 'year':2 |

Search using the new column

Now that we've created and populated our index, we can search it using the same techniques as before:

select
  *
from
  books
where
  fts @@ to_tsquery('little & big');

const { data, error } = await supabase.from('books').select().textSearch('fts', `'little' & 'big'`)

final result = await client
  .from('books')
  .select()
  .textSearch('fts', "'little' & 'big'");

let response = try await client
  .from("books")
  .select()
  .textSearch("fts", value: "'little' & 'big'")
  .execute()

val data = supabase.from("books").select {
    filter {
        textSearch("fts", "'title' & 'big'", TextSearchType.NONE)
    }
}

data = client.from_('books').select().text_search('fts', "'little' & 'big'").execute()

id	title	author	description	fts
3	Tootle	Gertrude Crampton	Little toy train has big dreams.	'big':5 'dream':6 'littl':1 'tootl':7 'toy':2 'train':3

Query operators

Visit Postgres: Text Search Functions and Operators to learn about additional query operators you can use to do more advanced full text queries, such as:

Proximity: `<->` [#proximity]

The proximity symbol is useful for searching for terms that are a certain "distance" apart. For example, to find the phrase big dreams, where the a match for "big" is followed immediately by a match for "dreams":

select
  *
from
  books
where
  to_tsvector(description) @@ to_tsquery('big <-> dreams');

const { data, error } = await supabase
  .from('books')
  .select()
  .textSearch('description', `'big' <-> 'dreams'`)

final result = await client
  .from('books')
  .select()
  .textSearch('description', "'big' <-> 'dreams'");

let response = try await client
  .from("books")
  .select()
  .textSearch("description", value: "'big' <-> 'dreams'")
  .execute()

val data = supabase.from("books").select {
    filter {
        textSearch("description", "'big' <-> 'dreams'", TextSearchType.NONE)
    }
}

data = client.from_('books').select().text_search('description', "'big' <-> 'dreams'").execute()

We can also use the <-> to find words within a certain distance of each other. For example to find year and school within 2 words of each other:

select
  *
from
  books
where
  to_tsvector(description) @@ to_tsquery('year <2> school');

const { data, error } = await supabase
  .from('books')
  .select()
  .textSearch('description', `'year' <2> 'school'`)

final result = await client
  .from('books')
  .select()
  .textSearch('description', "'year' <2> 'school'");

let response = try await supabase
  .from("books")
  .select()
  .textSearch("description", value: "'year' <2> 'school'")
  .execute()

val data = supabase.from("books").select {
    filter {
        textSearch("description", "'year' <2> 'school'", TextSearchType.NONE)
    }
}

data = client.from_('books').select().text_search('description', "'year' <2> 'school'").execute()

Negation: `!` [#negation]

The negation symbol can be used to find phrases which don't contain a search term. For example, to find records that have the word big but not little:

select
  *
from
  books
where
  to_tsvector(description) @@ to_tsquery('big & !little');

const { data, error } = await supabase
  .from('books')
  .select()
  .textSearch('description', `'big' & !'little'`)

final result = await client
  .from('books')
  .select()
  .textSearch('description', "'big' & !'little'");

let response = try await client
  .from("books")
  .select()
  .textSearch("description", value: "'big' & !'little'")
  .execute()

val data = supabase.from("books").select {
    filter {
        textSearch("description", "'big' & !'little'", TextSearchType.NONE)
    }
}

data = client.from_('books').select().text_search('description', "'big' & !'little'").execute()

Ranking search results [#ranking]

Postgres provides ranking functions to sort search results by relevance, helping you present the most relevant matches first. Since ranking functions need to be computed server-side, use RPC functions and generated columns.

Creating a search function with ranking [#search-function-ranking]

First, create a Postgres function that handles search and ranking:

create or replace function search_books(search_query text)
returns table(id int, title text, description text, rank real) as $$
begin
  return query
  select
    books.id,
    books.title,
    books.description,
    ts_rank(to_tsvector('english', books.description), to_tsquery(search_query)) as rank
  from books
  where to_tsvector('english', books.description) @@ to_tsquery(search_query)
  order by rank desc;
end;
$$ language plpgsql;

Now you can call this function from your client:

const { data, error } = await supabase.rpc('search_books', { search_query: 'big' })

final result = await client
  .rpc('search_books', params: { 'search_query': 'big' });

data = client.rpc('search_books', { 'search_query': 'big' }).execute()

select * from search_books('big');

Ranking with weighted columns [#weighted-ranking]

Postgres allows you to assign different importance levels to different parts of your documents using weight labels. This is especially useful when you want matches in certain fields (like titles) to rank higher than matches in other fields (like descriptions).

