npm package discovery and stats viewer.

Discover Tips

  • General search

    [free text search, go nuts!]

  • Package details

    pkg:[package-name]

  • User packages

    @[username]

Sponsor

Optimize Toolset

I’ve always been into building performant and accessible sites, but lately I’ve been taking it extremely seriously. So much so that I’ve been building a tool to help me optimize and monitor the sites that I build to make sure that I’m making an attempt to offer the best experience to those who visit them. If you’re into performant, accessible and SEO friendly sites, you might like it too! You can check it out at Optimize Toolset.

About

Hi, 👋, I’m Ryan Hefner  and I built this site for me, and you! The goal of this site was to provide an easy way for me to check the stats on my npm packages, both for prioritizing issues and updates, and to give me a little kick in the pants to keep up on stuff.

As I was building it, I realized that I was actually using the tool to build the tool, and figured I might as well put this out there and hopefully others will find it to be a fast and useful way to search and browse npm packages as I have.

If you’re interested in other things I’m working on, follow me on Twitter or check out the open source projects I’ve been publishing on GitHub.

I am also working on a Twitter bot for this site to tweet the most popular, newest, random packages from npm. Please follow that account now and it will start sending out packages soon–ish.

Open Software & Tools

This site wouldn’t be possible without the immense generosity and tireless efforts from the people who make contributions to the world and share their work via open source initiatives. Thank you 🙏

© 2024 – Pkg Stats / Ryan Hefner

reactive-postgres

v0.2.0

Published

Reactive queries for PostgreSQL

Downloads

26

Readme

reactive-postgres

This node.js package brings reactive (or live) queries to PostgreSQL. You can take an arbitrary SELECT query, using multiple joins, data transformations, and even custom functions, and besides the initial set of results also get real-time updates about any changes to those results. This can enable you to keep UI in sync with the database in a reactive manner.

Installation

This is a node.js package. You can install it using NPM:

$ npm install reactive-postgres

This package uses PostgreSQL. See documentation for more information how to install and use it.

Requires PostgreSQL 11 or newer.

reactive-postgres for enterprise

Available as part of the Tidelift Subscription.

The maintainers of reactive-postgres and thousands of other packages are working with Tidelift to deliver commercial support and maintenance for the open source dependencies you use to build your applications. Save time, reduce risk, and improve code health, while paying the maintainers of the exact dependencies you use. Learn more.

Usage

As an event emitter:

const {Manager} = require('reactive-postgres');

const manager = new Manager({
  connectionConfig: {
    user: 'dbuser',
    host: 'database.server.com',
    database: 'mydb',
    password: 'secretpassword',
    port: 3211,
  },
});

(async () => {
  await manager.start();

  const handle = await manager.query(`SELECT * FROM posts`, {
    uniqueColumn: 'id',
    mode: 'changed',
  });

  handle.on('start', () => {
    console.log("query has started");
  });

  handle.on('ready', () => {
    console.log("initial data have been provided");
  });

  handle.on('refresh', () => {
    console.log("all query changes have been provided");
  });

  handle.on('insert', (row) => {
    console.log("row inserted", row);
  });

  handle.on('update', (row, columns) => {
    console.log("row updated", row, columns);
  });

  handle.on('delete', (row) => {
    console.log("row deleted", row);
  });

  handle.on('error', (error) => {
    console.error("query error", error);
  });

  handle.on('stop', (error) => {
    console.log("query has stopped", error);
  });

  handle.start();
})().catch((error) => {
  console.error("async error", error);
});

process.on('SIGINT', () => {
  manager.stop().catch((error) => {
    console.error("async error", error);
  }).finally(() => {
    process.exit();
  });
});

As a stream:

const {Manager} = require('reactive-postgres');
const through2 = require('through2');

const manager = new Manager({
  connectionConfig: {
    user: 'dbuser',
    host: 'database.server.com',
    database: 'mydb',
    password: 'secretpassword',
    port: 3211,
  },
});

const jsonStream = through2.obj(function (chunk, encoding, callback) {
  this.push(JSON.stringify(chunk, null, 2) + '\n');
  callback();
});

(async () => {
  await manager.start();

  const handle = await manager.query(`SELECT * FROM posts`, {
    uniqueColumn: 'id',
    mode: 'changed',
  });

  handle.on('error', (error) => {
    console.error("stream error", error);
  });

  handle.on('close', () => {
    console.log("stream has closed");
  });

  handle.pipe(jsonStream).pipe(process.stdout);
})().catch((error) => {
  console.error("async error", error);
});

process.on('SIGINT', () => {
  manager.stop().catch((error) => {
    console.error("async error", error);
  }).finally(() => {
    process.exit();
  });
});

Design

Reactive queries are implemented in the following manner:

  • For every reactive query, a TEMPORARY TABLE is created in the database which serves as cache for query results.
  • Moreover, the query is PREPAREd, so that it does not have to be parsed again and again.
  • Triggers are added to all query sources for the query, so that when any of sources change, this package is notified using LISTEN/NOTIFY that a source has changed, which can potentially influence the results of the query.
  • Package waits for source changed events, and throttles them based on refreshThrottleWait option. Once the delay expires, the package creates a new temporary table with new query results. It compares the old and new table and computes changes using another database query.
  • Changes are returned the package's client and exposed to the user.
  • The old temporary table is dropped and the new one is renamed to take its place.

