Error Handling Skill

Unified error handling system for DF. Use when adding API endpoints, modifying error handling, or adding frontend API calls.

Prerequisites: Read docs/dev-guides/7-unified-error-handling.md before changing API error behavior.
Read docs/dev-guides/2-log-sanitization.md when the work involves logging, credentials, external services, or DataLoaders.
If your work introduces new error handling patterns or conventions, update this file and related dev-guides accordingly.

Architecture Overview

Frontend                              Backend
────────                              ───────
apiClient.ts                          errors.py
├── apiRequest()    ←── JSON ────     ├── ErrorCode (enum)
├── streamRequest() ←── NDJSON ──     └── AppError (exception)
└── parseStreamLine()
                                      error_handler.py
errorCodes.ts                         ├── register_error_handlers(app)
└── getErrorMessage()                 ├── classify_and_wrap_llm_error()
                                      └── stream_error_event()
errorHandler.ts
└── handleApiError()                  security/sanitize.py
                                      └── classify_llm_error() (internal)
MessageSnackbar ← dfSlice.messages

Protocol Snapshot

Use this contract for all new or reworked DF APIs:

Do not use HTTP 400/422 for application validation errors in new code. Do not convert in-flight NDJSON errors to status: "error"; once the stream has started, event type is the protocol discriminator.

Backend: Adding a New API Endpoint

HTTP Status Code Policy

Application-controlled business and validation errors return HTTP 200 with status: "error" in the body. Only these use non-200:

• 401/403 — auth errors (AUTHREQUIRED, AUTHEXPIRED, ACCESS_DENIED)
• 404 — no matching Flask route
• 413 — WSGI body limit exceeded
• 500 — unhandled exception (program bug)

Non-streaming endpoint

from data_formulator.errors import AppError, ErrorCode
from data_formulator.error_handler import json_ok

@bp.route('/my-endpoint', methods=['POST'])
def my_endpoint():
    content = request.get_json()
    if not content.get('required_field'):
        raise AppError(ErrorCode.INVALID_REQUEST, "Missing required_field")

    try:
        result = do_work(content)
    except SomeBusinessError as e:
        raise AppError(ErrorCode.DATA_LOAD_ERROR, "Failed to load data") from e
    except Exception as e:
        from data_formulator.error_handler import classify_and_wrap_llm_error
        raise classify_and_wrap_llm_error(e) from e

    return json_ok(result)
# Global handler returns: HTTP 200 + {"status": "error", "error": {code, message, retry}}
# Auth errors (AUTH_REQUIRED/AUTH_EXPIRED/ACCESS_DENIED) return 401/403

Legacy {"status": "error", "message": "..."}, errormessage, bare {error}, and status: "ok" responses are historical formats. Do not add new compatibility branches for them; migrate the route to jsonok() / AppError before using apiRequest().

Streaming endpoint

Validation MUST be outside the generator. Failures return 200 JSON (not NDJSON).

from data_formulator.errors import AppError, ErrorCode
from data_formulator.error_handler import (
    classify_and_wrap_llm_error,
    stream_error_event,
    stream_preflight_error,
)

@bp.route('/my-stream', methods=['POST'])
def my_stream():
    if not request.is_json:
        return stream_preflight_error(
            AppError(ErrorCode.INVALID_REQUEST, "Invalid request")
        )

    content = request.get_json()
    client = get_client(content['model'])

    def generate():
        try:
            for event in agent.run(...):
                yield json.dumps(event, ensure_ascii=False) + "\n"
        except Exception as e:
            yield stream_error_event(classify_and_wrap_llm_error(e))

    return Response(stream_with_context(generate()), mimetype='application/x-ndjson')

Streaming runtime errors intentionally use {"type": "error", "error": ...}. They cannot use a top-level status envelope because the HTTP response and NDJSON event stream have already started.

Frontend: Consuming an API

Non-streaming

import { apiRequest } from '../app/apiClient';
import { handleApiError } from '../app/errorHandler';

try {
    const { data } = await apiRequest<ResponseType>(getUrls().MY_ENDPOINT, {
        method: 'POST',
        body: JSON.stringify(payload),
        headers: { 'Content-Type': 'application/json' },
    });
} catch (e) {
    handleApiError(e, 'MyComponent');
}

For UI loading state, model the request lifecycle explicitly with LoadableState from src/app/loadableState.ts. Do not infer loading from missing data (!data), because failed requests may legitimately leave data empty while loading has ended.

Streaming

import { streamRequest } from '../app/apiClient';
import { handleApiError } from '../app/errorHandler';

try {
    for await (const event of streamRequest(url, options, abortController.signal)) {
        switch (event.type) {
            case 'text_delta':
                break;
            case 'error':
                // Error arrived mid-stream — show inline in component.
                break;
            case 'done':
                break;
        }
    }
} catch (e) {
    handleApiError(e, 'MyComponent');
}

With callbacks

handleApiError(e, 'MyComponent', {
    onAuth: () => redirectToLogin(),        // AUTH_REQUIRED / AUTH_EXPIRED
    onRetryable: () => retryOperation(),    // LLM_RATE_LIMIT / LLM_TIMEOUT
    silent: true,                           // don't show Snackbar (component handles display)
});

Migration and special cases

DF API consumers should use apiRequest() / streamRequest() and handleApiError(). Direct fetchWithIdentity() is for lower-level client helpers and explicit protocol exceptions such as file downloads, blob/CSV responses, OIDC redirects, SPA fallback, or third-party URLs.

