Offset Pagination based on cursor system to enhance the paginating speed to the point of Cursor Pagination.
More explanation about the package, you can read the medium article here.
npm install prisma-offset-pagination
yarn add prisma-offset-pagination
import { prismaOffsetPagination } from 'prisma-offset-pagination';
...
const result = prismaOffsetPagination({
model: User,
cursor: <cursor>,
size: 5,
buttonNum: 7,
orderBy: 'id',
orderDirection: 'desc',
where: {
email: {
contains: 'smallbee',
},
},
});
query {
users(
cursor: "c2FsdHlzYWx0Y2tjYTJxOGZqMDA2MGI5cnc1bTYyaHVveg=="
size: 5
buttonNum: 7
orderBy: "createdAt"
orderDirection: "asc"
where: email: {
contains: 'smallbee',
}
) {
pageEdges {
cursor
node {
id
email
createdAt
}
}
pageCursors {
first {
cursor
page
isCurrent
}
previous {
cursor
page
isCurrent
}
around {
cursor
page
isCurrent
}
next {
cursor
page
isCurrent
}
last {
cursor
page
isCurrent
}
}
}
}
{
"data": {
"users": {
"pageEdges": [
{
"cursor": "c2FsdHlzYWx0Y2tjYTJxOGZqMDA2MGI5cnc1bTYyaHVveg==",
"node": {
"id": "ckca2q8fj0060b9rw5m62huoz",
"email": "[email protected]",
"createdAt": "2020-07-06T05:38:34.927Z"
}
},
{
"cursor": "c2FsdHlzYWx0Y2tjYTJxOWZxMDA2NWI5cnd4eHU2b3V6bg==",
"node": {
"id": "ckca2q9fq0065b9rwxxu6ouzn",
"email": "[email protected]",
"createdAt": "2020-07-06T05:38:36.230Z"
}
}
],
"pageCursors": {
"first": null,
"previous": {
"cursor": "c2FsdHlzYWx0Y2tjYTJxNnVpMDA1MGI5cndreGEydHY4Mw==",
"page": 2,
"isCurrent": false
},
"around": [
{
"cursor": "c2FsdHlzYWx0Y2tjOXNtYTFoMDAwNDY2cnduZW15Zzlrbg==",
"page": 1,
"isCurrent": false
},
{
"cursor": "c2FsdHlzYWx0Y2tjYTJxNnVpMDA1MGI5cndreGEydHY4Mw==",
"page": 2,
"isCurrent": false
},
{
"cursor": "c2FsdHlzYWx0Y2tjYTJxOGZqMDA2MGI5cnc1bTYyaHVveg==",
"page": 3,
"isCurrent": true
},
{
"cursor": "c2FsdHlzYWx0Y2tjYTJxYW5wMDA3MGI5cnczeTI3dHA4OQ==",
"page": 4,
"isCurrent": false
}
],
"next": {
"cursor": "c2FsdHlzYWx0Y2tjYTJxYW5wMDA3MGI5cnczeTI3dHA4OQ==",
"page": 4,
"isCurrent": false
},
"last": {
"cursor": "c2FsdHlzYWx0Y2tjYnBjODZ4MDAzMHltcndjNHE1cXJ5bw==",
"page": 7,
"isCurrent": false
}
},
totalCount
}
}
}
model
: Receive model object that you want to implement pagination
cursor
(optional)
: Receive cursor of the page you want to navigate
(* If cursor is null, return the result of the first page)
size
: Receive the number of row in the page
buttonNum
: Receive the number of button in the bottom for pagination
orderBy
(optional)
: Receive one of model field
orderDirection
(optional)
: Receive 'asc' or 'desc'
include
(optional)
: Receive include option of Prisma
prisma
: Receive prisma object
ex)
import { PrismaClient } from '@prisma/client';
const prismaObject = new PrismaClient();
const result = prismaOffsetPagination({
...
prisma: prismaObject,
});
where
(optional)
: Receive Prisma2-style where options
For the more information, look up Prisma2's documentation about filtering
https://www.prisma.io/docs/reference/tools-and-interfaces/prisma-client/filtering
The value should be a valid JSON object type.
