// ---------------------------------------------------------------------------// Database service — Postgres via postgres.js (porsager)// ---------------------------------------------------------------------------//// We use `postgres` (not `pg`) because Cloudflare Workers forbids sharing// I/O objects across request handlers, and `pg`'s CloudflareSocket silently// hangs when its Client is reused across requests. postgres.js creates a// fresh TCP socket per Effect scope, which aligns with Workers' per-request// I/O model. See personal-notes/pg-cloudflare-sockets-dev.md.//// Migrations are run out-of-band (e.g. via a separate script or CI step),// not at request time — Cloudflare Workers cannot read the filesystem.import { env } from "cloudflare:workers";import { Context, Effect, Layer } from "effect";import { drizzle } from "drizzle-orm/postgres-js";import type { PgDatabase } from "drizzle-orm/pg-core";import postgres, { type Sql } from "postgres";import * as cloudSchema from "./schema";import * as executorSchema from "./executor-schema";// Exported so every drizzle() call in the cloud app shares one schema// object. Historically `mcp-session.ts` built its own and forgot to spread// `executorSchema`, producing runtime "unknown model source" errors that// only surfaced in prod. See apps/cloud/src/services/db.schema.test.ts.export const combinedSchema = { ...cloudSchema, ...executorSchema };// eslint-disable-next-line @typescript-eslint/no-explicit-anyexport type DrizzleDb = PgDatabase<any, any, any>;export type DbServiceShape = { readonly sql: Sql; readonly db: DrizzleDb;};const resolveConnectionString = () => { // In local dev prefer an explicit DATABASE_URL (direct connection to // the PGlite socket server) so we bypass Miniflare's Hyperdrive proxy. // In production fall back to the Hyperdrive binding. return env.DATABASE_URL || env.HYPERDRIVE?.connectionString || "";};const makeSql = (): Sql => postgres(resolveConnectionString(), { // max=1 is correct for Hyperdrive: one request, one connection. The // earlier deadlock under ctx.transaction (outer sql.begin holding the // only connection while nested writes pulled fresh ones) is fixed in // @executor/sdk — nested writes now thread through the active tx // handle via a FiberRef in buildAdapterRouter, so they reuse the same // connection and never contend with the outer sql.begin. max: 1, idle_timeout: 0, max_lifetime: 60, connect_timeout: 10, onnotice: () => undefined, });export class DbService extends Context.Tag("@executor/cloud/DbService")< DbService, DbServiceShape>() { static Live = Layer.scoped( this, Effect.acquireRelease( Effect.sync((): DbServiceShape => { const sql = makeSql(); return { sql, db: drizzle(sql, { schema: combinedSchema }) as DrizzleDb }; }), ({ sql }) => // Fire-and-forget: the Terminate round-trip sometimes hangs, and // we don't need to block scope close waiting for it. Effect.sync(() => { sql.end({ timeout: 0 }).catch(() => undefined); }), ), );}
// ---------------------------------------------------------------------------
// Database service — Postgres via postgres.js (porsager)
// ---------------------------------------------------------------------------
//
// We use `postgres` (not `pg`) because Cloudflare Workers forbids sharing
// I/O objects across request handlers, and `pg`'s CloudflareSocket silently
// hangs when its Client is reused across requests. postgres.js creates a
// fresh TCP socket per Effect scope, which aligns with Workers' per-request
// I/O model. See personal-notes/pg-cloudflare-sockets-dev.md.
//
// Migrations are run out-of-band (e.g. via a separate script or CI step),
// not at request time — Cloudflare Workers cannot read the filesystem.
import { env } from "cloudflare:workers";
import { Context, Effect, Layer } from "effect";
import { drizzle } from "drizzle-orm/postgres-js";
import type { PgDatabase } from "drizzle-orm/pg-core";
import postgres, { type Sql } from "postgres";
import * as cloudSchema from "./schema";
import * as executorSchema from "./executor-schema";
// Exported so every drizzle() call in the cloud app shares one schema
// object. Historically `mcp-session.ts` built its own and forgot to spread
// `executorSchema`, producing runtime "unknown model source" errors that
// only surfaced in prod. See apps/cloud/src/services/db.schema.test.ts.
export const combinedSchema = { ...cloudSchema, ...executorSchema };
// eslint-disable-next-line @typescript-eslint/no-explicit-any
export type DrizzleDb = PgDatabase<any, any, any>;
export type DbServiceShape = {
readonly sql: Sql;
readonly db: DrizzleDb;
};
const resolveConnectionString = () => {
// In local dev prefer an explicit DATABASE_URL (direct connection to
// the PGlite socket server) so we bypass Miniflare's Hyperdrive proxy.
// In production fall back to the Hyperdrive binding.
return env.DATABASE_URL || env.HYPERDRIVE?.connectionString || "";
};
const makeSql = (): Sql =>
postgres(resolveConnectionString(), {
// max=1 is correct for Hyperdrive: one request, one connection. The
// earlier deadlock under ctx.transaction (outer sql.begin holding the
// only connection while nested writes pulled fresh ones) is fixed in
// @executor/sdk — nested writes now thread through the active tx
// handle via a FiberRef in buildAdapterRouter, so they reuse the same
// connection and never contend with the outer sql.begin.
max: 1,
idle_timeout: 0,
max_lifetime: 60,
connect_timeout: 10,
onnotice: () => undefined,
});
export class DbService extends Context.Tag("@executor/cloud/DbService")<
DbService,
DbServiceShape
>() {
static Live = Layer.scoped(
this,
Effect.acquireRelease(
Effect.sync((): DbServiceShape => {
const sql = makeSql();
return { sql, db: drizzle(sql, { schema: combinedSchema }) as DrizzleDb };
}),
({ sql }) =>
// Fire-and-forget: the Terminate round-trip sometimes hangs, and
// we don't need to block scope close waiting for it.
Effect.sync(() => {
sql.end({ timeout: 0 }).catch(() => undefined);
}),
),
);
}