Do not apply the normal JSON API protocol mechanically to file downloads / CSV streaming, SPA fallback, OIDC redirect flows, frontend fetches to third-party URLs, or errors after a streaming response has already started. Check the route's protocol first, then preserve safe error bodies and avoid str(exc) exposure.

Adding a New Error Code

1. Backend — Add to py-src/data_formulator/errors.py ErrorCode:

``python MYNEWERROR = "MYNEWERROR" ` No HTTP mapping needed — defaults to HTTP 200. Only add to ERRORCODEHTTP_STATUS` if it's an auth code.

1. Frontend mapping — Add to src/app/errorCodes.ts ERRORCODEI18N_MAP:

``typescript MYNEWERROR: 'errors.myNewError', ``

1. Translations — Add to both locale files:

- src/i18n/locales/en/errors.json: "myNewError": "English message" - src/i18n/locales/zh/errors.json: "myNewError": "中文消息"

Migrated Endpoints Reference

All streaming endpoints are now on the unified protocol:

Non-streaming endpoints:

Empty Catch Policy

Not all .catch(() => {}) are bugs. Use this decision tree:

1. User-initiated action (delete, refresh, submit) → must notify with addMessages or handleApiError()
2. Background/best-effort fetch (connector list on mount, session list) → OK to swallow, but add a comment
3. RTK thunks → always add .rejected handler with addMessages
4. AbortError → filter out with if (action.error?.name !== 'AbortError')

Frontend Stream Parsing Pattern

When consuming a migrated streaming endpoint, handle the current NDJSON event format directly:

const data = JSON.parse(line);
if (data.type === 'error') {
    const errMsg = data.error?.message || 'Unknown error';
    // show to user...
}

Backend: Database/Workspace Errors (tables.py)

For table CRUD endpoints, use the specialized classifier:

from data_formulator.routes.tables import classify_and_raise_db_error

@tables_bp.route('/my-table-op', methods=['POST'])
def my_table_op():
    try:
        result = workspace.do_something()
        return jsonify({"status": "success", "data": result})
    except Exception as e:
        classify_and_raise_db_error(e)

classifyandraisedberror maps common DB errors to appropriate AppError codes (returned as HTTP 200 by the global handler, except ACCESS_DENIED → 403):

• "Table does not exist" → TABLENOTFOUND (HTTP 200)
• "Table already exists" → INVALID_REQUEST (HTTP 200)
• "Permission denied" → ACCESS_DENIED (HTTP 403)
• Other → CONNECTOR_ERROR (HTTP 200)

Backend: Connector Errors (data_connector.py)

For connector endpoints, use:

from data_formulator.data_connector import classify_and_raise_connector_error

except Exception as e:
    classify_and_raise_connector_error(e, operation="preview")

Connector/DataLoader classification is intentionally simple and lives in dataformulator.dataloader.connectorerrors. It maps common failures to a small stable set: INVALIDREQUEST, CONNECTORAUTHFAILED, AUTHEXPIRED, ACCESSDENIED, DBCONNECTIONFAILED, DBQUERYERROR, DATALOADERROR, or CONNECTOR_ERROR. Do not add endpoint-local string matching unless the classifier cannot reasonably cover the category.

All JSON errors include error.requestid and an X-Request-Id response header. Show/copy this ID for users when reporting backend failures; do not show raw exception text in production. Unhandled 500 responses must never include raw tracebacks, even in debug mode; return a safe category plus requestid and keep full stack traces in server logs only.

Debugging Error Propagation

When an error isn't reaching the frontend:

1. Check backend logs — is the error logged?
2. Check response format:

- Non-streaming: {"status": "error", "error": {"code": ..., "message": ...}} - Streaming: one line {"type": "error", "error": {"code": ..., "message": ...}}

1. Check Content-Type — streaming must be application/x-ndjson, not application/json or text/event-stream
2. Check frontend parser — is the consumer looking for data.type === 'error'?
3. Check global handler — verify registererrorhandlers(app) is called in app.py
4. Check blueprint handlers — blueprint-level errorhandler(Exception) takes priority over global handlers

Legacy message / error_message bodies are protocol violations on migrated API paths.

Log Sanitization (Sensitive Data in Server Logs)

Server-side logs must never leak passwords, tokens, API keys, or connection strings. The project uses a defense-in-depth approach with two layers.

Layer 1: Explicit Utilities (call-site)

from data_formulator.security.log_sanitizer import (
    sanitize_url, sanitize_params, redact_token,
)

# Dict with credentials → sanitize_params()
log.info("Connecting with: %s", sanitize_params(params))

# URL that may embed credentials → sanitize_url()
logger.info("Issuer: %s", sanitize_url(issuer_url))

# Token/API key → redact_token()
logger.debug("Token: %s", redact_token(token))

Layer 2: SensitiveDataFilter (global safety net)

Registered in app.py:configure_logging(). Automatically redacts:

• URL credentials (://user:pass@host)
• Bearer tokens
• password=xxx, api_key=xxx, secret=xxx patterns
• JWT-like base64 strings
• Python dict repr with sensitive keys

Disable with LOG_SANITIZE=false for local debugging only.

When to Use What

New Module Checklist

When adding a module that handles credentials or external services:

1. Audit all logger.*() calls for credential/URL/token logging
2. Use sanitizeparams() for dicts, sanitizeurl() for URLs, redact_token() for tokens
3. Prefer type(exc).name over str(exc) in warning-level logs
4. If introducing new credential key names, add to SENSITIVEKEYS in logsanitizer.py

Key Files