Implement HTTP service with guild routing

[vcarl]

Context

Part of load balancer architecture work (see PR #228 for full analysis).

Risk: Medium | Reward: Medium | Code Changes: Required

Architecture Change

Uses config service for guild lookups → Deterministic sharding

Load Balancer → HTTP Service (stateless, 2-10 pods)
                    ↓
              Calculate: hash(guildId) % numPods
                    ↓
              Gateway Pod N (via headless service DNS)

Implementation

Deterministic Routing

// app/helpers/guildRouting.ts
import crypto from 'crypto';

const NUM_GATEWAY_PODS = parseInt(process.env.NUM_GATEWAY_PODS || '3', 10);

export function getGatewayPodForGuild(guildId: string): number {
  // Consistent hash - same guild always routes to same pod
  const hash = crypto.createHash('md5').update(guildId).digest();
  const num = hash.readUInt32BE(0);
  return num % NUM_GATEWAY_PODS;
}

export function getGatewayUrl(guildId: string): string {
  const podOrdinal = getGatewayPodForGuild(guildId);
  // k8s headless service DNS
  return `http://gateway-${podOrdinal}.gateway-internal:3000`;
}

Request Routing

// In HTTP service handlers
import { getGatewayUrl } from './guildRouting';

async function routeToGateway(guildId: string, path: string, options: RequestInit) {
  const gatewayUrl = getGatewayUrl(guildId);
  return fetch(`${gatewayUrl}${path}`, options);
}

Service Discovery

Uses Kubernetes headless service DNS:

• Gateway pods: gateway-0.gateway-internal, gateway-1.gateway-internal, etc.
• No external config service needed

Tasks

• Implement getGatewayPodForGuild() deterministic hash function
• Add NUM_GATEWAY_PODS environment variable
• Implement guild routing in HTTP handlers
• Handle cases where target pod is unavailable (retry/failover)
• Deploy cluster/proposed/http-service.yaml (remove CONFIG_SERVICE_URL)
• Configure HPA for auto-scaling (2-10 pods)
• Test request routing in staging
• Load test with multiple HTTP pods

Dependencies

• Split monolith into smaller gateway/http services #232 (SERVICE_MODE support)

References

• PR Architecture analysis: Load balancing with SQLite constraint #228 (architecture analysis)
• Closed: Build config service for guild-to-pod assignments #233 (config service - no longer needed)