feat:improve performance improvements

[Samuel1505]

Performance improvements: DB partitioning, optimistic UI, CDN edge caching, PWA offline queue

Summary

This PR implements four performance issues (#501–#504) to improve query performance, perceived latency, API response time for distributed users, and offline resilience.

---

closes #504
closes #503
closes #502
closes #501

#504 — Database Partitioning by Tenant

Files changed

• backend/prisma/migrations/20260627000000_tenant_partitioning/migration.sql
• backend/src/db/partition-manager.ts

What was done

Added PostgreSQL LIST partitioning by tenant_id for the four high-volume tables (payments, invoices, audit_logs, webhook_logs).

• audit_logs gains a tenant_id column with backfill from users.tenant_id (falls back to 'system' for non-user actors).
• A tenant_partitions registry table tracks every partition with row counts, byte sizes, and last-analyzed timestamps.
• Two reusable PostgreSQL functions: create_tenant_partition(table, tenant) (idempotent) and refresh_partition_stats().
• Partitioned shadow tables (payments_partitioned, invoices_partitioned, audit_logs_partitioned, webhook_logs_partitioned) are seeded from the originals with ON CONFLICT DO NOTHING.
• PartitionManager TypeScript class exposes:
  • ensurePartitionsForTenant(tenantId) — call when a new tenant registers
  • dropTenantPartitions(tenantId) — archive/delete a tenant's data by dropping their partition
  • getPartitionStats([tenantId]) — monitoring: size and row counts per partition
  • getPartitionDistribution() — aggregate by table for dashboards
  • migrateExistingTenantData(tenantId, table) — backfill per-tenant with error result
  • migrateAllTenants(table) — full table backfill
  • getQueryMetrics(tenantId, table) — pruning efficiency metric

Zero-downtime strategy: shadow tables run in parallel with originals. The cutover (rename) is deferred to a maintenance window and driven by PartitionManager, not the migration itself, so rollback is a single rename.

---

#503 — Optimistic UI Updates with Background Revalidation

Files changed

• frontend/src/hooks/mutations/usePaymentMutations.ts
• frontend/src/components/common/optimistic-status.tsx
• frontend/lib/query-keys.ts

What was done

TanStack Query v5 optimistic update pattern applied to all payment mutations.

• useCreatePayment — immediately prepends an optimistic payment to the cache list; skips optimistic path for amounts >= OPTIMISTIC_THRESHOLD (10 000) to avoid false positives on critical transactions.
• useUpdatePayment — patches both list and detail caches; onError restores both from snapshot.
• useCancelPayment — sets status to 'cancelled' optimistically; rolls back on error.
• useRetryPayment — no optimistic update (server drives the retry outcome).
• All hooks: onMutate to snapshot and apply, onError to rollback, onSettled to invalidate for server revalidation.
• OptimisticStatus badge component: shows spinner/label during pending, checkmark on success, error label on error; auto-hides after configurable duration; marks optimistic items with (optimistic) suffix.
• StaleBadge and ConflictBanner components for stale-data and server-vs-optimistic conflict UX.
• useMutationState helper derives MutationState from TanStack Query's boolean flags.
• query-keys.ts extended with invoices, webhooks, and payments.infinite namespaces.

---

#502 — CDN-Powered Geo-Distributed API Edge Caching

Files changed

• backend/src/middleware/cache-headers.ts
• infra/cdn.tf

What was done

Adds Cache-Control / Surrogate-Control header management and a CloudFront distribution.

Middleware (cache-headers.ts)

• cacheHeaders(options) middleware sets Cache-Control and Surrogate-Control with configurable TTL, stale-while-revalidate, and stale-if-error.
• Authorization header is SHA-256 hashed into X-Auth-Hash; Vary includes this hash so CDN caches per-user without storing raw tokens in cache keys.
• Surrogate-Key / Cache-Tag headers enable tag-based purge after mutations.
• purgeCdnCache({ provider, paths?, surrogateKeys? }) supports CloudFront, Cloudflare, and Fastly.
• Pre-built presets: cdnCache.none(), cdnCache.realtime(), cdnCache.userData(), cdnCache.staticData().
• getCacheMetrics() and getCacheHitRatioSummary() for monitoring dashboards.

Terraform (cdn.tf)

• CloudFront distribution with origin shield (reduces origin load by coalescing edge requests).
• Three cache policies: no-store (mutations/auth), 30 s (user data), 5 min (static config).
• Separate ordered behaviours for /api/v1/payments*, /api/v1/invoices*, /api/v1/config*, /api/v1/currencies*.
• CloudFront Function replaces the Authorization header with its hash before the cache key is computed.
• S3 log bucket with 90-day lifecycle expiry for cache hit/miss analysis.
• cdn_distribution_id, cdn_domain_name, cdn_hosted_zone_id outputs for Route53 alias wiring.

