Skip to content

Open-source manufacturing ERP, MES and QMS. Quoting, MRP, inventory, shop floor, quality and lot/serial traceability on one Postgres schema, with a REST API and MCP server. Self-host or use Carbon Cloud.

refactor: remove the retired shared private bucket from code

aashuf499a2d23a8e
diff --git a/.claude/rules/audit-log-system.md b/.claude/rules/audit-log-system.md
index ad05b0fbdd..9c5441a2fe 100644
--- a/.claude/rules/audit-log-system.md
+++ b/.claude/rules/audit-log-system.md
@@ -147,8 +147,7 @@ lookup per target table), and writes via `client.rpc("insert_audit_log_batch", {
`auditArchiveFunction` in `packages/jobs/src/inngest/functions/scheduled/audit-archive.ts`
(Inngest id `audit-log-archive`, cron `0 2 * * *`): per company, fetch rows older than `retentionDays`,
gzip to JSONL, upload to the company's own bucket, record an `auditLogArchive` row, then delete the rows.
-`getArchiveDownloadUrl` reads the company bucket only. Archives written to the retired shared `private`
-bucket were moved there, and their `archivePath` rewritten, by `scripts/one-off/relocate-legacy-audit-archives.ts`.
+`getArchiveDownloadUrl` reads the company bucket only.
GOTCHA: the runtime path is `{companyId}/audit-logs/{year}/{month}/{YYYY-MM-DD}.jsonl.gz`, which does
**not** match `auditConfig.archivePath` (`.../{month}.jsonl.gz`) — the job builds the path inline rather
diff --git a/.claude/rules/billing-system.md b/.claude/rules/billing-system.md
index 8e0c1dd654..ea5fb717e9 100644
--- a/.claude/rules/billing-system.md
+++ b/.claude/rules/billing-system.md
@@ -237,9 +237,8 @@ whom it deleted.
documents then fails, and is logged and skipped. So does the last company in a group,
because its group's system accounts refuse deletion and the group would be stranded.
- BEFORE the transaction, `removeCompanyFiles` drains and deletes the per-company
- bucket (already missing counts as done), legacy files under `private/<companyId>/` and
- legacy audit-log archives under `private/audit-logs/<companyId>/` (listed strictly, so a
- listing error is a failure). Never inside it: storage deletes
+ bucket (already missing counts as done; listed strictly, so a listing error is a
+ failure). Never inside it: storage deletes
its object rows on its own connection, `delete_orphaned_documents` then deletes the
matching `document` rows, and when the open purge transaction has already deleted
those rows storage waits on it until it times out (HTTP 544) and the purge rolls
diff --git a/.claude/rules/company-backup-restore.md b/.claude/rules/company-backup-restore.md
index 0f27251b2e..45607dff1f 100644
--- a/.claude/rules/company-backup-restore.md
+++ b/.claude/rules/company-backup-restore.md
@@ -207,8 +207,6 @@ Buckets (important, two different things):
under a `{companyId}/` prefix, one folder per backup (`exports/<name>/` — its
`manifest.json`, `tables/`, `assets/{companyId}/…`), and pre-restore snapshots under
the same prefix — see `client.storage.from(companyId)`.
-- `LEGACY_SHARED_BUCKET = "private"` — the retired shared bucket. Export, backup and
- restore never read it; only the company purge still deletes what a company left there.
- `TEMPLATE_BUCKET = "company-templates"`, `TEMPLATE_ASSET_PREFIX = "_templates"`.
## Export — `company-export.ts`
diff --git a/apps/erp/app/modules/documents/documents.paths.ts b/apps/erp/app/modules/documents/documents.paths.ts
index 13d6b59d1b..286cfa0d9e 100644
--- a/apps/erp/app/modules/documents/documents.paths.ts
+++ b/apps/erp/app/modules/documents/documents.paths.ts
@@ -6,7 +6,7 @@ import { stripSpecialCharacters } from "@carbon/utils";
/**
* Build the canonical storage path for an uploaded document:
- * `${companyId}/${folder}/${entityId}/${sanitizedName}` in the `"private"` bucket.
+ * `${companyId}/${folder}/${entityId}/${sanitizedName}` in the company's own bucket.
* Single source of truth for the path convention every `use*Documents` hook
* inlines client-side; shared so the server-side (MCP) upload methods produce
* paths the entity document panels list from.
diff --git a/apps/erp/app/modules/documents/documents.server.ts b/apps/erp/app/modules/documents/documents.server.ts
index 2deb5584dc..3178e29ce4 100644
--- a/apps/erp/app/modules/documents/documents.server.ts
+++ b/apps/erp/app/modules/documents/documents.server.ts
@@ -488,7 +488,7 @@ export const updateDocument = defineServiceFn({
});
/**
- * Create a presigned upload URL for a document in the "private" bucket
+ * Create a presigned upload URL for a document in the company's bucket
*
* @mcp
*/
diff --git a/apps/erp/app/modules/documents/documents.service.ts b/apps/erp/app/modules/documents/documents.service.ts
index 5cfbd5fd5f..ceced41bbe 100644
--- a/apps/erp/app/modules/documents/documents.service.ts
+++ b/apps/erp/app/modules/documents/documents.service.ts
@@ -171,7 +171,7 @@ export async function upsertDocument(
}
/**
- * Create a presigned upload URL for a document in the "private" bucket. First step
+ * Create a presigned upload URL for a document in the company's bucket. First step
* of the two-step upload flow: PUT the file bytes to the returned `signedUrl` (or
* use supabase-js `uploadToSignedUrl(path, token, file)`), then call
* `insertUploadedDocument` with the returned `path` to register the metadata row.
diff --git a/apps/erp/app/routes/download.$token.tsx b/apps/erp/app/routes/download.$token.tsx
index 97368036a7..712f6b7149 100644
--- a/apps/erp/app/routes/download.$token.tsx
+++ b/apps/erp/app/routes/download.$token.tsx
@@ -51,8 +51,8 @@ export async function loader({ params }: LoaderFunctionArgs) {
if (doc.error || !doc.data?.path) return fail("unavailable");
// Access already proven above; fetch the bytes with the service role.
