Here is my situation:
I'm using postgres 9.4, Sequelize ORM and have following models:
Service
serviceCode - primary key, string of 6 characters
serviceTitle - string
ServiceGroup
serviceCodePrefixes - array of strings that are prefixes for Service.serviceCode
serviceGroupTitle - string
Task
serviceCode - reference to Service
I need to build Task object populated with Service and ServiceGroup objects. Example:
In database:
Service {
serviceCode: '123232',
serviceTitle: 'svc title #1',
}
ServiceGroup {
serviceCodePrefix: ['12', '13', '92', ...],
serviceGroupTitle: 'svc grp title #1',
}
Task {
serviceCode: '123232',
}
Result:
Task {
service: {
serviceTitle: 'svc title #1',
},
serviceGroup: {
serviceGroupTitle: 'svc grp title #1',
},
}
The problem is that serviceCodePrefix contains not simple IDs, which can be used to create association using hasOne/belongsTo/etc., but prefix for ID.
So questions is: how this can be done without raw sql?
Turns out that right now Sequelize has experimental feature: 'on' option for 'include'. This option allows users to customize joining conditions. So my problem can be solved this way:
const Service = sequelize.define('service', {
serviceTitle: Sequelize.STRING,
serviceCode: Sequelize.STRING,
});
const ServiceGroup = sequelize.define('service_group', {
serviceGroupTitle: Sequelize.STRING,
// Array of prefixes (e.g. ['01%', '023%'])
serviceCodePrefix: Sequelize.ARRAY(Sequelize.STRING),
});
const Task = sequelize.define('task', {
taskTitle: Sequelize.STRING,
serviceCode: Sequelize.STRING,
});
Task.belongsTo(Service, { foreignKey: 'serviceCode' });
// Hack needed to allow 'include' option to work
Task.hasMany(ServiceGroup, { foreignKey: 'serviceCodePrefix', constraints: false });
// And finally
Task.findAll({
include: [
{ model: Service },
{
model: ServiceGroup,
on: [' "task"."serviceCode" LIKE ANY("serviceGroup"."serviceCodePrefix") '],
},
],
});
Not sure about the performance though.
Related
I am trying to create an audit trail using Apollo Server and Mongoose. When a user initially registers, I create a document in the users collection and a document in the history collection for each piece of data they provided (username, password, email, etc) . For each history collection document, I include the id for the user document to create a relationship. Works perfectly.
However, when I add a transaction in (see below), the userId for the user document comes back as undefined, so I cannot add it to the history entry documents. I am assuming that the id for a document does not get created until the entire transaction has been completed?
Any ideas?
Mutation: {
register: async (_, { data }) => {
// Start a mongo session & transaction
const session = await mongoose.startSession()
session.startTransaction()
try {
// Hash password and create user
const hashedPassword = await bcrypt.hash(data.password, 12)
const user = await User.create(
[{ ...data, password: hashedPassword }],
{ session }
)
// Add history entries
HistoryEntry.create([
{
user: user.id,
action: 'registered'
},
{
user: user.id,
action: 'set',
object: 'profile',
instance: user.id,
property: 'firstName',
value: firstName
},
{
user: user.id,
action: 'set',
object: 'profile',
instance: user.id,
property: 'lastName',
value: lastName
},
{
user: user.id,
action: 'set',
object: 'profile',
instance: user.id,
property: 'password'
}
])
if (loginType === 'email') {
HistoryEntry.create({
user: user.id,
action: 'set',
object: 'profile',
instance: user.id,
property: 'email',
value: login
})
}
if (loginType === 'mobile') {
HistoryEntry.create({
user: user.id,
action: 'set',
object: 'profile',
instance: user.id,
property: 'mobile',
value: login
})
}
// commit the changes if everything was successful
await session.commitTransaction()
return {
ok: true,
user
}
} catch (err) {
// if anything fails above, rollback the changes in the transaction
await session.abortTransaction()
return formatErrors(err)
} finally {
// end the session
session.endSession()
}
}
}
If you think about it, how can you add a HistoryEntry if you haven't added User yet? It's not a 'history' as you are currently doing it. I believe you got two options here - set _id on User manually new Schema({ _id: { type: Schema.ObjectId, auto: true }}) and then generate it within the transaction: var userId = ObjectId(); and use for both User and History Entries.
And the second option, more semantically correct in this context, I believe - you should attach to post-save hook:
schema.post('save', function(doc) {
console.log('%s has been saved', doc._id);
});
So, whenever an User is created, a post-save hook is fired to update History.
Came across the same issue recently, hope you have figured it out already. I may add this for future seekers.