---

#501 — Progressive Web App with Offline Transaction Queuing

Files changed

• frontend/src/lib/offline-queue.ts
• frontend/src/hooks/useOnlineStatus.ts
• frontend/public/sw.js

What was done

IndexedDB-backed offline transaction queue with automatic background sync.

offline-queue.ts

• enqueue(input) — writes to IndexedDB, registers a SyncManager tag, and dispatches a CustomEvent for cross-component reactivity.
• flush(apiBaseUrl) — submits all due items (respects retryAt), marks synced or applies exponential backoff (5 s -> 15 s -> 1 min -> 5 min -> 10 min); after 5 retries status is 'failed'.
• HTTP 409 responses are treated as success (idempotent replay: server already processed the request).
• getSnapshot() — counts by status for UI indicators.
• subscribeToQueue(listener) — event-based subscription; unsubscribe function returned.

useOnlineStatus.ts

• Tracks navigator.onLine via online / offline window events.
• On reconnect (online event): calls flush() automatically.
• Listens for PAYMENT_QUEUE_CHANGED and PAYMENT_QUEUE_SYNCED messages from the service worker.
• Exposes flushNow() for manual sync (e.g. a "Sync now" button in the UI).
• Returns { isOnline, wasOffline, queue, isFlushing, lastFlushResult, flushNow }.

sw.js updates

• Bumped APP_VERSION to 2026-06-27.1 to invalidate old caches on deploy.
• Added /dashboard/payments and /dashboard/transactions to the precache App Shell.
• IndexedDB upgraded to v2 with retryAt index for efficient due-item queries.
• flushPaymentQueue filters by retryAt <= now() before submitting; applies same exponential backoff; stops retrying on offline detection.

---

Test plan

• Performance: Add Database Partitioning by Tenant for Multi-Tenant Isolation #504: PartitionManager.ensurePartitionsForTenant('test-tenant') creates four partitions; EXPLAIN (ANALYZE) SELECT * FROM payments_partitioned WHERE tenant_id = 'test-tenant' shows Bitmap Index Scan on the single partition only.
• Performance: Add Database Partitioning by Tenant for Multi-Tenant Isolation #504: dropTenantPartitions('test-tenant') drops all four partitions and removes registry rows.
• Performance: Implement Optimistic UI Updates with Background Revalidation #503: Creating a payment reflects in the list immediately before the server responds; cancelling the mutation (e.g. network error) rolls back to the prior list.
• Performance: Implement Optimistic UI Updates with Background Revalidation #503: Payment amount >= 10 000 does not show optimistic update — list only updates after server confirms.
• Performance: Implement Optimistic UI Updates with Background Revalidation #503: OptimisticStatus shows spinner during mutation, checkmark on success, auto-hides after 2 s.
• Performance: Add CDN-Powered Geo-Distributed API Edge Caching #502: GET /api/v1/payments response has Cache-Control: private, max-age=30, stale-while-revalidate=60.
• Performance: Add CDN-Powered Geo-Distributed API Edge Caching #502: POST /api/v1/payments response has Cache-Control: no-store.
• Performance: Add CDN-Powered Geo-Distributed API Edge Caching #502: Mutation triggers purgeCdnCache for the relevant surrogate key (check Cloudflare/CloudFront purge logs).
• Performance: Add CDN-Powered Geo-Distributed API Edge Caching #502: Terraform plan produces a CloudFront distribution with origin shield and three cache policies.
• Performance: Implement Progressive Web App with Offline Transaction Queueing #501: Going offline and creating a payment queues it to IndexedDB; getSnapshot() shows pending: 1.
• Performance: Implement Progressive Web App with Offline Transaction Queueing #501: Reconnecting flushes the queue; the payment appears on the server and getSnapshot() shows total: 0.
• Performance: Implement Progressive Web App with Offline Transaction Queueing #501: A failing endpoint is retried with backoff and eventually marked 'failed' after 5 attempts.
• Performance: Implement Progressive Web App with Offline Transaction Queueing #501: Service worker installs cleanly after version bump; old caches are evicted on activate.

[vercel]

@Samuel1505 is attempting to deploy a commit to the smartdevs17's projects Team on Vercel.

A member of the Team first needs to authorize it.

[drips-wave]

@Samuel1505 Great news! 🎉 Based on an automated assessment of this PR, the linked Wave issue(s) no longer count against your application limits.

You can now already apply to more issues while waiting for a review of this PR. Keep up the great work! 🚀

Learn more about application limits