- // Documents live in the company's private bucket (legacy `private`
- // fallback), keyed directly by `document.path` (mirrors useDocument.ts).
+ // Documents live in the company's own bucket, keyed directly by
+ // `document.path` (mirrors useDocument.ts).
const file = await storage(getCarbonServiceRole())
.company(companyId)
.download(doc.data.path);
diff --git a/contrib/building/examples/upload-3d-model/app/lib/carbon.server.ts b/contrib/building/examples/upload-3d-model/app/lib/carbon.server.ts
index 59baba4c27..62c7570b92 100644
--- a/contrib/building/examples/upload-3d-model/app/lib/carbon.server.ts
+++ b/contrib/building/examples/upload-3d-model/app/lib/carbon.server.ts
@@ -38,7 +38,7 @@ class CarbonClient {
const fileName = `${this.companyId}/models/${modelId}.${fileExtension}`;
const [fileUpload, recordInsert] = await Promise.all([
- this.client.storage.from("private").upload(fileName, file),
+ this.client.storage.from(this.companyId).upload(fileName, file),
this.client.from("modelUpload").insert({
id: modelId,
modelPath: fileName,
@@ -78,7 +78,7 @@ class CarbonClient {
const thumbnailPath = `${this.companyId}/thumbnails/${thumbnailId}.png`;
const thumbnailUpload = await this.client.storage
- .from("private")
+ .from(this.companyId)
.upload(thumbnailPath, file, {
upsert: true,
contentType: "image/png"
diff --git a/packages/database/supabase/migrations/20261020051356_drop-shared-private-bucket-policy.sql b/packages/database/supabase/migrations/20261020051356_drop-shared-private-bucket-policy.sql
new file mode 100644
index 0000000000..6c5661c263
--- /dev/null
+++ b/packages/database/supabase/migrations/20261020051356_drop-shared-private-bucket-policy.sql
@@ -0,0 +1,10 @@
+-- The legacy shared `private` bucket is retired: every company's files live in
+-- its own bucket, and nothing reads or writes `private` any more. Drop its
+-- storage policy so the bucket itself can be emptied and deleted.
+--
+-- The bucket is not deleted here: Storage owns its objects, so it is emptied
+-- and removed from the dashboard (or the Storage API) after this ships.
+--
+-- `company_id_not_reserved_bucket_id` still reserves the id `private`, so no
+-- company can ever be created with that id and claim a bucket of that name.
+DROP POLICY IF EXISTS "Shared Private Bucket" ON storage.objects;
diff --git a/packages/files/AGENTS.md b/packages/files/AGENTS.md
index ba73150ed1..30da847945 100644
--- a/packages/files/AGENTS.md
+++ b/packages/files/AGENTS.md
@@ -6,7 +6,7 @@ in the codebase, shared by the browser and Node.
| Subpath | Provides |
|---------|----------|
-| `.` (root) | Cross-type helpers: `MEDIA_CONTENT_TYPES` + `getContentType` (the one extension→MIME map every file-serving route uses), `getFileExtension` / `effectiveExtension` (`.zst` unwrap), `getDocumentType` / `documentTypes` / `isPreviewableDocumentType` (file classification for the Documents UI), `convertKbToString`, `downloadUrl(url, filename)` (fetches first and throws on a failed response, so an expired session never saves an error page under the file's name), and `downloadBlob` / `downloadText` — THE browser "save this as a file" sequence (object URL → anchor click → revoke); never hand-roll it. **`storage(client)`** — the Carbon storage client, supabase-shaped: `.from(bucket)` is plain supabase for `public` / `temp-staging`; `.company(companyId)` is that company's private bucket (bucket id = companyId, THROWS on an empty id). Every read and write goes to the company bucket only. The retired shared `private` bucket is not read: the one-off copies (`scripts/one-off/migrate-private-buckets.ts`, `recopy-private-buckets.ts`, `relocate-legacy-audit-archives.ts`) moved every live company's files out of it, and a read check of production on 2026-10-17 found no live file left there. `list(folder)` takes no options: it reads the cursor-paged `listV2` endpoint to the end and returns every entry directly inside the folder, sorted by name, in the old `list` shape (`name` is the entry's own name; a sub-folder has a null `id`). Every key must sit under `${companyId}/` and pass `isUnsafeStoragePath`; anything else gets `{ data: null, error }` without touching storage. **`isUnsafeStoragePath(path)`** is THE storage-key traversal guard (dot segments, `\`, `?`, `#`, control characters, checked as given and after every decode so `%2e%2e` / `%252e%252e` are caught; malformed escapes refused) — every route that takes a key from a URL or form calls it rather than a local copy; **`safeStorageFileName(name)`** turns a caller-supplied file name into a safe last key segment (basename, no `%`) or null. This is the contract for document keys (`${companyId}/parts/…`, `/job/…`, `/tmp/…`, everything `getPrivateUrl` serves); the backup and audit-archive layouts (`exports/…`, `audit/…`) share the bucket under their own keys and keep using plain `client.storage.from(companyId)`. `getCompanyPrivateBucket` / `hasCompanyPrivateObjectPathPrefix` are exported for the few places that need the bucket *name* (MediaUploader staging, route authorization, the assembler's bucket resolver). `PRIVATE_PREVIEW_ALIAS` (`"private"`) is a stored URL segment, not a bucket: `/file/preview/private/<path>` links in rich text and inspection documents are served from the caller's own company bucket |