Understanding weight labels

Postgres uses four weight labels: A, B, C, and D, where:

A = Highest importance (weight 1.0)
B = High importance (weight 0.4)
C = Medium importance (weight 0.2)
D = Low importance (weight 0.1)

Creating weighted search columns

First, create a weighted tsvector column that gives titles higher priority than descriptions:

-- Add a weighted fts column
alter table books
add column fts_weighted tsvector
generated always as (
  setweight(to_tsvector('english', title), 'A') ||
  setweight(to_tsvector('english', description), 'B')
) stored;

-- Create index for the weighted column
create index books_fts_weighted on books using gin (fts_weighted);

Now create a search function that uses this weighted column:

create or replace function search_books_weighted(search_query text)
returns table(id int, title text, description text, rank real) as $$
begin
  return query
  select
    books.id,
    books.title,
    books.description,
    ts_rank(books.fts_weighted, to_tsquery(search_query)) as rank
  from books
  where books.fts_weighted @@ to_tsquery(search_query)
  order by rank desc;
end;
$$ language plpgsql;

Custom weight arrays

You can also specify custom weights by providing a weight array to ts_rank():

create or replace function search_books_custom_weights(search_query text)
returns table(id int, title text, description text, rank real) as $$
begin
  return query
  select
    books.id,
    books.title,
    books.description,
    ts_rank(
      '{0.0, 0.2, 0.5, 1.0}'::real[], -- Custom weights {D, C, B, A}
      books.fts_weighted,
      to_tsquery(search_query)
    ) as rank
  from books
  where books.fts_weighted @@ to_tsquery(search_query)
  order by rank desc;
end;
$$ language plpgsql;

This example uses custom weights where:

A-labeled terms (titles) have maximum weight (1.0)
B-labeled terms (descriptions) have medium weight (0.5)
C-labeled terms have low weight (0.2)
D-labeled terms are ignored (0.0)

Using the weighted search

// Search with standard weighted ranking
const { data, error } = await supabase.rpc('search_books_weighted', { search_query: 'Harry' })

// Search with custom weights
const { data: customData, error: customError } = await supabase.rpc('search_books_custom_weights', {
  search_query: 'Harry',
})

# Search with standard weighted ranking
data = client.rpc('search_books_weighted', { 'search_query': 'Harry' }).execute()

# Search with custom weights
custom_data = client.rpc('search_books_custom_weights', { 'search_query': 'Harry' }).execute()

-- Standard weighted search
select * from search_books_weighted('Harry');

-- Custom weighted search
select * from search_books_custom_weights('Harry');

Practical example with results

Say you search for "Harry". With weighted columns:

"Harry Potter and the Goblet of Fire" (title match) gets weight A = 1.0
Books mentioning "Harry" in description get weight B = 0.4

This ensures that books with "Harry" in the title ranks significantly higher than books that only mention "Harry" in the description, providing more relevant search results for users.

Using ranking with indexes [#ranking-with-indexes]

When using the fts column you created earlier, ranking becomes more efficient. Create a function that uses the indexed column:

create or replace function search_books_fts(search_query text)
returns table(id int, title text, description text, rank real) as $$
begin
  return query
  select
    books.id,
    books.title,
    books.description,
    ts_rank(books.fts, to_tsquery(search_query)) as rank
  from books
  where books.fts @@ to_tsquery(search_query)
  order by rank desc;
end;
$$ language plpgsql;

const { data, error } = await supabase.rpc('search_books_fts', { search_query: 'little & big' })

final result = await client
  .rpc('search_books_fts', params: { 'search_query': 'little & big' });

data = client.rpc('search_books_fts', { 'search_query': 'little & big' }).execute()

select * from search_books_fts('little & big');

Using web search syntax with ranking [#websearch-ranking]

You can also create a function that combines websearch_to_tsquery() with ranking for user-friendly search:

create or replace function websearch_books(search_text text)
returns table(id int, title text, description text, rank real) as $$
begin
  return query
  select
    books.id,
    books.title,
    books.description,
    ts_rank(books.fts, websearch_to_tsquery('english', search_text)) as rank
  from books
  where books.fts @@ websearch_to_tsquery('english', search_text)
  order by rank desc;
end;
$$ language plpgsql;

// Support natural search syntax
const { data, error } = await supabase.rpc('websearch_books', {
  search_text: '"little puppy" or train -vegetables',
})

select * from websearch_books('"little puppy" or train -vegetables');

Resources

Postgres: Text Search Functions and Operators

33 KiB Raw Blame History

Full-Text Search and VFS Indexing Architecture

1. Indexing & Text Extraction Pipeline

File Indexing Sequence

Core Components

2. Searching & WebSearch Concurrency

File Search Sequence

Core Components

3. TODOs & Future Enhancements

Full Text Search

Preparation

Usage

to_tsvector() [#to-tsvector]

to_tsquery() [#to-tsquery]

Match: @@ [#match]

Basic full text queries

Search a single column

Search multiple columns

Match all search words

Match any search words

Partial search

Implementing partial search

Extending functionality with RPC

Handling spaces in queries

Web search syntax with websearch_to_tsquery() [#websearch-to-tsquery]

Basic usage

Quoted phrases

OR searches

Negation

Complex queries

Creating indexes

Searchable columns

Search using the new column

Query operators

Proximity: <-> [#proximity]

Negation: ! [#negation]

Ranking search results [#ranking]

Creating a search function with ranking [#search-function-ranking]

Ranking with weighted columns [#weighted-ranking]

Understanding weight labels

Creating weighted search columns

Custom weight arrays

Using the weighted search

Practical example with results

Using ranking with indexes [#ranking-with-indexes]

Using web search syntax with ranking [#websearch-ranking]

Resources

33 KiB

Raw Blame History

`to_tsvector()` [#to-tsvector]

`to_tsquery()` [#to-tsquery]

Match: `@@` [#match]

Web search syntax with `websearch_to_tsquery()` [#websearch-to-tsquery]

Proximity: `<->` [#proximity]

Negation: `!` [#negation]