ex)
const where = {
posts: {
some: {
published: true,
}
}
}
const where = {
gender: {
in: ['female', 'male'],
}
}
where: {
createdAt: {
gte: '2020-07-07T08:58:57.001Z',
}
}
( * When you filter time date, should use 'ISO 8601' type as above. Otherwise, Prisma2 doesn't understand the DateTime value. You can make 'ISO 8601' type date using "Date.prototype.toISOString()" )
const where = {
OR: [
{
email: {
contains: "smallbee",
}
},
{
createdAt: {
gt: "2020-07-07T03:40:00.000Z"
}
}
]
}
When implementing offset pagination, normally use offset
to get the data.
offset
is the terminology of SQL, and use like this:
SELECT *
FROM
users
WHERE
createdAt < {targetTime}
AND ...
LIMIT {pageSize}
OFFSET {offsetSize}
which could be written with Prisma2 as this:
prisma.user.findMany({
where: {
createdAt : {
lt: {targetTime}
}
...
},
skip: {offsetSize},
take: {pageSize},
})
But, the usage of offset
(skip) brings a bad result when paginating large data as shown in the below graph.
The common solution for this offset issue is using the clustering index. But, the problem is that Prisma2 does not support clustering index at the moment.
And another solution is using table joining with subquery as below:
SELECT id, name, email, ...
FROM (
SELECT id
FROM {targetTable}
LIMIT 2000000, 1000
) q
JOIN {targetTable} p
ON p.id = q.id
This is quite tricky to implement with Prisma2 and the code is more complicated. This leads to imperative programming rather than declarative, which is one of our coding principles.
Cursor pagination doesn't have the problems Offset Pagination has. It is because Cursor pagination is using the cursor value to get data, instead of offset, which makes a big difference in performance. Cursor should be unique and sequential value for good performance. However, with Cursor Pagination, the user cannot jump to a specific page but only jump to the previous or next page. As we want to make a Offset style pagination, Cursor pagination is not a possible option for us.
So, what comes out in this situation is Cursor-based Offset Pagination that mixed the strength of two paginations. So, Cursor-based Offset Pagination is using a cursor to navigate pages as fas as Cursor pagination. At the same time, it gives certain pages where user can jump. This is possible thanks to Prisma Client's cursor system.
With Prism client, we can easily move to certain page like this:
const secondQuery = prisma.post.findMany({
take: 4,
cursor: {
id: myCursor,
},
...
})
With this, we can figure out two pages away from the cursor as below:
const result = await prisma.post.findMany({
cursor: {
id: {targetObjectId},
},
skip: 2 * {pageSize},
take: 1,
});
And also Prisma Client offers the way to fetch data backward, which allows jumping to two pages backward.
const result = await prisma.post.findMany({
cursor: {
id: {targetObjectId},
},
skip: 2 * {pageSize},
take: -1,
});
With this strategy, we could know cursors of each around page of the current page's cursor like this:
{
"cursor": "c2FsdHlzYWx0Y2tjOXNtYTFoMDAwNDY2cnduZW15Zzlrbg==",
"page": 1,
"isCurrent": false
},
{
"cursor": "c2FsdHlzYWx0Y2tjYnBibmkzMDAwMHltcndhNTRoeGxjOQ==",
"page": 2,
"isCurrent": false
},
{
"cursor": "c2FsdHlzYWx0Y2tjYnBicHVwMDAxMHltcnd0aDhtcjR4ag==",
"page": 3,
"isCurrent": true
},
{
"cursor": "c2FsdHlzYWx0Y2tjYnBjNm8xMDAyMHltcnc5OG10aTdiNQ==",
"page": 4,
"isCurrent": false
}
This paginator is as fast as Cursor pagination if you follow this guide well. It is approximately O(1).
As paginating using cursor, there will be no duplicated data showed up on the list.
As Prisma client is still fast with complex where filtering options and ordering, this paginator is fast too with those conditions.