Blog post

dbdev: PostgreSQL Package Manager

2023-04-14

11 minute read

dbdev: PostgreSQL Package Manager

Today we’re publicly previewing database.dev, a PostgreSQL package manager. At this stage the package registry is read-only. We've preloaded it with a handful of packages, or pglets (PostGres appLETs), to showcase some of the more interesting possibilities.

db dev landing page

dbdev fills the same role for PostgreSQL as npm for JavaScript, pip for Python and cargo for Rust in that it enables publishing libraries and applications for repeatable deployment. We'll be releasing the tooling necessary for third-parties to publish pglets to the registry once we’ve collected some community feedback and incorporate any great new ideas. Our goal is to create an open ecosystem for packaging and discovering SQL.

The initial preview is compatible with new projects on the Supabase platform. It can also be installed on any PostgreSQL instance that support pg_tle and pgsql-http.

Get Started with dbdev

The in-database client is the easiest way to get started. You can setup the installer by executing the SQL snippet available at database.dev/installer.

Once the dbdev client is present, pglets can be installed from the registry as shown below:


_10
-- Load the package from the package index
_10
select
_10
dbdev.install ('olirice-asciiplot');
_10
_10
-- Enable the extension
_10
create extension "olirice-asciiplot" version '0.2.1';

You can explore all available pglets on database.dev.

Notice that PostgreSQL sees the olirice-asciiplot pglet as a native extension, rather than a raw snippet of SQL. That approach allows us to leverage PostgreSQL's builtin tooling for extension management.

With our extension installed, you can use it like any other PostgreSQL extension. Continuing with the olirice-asciiplot example, we can call the scatter function it provides to create an ASCII scatterplot:


_28
select
_28
scatter(
_28
val::numeric, -- x
_28
val::numeric, -- y
_28
'stonks!', -- title
_28
15, -- height
_28
50 -- width
_28
)
_28
from
_28
generate_series(1,10) z(val);
_28
/*
_28
stonks!
_28
----------------------------------------------
_28
| *
_28
|
_28
| *
_28
| *
_28
|
_28
| *
_28
|
_28
| *
_28
| *
_28
|
_28
| *
_28
|
_28
| *
_28
| *
_28
*/

PostgreSQL's extension tooling is excellent, but it predates some practices learned from best-in-class package indexes like crates.io. To give developers a more modern development experience, we opted to layer additional strictness on top of what PostgreSQL imposes:

Versioning

The extension system has full support for versioning and migrations. Officially, PostgreSQL has loose constraints for version names. We made the choice to enforce a lite version of Semantic Versioning that restricts version numbers to major.minor.patch so authors can communicate bug-fixes, features, and breaking changes in a familiar way.

Namespaces

Two common challenges faced by package indexes are name squatting and typo squatting.

  • Name squatting: reserving names for future use
  • Typo squatting: reserving misspelling of existing package

The ethics of name squatting get dicey at scale while typo squatting is widely viewed as malicious behavior. To mitigate both issues, all pglets published to database.dev are namespaced to their owning organization or user’s handle. For example a pglet named olirice-index_advisor was created by the account olirice under the name index_advisor. If another user, some_user, forks and republishes the project, it would be available under some_user-index_advisor. Problem solved ✅

Running on Supabase

database.dev is not coupled to the Supabase platform. dbdev can load SQL libraries on any PostgreSQL instance with the required base extensions. However, using dbdev in tandem with Supabase yields some extra possibilities.

Supabase reflects APIs directly from your database’s structure, so a pglet can contain an entire stateful application, pre-configured with authentication, REST, GraphQL, and realtime change data capture all baked in!

For example, our friends at LangChain published a Supabase backend for their docs search tool that uses a hybrid of document embeddings and full text search to find relevant documents for a user’s query

Its available at langchain-hybrid_search and here’s how you’d set it up:


_10
select
_10
dbdev.install ('langchain-hybrid_search');
_10
_10
create extension if not exists vector;
_10
_10
create extension "langchain-hybrid_search" schema public version '1.0.0';

That creates the relevant documents table and associated search functions. Then, you can immediately hit it from your front end for best-in-class document search.


_29
import { OpenAIEmbeddings } from 'langchain/embeddings/openai'
_29
import { createClient } from '@supabase/supabase-js'
_29
import { SupabaseHybridSearch } from 'langchain/retrievers/supabase'
_29
_29
const privateKey = process.env.SUPABASE_PRIVATE_KEY
_29
if (!privateKey) throw new Error(`Expected env var SUPABASE_PRIVATE_KEY`)
_29
_29
const url = process.env.SUPABASE_URL
_29
if (!url) throw new Error(`Expected env var SUPABASE_URL`)
_29
_29
export const run = async () => {
_29
const client = createClient(url, privateKey)
_29
_29
const embeddings = new OpenAIEmbeddings()
_29
_29
const retriever = new SupabaseHybridSearch(embeddings, {
_29
client,
_29
// Below are the defaults, expecting that you set up your supabase table and functions according to the guide above. Please change if necessary.
_29
similarityK: 2,
_29
keywordK: 2,
_29
tableName: 'documents',
_29
similarityQueryName: 'match_documents',
_29
keywordQueryName: 'kw_match_documents',
_29
})
_29
_29
const results = await retriever.getRelevantDocuments('hello bye')
_29
_29
console.log(results)
_29
}

Package Highlights

That's it for the dbdev announcement, but a package index is less interesting than what you can do with it! In that vein, the following highlights a few of packages I thought were interesting enough to callout:

burggraf-pg_headerkit

burggraf-pg_headerkit is a toolkit for adding advanced features to PostgREST APIs (including Supabase REST):

  • rate limiting
  • IP allowlisting/denylisting
  • request logging

and more.