Following create function returns an array of created documents.
const user = await User.create(
[{ ...data, password: hashedPassword }],
{ session }
);
Therefore access the user id as user[0]._id
Pass the session also to HistoryEntry.create()
HistoryEntry.create([{...},{...}], {session})
Note: In this use case, I personally prefer #marek second option to use a post-save hook.
GraphQL has mutations, Postgres has INSERT; GraphQL has queries, Postgres has SELECT's; etc., etc.. I haven't found an example showing how you could use both in a project, for example passing all the queries from front end (React, Relay) in GraphQL, but to a actually store the data in Postgres.
Does anyone know what Facebook is using as DB and how it's connected with GraphQL?
Is the only option of storing data in Postgres right now to build custom "adapters" that take the GraphQL query and convert it into SQL?
GraphQL is database agnostic, so you can use whatever you normally use to interact with the database, and use the query or mutation's resolve method to call a function you've defined that will get/add something to the database.
Without Relay
Here is an example of a mutation using the promise-based Knex SQL query builder, first without Relay to get a feel for the concept. I'm going to assume that you have created a userType in your GraphQL schema that has three fields: id, username, and created: all required, and that you have a getUser function already defined which queries the database and returns a user object. In the database I also have a password column, but since I don't want that queried I leave it out of my userType.
// db.js
// take a user object and use knex to add it to the database, then return the newly
// created user from the db.
const addUser = (user) => (
knex('users')
.returning('id') // returns [id]
.insert({
username: user.username,
password: yourPasswordHashFunction(user.password),
created: Math.floor(Date.now() / 1000), // Unix time in seconds
})
.then((id) => (getUser(id[0])))
.catch((error) => (
console.log(error)
))
);
// schema.js
// the resolve function receives the query inputs as args, then you can call
// your addUser function using them
const mutationType = new GraphQLObjectType({
name: 'Mutation',
description: 'Functions to add things to the database.',
fields: () => ({
addUser: {
type: userType,
args: {
username: {
type: new GraphQLNonNull(GraphQLString),
},
password: {
type: new GraphQLNonNull(GraphQLString),
},
},
resolve: (_, args) => (
addUser({
username: args.username,
password: args.password,
})
),
},
}),
});
Since Postgres creates the id for me and I calculate the created timestamp, I don't need them in my mutation query.
The Relay Way
Using the helpers in graphql-relay and sticking pretty close to the Relay Starter Kit helped me, because it was a lot to take in all at once. Relay requires you to set up your schema in a specific way so that it can work properly, but the idea is the same: use your functions to fetch from or add to the database in the resolve methods.
One important caveat is that the Relay way expects that the object returned from getUser is an instance of a class User, so you'll have to modify getUser to accommodate that.
The final example using Relay (fromGlobalId, globalIdField, mutationWithClientMutationId, and nodeDefinitions are all from graphql-relay):
/**
* We get the node interface and field from the Relay library.
*
* The first method defines the way we resolve an ID to its object.
* The second defines the way we resolve an object to its GraphQL type.
*
* All your types will implement this nodeInterface
*/
const { nodeInterface, nodeField } = nodeDefinitions(
(globalId) => {
const { type, id } = fromGlobalId(globalId);
if (type === 'User') {
return getUser(id);
}
return null;
},
(obj) => {
if (obj instanceof User) {
return userType;
}
return null;
}
);
// a globalId is just a base64 encoding of the database id and the type
const userType = new GraphQLObjectType({
name: 'User',
description: 'A user.',
fields: () => ({
id: globalIdField('User'),
username: {
type: new GraphQLNonNull(GraphQLString),
description: 'The username the user has selected.',
},
created: {
type: GraphQLInt,
description: 'The Unix timestamp in seconds of when the user was created.',
},
}),
interfaces: [nodeInterface],
});
// The "payload" is the data that will be returned from the mutation
const userMutation = mutationWithClientMutationId({
name: 'AddUser',
inputFields: {
username: {
type: GraphQLString,
},
password: {
type: new GraphQLNonNull(GraphQLString),
},
},
outputFields: {
user: {
type: userType,
resolve: (payload) => getUser(payload.userId),
},
},
mutateAndGetPayload: ({ username, password }) =>
addUser(
{ username, password }
).then((user) => ({ userId: user.id })), // passed to resolve in outputFields
});
const mutationType = new GraphQLObjectType({
name: 'Mutation',
description: 'Functions to add things to the database.',
fields: () => ({
addUser: userMutation,
}),
});
const queryType = new GraphQLObjectType({
name: 'Query',
fields: () => ({
node: nodeField,
user: {
type: userType,
args: {
id: {
description: 'ID number of the user.',
type: new GraphQLNonNull(GraphQLID),
},
},
resolve: (root, args) => getUser(args.id),
},
}),
});
We address this problem in Join Monster, a library we recently open-sourced to automatically translate GraphQL queries to SQL based on your schema definitions.
