We study liquidation in onchain perpetual markets as a stochastic control problem with delayed marks and finite liquidity. We evaluate batch liquidation (BALI), a solvency-constrained mechanism with stress triggers, uniform-price clearing, concentration caps, and deterministic fallback, against a mathematically comparable continuous liquidation baseline. Results are conditional: BALI consistently reduces cascade depth, while impact and bad debt improve only when auction participation is sufficiently elastic and fallback activates before uncovered residuals approach bankruptcy. Under weaker participation or larger effective delay, continuous liquidation can perform better on impact or bad debt. These findings position batch liquidation (BALI) as a state-contingent control layer rather than a universal replacement.