NPM Package Build Status Dependencies

simply elegant sql for nodejs

this documentation targets the upcoming mesa@1.0.0 release currently in alpha and available on npm as mesa@1.0.0-alpha.*. it's already used in production, is extremely useful, well tested and quite stable !

mesa is a moving target. we are using it in production and it grows and changes with the challenges it helps us solve.

mesa@1.0.0 will be released when it's done !

this documentation does not yet represent everything that is possible with mesa. feel free to look at the code. it's just around 600 lines.

click here for documentation and code of mesa@0.7.1 which will see no further development.


install latest:

npm install --save mesa

mesa needs node-postgres:

npm install --save pg

require both:

var mesa = require('mesa');
var pg = require('pg');


let's tell mesa how to get a database connection for a query:

var database = mesa
  .setConnection(function(cb) {
    pg.connect('postgres://localhost/your-database', cb);

a call to setConnection is the (only) thing tying/coupling the database mesa-object to the node-postgres library and to the specific database.

core ideas and configuration

calling setConnection(callbackOrConnection) has returned a new object. the original mesa-object is not modified:

assert(database !== mesa);

mesa embraces functional programming: no method call on a mesa-object modifies that object. mesa configuration methods are pure: they create a NEW mesa-object copy all OWN properties over to it, set some property and return it.

this has no effect:

  .setConnection(function(cb) {
    pg.connect('postgres://localhost/your-database', cb);

it creates a new object that is not used anywhere and eventually gets garbage collected.

let's configure some tables:

var movieTable = database.table('movie');

var personTable = database.table('person');

there are no special database-objects, table-objects or query-objects in mesa. only mesa-objects that all have the same methods. order of configuration method calls does not matter. you can change anything at any time:

var personTableInOtherDatabase = personTable
  .setConnection(function(cb) {
    pg.connect('postgres://localhost/your-other-database', cb);

it naturally follows that method calls on mesa-objects are chainable !

var rRatedMoviesOfThe2000s = movieTable
  // `where` accepts raw sql and optional parameter bindings
  .where('year BETWEEN ? AND ?', 2000, 2009)
  // repeated calls to where are 'anded' together
  // `where` accepts objects that describe conditions
  .where({rating: 'R'});


the .where() and .having() methods take exactly the same arguments as criterion...

we can always get the SQL and parameter bindings of a mesa-object:

// -> 'SELECT * FROM "movie" WHERE (year BETWEEN ? AND ?) AND (rating = ?)'
// -> [2000, 2009, 'R']



mesa uses mohair to generate sql which it then sends to the database. in addition to it's own methods every mesa-object has the entire interface of a mohair-object. for this reason the mohair methods are not documented in this readme. consult the mohair documentation as well to get the full picture.

mesa supports all methods supported by mohair with some additions. look into mohairs documentation to get the full picture of what's possible with mesa.

**mohair powers mesa's .where


mesa's .where method is one such method that is implemented by mohair mohair uses criterion

for this reason the criterion methods are not documented in this readme.

**criterion powers/documents mesa's .where and .having

we can refine:

var top10GrossingRRatedMoviesOfThe2000s = rRatedMoviesOfThe2000s
  .order('box_office_gross_total DESC')

because every mesa-object gets a copy of all a method added to a mesa-object is available on all mesa-objects down the chain.

this makes it very easy to extend the chainable interface...

movieTable.betweenYears = function(from, to) {
  return this
    .where('year BETWEEN ? AND ?', from to);
}; = function(page, perPage) {
  perPage = perPage ? perPage : 10;
  return this
    .offset(page * perPage);

var paginatedTopGrossingPG13RatedMoviesOfThe90s = movieTable
  // we can freely chain and mix build-in and custom methods !
  .order('box_office_gross_total DESC')
  .where({rating: 'PG13'})
  .betweenYears(1990, 1999);

we see how pure functions and immutability lead to simplicity, reusability and composability !

select queries

we can run a select query on a mesa object and return all results:

  // run a select query and return all results
  // running a query always returns a promise
  .then(function(top10Movies) {

running a query always returns a promise !

we can run a select query on a mesa object and return only the first result:

  // run a select query and return only the first result
  // `first` automatically calls `.limit(1)` to be as efficient as possible
  // running a query always returns a promise
  .then(function(topMovie) {


we can also simply check whether a query returns any records:

  .where({name: 'Moon'})
  // running a query always returns a promise
  .then(function(exists) {


insert queries

we can run an insert query on a mesa object:

  // whitelist some properties to prevent mass assignment
  .insert({name: 'Moon'})
  // running a query always returns a promise
  // if insert is called with a single object only the first inserted object is returned
  .then(function(insertedMovie) {

before running insert queries

if you have control over the properties of the inserted objects and can ensure that no properties can disable this by calling .unsafe(). you can reenable it by calling .unsafe(false).

you can insert multiple records by passing multiple arguments and/or arrays to insert:

  // disable mass-assignment protection
  // running a query always returns a promise
    {name: ''}
      {name: ''}
      {name: ''}
    {name: ''}
  .then(function(insertedMovies) {

you see that mesa returns the inserted records by default

update queries

This part is coming soon.

delete queries

This part is coming soon.

sql fragments

an sql fragment is any object with the following properties:

an sql([escape]) method which returns an sql string takes an optional parameter escape which is a function to be used to escape column and table names in the resulting sql a params() method which returns an array

at the heart of mesa is the query method

if you pass mesa (which is an sql fragment) into the query function...

connections revisited

This part is coming soon.

setConnection either accepts


all of sql

down to the metal


mesaWithDebug = mesa.debug(function( , detail, state, verboseState, instance)

only on refined versions

intermediary results

debugging per table, per query, ...