This GraphQL Starter Kit can be used for experimenting with GraphQL.js and PostgreSQL:
https://github.com/kriasoft/graphql-starter-kit - Node.js, GraphQL.js, PostgreSQL, Babel, Flow
(disclaimer: I'm the author)
Have a look at graphql-sequelize for how to work with Postgres.
For mutations (create/update/delete) you can look at the examples in the relay repo for instance.
Postgraphile https://www.graphile.org/postgraphile/ is Open Source
Rapidly build highly customisable, lightning-fast GraphQL APIs
PostGraphile is an open-source tool to help you rapidly design and
serve a high-performance, secure, client-facing GraphQL API backed
primarily by your PostgreSQL database. Delight your customers with
incredible performance whilst maintaining full control over your data
and your database. Use our powerful plugin system to customise every
facet of your GraphQL API to your liking.
You can use an ORM like sequelize if you're using Javascript or Typeorm if you're using Typescript
Probably FB using mongodb or nosql in backend. I've recently read a blog entry which explain how to connect to mongodb. Basically, you need to build a graph model to match the data you already have in your DB. Then write resolve, reject function to tell GQL how to behave when posting a query request.
See https://www.compose.io/articles/using-graphql-with-mongodb/
Have a look at SequelizeJS which is a promise based ORM that can work with a number of dialects; PostgreSQL, MySQL, SQLite and MSSQL
The below code is pulled right from its example
const Sequelize = require('sequelize');
const sequelize = new Sequelize('database', 'username', 'password', {
host: 'localhost',
dialect: 'mysql'|'sqlite'|'postgres'|'mssql',
pool: {
max: 5,
min: 0,
acquire: 30000,
idle: 10000
},
// SQLite only
storage: 'path/to/database.sqlite',
// http://docs.sequelizejs.com/manual/tutorial/querying.html#operators
operatorsAliases: false
});
const User = sequelize.define('user', {
username: Sequelize.STRING,
birthday: Sequelize.DATE
});
sequelize.sync()
.then(() => User.create({
username: 'janedoe',
birthday: new Date(1980, 6, 20)
}))
.then(jane => {
console.log(jane.toJSON());
});
I have three schemas, one which references two others:
userSchema
{ name: String }
postSchema
{ content: String }
commentSchema
{
content: String,
user: { ObjectID, ref: 'User' },
post: { ObjectID, ref: 'Post' }
}
How can I seed this database in a sane, scalable way? Even using bluebird promises it quickly becomes a nightmare to write.
My attempt so far involves multiple nested promises and is very hard to maintain:
User
.create([{ name: 'alice' }])
.then(() => {
return Post.create([{ content: 'foo' }])
})
.then(() => {
User.find().then(users => {
Post.find().then(posts => {
// `users` isn't even *available* here!
Comment.create({ content: 'bar', user: users[0], post: posts[0] })
})
})
})
This is clearly not the correct way of doing this. What am I missing?
Not sure about bluebird, but the nodejs Promise.all should do the job:
Promise.all([
User.create([{ name: 'alice' }]),
Post.create([{ content: 'foo' }])
]).then(([users, posts]) => {
const comments = [
{ content: 'bar', user: users[0], post: posts[0] }
];
return Comment.create(comments);
})
If you want to seed database with automatically references, use Seedgoose.
This is the easiest seeder for you to use. You don't need to write any program files, but only data files. And Seedgoose handles smart references for you. And by the way, I'm the author and maintainer of this package.
Try this it will work fine:
Note: Node Promise.all will make sure that the both query is executed properly and then return the result in Array:[Users, Posts],
If you get any error during execution of any query, it will be handle by catch block of the Promise.all.
let queryArray = [];
queryArray.push(User.create([{ name: 'alice' }]));
queryArray.push(Post.create([{ content: 'foo' }]));
Promise.all(queryArray).then(([Users, Posts]) => {
const comments = [
{ content: 'bar', user: Users[0], post: posts[0] }
];
return Comment.create(comments);
}).catch(Error => {
console.log("Error: ", Error);
})
My SQL query is
SELECT t."id", t."topicText", COUNT(t."id") AS topicCount
FROM "WorkbookQuestions" wq, "Workbooks" w, "Questions" q, "Topics" t
WHERE w."StudentId" = 3
AND wq."QuestionId" = q."id"
AND wq."WorkbookId" = w."id"
AND q."TopicId" = t."id"
GROUP BY "t"."id"
ORDER BY "topiccount" DESC
How can I write this in Sequelize code so I can avoid writing SQL directly? Or is it not possible?
