Your bot closes a winning trade at 3 AM. The exchange confirms the fill. Freqtrade logs it. And then the Kubernetes node gets recycled during a cluster upgrade — and the trade disappears. No history. No PnL record. No data for the next backtest.
A bot that misses a stop-loss because its container restarted is not just an inconvenience — it can convert a -3% planned exit into a -12% surprise. The hosting model under your bot is the second-most consequential decision in algorithmic trading after the strategy itself, and most traders pick it by accident on day one. This guide walks through the trade-offs between cloud-managed runners and self-hosted infrastructure so you can pick the right model deliberately.