Sidecar Communication

Handlers access external infrastructure (databases, caches, storage) through HTTP calls to sidecar services. This keeps credentials out of WASM modules and enables policy enforcement.

Architecture

Handler (WASM) ──HTTP──> Sidecar (native) ──> Database

Sidecars are native processes that:

• Hold database credentials securely
• Translate HTTP requests to infrastructure protocols
• Enforce rate limits, auth, and auditing
• Can implement caching, retries, circuit breakers

Transport Options

Transport	Latency	Use Case
TCP HTTP	~100-500μs	Remote sidecars, Kubernetes
Unix Socket HTTP	~50-150μs	Local sidecars, same machine

For local deployments, use Unix sockets for ~2x lower latency while keeping the same HTTP protocol.

Configuration

# Local deployment - use Unix sockets for speed
[sidecars]
db = "unix:///var/run/sidecars/db.sock"
cache = "unix:///var/run/sidecars/cache.sock"
storage = "unix:///var/run/sidecars/storage.sock"

# Kubernetes/remote - use TCP
[sidecars]
db = "http://db-sidecar:8080"
cache = "http://cache-sidecar:8080"
storage = "http://storage-sidecar:8080"

Calling Sidecars from Handlers

Use the mik_sdk SDK’s HTTP client:

use mik_sdk::http_client as client;

// GET request
let response = client::get("http://db-sidecar/users/123")?;
let user = response.text();

// POST with JSON
let response = client::request()
    .post("http://cache-sidecar/set")
    .header("X-Cache-TTL", "3600")
    .json(b"{\"key\": \"value\"}")
    .send()?;

The sidecar URL comes from environment variables injected by the runtime.

Deployment Recommendations

Environment	Transport	Why
Development	TCP (localhost:8080)	Simplicity, easy debugging
Production (same host)	Unix sockets	~2x lower latency
Production (Kubernetes)	TCP to service	Cross-pod communication

Why HTTP (not ZeroMQ or other IPC)?

We evaluated faster alternatives:

Option	Latency	Trade-off
ZeroMQ	~10-50μs	Requires ZMQ client in WASM, breaks handler portability
Shared Memory	~1-5μs	Complex, not WASM-friendly, security concerns
gRPC	~50-150μs	Heavy dependency, similar speed to Unix sockets
Cap’n Proto	~10-30μs	Niche ecosystem, breaks portability

Key insight: Actual work (DB queries, external APIs) dominates latency. Saving 50μs on transport doesn’t matter when the DB query takes 5ms.

Benefits of HTTP

Unix sockets with HTTP preserve:

• Handler portability - Standard HTTP works on any WASI Preview 2 runtime
• Debuggability - Test with curl --unix-socket /path/to/sock http://localhost/endpoint
• Observability - Standard HTTP logging, tracing, and metrics
• Simplicity - No special client libraries needed in WASM

Example: Database Sidecar

A typical database sidecar exposes a REST API:

GET  /users/:id       → SELECT * FROM users WHERE id = ?
POST /users           → INSERT INTO users ...
PUT  /users/:id       → UPDATE users SET ... WHERE id = ?
DELETE /users/:id     → DELETE FROM users WHERE id = ?

The sidecar holds the connection pool and credentials. Handlers only see HTTP.