Performance and resilience

Performance is the user experience under expected load. Resilience is useful behavior when dependencies, instances, or networks fail.

Measure first

Load-test realistic traffic, payloads, database size, cache state, and failure. Measure latency percentiles, throughput, errors, CPU, memory, garbage collection, thread pools, connection pools, queue depth, and dependency time.

Protect resources

Configure HTTP, database, broker, and executor pools from total instance capacity. Bound queues. A huge queue hides overload and increases latency. Use bulkheads so one dependency cannot consume every thread.

Remote-call policy

Every remote call needs a timeout. Retry only transient and safe operations, with a small bound, exponential backoff, and jitter. Use circuit breakers to stop repeated calls during an outage. Provide graceful degradation for optional features.

JVM and native choices

Tune only after profiling. Modern JVM ergonomics are strong. Native images can improve startup and memory for suitable workloads but change build, reflection/resource configuration, debugging, and peak-throughput trade-offs.

Feynman check

A timeout is a clock, a retry is another attempt, a circuit breaker closes the road, and a bulkhead keeps one flooded room from sinking the whole ship.