LLM-Driven Extraction of Unstructured Data — Built for API Deployments & ETL Pipeline Workflows
Python ai-agentsdata-engineeringdocument-aigenerative-aiidpjson-extractionllmmcp-serverocrpdf-extractionprompt-engineeringstructured-output
UN-4069 [FIX] Raise upload limit on prompt_output to 10 MB with clear error message (#2331)
harini-venkataraman99dbcff5273b
diff --git a/backend/prompt_studio/prompt_studio_core_v2/internal_views.py b/backend/prompt_studio/prompt_studio_core_v2/internal_views.pyindex 7d93e298a..3fcdba182 100644--- a/backend/prompt_studio/prompt_studio_core_v2/internal_views.py+++ b/backend/prompt_studio/prompt_studio_core_v2/internal_views.py@@ -20,6 +20,10 @@ from django.views.decorators.csrf import csrf_exemptfrom django.views.decorators.http import require_http_methodsfrom rest_framework import status+# 10 MB — agentic table extraction output with highlight coordinates+# can exceed Django's default 2.5 MB DATA_UPLOAD_MAX_MEMORY_SIZE.+_PROMPT_OUTPUT_MAX_BODY = 10 * 1024 * 1024+logger = logging.getLogger(__name__)_ERR_INVALID_JSON = "Invalid JSON"@@ -126,10 +130,31 @@ def prompt_output(request):"profile_manager_id": str | null,"metadata": dict}++ Uses ``request.read()`` instead of ``request.body`` to bypass+ Django's ``DATA_UPLOAD_MAX_MEMORY_SIZE`` check. Agentic table+ extraction output with highlight coordinates can exceed the default+ 2.5 MB limit; this view enforces its own 10 MB cap instead."""- data, err = _parse_json_body(request)- if err:- return err+ content_length = int(request.META.get("CONTENT_LENGTH") or 0)+ if content_length > _PROMPT_OUTPUT_MAX_BODY:+ return JsonResponse(+ {+ "success": False,+ "error": (+ f"Request body too large ({content_length} bytes, "+ f"limit {_PROMPT_OUTPUT_MAX_BODY})"+ ),+ },+ status=status.HTTP_413_REQUEST_ENTITY_TOO_LARGE,+ )+ try:+ data = json.loads(request.read())+ except json.JSONDecodeError:+ return JsonResponse(+ {"success": False, "error": _ERR_INVALID_JSON},+ status=status.HTTP_400_BAD_REQUEST,+ )run_id = data.get("run_id", "")prompt_ids = data.get("prompt_ids", [])diff --git a/backend/prompt_studio/prompt_studio_core_v2/tests/test_prompt_output_outputs_validation.py b/backend/prompt_studio/prompt_studio_core_v2/tests/test_prompt_output_outputs_validation.pyindex ba81f9785..d2992eb50 100644--- a/backend/prompt_studio/prompt_studio_core_v2/tests/test_prompt_output_outputs_validation.py+++ b/backend/prompt_studio/prompt_studio_core_v2/tests/test_prompt_output_outputs_validation.py@@ -35,7 +35,7 @@ def _request(outputs):# require_http_methods reads request.method; a MagicMock attribute is not# the string "POST" and the view short-circuits with 405.request.method = "POST"- request.body = json.dumps(+ body = json.dumps({"run_id": "run-1","prompt_ids": [_PROMPT_ID],@@ -43,7 +43,10 @@ def _request(outputs):"document_id": _DOC_ID,"is_single_pass_extract": True,}- )+ ).encode()+ request.body = body+ request.read.return_value = body+ request.META = {"CONTENT_LENGTH": str(len(body))}return request@@ -79,14 +82,17 @@ def test_non_mapping_metadata_rejected():`if not prompts` early exit — so it 500s even with valid outputs."""request = MagicMock()request.method = "POST"- request.body = json.dumps(+ body = json.dumps({"prompt_ids": [_PROMPT_ID],"document_id": _DOC_ID,"outputs": {"invoice_number": "INV-001"},"metadata": [],}- )+ ).encode()+ request.body = body+ request.read.return_value = body+ request.META = {"CONTENT_LENGTH": str(len(body))}response = prompt_output(request)assert response.status_code == 400assert "metadata must be a JSON object" in _body(response)["error"]@@ -136,7 +142,10 @@ def test_missing_outputs_defaults_to_empty_dict_and_is_accepted():"""Absent `outputs` has always meant {}; the guard must not change that."""request = MagicMock()request.method = "POST"- request.body = json.dumps({"prompt_ids": [_PROMPT_ID], "document_id": _DOC_ID})+ body = json.dumps({"prompt_ids": [_PROMPT_ID], "document_id": _DOC_ID}).encode()+ request.body = body+ request.read.return_value = body+ request.META = {"CONTENT_LENGTH": str(len(body))}with patch("prompt_studio.prompt_studio_v2.models.ToolStudioPrompt.objects") as prompts, patch(