Last fall I (finally) read CouchDB’s Documentation completely. And since I’ve read its security and design documents parts, I can’t get the idea out of my head, that all you need for a CRUD web app is a front end with React + React-Hooks + PouchDB hosted on a static file host. And CouchDB paired with some serverless functions (function as a service) as the backend end.
So I written those React hooks and published them as usePouchDB to npm. This post will go into some design decisions and how I think they could be used. For a tutorial and API docs please visit https://christopher-astfalk.de/use-pouchdb/.
If you use a service-worker, then your CRUD web app becomes, with some small changes, even offline first! Create-react-app for example, comes with a service-worker out of the box.
Main concepts
The target App I had in mind while developing usePouchDB is a Web-App that solely uses function components and hooks. With as little as possible server side functionality. Everything on the server should be done by CouchDB or some serverless functions. The client should be able to be hosted on a static file host. In short: A modified version of JAMStack.
Other Apps are possible, even encouraged, and can be achieved with Redux and redux-pouchdb combined with usePouchDB. But they are not the focus.
Provider
A Context Provider, that provides access to one or multiple PouchDB databases, is required. With plugins and settings and different database names, there is just to much that can be done using PouchDB to expose everything.
The concept is that the package user would create a Databases component, which renders the <Provider>. It would be rendered on the top of the component tree and makes all databases available to all components. It could also be used as a bridge to a Redux store.
The Databases component’s main job would be to setup the local and remote databases, handle sessions (sign up, login and logout), and synchronization. That’s a lot to do for a single component, but all this is linked together. On login you what to sync with that users database. In a couch_peruser setup you even have to know the username! Similar, on logout you want to stop syncing and maybe destroy the local databases.
Extending Context and multiple databases
If a <Provider> embedded in the component tree of another <Provider> would extend the context, then you could use special databases that are only needed in a subtree. And combined with React-Router and Loadable Components databases could be created for routes. For example a database for sharing between a group of people would only be rendered if its content would be displayed. With a <Provider> and a Databases component for that route, that database would be created on entering that route and closed once the user visits another route.
Every hook has an option to select the used database by a key. That key can be changed and the hook will re-query its content. By default the database’s name is used for its key. But a <Provider> can also overwrite a database’s key. Then that key will be used to select a database, and not it’s name. This is useful for remote databases, their name is their full URL.
There is also the concept of a default database: it gets used when no database name/key is provided. The <Provider> sets which of the databases is the default. A child <Provider> overwrites the default db for its subtree.
This can be combined to create seamless switching between two databases: For example when your user visits your page the remote database (on CouchDB) is the default db. The user can load and interact with all your content. While in the background you create a local database and replicate the remote database to the local one. And once the local database, all Mango indexes and views are up to date, switch the default database to be the local one. Then all hooks will automatically re-query.
Hooks that are re-querying because an option or a database did change, show the last result until the query finishes. The sole exception is useDoc (hook version of db.get()). When it has a initial value, then it returns the initial value until the query finishes. To reset hooks, embed them in Components and use those options, which should reset the hook, in its key prop.
Hooks
usePouchDB is a set of React-Hooks that re-implement some of PouchDB’s API. Also every hook subscribes by default to changes. That way every component will become a function of your database (+ some local state) that returns your UI. Similar to Redux, where your components became functions of your Redux store.
In future updates I will provide a way to opt out of auto subscription and add a method to manual setoff an update. But I don’t know how this will look like, yet.
Many of PouchDBs/CouchDBs APIs have objects or arrays as arguments. I wanted to be able to use them without having to use useMemo for those arguments, if I didn’t want to re-query on every render. That’s why I decided to compare their value, and only re-query if they change by value.
Updates and your own hooks
With usePouch usePouchDB provides a hook to access a database. It is intended to serve as a escape hatch if the build in hooks don’t provide what you need.
It can also be used to create your own hooks.
One possible hook category are hooks that return functions to update your data:
import { useCallback } from 'react'
import { usePouch } from 'use-pouchdb'
export function useAddTodo(dbKey = '_default') {
const db = usePouch(dbKey)
return useCallback(async text => {
const result = await db.put({
_id: `todo_${new Date().toJSON()}`,
type: 'todo',
text: text,
completed: false
})
return result
}, [db])
}
This hook returns a callback. When called with a string, that callback will create a new todo-item in your database. Those hooks fill the role of actions in a Redux-App: They update your data.