+| `.` (root) | Cross-type helpers: `MEDIA_CONTENT_TYPES` + `getContentType` (the one extension→MIME map every file-serving route uses), `getFileExtension` / `effectiveExtension` (`.zst` unwrap), `getDocumentType` / `documentTypes` / `isPreviewableDocumentType` (file classification for the Documents UI), `convertKbToString`, `downloadUrl(url, filename)` (fetches first and throws on a failed response, so an expired session never saves an error page under the file's name), and `downloadBlob` / `downloadText` — THE browser "save this as a file" sequence (object URL → anchor click → revoke); never hand-roll it. **`storage(client)`** — the Carbon storage client, supabase-shaped: `.from(bucket)` is plain supabase for `public` / `temp-staging`; `.company(companyId)` is that company's private bucket (bucket id = companyId, THROWS on an empty id). Every read and write goes to the company bucket only; there is no shared `private` bucket. `list(folder)` takes no options: it reads the cursor-paged `listV2` endpoint to the end and returns every entry directly inside the folder, sorted by name, in the old `list` shape (`name` is the entry's own name; a sub-folder has a null `id`). Every key must sit under `${companyId}/` and pass `isUnsafeStoragePath`; anything else gets `{ data: null, error }` without touching storage. **`isUnsafeStoragePath(path)`** is THE storage-key traversal guard (dot segments, `\`, `?`, `#`, control characters, checked as given and after every decode so `%2e%2e` / `%252e%252e` are caught; malformed escapes refused) — every route that takes a key from a URL or form calls it rather than a local copy; **`safeStorageFileName(name)`** turns a caller-supplied file name into a safe last key segment (basename, no `%`) or null. This is the contract for document keys (`${companyId}/parts/…`, `/job/…`, `/tmp/…`, everything `getPrivateUrl` serves); the backup and audit-archive layouts (`exports/…`, `audit/…`) share the bucket under their own keys and keep using plain `client.storage.from(companyId)`. `getCompanyPrivateBucket` / `hasCompanyPrivateObjectPathPrefix` are exported for the few places that need the bucket *name* (MediaUploader staging, route authorization, the assembler's bucket resolver). `PRIVATE_PREVIEW_ALIAS` (`"private"`) is a stored URL segment, not a bucket: `/file/preview/private/<path>` links in rich text and inspection documents are served from the caller's own company bucket |
| `./csv` | `encodeCsv` (object rows; `header: false` encodes a later page of a streamed file — pages join with `\r\n`, see the Sync Activity export) / `encodeCsvTable` (header + cells) — the ONE CSV encoder, injection-safe by construction (`stripCsvFormulaPrefix` on every string cell; `formulaSafe: false` turns that off for CSV a program reads and no spreadsheet opens, which today is only the MCP server's tool results); `parseCsv` (text) / `parseCsvFile` (browser File, streamed) / `parseCsvLines` (text → records with the line each starts on, blank and delimiter-only records left out; for messages that name a line); `downloadCsv`; `CSV_CONTENT_TYPE`. papaparse underneath, in every runtime. The Table export button, report exports, import templates, import-error re-exports and the sales CSV import all go through here — before this there were two libraries (`json-2-csv` + papaparse) and two hand-rolled encoders, and only ONE of the five export sites protected against formula injection |
| `./media` | Images. **`MediaUploader`** — the configured client (constructor takes the storage client + HEIC staging location once, files-sdk style): `prepareForUpload(files)` (HEIC → JPEG sequentially + duplicate-name refusal, throws `DuplicateFileNameError`), `convertHeic(file)`, `prepareImage(file, shape)`. Underneath: `processImage` (the pipeline), `prepareImageUpload` (pipeline → native-decode fallback → imgproxy fallback), `convertHeicToJpeg` / `convertHeicFiles` / `findDuplicateFileName`, `transformImageViaStorage` (imgproxy round-trip), `isHeic`, `IMAGE_UPLOAD_MIME_TYPES`, plus the storage-path helpers `getPrivateUrl` / `getRawModelUrl` / `parseJobFilePath` |
| `./media/node` | `initNodeImageCodecs()` — Node-only wasm pre-instantiation; call once before the pipeline in Node (jobs, paperless, vitest). The wasm comes from the BUNDLER (`?inline` data URIs), never off disk — so this module is consumable only from a Vite pipeline, and that pipeline needs `assetsInclude: ["**/*.wasm"]` (already set in `apps/erp/vite.config.ts` and `@carbon/config/vitest`). `renderLabelLogo(bytes, ext, { widthDots, threshold?, crop? })` — a logo as a mono PNG data URL + ZPL `^GFA` field (inits the codecs itself; `@carbon/documents/labels` calls it). Never import from client code |
diff --git a/packages/files/src/storage.ts b/packages/files/src/storage.ts
index f9ef4d2a69..9369c50525 100644
--- a/packages/files/src/storage.ts
+++ b/packages/files/src/storage.ts
@@ -26,8 +26,8 @@ import {
/**
* The URL segment `/file/preview/private/<path>` — a stored link format
* (rich text, inspection documents) that the preview routes serve from the
- * caller's own company bucket. The shared `private` bucket it once named is
- * no longer read.
+ * caller's own company bucket. The shared `private` bucket it once named no
+ * longer exists.
