StalledFileDetail
A file that never got a chance to be (re-)attempted this migration. Nightshift processes one file at a time; when another file's permanent failure flips the whole migration to `failed` first, any file whose message hadn't been picked up yet is skipped without ever being attempted (see `reembed_file_for_migration`'s early exit when the migration is no longer `migrating` - required to avoid a duplicate-processing race against a concurrent retry). It stays `pending` forever unless surfaced explicitly here; a `retry` re-dispatches it same as any other non-completed file.
A file that never got a chance to be (re-)attempted this migration. Nightshift processes one file at a time; when another file's permanent failure flips the whole migration to failed first, any file whose message hadn't been picked up yet is skipped without ever being attempted (see reembed_file_for_migration's early exit when the migration is no longer migrating - required to avoid a duplicate-processing race against a concurrent retry). It stays pending forever unless surfaced explicitly here; a retry re-dispatches it same as any other non-completed file.
| Property | Type | Required | Description |
|---|---|---|---|
vector_store_id | string | Yes | A 24-character hexadecimal MongoDB ObjectId string. |
file_id | string | Yes | A 24-character hexadecimal MongoDB ObjectId string. |
We use cookies on our website. We use them to ensure proper functioning of the site and, if you agree, for purposes such as analytics, marketing, and targeting ads.
SourceFile
One source document as it was when a run started; files are immutable, so ID and size identify it.
StartMigrationRequest
Request body for `POST /v1/migrate`. Deliberately has no `project_id` field: the project is always the caller's own authenticated project (from `ProjectAuth`, resolved from the request's auth headers), never a client-supplied value — otherwise a caller could trigger a migration for a project that isn't theirs by simply naming a different id. Has no `max_context_size` either, unlike `MigrationTarget`: that's resolved server-side from infra (`GET /v1/models/{model}?additional_data=true`), the same as `create_vector_store` does - not taken from the client, which would otherwise let a wrong or stale value get written into `embedding_config` at cutover (silently capping chunk size and query length below the target model's real context window).