This is what I've tried (unsuccessfully) so far:
db.WorkbookQuestion.findAll({
attributes: ['Topics.id', [db.sequelize.fn('COUNT', db.sequelize.col('Topics.id'))], 'TopicCount'],
include: [
{
model: db.Workbook,
where: {
StudentId: newWorkbook.StudentId,
SubjectId: newWorkbook.SubjectId
}
}, {
model: db.Question,
include: [
{
model: db.Topic
}
]
}
],
group: ['Topics.id']
});
Just by looking at it, i would write the code like this
db.Workbook.findAll({
attributes: ['wbquestion.question.topic.id', [db.sequelize.fn('COUNT', db.sequelize.col('wbquestion.question.topic.id'))], 'topiccount'],
include: [
{
model: db.WorkbookQuestion,
as: wbquestion,
where: {
id: wbquestion.workbookId
},
include: [
model: db.Question,
as: question,
where: {
id: wbquestion.questionId
},
include: [
model: db.Topic,
as: topic,
where: {
id: question.topicId
}
]
]
}
],
where: {
studentId : 3
}
group: ['topic.id']
});
Maybe you can get more readable code by defining the models and relations separately and using eager loading afterwards.
Edit:
Generated the migrations and models
node_modules/.bin/sequelize model:create --name Student --attributes firstName:string,lastName:string
node_modules/.bin/sequelize model:create --name Topic --attributes topicText:string
node_modules/.bin/sequelize model:create --name Question --attributes TopicId:integer
node_modules/.bin/sequelize model:create --name Workbook --attributes studentId:integer
node_modules/.bin/sequelize model:create --name WorkbookQuestion --attributes QuestionId:integer,WorkbookId:integer
A simple program for testing the query in console log.
var Sequelize = require("sequelize");
var models = require('./models/index');
models.WorkbookQuestion.belongsTo(models.Question, {foreignKey: 'QuestionId'});
models.WorkbookQuestion.belongsTo(models.Workbook, {foreignKey: 'WorkbookId'});
models.Question.hasOne(models.WorkbookQuestion);
models.Workbook.hasOne(models.WorkbookQuestion);
models.Workbook.belongsToMany(models.Question, {through: models.WorkbookQuestion});
models.Workbook.belongsToMany(models.Topic, {through: models.Question});
var res = models.Workbook.findAll({
attributes: ['topicId', Sequelize.fn('count', Sequelize.col('topiccount'))],
include: [models.WorkbookQuestion, models.Question, models.Topic],
where: {
studentId: 3
},
group: ['topic.id']
}).then(function(data) {
console.log(data);
}).catch(function(err) {
console.trace(err);
});
The result is:
SequelizeDatabaseError: column Topics.Question.WorkbookId does not exist
Maybe you can fix the relations between the models or you should create the query not using findAll functionality.
I have the following in my Sailsjs config/adapter.js:
module.exports.adapters = {
'default': 'postgres',
postgres : {
module : 'sails-postgresql',
host : 'xxx.compute-1.amazonaws.com',
port : 5432,
user : 'xxx',
password : 'xxx',
database : 'xxx',
ssl : true,
schema : true
}
};
And in models/Movie.js:
Movie = {
attributes: {
tableName: 'movies.movies',
title: 'string',
link: 'string'
}
};
module.exports = Movie;
In my controller:
Movie.query("SELECT * FROM movies.movies", function(err, movies) {
console.log('movies', movies.rows);
});
movies.rows DOES return the correct data
However:
Movie.find({ title: 'Frozen' }, function(err, movies) {
console.log('movies', movies)
});
movies returns an EMPTY ARRAY
So it seems all connections are good because the raw query works perfectly.
Could there be something I am doing wrong with setting up the Movie.find() or with models/Movie.js?
Does the tableName attribute not support postgresql schema_name.table_name?
First off, you need to move tableName out of attributes, since it's a class-level property. Second, sails-postgresql does have some (very undocumented) support for schemas, using the meta.schemaName option:
Movie = {
tableName: 'movies',
meta: {
schemaName: 'movie'
},
attributes: {
title: 'string',
link: 'string'
}
};
module.exports = Movie;
You can give that a try, and if it doesn't work, either move your table into the public schema, or nudge the author of the schemaName support for help.