*/
export const PRIVATE_PREVIEW_ALIAS = "private";
// Ephemeral staging for uploads too big for a company bucket's per-object cap
diff --git a/packages/jobs/manifests/schema.json b/packages/jobs/manifests/schema.json
index de7c108b31..5fd82b1077 100644
--- a/packages/jobs/manifests/schema.json
+++ b/packages/jobs/manifests/schema.json
@@ -1,11 +1,11 @@
{
"kind": "carbon-company-backup",
"version": 1,
- "schemaVersion": "20261019185808",
+ "schemaVersion": "20261020051356",
"sourceCompanyId": "schema-baseline",
"sourceCompanyGroupId": null,
"sourceCompanyName": null,
- "exportedAt": "2026-10-19T17:59:10.246Z",
+ "exportedAt": "2026-10-20T05:29:25.062Z",
"exportedBy": "schema-baseline",
"label": "schema baseline",
"includeStorage": "none",
@@ -479,10 +479,10 @@
"accountingCutoverDate",
"accountingActivatedAt",
"accountingActivatedBy",
- "requireBackflushStorageUnit",
"leaseIncrementalBorrowingRate",
"leaseShortTermExpedient",
"leaseReminderDays",
+ "requireBackflushStorageUnit",
"preventNegativeBackflush",
"receiptQuantitiesStartAtZero"
]
diff --git a/packages/jobs/src/inngest/functions/scheduled/cleanup.ts b/packages/jobs/src/inngest/functions/scheduled/cleanup.ts
index f5038d102a..f7fd73cd2f 100644
--- a/packages/jobs/src/inngest/functions/scheduled/cleanup.ts
+++ b/packages/jobs/src/inngest/functions/scheduled/cleanup.ts
@@ -406,13 +406,11 @@ export const cleanupFunction = inngest.createFunction(
return;
}
- // Rule 1 — RELOCATED: the compact pipeline copies survivors to `private`
- // under the same key (copy, not move — a move races concurrent jobs
- // holding a temp-staging source pointer). Once the durable copy exists,
- // the staged one is redundant regardless of size or references: every
- // reader probes/falls back to `private`.
- // A durable copy may live in the company's own bucket (current pipeline)
- // or the legacy shared `private` bucket (pre-migration relocations).
+ // Rule 1 — RELOCATED: the compact pipeline copies survivors to the
+ // company's own bucket under the same key (copy, not move — a move races
+ // concurrent jobs holding a temp-staging source pointer). Once the durable
+ // copy exists, the staged one is redundant regardless of size or
+ // references: every reader looks in the company bucket first.
// Object keys start with the companyId segment, and a key without one
// can't have a durable copy anywhere.
let relocated: Set<string>;
@@ -486,10 +484,8 @@ export const cleanupFunction = inngest.createFunction(
Date.now() - TMP_STAGING_TTL_HOURS * 60 * 60 * 1000
).toISOString();
- // Staging now lives in each company's own bucket (bucket id = companyId),
- // with the legacy shared `private` bucket still holding pre-migration
- // objects — so select the bucket alongside the name and prune per bucket
- // rather than assuming one shared bucket.
+ // Staging lives in each company's own bucket (bucket id = companyId), so
+ // select the bucket alongside the name and prune per bucket.
const stale = await listStorageObjects(
sql`name LIKE '%/tmp/%' AND created_at < ${cutoff}`
);
diff --git a/packages/jobs/src/inngest/functions/scheduled/company-cleanup.ts b/packages/jobs/src/inngest/functions/scheduled/company-cleanup.ts
index 00e82f218e..219bacd62c 100644
--- a/packages/jobs/src/inngest/functions/scheduled/company-cleanup.ts
+++ b/packages/jobs/src/inngest/functions/scheduled/company-cleanup.ts
@@ -243,15 +243,12 @@ export async function deleteCompanies(
continue;
}
// Outside the transaction: see `removeCompanyFiles`.
- const failures = await removeCompanyFiles(serviceRole, company.id);
- if (failures.length > 0) {
- for (const failure of failures) {
- logger.error("Failed to remove company {failurePart}", {
- failurePart: failure.part,
- ...company,
- error: failure.error
- });
- }
+ const filesError = await removeCompanyFiles(serviceRole, company.id);
+ if (filesError) {
+ logger.error("Failed to remove company bucket", {
+ ...company,
+ error: filesError
+ });
results.push({ id: company.id, deleted: false });
continue;
}
diff --git a/packages/jobs/src/inngest/functions/scheduled/purge-company.test.ts b/packages/jobs/src/inngest/functions/scheduled/purge-company.test.ts
index 430f0f796a..c8d126d939 100644
--- a/packages/jobs/src/inngest/functions/scheduled/purge-company.test.ts
+++ b/packages/jobs/src/inngest/functions/scheduled/purge-company.test.ts
@@ -58,37 +58,26 @@ function fakeStorage(
}
describe("removeCompanyFiles", () => {
- it("drains the company bucket before deleting it, and the legacy folders", async () => {
+ it("drains the company bucket before deleting it", async () => {
const { serviceRole, buckets } = fakeStorage({
co1: ["parts/a.png", "job/j1/b.pdf"],
- private: [
- "co1/docs/c.pdf",
- "co12/docs/keep.pdf",
- "audit-logs/co1/2026/01/2026-01-31.jsonl.gz",
- "audit-logs/co12/2026/01/keep.jsonl.gz"
- ]
- });
- expect(await removeCompanyFiles(serviceRole, "co1")).toEqual([]);
- expect(buckets).toEqual({
- private: ["co12/docs/keep.pdf", "audit-logs/co12/2026/01/keep.jsonl.gz"]
+ co12: ["docs/keep.pdf"]
});
+ expect(await removeCompanyFiles(serviceRole, "co1")).toBeNull();
+ expect(buckets).toEqual({ co12: ["docs/keep.pdf"] });
});
it("treats a bucket that is already gone as done", async () => {
- const { serviceRole } = fakeStorage({ private: [] });
