REDWEBAPI guide

Documentation for Redweb 0.13.0. Install that exact version when following these examples.

0.9.0 verification evidence

This is historical evidence for the 0.9.0 candidate, not certification of a newer release. Current commands, measurement scopes and unresolved gates are recorded in operations and the development release checklist.

Release-candidate measurements were taken on Windows x64, Node 22.21.0, and an AMD Ryzen 7 7800X3D. Performance numbers are machine-specific; the scripts and thresholds are the durable contract.

Automated correctness

  • 290 unit and mock-free integration/fuzz tests pass on Node 18, 20, and 22.
  • Statements, branches, functions, and lines are each 100% covered.
  • Type declarations compile and generated protocol declarations match their schema.

Resource and failure gates

  • Real-socket load: 32 concurrent clients, 3,200 request/response messages, 6,874 messages/second, 5.87 ms p99, with a paused slow consumer disconnected by the outbound-buffer policy.
  • Reconnect recovery: 200 warm connections followed by 1,200 storm connections; retained heap recovered to 104.33% of warm baseline and connection, room, and session registries returned to zero.
  • Fully enabled idle-route metadata: 1,756.93 bytes per connection across the median of three 500-connection trials, below the 2,048-byte gate.
  • Disabled-feature comparison with Redweb 0.8: throughput improved 1.09% and p99 regressed 2.00% (limits: 3% and 5%), using five alternating 20,000-message trials at concurrency 128.
  • npm audit reports zero vulnerabilities after upgrading Express to 4.22.2, ws to 8.21.3, and patched transitive dependencies.
  • The corrected 60-minute soak sent 2,099,717 messages across 64 rotating clients and received 2,099,565 responses (99.993%). Steady-state clients, rooms, sessions, in-flight work, queued work, listeners, and Redweb-owned timers showed no sustained growth; late-window heap was 3.24% above the early window, within the 10% gate. Final heap was 100.04% of the warmed baseline; all registries returned to zero; active handles stayed within the one-handle allowance.

All measurements above were rerun on the final release candidate. Shortened soak smoke runs are not counted as release evidence.