langflow-ai/langflow · error · UserComponentError

code size bytes exceeds limit of bytes

Error message

code size {encoded_size} bytes exceeds limit of {MAX_COMPONENT_SOURCE_BYTES} bytes

What it means

Raised when the UTF-8 encoded component source exceeds MAX_COMPONENT_SOURCE_BYTES (1 MiB). Real generated components are far under 1 MB; the cap exists to catch runaway model outputs (repetition loops) and abuse. The size is measured on the encoded bytes, not characters.

Solutions

Move large embedded data out of the component and load it from file storage or a URL at runtime.
If the code is a runaway model output, regenerate with a shorter max-length constraint.
As a last resort raise MAX_COMPONENT_SOURCE_BYTES in your fork — but 1 MiB signals something wrong with the input.

Example fix

# before
code = 'class Big:\n    DATA = "' + base64.b64encode(huge_file).decode() + '"'
register_user_component(user_id=uid, class_name="Big", code=code)

# after
code = 'class Big:\n    def load(self):\n        return download_or_read_from_storage("big.bin")'
register_user_component(user_id=uid, class_name="Big", code=code)

Defensive patterns

Strategy: validation

Validate before calling

MAX_BYTES = 1 * 1024 * 1024

def code_within_limit(code: str) -> bool:
    return len(code.encode("utf-8")) <= MAX_BYTES

Try / catch

try:
    register_user_component(user_id=uid, class_name=name, code=code)
except UserComponentError as e:
    if "exceeds limit" in str(e):
        code = truncate_or_externalize(code)  # move data blobs out

Prevention

Constrain model output max_tokens for component generation.
Never embed base64/data URIs in component source; load data at runtime.
Check encoded byte size (not character count) before registering.

When it happens

Trigger: register_user_component with a code payload larger than 1048576 bytes — e.g. a model stuck in a repetition loop emitting megabytes of similar lines, or embedded base64 blobs/data URIs inside the component.

Common situations: LLM degenerate repetition during long streams; developers embedding large lookup tables or serialized model weights directly in component code instead of loading from storage.

Related errors

AI-assisted analysis of langflow-ai/langflow@976ec789d2 (2026-08-14). Data as JSON: /api/errors/4340768e10c7dfef. Report an issue: GitHub.

Appendix: source

Thrown at src/backend/base/langflow/agentic/services/user_components.py:243

    if not _CLASS_NAME_RE.fullmatch(class_name):
        msg = (
            f"class_name must be a CamelCase identifier "
            f"(letters/digits/underscores, leading uppercase). Got: {class_name!r}"
        )
        raise UserComponentError(msg)
    if class_name.upper() in _WINDOWS_RESERVED_DEVICES:
        msg = f"class_name {class_name!r} is a Windows-reserved device name"
        raise UserComponentError(msg)


def _validate_code(code: str) -> None:
    if not code or not code.strip():
        msg = "code must be a non-empty string"
        raise UserComponentError(msg)
    encoded_size = len(code.encode("utf-8"))
    if encoded_size > MAX_COMPONENT_SOURCE_BYTES:
        msg = f"code size {encoded_size} bytes exceeds limit of {MAX_COMPONENT_SOURCE_BYTES} bytes"
        raise UserComponentError(msg)


def _resolve_components_dir(*, user_id: str | None) -> Path:
    """Resolve and create ``<sandbox>/.components/`` for the given user.

    Reuses the FS tool's authoritative sandbox resolver so the hash
    function, pepper handling, AUTO_LOGIN dispatch, and no-user refusal
    stay in one place. The reserved-segment guard does NOT apply here —
    this helper is the privileged writer that the guard is protecting.
    """
    component = FileSystemToolComponent()
    if user_id is not None:
        component._user_id = user_id  # noqa: SLF001 — privileged binding seam
    try:
        sandbox_root = component._validate_root()  # noqa: SLF001
    except PermissionError as exc:
        # PermissionError from _validate_root happens in two cases:
        # 1. AUTO_LOGIN=False and no user_id → translate to our domain error.

View on GitHub (pinned to 976ec789d2)