- expect(await removeCompanyFiles(serviceRole, "co1")).toEqual([]);
+ const { serviceRole } = fakeStorage({});
+ expect(await removeCompanyFiles(serviceRole, "co1")).toBeNull();
});
- it("reports each part that failed, and leaves the bucket", async () => {
+ it("reports a failed drain, and leaves the bucket", async () => {
const { serviceRole, calls } = fakeStorage(
- { co1: ["a.png"], private: ["co1/b.pdf"] },
+ { co1: ["a.png"] },
{ failRemove: true }
);
- const failures = await removeCompanyFiles(serviceRole, "co1");
- expect(failures.map((f) => f.part)).toEqual([
- "company bucket",
- "legacy private files"
- ]);
+ expect(await removeCompanyFiles(serviceRole, "co1")).toBeInstanceOf(Error);
expect(calls).not.toContain("deleteBucket co1");
});
});
diff --git a/packages/jobs/src/inngest/functions/scheduled/purge-company.ts b/packages/jobs/src/inngest/functions/scheduled/purge-company.ts
index 0b9a3e65a6..7cfd1f0685 100644
--- a/packages/jobs/src/inngest/functions/scheduled/purge-company.ts
+++ b/packages/jobs/src/inngest/functions/scheduled/purge-company.ts
@@ -8,11 +8,7 @@ import { getCompanyPrivateBucket, isStorageNotFound } from "@carbon/files";
import { chunkArray } from "@carbon/utils";
import { type Kysely, sql } from "kysely";
import { listBucketFilesRecursive } from "../../../backups/storage";
-import {
- type Catalog,
- LEGACY_SHARED_BUCKET,
- wipeScopedData
-} from "../tasks/company-backup";
+import { type Catalog, wipeScopedData } from "../tasks/company-backup";
/**
* Delete one company and everything it owns, inside the caller's transaction.
@@ -158,9 +154,8 @@ export async function dropOrphanCompanyTables(
}
/**
- * A company's files: its own storage bucket, and pre-bucket-migration files
- * under `<companyId>/` in the shared private bucket. Returns the failures; each
- * part is attempted regardless.
+ * A company's files: its own storage bucket, drained and then deleted. Returns
+ * the error that stopped it, or null once the bucket is gone.
*
* Run BEFORE the purge transaction, never inside it. Storage deletes its object
* rows on its own connection, and `delete_orphaned_documents` then deletes the
@@ -175,9 +170,7 @@ export async function dropOrphanCompanyTables(
export async function removeCompanyFiles(
serviceRole: ReturnType<typeof getCarbonServiceRole>,
companyId: string
-): Promise<{ part: string; error: unknown }[]> {
- const failures: { part: string; error: unknown }[] = [];
-
+): Promise<unknown> {
// Drained by listing, not `emptyBucket`: that only queues the deletes, so the
// `deleteBucket` after it is refused as "not empty" for any bucket with files.
// A bucket already gone (an earlier attempt removed it, then failed) is done.
@@ -186,25 +179,8 @@ export async function removeCompanyFiles(
if (!bucketError) {
bucketError = (await serviceRole.storage.deleteBucket(bucket)).error;
}
- if (bucketError && !(await isBucketGone(bucketError))) {
- failures.push({ part: "company bucket", error: bucketError });
- }
-
- // The retired shared bucket held a company's files under `<companyId>/` and
- // its audit-log archives under `audit-logs/<companyId>/`.
- for (const [part, prefix] of [
- ["legacy private files", companyId],
- ["legacy audit-log archives", `audit-logs/${companyId}`]
- ] as const) {
- const legacyError = await drainFolder(
- serviceRole,
- LEGACY_SHARED_BUCKET,
- prefix
- );
- if (legacyError) failures.push({ part, error: legacyError });
- }
-
- return failures;
+ if (bucketError && !(await isBucketGone(bucketError))) return bucketError;
+ return null;
}
const isBucketGone = async (error: unknown) =>
diff --git a/packages/jobs/src/inngest/functions/tasks/company-backup.ts b/packages/jobs/src/inngest/functions/tasks/company-backup.ts
index 3760af259e..ce079fe06b 100644
--- a/packages/jobs/src/inngest/functions/tasks/company-backup.ts
+++ b/packages/jobs/src/inngest/functions/tasks/company-backup.ts
@@ -46,13 +46,6 @@ export const EXPORTS_PREFIX = "exports";
/** Shared, env-agnostic bucket holding the onboarding backup templates. */
export const TEMPLATE_BUCKET = "company-templates";
-/**
- * The retired shared bucket that held every company's files under a
- * `{companyId}/` prefix before each company got its own bucket. Nothing reads
- * or writes it; a company purge still deletes what that company left there.
- */
-export const LEGACY_SHARED_BUCKET = "private";
-
/**
* TEXT columns holding a storage path (`{companyId}/…/{id}.ext`). A cross-company
* restore changes both the companyId and the embedded ids, so these are rewritten
diff --git a/packages/jobs/src/inngest/functions/tasks/company-import.ts b/packages/jobs/src/inngest/functions/tasks/company-import.ts
index ae1df192e6..e005a4eecc 100644
--- a/packages/jobs/src/inngest/functions/tasks/company-import.ts
+++ b/packages/jobs/src/inngest/functions/tasks/company-import.ts
@@ -440,7 +440,7 @@ export const companyImportFunction = inngest.createFunction(
// Storage files travel outside the transaction — failures here leave the
// imported rows intact and are surfaced as warnings. Copy them server-side
- // from the artifact's `.assets/` folder into the shared `private` bucket,
+ // from the artifact's `.assets/` folder into the company's own bucket,
// with the path rewritten for the target company (and remapped ids on
// reseed) to match the path columns rewritten above.