directly before a query debug will just for that specific query

just display sql

mesa = mesa.debug(function(topic, query, data)
  if (topic === 'query' && event === 'before') {
    console.log('QUERY', data.sql, data.params);

the topics are connection, query, transaction, find, embed

that function will be called with five arguments

the first argument is

the fifth argument is the instance

the fourth argument contains ALL additional local state that is connection, arguments

here is a quick overview:

look into the source to see exactly which


often you want to for all tables, a specific table a specific do something to the records

you can configure mesa instances.

you can add functions to the queues with the following ways

hooks either run on a the array of all items or item

array queues are run before

functions in queues are run in the order they were added.

there are the following queues:

  • queueBeforeInsert run before insert on array of items

  • there is no queueBeforeUpdate because update always operates on a single item. use queueBeforeEachUpdate

  • queueBeforeEachInsert run before insert on each item

  • queueBeforeEachUpdate run before update on each item

  • queueBeforeEach run before update or insert on each item

  • queueAfterSelect run after find or first on array of items

  • queueAfterInsert run after insert on array of items

  • queueAfterUpdate run after update on array of items

  • queueAfterDelete run after delete on array of items

  • queueAfter run after find, first, insert, update and delete on array of items

  • queueAfterEachSelect run after find or first on each item

  • queueAfterEachInsert run after insert on each item

  • queueAfterEachUpdate run after update on each item

  • queueAfterEachDelete run after delete on each item

  • queueAfterEach run after find, first, insert, update and delete on each item

nice things you can to with queueing:

omit password property when a user is returned

var _ = require('lodash');

  .queueAfterEachSelect(_.omit, 'password')
  .where({id: 3})
  .first(function(user) {

hash password before user is inserted or updated

var Promise = require('bluebird');
var bcrypt = Promise.promisifyAll(require('brypt'));

var hashPassword = function(record) {
  if (record.password) {
    bcrypt.genSaltAsync(10).then(function(salt) {
      return bcrypt.hashAsync(password, salt);
  } else {
    return Promise.resolve(null);

userTable = userTable.queueBeforeEach(hashPassword);

convert property names between camelCase and snake_case

see example/

set columns like created_at and updated_at automatically

userTable = userTable
  .queueBeforeEach(function(record) {
    record.updated_at = new Date();
    return record;
  .queueBeforeInsert(function(record) {
    record.created_at = new Date();
    return record;

fetch associated data

see includes

protect from mass assignment

mesa comes with a very powerful mechanism to manipulate records before they are sent to the database or after they were received from the database and before returning them.

if you are familiar with the active record pattern prefer a more object-oriented style here is how you would use mesa to implement it as the foundation as the building blocks

if you want to use camelcased property names in your program and underscored in your database you can automate the translation

add them to the mesa instance and have it work for all your tables

by setting the order you ensure that the other hooks see camelcased properties !!!


includes are a NEW feature and may not be as stable as the rest

in any rows in different tables are linked via foreign keys.

includes make it easy to fetch those linked rows and add them to our data:

lets assume, for a moment, the following tables and relationships:

  • user with columns id, name. has one address, has many orders
  • address with columns id, street, city, user_id. belongs to user via foreign key user_id ->
  • order with columns id, status. belongs to user
userTable = database.table('user');
addressTable = database.table('address');
orderTable = database.table('order');

we can now find some users and include the orders in each of them:

has many relationship

  .find(function(users) {


a lot is happening here. let's break it down:

include has no side-effects and does not fetch any data. instead it queues a function to be executed on all results (if any) of first, find, insert, delete and update queries further down the chain.

in this case that function will will run a query on orderTable to fetch all orders where order.user_id is in the list of all id values in users. it will then for every user add as property orders the list of all orders where === order.user_id.

by default include queues a fetch of a has-many relationship

the above code snippet is equivalent to this:

    left: 'id',
    right: 'user_id',
    forward: true,
    first: false,
    as: 'orders'
  }, orderTable)
  .find(function(users) {


the first argument to

in case that link-object is missing or any properties are missing (and only those fields) mesa will autocomplete it from table names , primary keys set with .primaryKey(key)

belongs to relationship

  .include({forward: false, first: true}, userTable)
  .find(function(users) {


has many through

you can add as many additional link

you can modify, add conditions

you can nest

using an explicit link object:

you get the idea

includes are intentionally very flexible. they work with any two tables where the values in whose values match up.

if you are using primary keys other than id

fetch a one-to-one association (in a single additional query)

the implementation uses the hooks its surprisingly simple

using the same connection as the

use one additional query to fetch all and then associate them with the records

order and conditions and limits on the other tables have their full effects


using mesa you'll often find yourself calling methods only when certain conditions are met:

var dontFindDeleted = true;
var pagination = {page: 4, perPage: 10};

var tmp = userTable;

if (dontFindDeleted) {
  tmp = userTable.where({is_deleted: false});

if (pagination) {
  tmp = tmp
    .offset( * pagination.perPage);

tmp.find(function(users) {

all those temporary objects are not very nice.

fortunately there is another way:

  .when(dontFindDeleted, userTable.where, {is_deleted: false})
  .when(pagination, function() {
    return this
      .offset( * pagination.perPage);
  .find(function(users) {

mesa by example


TL;DR: bugfixes, issues and discussion are always welcome. ask me before implementing new features.

i will happily merge pull requests that fix bugs with reasonable code.

i will only merge pull requests that modify/add functionality if the changes align with my goals for this package and only if the changes are well written, documented and tested.

communicate: write an issue to start a discussion before writing code that may or may not get merged.

license: MIT