Performance

  • Memory use of node.js process is very low, is constant and does not grow with the number of rows in query results. Moreover, node.js process also does no heavy computation and mostly just passes data around. All this is achieved by caching a query using a temporary table in the database instead of the client, and using a database query to compare new and old query results.
  • Computing changes is done through one query, a very similar query to the one used internally by REFRESH MATERIALIZED VIEW CONCURRENTLY PostgreSQL command.
  • Based on the design, the time to compute changes and provide them to the client seems to be the lowest when compared with other similar packages. For more information about performance comparisons of this package and related packages, see this benchmark tool and results at the end.
  • Because this package uses temporary tables, consider increasing temp_buffers PostgreSQL configuration so that there is more space for temporary tables in memory. Consider creating a dedicated tablespace and configuring temp_tablespaces.
  • You might consider increasing refreshThrottleWait for reactive queries for which you can tolerate lower refresh rate and higher update latency, to decrease load on the database. Making too many refreshes for complex queries can saturate the database which then leads to even higher delays. Paradoxically, having higher refreshThrottleWait could give you lower delay in comparison with a saturated state.
  • Multiple reactive queries share the same connection to the database. So correctly configuring maxConnections is important. More connections there are, higher load is on the database, but over more connections reactive queries can spread. But higher load on the database can lead to its saturation.
  • Currently, when any of sources change in any manner, whole query is rerun and results compared with cached results (after a throttling delay). To improve this, refresh could be done only when it is known that a source change is really influencing the results. Ideally, we could even compute changes to results directly based on changes to sources. See #7 for more information.

Limitations

  • Queries require an unique column which serves to identify rows and changes to them. Which column this is is configured through uniqueColumn query option. By default is id.
  • Order of rows in query results are ignored when determining changes. Order still matters when selecting which rows are in query results through ORDER BY X LIMIT Y pattern. If you care about order of query results, order rows on the client.
  • Queries cannot contain placeholders or be prepared. You can use client.escapeLiteral(...) function to escape values when constructing a query.
  • To be able to determine if an UPDATE query really changed any source tables used in your reactive query, source tables should have only columns of types which have an equality operator defined (e.g., json column type does not). If this is not so, it is just assumed that every UPDATE query makes a change.

API

Manager

Manager manages a pool of connections to the database and organizes reactive queries over them. You should initialize one instance of it and use it for your reactive queries

constructor([options])

Available options:

  • maxConnections, default 10: the maximum number of connections to the database for reactive queries, the final number of connections is maxConnections + 1, for the extra manager's connection
  • connectionConfig, default {}: PostgreSQL connection configuration, more information
  • handleClass: default ReactiveQueryHandle: a class to use for reactive query handles returned by query method
async start()

Initializes the manager, the database, and establishes its connection. It emits 'start' event.

async stop([error])

Stops all reactive queries and the manager itself. It emits 'stop' event. Accepts optional error argument which is passed as payload in 'stop' event.

async query(query[, options={}])

Constructs a new reactive query and returns a handle.

query can be any arbitrary SELECT, TABLE, or VALUES command, adhering to limitations.

Available options:

  • uniqueColumn, default id: the name of an unique column in query results, used to identify rows and changes to them
  • refreshThrottleWait, default 100: this option controls that refresh can happen at most once per every refreshThrottleWait milliseconds
    • this introduces a minimal delay between a source change and a refresh, you can control this delay based on requirements for a particular query
    • lower this value is, higher the load on the database will be, higher it is, lower the load will be
  • mode, default changed: in which mode to operate, it can be:
    • columns: for every query results change, provide only which row and columns changed
    • changed: for every query results change, provide new values for changed columns, too
    • full: for every query results change, provide full changed rows, both columns which have changed and those which have not
  • types, default null: custom type parsers
'start' event ()

Event emitted when manager starts successfully.

'connect' event (client)

Event emitted when a new PostgreSQL client is created and connected. client is provided as an argument.

'disconnect' event (client)

Event emitted when a PostgreSQL client is disconnected. client is provided as an argument.

'error' event (error[, client])

Event emitted when there is an error at the manager level. error is provided as an argument. If the error is associated with a PostgreSQL client, the client is provided as well.

