Skip to content

backend-performance-wave5.md

Latest commit

 

History

History
73 lines (53 loc) · 2.97 KB

backend-performance-wave5.md

File metadata and controls

73 lines (53 loc) · 2.97 KB
Raw

Backend Performance — Wave 5

Overview

Wave 5 performance work reduces RPC load, caps in-memory read-cache growth, tunes the Postgres pool, and fixes RPC retry behavior so transient Soroban failures recover correctly.

Implementation plan

Priority Change Files
High Bounded TTL read cache (max entries + expiry) backend/src/services/read-cache.ts, stellar.ts
High Fix executeWithRetry to await Promise.race (retries actually run) backend/src/services/stellar.ts
Medium Configurable cache and pool via env backend/src/config/env.ts, backend/.env.example
Medium Postgres connection pool max via TypeORM extra backend/src/services/database.ts
Medium Unit tests for cache behavior backend/src/services/read-cache.test.ts

Operational impact

Read cache

  • Before: Unbounded Map for simulation reads; risk of memory growth on long-lived processes.
  • After: Default max 500 entries, TTL 30s (configurable). Oldest entry evicted when full.
  • Tuning: READ_CACHE_TTL_MS, READ_CACHE_MAX_ENTRIES in environment.
  • Invalidation: Unchanged — write routes still call invalidateCache / invalidateCacheByPrefix.

Database pool

  • Default: DATABASE_POOL_MAX=10 connections per process.
  • Tuning: Raise for high concurrency; lower for small containers.

RPC retries

  • Transient errors now retry with exponential backoff (structured logger.warn).

Rollback

  1. Revert the merge commit and redeploy the previous image.
  2. No schema migrations in this wave.
  3. Remove optional env vars to restore prior defaults (30s TTL, unbounded cache behavior replaced by bounded — revert code to restore old semantics).

Monitoring after deploy

  • Process RSS / heap (cache cap should stabilize growth)
  • Soroban RPC request rate (should drop on read-heavy endpoints)
  • API p95 latency on GET split/project routes
  • Postgres active connections vs DATABASE_POOL_MAX

Local CI (.github/workflows/ci.yml backend job)

export CI=true NODE_ENV=test \
  DATABASE_URL=postgresql://splitnaira:splitnaira@localhost:5432/splitnaira_ci \
  HORIZON_URL=https://horizon-testnet.stellar.org \
  SOROBAN_RPC_URL=https://soroban-testnet.stellar.org \
  SOROBAN_NETWORK_PASSPHRASE="Test SDF Network ; September 2015" \
  CONTRACT_ID=CAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA \
  SIMULATOR_ACCOUNT=GAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAWHF \
  NEXT_PUBLIC_CONTRACT_ID=CAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA

npm ci
npm run deps:check -w backend
npm run migration:run -w backend   # requires Postgres on :5432
npm run lint -w backend
npm run build -w backend
npm run test:compat -w backend
npm run test -w backend

Pre-deployment checklist

  • Full backend test suite green
  • Set READ_CACHE_* / DATABASE_POOL_MAX in staging if defaults are insufficient
  • Confirm no regression on write paths (cache invalidation still called)