//
diff --git a/scripts/one-off/migrate-private-buckets.ts b/scripts/one-off/migrate-private-buckets.ts
deleted file mode 100644
index 9c1f6a124e..0000000000
--- a/scripts/one-off/migrate-private-buckets.ts
+++ /dev/null
@@ -1,201 +0,0 @@
-// SPDX-License-Identifier: AGPL-3.0-only
-// Carbon (github.com/crbnos/carbon). Modified or adapted versions of this file,
-// including ports, remain AGPLv3; serving them over a network requires releasing their source.
-
-import { createRequire } from "node:module";
-import { readLocalScriptConfig } from "../lib/local-script-config";
-
-// One-off, idempotent migration: copy the legacy shared `private` bucket's
-// objects into each company's own bucket (bucket id = companyId). Object keys
-// are identical in both buckets (they start with `${companyId}/`).
-// Nothing is ever deleted. Re-running skips objects that already exist.
-//
-// COVERAGE: only objects under a LIVE company's `${companyId}/` prefix are
-// copied. Pre-migration audit archives at `audit-logs/{companyId}/...` are moved
-// by `relocate-legacy-audit-archives.ts`; prefixes of deleted companies are not
-// read by anything.
-//
-// Usage:
-// pnpm exec tsx scripts/one-off/migrate-private-buckets.ts # migrate
-// pnpm exec tsx scripts/one-off/migrate-private-buckets.ts --dry-run # read-only
-
-const LEGACY_BUCKET = "private";
-const BUCKET_FILE_SIZE_LIMIT = 52428800; // 50 MB
-const LIST_PAGE_SIZE = 1000;
-
-const isDryRun = process.argv.includes("--dry-run");
-
-const { SUPABASE_URL: supabaseUrl, SUPABASE_SERVICE_ROLE_KEY: serviceRoleKey } =
- readLocalScriptConfig(
- ["SUPABASE_URL", "SUPABASE_SERVICE_ROLE_KEY"],
- process.env
- );
-
-function createServiceRoleClient(url: string, serviceRole: string) {
- // The database workspace declares and pins this dependency; root scripts do not.
- const fromDatabase = createRequire(
- new URL("../../packages/database/package.json", import.meta.url)
- );
- const { createClient } = fromDatabase("@supabase/supabase-js");
- return createClient(url, serviceRole, {
- auth: { autoRefreshToken: false, persistSession: false },
- });
-}
-
-const client = createServiceRoleClient(supabaseUrl, serviceRoleKey);
-
-type CompanySummary = { copied: number; skipped: number; failed: number };
-
-function isAlreadyExistsError(error: {
- message?: string;
- statusCode?: string | number;
-}) {
- const message = error.message ?? "";
- return (
- /already exists/i.test(message) ||
- /duplicate/i.test(message) ||
- String(error.statusCode) === "409"
- );
-}
-
-async function getCompanyIds(): Promise<string[]> {
- const ids: string[] = [];
- for (let offset = 0; ; offset += LIST_PAGE_SIZE) {
- const { data, error } = await client
- .from("company")
- .select("id")
- .order("id")
- .range(offset, offset + LIST_PAGE_SIZE - 1);
- if (error)
- throw new Error(`Failed to read company table: ${error.message}`);
- ids.push(...data.map((row: { id: string }) => row.id));
- if (data.length < LIST_PAGE_SIZE) break;
- }
- return ids;
-}
-
-// Recursively list every object key under `prefix` in `bucket`.
-// Storage list returns folders as entries with id === null; recurse into them.
-async function listObjectKeys(
- bucket: string,
- prefix: string
-): Promise<string[]> {
- const keys: string[] = [];
- for (let offset = 0; ; offset += LIST_PAGE_SIZE) {
- const { data, error } = await client.storage
- .from(bucket)
- .list(prefix, { limit: LIST_PAGE_SIZE, offset });
- if (error) {
- throw new Error(`Failed to list ${bucket}/${prefix}: ${error.message}`);
- }
- for (const entry of data as Array<{ name: string; id: string | null }>) {
- if (entry.id === null) {
- keys.push(...(await listObjectKeys(bucket, `${prefix}/${entry.name}`)));
- } else {
- keys.push(`${prefix}/${entry.name}`);
- }
- }
- if (data.length < LIST_PAGE_SIZE) break;
- }
- return keys;
-}
-
-// Look before creating: Postgres logs every rejected duplicate insert as an
-// error, so creating unconditionally logged one `buckets_pkey` error per
-// company on every run.
-async function bucketExists(companyId: string): Promise<boolean> {
- const { error } = await client.storage.getBucket(companyId);
- if (!error) return true;
- if (String(error.statusCode) === "404") return false;
- throw new Error(`Failed to read bucket ${companyId}: ${error.message}`);
-}
-
-async function createBucket(companyId: string) {
- const { error } = await client.storage.createBucket(companyId, {
- public: false,
- fileSizeLimit: BUCKET_FILE_SIZE_LIMIT,
- });
- if (error && !isAlreadyExistsError(error)) {
- throw new Error(`Failed to create bucket ${companyId}: ${error.message}`);
- }
-}
-
-async function migrateCompany(companyId: string): Promise<CompanySummary> {
- const summary: CompanySummary = { copied: 0, skipped: 0, failed: 0 };
-
- const exists = await bucketExists(companyId);
- if (!exists && !isDryRun) {
- await createBucket(companyId);
- }
-
- const keys = await listObjectKeys(LEGACY_BUCKET, companyId);
- // Keys already in the company bucket are skipped rather than copied again:
- // a duplicate copy is another rejected insert, and another logged error.
- // A bucket that did not exist yet has nothing in it to list.
