nautechsystems/nautilus_trader · error

Python on_save failed: {e}

Error message

Python on_save failed: {e}

What it means

The strategy's Python-level `on_save` callback raised an exception (or returned data that could not be converted). `dispatch_on_save()` (strategy.rs:384) calls `on_save` on the Python instance, then `cast_bound::<PyDict>` and `pydict_to_state` to convert the returned dict into `IndexMap<String, Vec<u8>>`; any failure is wrapped at strategy.rs:1096 as `Python on_save failed: {e}`.

Source

Thrown at crates/trading/src/python/strategy.rs:1096

    fn on_dispose(&mut self) -> anyhow::Result<()> {
        self.dispatch_on_dispose()
            .map_err(|e| anyhow::anyhow!("Python on_dispose failed: {e}"))
    }

    fn on_degrade(&mut self) -> anyhow::Result<()> {
        self.dispatch_on_degrade()
            .map_err(|e| anyhow::anyhow!("Python on_degrade failed: {e}"))
    }

    fn on_fault(&mut self) -> anyhow::Result<()> {
        self.dispatch_on_fault()
            .map_err(|e| anyhow::anyhow!("Python on_fault failed: {e}"))
    }

    fn on_save(&self) -> anyhow::Result<IndexMap<String, Vec<u8>>> {
        self.dispatch_on_save()
            .map_err(|e| anyhow::anyhow!("Python on_save failed: {e}"))
    }

    fn on_load(&mut self, state: IndexMap<String, Vec<u8>>) -> anyhow::Result<()> {
        self.dispatch_on_load(&state)
            .map_err(|e| anyhow::anyhow!("Python on_load failed: {e}"))
    }

    fn on_time_event(&mut self, event: &TimeEvent) -> anyhow::Result<()> {
        route_time_event(self, event);
        self.dispatch_on_time_event(event)
            .map_err(|e| anyhow::anyhow!("Python on_time_event failed: {e}"))
    }

    #[allow(unused_variables)]
    fn on_data(&mut self, data: &CustomData) -> anyhow::Result<()> {
        Python::attach(|py| {
            let py_data: Py<PyAny> = Py::new(py, data.clone())?.into_any();
            self.dispatch_on_data(py_data)

View on GitHub (pinned to 18893faf8b)

Solutions

  1. Check the `{e}` text: if it mentions a dict cast, ensure `on_save` returns `dict[str, bytes]` (encode strings, e.g. `value.encode()` or use pickle/json serialized to bytes).
  2. Ensure `on_save` explicitly returns the state mapping rather than None.
  3. Guard collection of each state key so a missing cache entry cannot raise.
  4. Test saving by calling the strategy's save path in a unit test before deploying.

Example fix

// before
def on_save(self):
    return {"position": str(self.position_qty)}  # str, not bytes

// after
def on_save(self):
    return {"position": str(self.position_qty).encode("utf-8")}
Defensive patterns

Strategy: validation

Validate before calling

def validate_on_save_state(state):
    assert isinstance(state, dict), "on_save must return a dict"
    for k, v in state.items():
        assert isinstance(k, str), f"key not str: {k!r}"
        assert isinstance(v, (bytes, bytearray)), f"value not bytes for key {k!r}: {type(v)}"

Type guard

def is_valid_state(state):
    return isinstance(state, dict) and all(
        isinstance(k, str) and isinstance(v, (bytes, bytearray))
        for k, v in state.items())

Try / catch

try:
    state = self.on_save()
    validate_on_save_state(state)
except Exception as e:
    self.log.error(f"on_save failed: {e}")
    state = {}

Prevention

When it happens

Trigger: A snapshot/save is requested (e.g. on stop or periodic state persistence) and the Python `on_save` raises, or returns something that is not a `dict` of `str` -> `bytes` (causing the PyDict cast or conversion to fail).

Common situations: `on_save` returning a plain dict of str->str/int instead of bytes; forgetting to `return` the state dict (returns None, cast fails); pickling state inside `on_save` with an unpicklable object; KeyError while collecting state from a cache that was never populated.

Related errors


AI-assisted analysis of nautechsystems/nautilus_trader@18893faf8b (2026-09-08). Data as JSON: /api/errors/c213e7fff47eb473. Report an issue: GitHub.