Hey 
Ok so that’s a complex topic. Vercel (And serverless) being what they are, you’ll need a few tweaks in the default configuration to make this work.
To be honest, my tests are super recent and as stated by @dogan.ay, we don’t offer official support of self-hosted serverless agent. So everything below is super experimental.
I’m not sure about what type of machine will run your code, so I’ll base the following from a free tier account:
- Vercel has a strong timeout limit, ~10s per call. Even though our server start is pretty fast, it is splitted in multiple operations that can sometimes being time consuming, especially on slow machines
- On server start, we connect the the database (If you database requires SSL, the is the SSL/TLS negociation)
- We run a bunch of SQL queries to get your overall schema
- We generate a
.forestadmin-schema.json
All of these, at least from my test on a free tier machine, takes more than 8/9s on a medium sized database (~12 tables, ~100k rows per table).
To mitigate these issues:
- Sadly, not so much can be done on the SSL/TLS & connection aspect. From my test, there is a bit of time that could be reduced here, but I didn’t push my tests further.
- We can drastically reduce time on the introspection step, by just not doing it on vercel. See in the code below.
- Same goes for the schema. We can avoid regenerating the schema by using
isProduction: true, and we can avoid sending it (Kind of error prone though) by using skipSchemaUpdate: true and finding a way to publish the schema using another mecanism (Like forest schema:apply, or using a CI process to publish the schema to our servers).
Overall, I’m running the following base code (No customizations)
index.ts
import type { SslMode } from '@forestadmin/datasource-sql';
import 'dotenv/config';
import { createAgent } from '@forestadmin/agent';
import { createSqlDataSource, introspect } from '@forestadmin/datasource-sql';
import pg from 'pg';
import fs from 'fs';
import express from 'express';
import type { Schema } from './typings';
async function buildServer() {
const app = express();
const dbCredentials = {
uri: process.env.DATABASE_URL,
schema: process.env.DATABASE_SCHEMA,
sslMode: process.env.DATABASE_SSL_MODE as SslMode,
};
let introspection;
try {
introspection = JSON.parse(
fs.readFileSync(`${__dirname}/my-database-introspection.json`, 'utf-8'),
);
} catch (e) {
// We should write the file on disk if it doesn't exist.
fs.writeFileSync(
'introspection.json',
JSON.stringify(await introspect(dbCredentials), null, 2),
);
}
const agent = createAgent<Schema>({
authSecret: process.env.FOREST_AUTH_SECRET!,
envSecret: process.env.FOREST_ENV_SECRET!,
// We are forced to be in production mode in Vercel, as the FS is readonly so schema cannot be written
isProduction: true,
schemaPath: `${__dirname}/.forestadmin-schema.json`,
// We need to remove the typings generation on Vercel, again as FS is readonly
// typingsPath: './typings.ts',
// typingsMaxDepth: 5,
instantCacheRefresh: false,
// This is kind of touchy, as this means that the schema won't be sent. In my case, I "tricked" the system to publish my schema on a Production environment from my dev env, but you could rely on our `forest schema:apply` command.
skipSchemaUpdate: true,
});
agent.addDataSource(
createSqlDataSource(
{
...dbCredentials,
dialect: 'pg',
dialectModule: pg,
},
{ introspection },
),
);
// I'm using an express server here, but you should be able to use mountOnStandAloneServer
agent.mountOnExpress(app);
await agent.start();
return app;
}
export default buildServer()
.then(app => {
app.listen(3310, () => {
console.log('Your admin panel is available here');
});
return app;
})
.catch(console.log);
vercel.json
{
"version": 2,
"builds": [
{
"src": "index.ts",
"use": "@vercel/node"
},
{
"src": "customizations/*.ts",
"use": "@vercel/node"
},
{
"src": ".forestadmin-schema.json",
"use": "@vercel/static"
},
{
"src": "my-database-introspection.json",
"use": "@vercel/static"
},
{
"src": "typings.ts",
"use": "@vercel/node"
}
],
"devCommand": "yarn start",
"framework": null,
"installCommand": "yarn install",
"routes": [
{
"src": "/(.*)",
"dest": "index.ts",
"methods": [
"GET",
"POST",
"PUT",
"PATCH",
"DELETE",
"OPTIONS"
],
"headers": {
"Access-Control-Allow-Origin": "https://app.forestadmin.com",
"Access-Control-Allow-Methods": "GET, POST, PATCH, PUT, DELETE, OPTIONS",
"Access-Control-Allow-Headers": "Authorization, Content-Type, X-CSRF-Token, X-Requested-With, Accept, Accept-Version, Content-Length, Content-MD5, Date, X-Api-Version",
"Access-Control-Allow-Credentials": "true"
}
}
]
}
This gives me a working standalone agent, even though obviously response time are a bit higher than what we’re use to in a stateful deployment (Between 100ms up to 3s, P95 ~900ms).
Regarding the NextJS aspect, having more informations regarding what you want to achieve would be useful. As stated by @dogan.ay, we do not offer anything specific regarding integrating directly our agent with NextJS. However, depending on your answer, you may be able to forward calls to your main app via HTTP through our Action API or Field API.
In any case, I would highly suggest to still deploy in a stateful manner whenever possible, but if you really want to make it work, the code above should allow you to run a serverless agent on vercel.
Let me know if that helps 