- const alreadyMigrated = new Set(
- exists ? await listObjectKeys(companyId, companyId) : []
- );
-
- for (const key of keys) {
- if (alreadyMigrated.has(key)) {
- summary.skipped += 1;
- continue;
- }
-
- if (isDryRun) {
- process.stdout.write(
- ` would copy ${LEGACY_BUCKET}/${key} -> ${companyId}/${key}\n`
- );
- summary.copied += 1;
- continue;
- }
-
- const { error } = await client.storage
- .from(LEGACY_BUCKET)
- .copy(key, key, { destinationBucket: companyId });
-
- if (!error) {
- summary.copied += 1;
- } else if (isAlreadyExistsError(error)) {
- summary.skipped += 1;
- } else {
- summary.failed += 1;
- process.stderr.write(
- ` FAILED ${companyId}: ${key} — ${error.message}\n`
- );
- }
- }
-
- return summary;
-}
-
-(async () => {
- if (isDryRun) {
- process.stdout.write("Dry run — no buckets created, no objects copied.\n");
- }
-
- const companyIds = await getCompanyIds();
- process.stdout.write(`Found ${companyIds.length} companies.\n`);
-
- const total: CompanySummary = { copied: 0, skipped: 0, failed: 0 };
-
- for (const companyId of companyIds) {
- process.stdout.write(`\nCompany ${companyId}\n`);
- const summary = await migrateCompany(companyId);
- total.copied += summary.copied;
- total.skipped += summary.skipped;
- total.failed += summary.failed;
- process.stdout.write(` ${JSON.stringify(summary)}\n`);
- }
-
- process.stdout.write(
- `\nTotal${isDryRun ? " (dry run)" : ""}: ${JSON.stringify(total)}\n`
- );
-
- if (total.failed > 0) {
- process.exitCode = 1;
- }
-})().catch((err) => {
- process.stderr.write(`Migration failed: ${(err as Error).message}\n`);
- process.exitCode = 1;
-});
diff --git a/scripts/one-off/recopy-private-buckets.ts b/scripts/one-off/recopy-private-buckets.ts
deleted file mode 100644
index a64a253971..0000000000
--- a/scripts/one-off/recopy-private-buckets.ts
+++ /dev/null
@@ -1,16 +0,0 @@
-// SPDX-License-Identifier: AGPL-3.0-only
-// Carbon (github.com/crbnos/carbon). Modified or adapted versions of this file,
-// including ports, remain AGPLv3; serving them over a network requires releasing their source.
-
-// Second pass of migrate-private-buckets.ts. The first pass ran while the
-// legacy `private` bucket was still writable, so files written there after it
-// exist only in `private`. 20261002022403_private-bucket-read-only.sql closes the
-// bucket to sessions and API keys in the same deploy, before this runs. The
-// service role bypasses that policy, so this pass is final only while no
-// service-role code writes to `private`.
-//
-// Same copy, new ledger name: the script is idempotent and skips objects that
-// are already in the company bucket.
-//
-// Safe to remove once every deployment has run it.
-import "./migrate-private-buckets";
diff --git a/scripts/one-off/relocate-legacy-audit-archives.ts b/scripts/one-off/relocate-legacy-audit-archives.ts
deleted file mode 100644
index f481377996..0000000000
--- a/scripts/one-off/relocate-legacy-audit-archives.ts
+++ /dev/null
@@ -1,168 +0,0 @@
-// SPDX-License-Identifier: AGPL-3.0-only
-// Carbon (github.com/crbnos/carbon). Modified or adapted versions of this file,
-// including ports, remain AGPLv3; serving them over a network requires releasing their source.
-
-import { createRequire } from "node:module";
-import { readLocalScriptConfig } from "../lib/local-script-config";
-
-// One-off, idempotent: move the audit-log archives written before each company
-// had its own bucket. They live in the retired shared `private` bucket at
-// `audit-logs/{companyId}/…`, and `auditLogArchive.archivePath` still points
-// there. Each is copied to `{companyId}/audit-logs/…` in the company's own
-// bucket (the layout new archives use), then its row is repointed. Nothing is
-// deleted. Re-running skips rows that already point at the company bucket.
-//
-// The apps stopped reading `private` in the same change that added this
-// script, so an archive this script has not moved cannot be downloaded.
-// Safe to remove once every deployment has run it. CI runs it once per
-// database after the migrations (scripts/one-off/README.md); `--dry-run`
-// lists what it would move.
-//
-// An archive whose file is gone is reported and left as it is: nothing can
-// bring it back, and failing on it would fail every later deploy. Any other
-// failure exits ONE_OFF_SCRIPT_DEFERRED, so the next deploy retries.
-
-/** ONE_OFF_SCRIPT_DEFERRED in ci/src/one-off-scripts.ts. */
-const DEFERRED = 75;
-const LEGACY_BUCKET = "private";
-const LEGACY_PREFIX = "audit-logs/";
-const PAGE_SIZE = 1000;
-
-const isDryRun = process.argv.includes("--dry-run");
-
-const { SUPABASE_URL: supabaseUrl, SUPABASE_SERVICE_ROLE_KEY: serviceRoleKey } =
- readLocalScriptConfig(
- ["SUPABASE_URL", "SUPABASE_SERVICE_ROLE_KEY"],
- process.env
- );
-
-function createServiceRoleClient(url: string, serviceRole: string) {
- // The database workspace declares and pins this dependency; root scripts do not.