'stop' event ([error])

Event emitted when the manager stops. If it stopped because of an error, the error is provided as an argument.

ReactiveQueryHandle

Reactive query handles can be used as an EventEmitter or a Readable stream, but not both.

Constructor is seen as private and you should not create instances of ReactiveQueryHandle yourself but always through Manager's query method. The method also passes all options to ReactiveQueryHandle.

EventEmitter API

There are more methods, properties, and options available through the EventEmitter base class.

async start()

Initializes the reactive query and starts observing the reactive query, emitting events for initial results data and later changes to results data. It emits 'start' event.

Having this method separate from constructing a reactive query allows attaching event handlers before starting, so that no events are missed.

async stop([error])

Stops the reactive query. It emits 'stop' event. Accepts optional error argument which is passed as payload in 'stop' event.

async refresh()

Forces the refresh of the reactive query, computation of changes, and emitting relevant events. This can override refreshThrottleWait option which otherwise controls the minimal delay between a source change and a refresh.

async flush()

When operating in changed or full mode, changes are batched together before a query to fetch data is made. This method forces the query to be made using currently known changes instead of waiting for a batch as configured by the batchSize option.

on(eventName, listener) and addListener(eventName, listener)

Adds the listener function to the end of the listeners array for the event named eventName. More information.

once(eventName, listener)

Adds a one-time listener function for the event named eventName. More information.

off(eventName, listener) and removeListener(eventName, listener)

Removes the specified listener from the listener array for the event named eventName. More information.

'start' event ()

Event emitted when reactive query starts successfully.

'ready' event ()

Event emitted when all events for initial results data have been emitted. Later events are about changes to results data.

'refresh' event ()

Event emitted when a refresh has finished and all events for changes to results data as part of one refresh have been emitted.

'insert' event (row)

Event emitted when a new row has been added to the reactive query. row is provided as an argument. In columns mode, row contains only the value of the unique column of the row which has been inserted. In changed and full modes, row contains full row which has been inserted.

'update' event (row, columns)

Event emitted when a row has been updated. row is provided as an argument. In columns mode, row contains only the value of the unique column of the row which has been updated. In changed mode, row contains also data for columns which have changed. In full mode, row contains the full updated row, both columns which have changed and those which have not. columns is a list of columns which have changed.

'delete' event (row)

Event emitted when a row has been deleted. row is provided as an argument. In columns mode, row contains only the value of the unique column of the row which has been deleted. In changed and full modes, row contains full row which has been deleted.

'error' event (error)

Event emitted when there is an error at the reactive query level. error is provided as an argument.

'stop' event ([error])

Event emitted when the reactive query stops. If it stopped because of an error, the error is provided as an argument.

Readable stream API

There are more methods, properties, and options available through the Readable stream base class.

When operating as a stream, the reactive handle is producing objects describing a change to the reactive query. They are of the following structure:

{
  op: <event name>,
  ... <payload> ...
}

event name matches names of events emitted when operating as an event emitter. Arguments of those events are converted to object's payload.

Stream supports backpressure and if consumer of the stream reads slower than what refreshThrottleWait dictates should be the minimal delay between refreshes, further refreshing is paused until stream is drained.

read([size])

Reads the next object describing a change to the reactive query, if available. More information.

pipe(destination[, options])

The method attaches the stream to the destination, using options. More information.

destroy([error])

Destroy the stream and stop the reactive query. It emits 'error' (if error argument is provided, which is then passed as payload in 'error' event) and 'close' events. More information.

'data' event (chunk)

Event emitted whenever the stream is relinquishing ownership of a chunk of data to a consumer. chunk is provided as an argument. More information.

'readable' event ()

Event emitted when there is data available to be read from the stream. More information.

'error' event (error)

Event emitted when there is an error at the reactive query level. error is provided as an argument. More information.

'close' event ()

Event emitted when the stream has been destroyed. More information.

Related projects

  • pg-live-select – when the query's sources change, the client reruns the query to obtain updated rows, which are identified through hashes of their full content and client then computes changes, on the other hand, this package maintains cached results of the query in a temporary table in the database and just compares new results with cached results and return updates to the client
  • pg-live-query – adds revision columns to sources and additional temporary table to store latest revisions for results, moreover, it rewrites queries to expose those additional revision columns, all this then allows simpler determination of what has changed when the query is refreshed, inside the database through a query, but in benchmarks it seems all this additional complexity does not really make things faster in comparison with just comparing new results with cached results directly, which is what this package does
  • pg-query-observer – it seems like a bit cleaned and updated version of pg-live-select, but buggy and does not work with multiple parallel queries
  • pg-reactivity-benchmark – a benchmark for this and above mentioned packages