Portfolio::update_position clones the full account (incl. unbounded events history) on every fill — O(n²) cost over a backtest

[ivannp]

Summary

Every position/fill event during a backtest clones the entire account object, which carries an append-only events: Vec that grows for the whole run. Since the clone copies that growing vector each time, per-fill cost grows with the number of prior events - O(n²) over the length of the backtest. It's invisible at the call site; it reads as "just cloning an account."

My understanding

BaseAccount (nautilus-model, src/accounts/base.rs) holds events: Vec, append-only (self.events.push(event)), untrimmed by default.
nautilus-portfolio::portfolio::update_position() (src/portfolio.rs, ~2562) calls cache.borrow().account_owned(&account_id) — an owned clone of the whole AccountAny / BaseAccount including that vector - on every position event, where most other accessors (Cache::account_ref) return a borrow.
Impact

Profiling a multi-year 1-minute backtest (~2.46M bars, ~1.1k trades) with samply, nautilus_model::accounts::{base,any}::clone accounted for ~15% of total backtest CPU time, scaling with backtest length.

Existing mitigation and its limit

ExecutionEngineConfig::purge_account_events_interval_mins / purge_account_events_lookback_mins (both default None = off) drive BaseAccount::base_purge_account_events to trim the history, bounding the O(n²) growth. But the clone still happens every fill - purging shrinks the payload, not the copy. For a backtest that never reads account.events(), the whole history is copied for data nothing consumes.

This was my solution and it works fine, but it took quite a bit to become aware of the issue and solve it.

Understood to be by design - the history backs last_event(), Portfolio::statistics(), and live reconciliation. Filing to surface the hidden cost and ask whether:

• the default could enable a bounded lookback (unbounded-by-default is a footgun for long backtests), and/or
• update_position could borrow instead of taking an owned clone, since it doesn't appear to need ownership past the RefCell borrow.