- const fromDatabase = createRequire(
- new URL("../../packages/database/package.json", import.meta.url)
- );
- const { createClient } = fromDatabase("@supabase/supabase-js");
- return createClient(url, serviceRole, {
- auth: { autoRefreshToken: false, persistSession: false }
- });
-}
-
-const client = createServiceRoleClient(supabaseUrl, serviceRoleKey);
-
-type Archive = { id: string; companyId: string; archivePath: string };
-
-/** `audit-logs/{companyId}/rest` → `{companyId}/audit-logs/rest`, or null. */
-function relocatedArchivePath(archive: Archive): string | null {
- const [root, companyId, ...rest] = archive.archivePath.split("/");
- if (`${root}/` !== LEGACY_PREFIX) return null;
- if (companyId !== archive.companyId || rest.length === 0) return null;
- return `${archive.companyId}/audit-logs/${rest.join("/")}`;
-}
-
-function isAlreadyExistsError(error: {
- message?: string;
- statusCode?: string | number;
-}) {
- return (
- /already exists|duplicate/i.test(error.message ?? "") ||
- String(error.statusCode) === "409"
- );
-}
-
-function isNotFoundError(error: {
- message?: string;
- statusCode?: string | number;
-}) {
- return (
- /not found/i.test(error.message ?? "") ||
- String(error.statusCode) === "404"
- );
-}
-
-async function legacyArchives(): Promise<Archive[]> {
- const rows: Archive[] = [];
- for (let offset = 0; ; offset += PAGE_SIZE) {
- const { data, error } = await client
- .from("auditLogArchive")
- .select("id, companyId, archivePath")
- .like("archivePath", `${LEGACY_PREFIX}%`)
- .order("id")
- .range(offset, offset + PAGE_SIZE - 1);
- if (error) {
- throw new Error(`Failed to read auditLogArchive: ${error.message}`);
- }
- rows.push(...(data as Archive[]));
- if (data.length < PAGE_SIZE) break;
- }
- return rows;
-}
-
-async function relocate(
- archive: Archive
-): Promise<"moved" | "missing" | "failed"> {
- const target = relocatedArchivePath(archive);
- if (!target) {
- process.stderr.write(
- ` FAILED ${archive.id}: ${archive.archivePath} is not under audit-logs/${archive.companyId}/\n`
- );
- return "failed";
- }
- if (isDryRun) {
- process.stdout.write(
- ` would move ${LEGACY_BUCKET}/${archive.archivePath} -> ${archive.companyId}/${target}\n`
- );
- return "moved";
- }
-
- const copied = await client.storage
- .from(LEGACY_BUCKET)
- .copy(archive.archivePath, target, { destinationBucket: archive.companyId });
- if (copied.error && isNotFoundError(copied.error)) {
- process.stderr.write(
- ` MISSING ${archive.id}: ${LEGACY_BUCKET}/${archive.archivePath} is gone; left as it is\n`
- );
- return "missing";
- }
- if (copied.error && !isAlreadyExistsError(copied.error)) {
- process.stderr.write(
- ` FAILED ${archive.id}: copy — ${copied.error.message}\n`
- );
- return "failed";
- }
-
- // Compare-and-set on the old path, scoped to the company.
- const updated = await client
- .from("auditLogArchive")
- .update({ archivePath: target })
- .eq("id", archive.id)
- .eq("companyId", archive.companyId)
- .eq("archivePath", archive.archivePath);
- if (updated.error) {
- process.stderr.write(
- ` FAILED ${archive.id}: repoint — ${updated.error.message}\n`
- );
- return "failed";
- }
- return "moved";
-}
-
-(async () => {
- if (isDryRun) {
- process.stdout.write("Dry run — nothing copied, no row changed.\n");
- }
- const archives = await legacyArchives();
- process.stdout.write(`Found ${archives.length} legacy audit archives.\n`);
-
- const total = { moved: 0, missing: 0, failed: 0 };
- for (const archive of archives) {
- total[await relocate(archive)] += 1;
- }
-
- process.stdout.write(
- `Total${isDryRun ? " (dry run)" : ""}: ${JSON.stringify(total)}\n`
- );
- if (total.failed > 0) process.exitCode = DEFERRED;
-})().catch((err) => {
- process.stderr.write(`Relocation failed: ${(err as Error).message}\n`);
- process.exitCode = 1;
-});
diff --git a/scripts/restore-database.sh b/scripts/restore-database.sh
index 8035429e0c..2f0db291f0 100644
--- a/scripts/restore-database.sh
+++ b/scripts/restore-database.sh
@@ -489,7 +489,6 @@ $PSQL_SA -v ON_ERROR_STOP=0 -c "
INSERT INTO storage.buckets (id, name, public) VALUES
('public', 'public', true),
('avatars', 'avatars', true),
- ('private', 'private', false),
('feedback', 'feedback', true),
('company-templates', 'company-templates', false)
ON CONFLICT (id) DO NOTHING;
@@ -501,9 +500,9 @@ cat "$WORK_DIR/buckets.err" >> "$RESTORE_LOG"
# Every company needs its bucket (uploads go there), so a short count is a
# broken restore, not a number to eyeball.
BUCKET_COUNT=$($PSQL_PG -At -c "SELECT count(*) FROM storage.buckets;" 2>/dev/null || echo 0)
-BUCKETS_EXPECTED=$($PSQL_PG -At -c "SELECT 5 + count(*) FROM public.company;" 2>/dev/null || echo 5)
+BUCKETS_EXPECTED=$($PSQL_PG -At -c "SELECT 4 + count(*) FROM public.company;" 2>/dev/null || echo 4)
if [[ "$BUCKET_COUNT" -ge "$BUCKETS_EXPECTED" ]]; then
- echo " ✓ $BUCKET_COUNT storage buckets present (5 fixed + one per company)"
+ echo " ✓ $BUCKET_COUNT storage buckets present (4 fixed + one per company)"
else
RESTORE_UNVERIFIED=1
echo " ✗ $BUCKET_COUNT of $BUCKETS_EXPECTED storage buckets present — the re-seed failed:"