The optional dbKey argument is to select the database that should be used. '_default' is a special key to select the default database.
If you create your own hooks in a similar way, then they can be used everywhere in your app. And they behave like the hooks from usePouchDB: they use the default database or optionally another one and they can switch databases.
// How the useAddTodo hook can be used
import React, { useState } from 'react'
import { useAddTodo } from './hooks'
export default function AddNewTodo() {
const [todoText, setTodoText] = useState('')
const addTodo = useAddTodo()
return (
<form
onSubmit={event => {
event.preventDefault()
addTodo(todoText).then(() => {
setTodoText('')
})
}}
>
<input
type='text'
placeholder='todo text'
value={todoText}
onChange={event => {
setTodoText(event.target.value)
}}
/>
<button>Add</button>
</form>
)
}
useFind
All hooks have a similar API to the method they use. But useFind is a combination of db.createIndex() and db.find().
The docs encourage to explicitly specify the index to use in a db.find() request. And db.createIndex() results with the id and name of the created Index, it also returns them if the index already exists.
It would also be complicated to create an API for two hooks: one to create an index and a second one to query it. And the second one should only query if the first did create its index. And what if the user did forget to use both hooks?
So I combined use_index from db.find() and the index-field of db.createIndex() into one index-field on the useFind-options:
- If the
index-field is a string or an array of strings,useFindwill use them asuse_indexfordb.find(). - But if the
index-field contains an object, that object will be used in theindex-field ofdb.createIndex(). And theidandnamein its result will be used in theuse_indexfield ofdb.find(). - And if the
index-field isnullorundefined, thendb.explain()will be used to find theidandname. - There is no
use_indexoption on theuseFindhook.
Once the id and name is know, useFind will not call db.createIndex() or db.explain() until an option changes.
This has the result, that all what you need to do to create, query and subscribe to a Mango index is this:
const { docs, loading, state, error } = useFind({
index: { // create index if it doesn't exist
fields: ['type', 'title'],
},
selector: {
type: 'story',
title: { $exists: true },
},
sort: ['title'],
fields: ['_id', 'title'],
})
As a nice bonus: when partial indexes get added to PouchDB, you can be sure that your query will use them!
Subscriptions
Internally every database gets a Subscription-Manager. It’s job is to optimize subscriptions.
When I first tested my hooks idea every hook directly subscribed to its database. But I quickly noticed that with even a small number of hooks PouchDB’s event-emitter will log a warning, that to many event-listeners are registered. Everything works just fine, this is just a functionality to help to find memory leaks.
So I created the Subscription-Manager. It subscribes to every type only once. Once for document changes and once for every map/reduce-view in use of an useView hook (hook version of db.query()).
At the moment a Subscription-Manager will subscribe to all document changes. I know this is like drinking from a fire hose, but it was easier to implement. Also in most cases every hook makes a full re-query. A potential, but complicated place for optimizations.
But optimizations will follow: if you only subscribe to some documents, then the Subscription-Manager should only subscribe to those. And once that list changes, resubscribe.
But there are some hooks that require a subscription to all documents:
useAllDocsthe hook version ofdb.allDocs(). But only if you don’t use thekeyorkeysoptions.useFindthe hook version ofdb.createIndex()anddb.find()combined.
This all results in following behavior: usePouchDB will subscribe to all changes. You can use as many hooks as you like; they result in only one connection for every remote database. But if you use useView you get an additional connection for every view. And often a hook will make a full query.
If you use a local database this all doesn’t matter. Because its all local! That is the reason behind my warning that usePouchDB is only optimized for local databases.
Conclusion
usePouchDB is probably useful. When I moved my experimentation App from Redux to only hooks it became much simpler.
In components I specified what data I needed in that component, and the hooks delivered. I also had some other hooks for other state and some to update my data. When the later ones where call, usePouchDBs hooks automatically updated!
But there is still work left to do.