For example, you could apply a deny listing to your API using hdr.in_deny_list() in a row level security policy or view:


_10
select
_10
*
_10
from app.memos
_10
where not hdr.in_deny_list();

olirice-index_advisor

olirice-index_advisor is one of the projects we cut from Launch Week 7. It is simple tool that takes a query and recommends indexes to minimize the “total_cost” according to the query’s explain plan.

We ultimately ran out of time to squeeze the feature in, but the optimizer works just fine:


_15
select dbdev.install('olirice-index_advisor');
_15
create extension if not exists hypopg;
_15
create extension "olirice-index_advisor";
_15
_15
-- Create a dummy table
_15
create table account(
_15
id int primary key,
_15
name text
_15
);
_15
_15
-- Search for indexes to optimize "select id from account where name = 'adsf'"
_15
select
_15
*
_15
from
_15
index_advisor($$select id from account where name = 'Foo'$$)

which shows


_10
| startup_cost_before | startup_cost_after | total_cost_before | total_cost_after | index_statements |
_10
| ------------------- | ------------------ | ----------------- | ---------------- | ----------------------------------------------------- |
_10
| 0.00 | 1.17 | 25.88 | 6.40 | {"CREATE INDEX ON public.account USING btree (name)"} |

In other words, it recommends the index CREATE INDEX ON public.account USING btree (name) which is expected to reduce the total cost from 25.88 to 6.40 for a 4x decrease.

olirice-index_advisor is compatible with tables, views, and materialized views. It can also see through views to find relevant indexes on underlying tables, and supports generic query arguments. For example, $1 in select id from account where name = $1, which makes it compatible with queries from pg_stat_statements and queries generated by the REST API.

Keep an eye open for it in Launch Week 8.

michelp-adminpack

michelp-adminpack is a collection of tools helpful for administrating your database that we often use internally at Supabase. It holds views for reviewing useful info for debugging and optimizing performance like duplicate indexes, index usage, and table size, to name a few.

For example, to identify potentially unused indexes that can be dropped, you could use the index_usage view, which has columns for:

ColumnType
schemanamename
tablenamename
num_rowsbigint
table_sizetext
index_namename
index_sizetext
uniquetext
number_of_scansbigint
tuples_readbigint
tuples_fetchedbigint

Limitations

There are several procedural languages (PL) that can be embedded in PostgreSQL and used to define functions. The ones that ship with stock PostgreSQL are SQL, and pl/pgSQL but there others that can be installed separately, including pl/v8 for JavaScript, or pl/perl for Perl. A trusted language has been restricted to remove potentially hazardous functionality like access to the network stack and file system. pl/v8 and pl/perl are examples of trusted languages. In contrast, pl/python3u is untrusted.

A Trusted Language Extension (TLE) is a PostgreSQL extension, written exclusively using trusted languages. In some ways that makes them less flexible than classic extensions, which can have C language components (more on that in a second). The advantage to TLEs is that they don't require direct access to the PostgreSQL server’s file system to install. That enables TLEs to be installed by end-users rather than by database administrators or hosting providers. TLEs are the enabling technology that allows a package manager like dbdev to function on hosted PostgreSQL platforms like Supabase.

For a more in-depth explanation of Trusted Language Extensions checkout AWS's pg_tle on Supabase blog post or dive into the code at github.com/aws/pg_tle.

A recent development in the PostgreSQL extension ecosystem is the 1.0 release of a new trusted language, pl/rust, allowing users to define SQL functions written in Rust. As a compiled language, pl/rust functions can execute an order of magnitude faster than pl/pgSQL for computationally heavy workloads. That closes the biggest capability gap between native extensions with C components and TLEs. pl/rust hasn’t released to Supabase yet, but we’re excited about rolling it out in the coming weeks.

Please Give Feedback

As this is a preview, we anticipate that there may be a few rough edges. If you do take the time to explore dbdev at this stage, please contribute to its development at github.com/supabase/dbdev.

We are particularly interested in hearing about:

  1. Any issues or bugs you encounter
  2. Feature requests and suggestions for improvement
  3. Contributions in the form of code, documentation, or testing
Share this article

Last post

Supabase Studio 2.0: help when you need it most

14 April 2023

Next post

Trusted Language Extensions for Postgres

14 April 2023

Related articles

Offline-first React Native Apps with Expo, WatermelonDB, and Supabase

Supabase Beta September 2023

Dynamic Table Partitioning in Postgres

Supabase Beta August 2023

pgvector v0.5.0: Faster semantic search with HNSW indexes

Build in a weekend, scale to millions