Genroc stores large values separately and doesn't return them directly when you communicate with the API.

What is a large value?

Genroc measures each stored value on its own. When one is over its limit, genroc moves its largest parts into objects until the rest fits.

Stored value	Where you see it	Limit
Process input	state.input	2 KiB
Each task's output, separately	state.outputs.<task id>	2 KiB
Process output	output	2 KiB
The last error	state.last_error	2 KiB
A raised error's data	error_data	2 KiB
External task input	external_input	2 KiB
A log entry's payload	the entry's data	2 KiB

Ten task outputs of 1.5 KB each are therefore all kept inline: each one is under its own limit.

A log entry's payload over --log-payload-max (default 1 MiB) is not stored at all. The entry's meta.payload_omitted_bytes says how big it was.

What it looks like

Say a task downloads a web page and keeps the HTML (about 180 KB) in its output. The instance detail doesn't contain the HTML. The body field is null, and the objects section says where it was and how to get it:

{
  "id": "5fy765kh",
  "status": "completed",
  "state": {
    "outputs": {
      "download": {
        "status": 200,
        "content_type": "text/html",
        "body": null
      }
    }
  },
  "objects": [
    {
      "path": ["state", "outputs", "download", "body"],
      "ref": "9f2ac1b4e7d05f38a61c0d2e4b7f9a13",
      "size": 184320
    }
  ]
}

path points to the null field, ref is the id of the stored value and size is its size in bytes. Only the large parts are moved out: status and content_type stay where they are.

To get the value, use the ref:

> curl http://localhost:8448/api/objects/9f2ac1b4e7d05f38a61c0d2e4b7f9a13
{"data": "\"<!doctype html><html>...\""}

data is the value encoded as JSON text, so parse it once more to get the original value.

The CLI does this for you: genctl detail shows {ref, size} in place of the null, genctl detail --resolve fills it in and genctl object <ref> prints one value.

Which endpoints

Endpoint	What can be moved out	Paths start at
GET /api/instances/{id}	output, error_data, external_input	the response
GET /api/instances/{id}/detail	anything in state	the response
GET /api/instances/{id}/logs	each entry's data	the log entry
POST /api/external-tasks/claim	each item's external_input	the item

No endpoint fills the values in for you; fetch each one with GET /api/objects/{ref}.

Fetching an object needs the read or worker permission.