storage::abortMultipartUpload
- bucketstringrequired
- keystringrequired
- upload_idstringrequired
- abortedbooleanrequired
Object storage across AWS S3, GCS, Cloudflare R2, and a native local filesystem backend. Streamed uploads, signed URLs, and observable object change events.
iii trigger compose::add worker=storage@0.2.3Run in your project with the iii engine running. Exact versions are immutable and digest-pinned.
Remove a bucket from the storage configuration. Objects, local data and secrets are never deleted; the worker may run with zero buckets.
Delete an object. No-op when the object does not exist.
Read one stored bucket entry for editing: references and templates intact, literal secret keys, session tokens, credentials JSON and API tokens as "__redacted__".
Bucket name (the key under `buckets`).
The entry as stored: references and `${VAR}` templates intact, literal secret keys, session tokens, credentials JSON and API tokens as `"__redacted__"`.
Download a small object, or a byte range of any object (offset, length), inline as base64. At most 10 MiB per call; use a presignUrl GET for files when the client can reach the provider, ranged reads when it cannot.
How many bytes to return at most (clamped to the end of the object). Default: to the end.
First byte to return. Default 0. Past the end of the object → `INVALID_RANGE`; equal to its size → an empty body.
How many bytes were returned.
Where the returned bytes start in the object.
The object's total size.
Fetch object metadata (size, ETag, content-type, last-modified) without downloading the body.
List the configured buckets with their provider, provider-side bucket, region or endpoint, display, whether they use secrets and allow direct downloads, and the error of any bucket that could not be built.
r2 only: the literal account id, or the reference as stored.
The provider-side bucket.
A presigned GET (`storage::presignUrl`) gives a URL a client can fetch: false for gcs without signing credentials or with an `endpoint_url` override (read through ranged `storage::getObject`).
How clients draw the bucket (well-formed parts only).
`#RRGGBB`.
A Lucide icon name (`hard-drive`, `cloud`, …): `[a-z0-9-]`, at most 64.
The custom endpoint (S3-compatible stores, R2/GCS overrides), userinfo stripped.
Why the bucket has no backend (its calls answer `BUCKET_UNAVAILABLE`).
`config` | `secrets` | `auth`.
The worker-facing name (the key under `buckets`).
Why the bucket's notifications source is not running (the bucket works; its object events are not called).
`s3` | `r2` | `gcs` | `local` (`unknown` for an entry that names none).
s3 only.
Some credential field is a `secret://` / `env://` reference.
The configuration entry the buckets live in.
List objects and common prefixes in a bucket with provider-native pagination.
Directory separator. Use `/` for the bucket/folder explorer; omit it for a flat recursive listing.
Read the configured bucket's multipart wire contract after a successful bucket HEAD; no object bodies or writes.
Internal: apply the authoritative storage configuration without restarting the worker.
Called once per object written to a configured bucket whose notifications source is wired up; returns the event { bucket, key, size, etag, content_type, created_at, raw_event_id }. Observers pass args { bucket, handler_timeout_ms? } (default 60000); each target receives the event and acks it by returning null or { ack: true }. The caller gets { acked }: true when every matching target acked, and the notifications source acks the upstream message only then. Only the storage worker calls it.
Called once per object removed from a configured bucket whose notifications source is wired up; returns the event { bucket, key, deleted_at, raw_event_id }. Observers pass args { bucket, handler_timeout_ms? } (default 60000); each target receives the event and acks it by returning null or { ack: true }. The caller gets { acked }: true when every matching target acked, and the notifications source acks the upstream message only then. Only the storage worker calls it.
Internal: re-read a rotated or revoked secret:// / env:// value and rebuild the buckets that reference it.
The handler the call was made from; absent for a top-level call.
The caller of that handler, one level only.
`created` | `rotated` | `deleted` | `access_changed` (advisory).
Issue a short-lived multipart/form-data POST for direct browser upload. The file never buffers in the engine or worker RPC path.
Optional upload cap enforced while the HTTP body is streamed.
Issue a short-lived URL the browser can hit directly to PUT or GET an object.
GET-only: override `Content-Disposition` header on the served response. Rejected with `INVALID_PRESIGN_PARAMS` if set on a PUT presign.
GET-only: override `Content-Type` header on the served response. Rejected with `INVALID_PRESIGN_PARAMS` if set on a PUT presign.
Upload a small object inline as base64 (10 MiB hard limit). This buffers and inflates the payload; use presignPost or a presignUrl PUT for files and large objects.
Base64-encoded small body (10 MiB decoded hard limit). The legacy field name `body` is accepted as an alias for backward compatibility.
Create, update (previous_name = name) or rename (previous_name = old) a bucket in the storage configuration; "__redacted__" keeps a stored value, credentials go as secret:// references or in `credentials`, create_bucket creates an s3/r2 bucket that does not exist. Returns after the runtime reloaded, with the bucket's error if any.
The entry (`provider` and its fields). `"__redacted__"` keeps the stored value of that field. Sensitive fields hold references here; literals go in `credentials`.
s3 / r2: create the provider-side bucket when it does not exist. Local buckets always ensure their storage.
Sensitive values sent apart from `bucket` and merged over it: `access_key_id`, `secret_access_key`, `session_token`, `credentials_json`, `notifications_api_token`. Prefer `secret://` references in `bucket`, which are safe anywhere.
The name to save under: `[a-z0-9][a-z0-9_-]{0,62}`.
The bucket's current name when editing an existing one (equal to `name` to update in place, different to rename). Absent creates a new bucket, refused when `name` is taken.
With `create_bucket`: whether this save created the provider-side bucket (false: it already existed).
Why the saved bucket has no backend after the reload, or why its creation failed (it is saved regardless). Absent when it is usable.
Test a candidate bucket entry (or a stored one by name) with a throwaway client and freshly resolved secrets: lists one object at the root. Live buckets are untouched; reports ok/provider/latency, the failing stage (config, secrets, endpoint, auth, bucket, list) and bucket_missing; failures are data, not errors.
The candidate entry (`provider` and its fields), as `saveBucket` takes it. Absent: test the stored bucket `name` as it is.
Sensitive values apart from `bucket` (`access_key_id`, `secret_access_key`, `session_token`, `credentials_json`, `notifications_api_token`), merged over it.
The stored bucket `"__redacted__"` values come from (and the default provider-side bucket when the entry has no `bucket`).
Overall budget. Default 10000, capped at 30000.
The provider said the bucket does not exist (`saveBucket` with `create_bucket: true` creates it, for s3 and r2).
Whether presigned GETs from this bucket are fetchable by clients. Present once the backend was built.
Wall time of the whole attempt.
Why the probe failed (credentials scrubbed). Absent on success.
Whether the bucket was reached and listed.
`s3` | `r2` | `gcs` | `local` | `unknown`.
`config` | `secrets` | `endpoint` | `auth` | `bucket` | `list`. Absent on success.
Serve the storage worker's injected console UI assets (the target of its ade::ui::asset observers; the console calls it through each asset's observer with {path}).
The asset path from the observer's args (e.g. `state/page.js`).
The asset source, verbatim.
MIME type the console should serve the asset with.
Find workers, kits and executables, or jump to a page.