proof-server:proof-server-architecture

Configure zero-knowledge proof concurrency by tuning --num-workers for parallelism and --job-capacity for queue size. The channel-based async job queue manages backpressure, ensuring controlled concurrent job processing without overwhelming system resources.

HTTP 429 responses occur when the proof server job queue reaches full capacity, applying backpressure to prevent unbounded memory growth. Handle 429s by increasing --job-capacity or scaling worker pools horizontally across multiple deployment instances.

Zero-knowledge proof key prefetching occurs at server startup, pre-loading proving keys and circuit parameters via MidnightDataProvider. This on-demand fetch mechanism prevents missing key errors during initial transaction proving requests and reduces first-request latency.

Deploy proof servers using musl static builds packaged as multi-arch Docker images for portable production environments. This approach ensures consistent binary serialization across architectures, enabling portable deployments without runtime dependency conflicts.

ZKIR versioning routes V2 and V3 zero-knowledge proofs to appropriate handlers using versioned IR dispatch. The /proof-versions endpoint reports supported circuit versions, ensuring correct handler assignment and backward compatibility across proof generations.

Use TTL-based garbage collection when managing concurrent job queues to prevent unbounded memory growth from abandoned or timed-out jobs. Configure --job-timeout to trigger cleanup of stale zero-knowledge proof jobs, maintaining stable long-running production capacity.