Bulkhead pattern: isolating failures in multi-tenant systems

we recently had a major outage where one tenant's large data import operation consumed all available database connections, effectively taking down the entire multi-tenant application. this highlighted a critical need for implementing the bulkhead pattern. we're looking at various options: separate database connection pools for different tenant tiers, aggressive rate limiting per tenant, or even dynamically provisioning separate compute resources for high-impact operations. the challenge is balancing true isolation with operational efficiency and cost. what strategies have people found most effective for isolating failures in multi-tenant environments to prevent a noisy neighbor from impacting everyone else?