Deployment, Monitoring, and Restart Discipline

Process supervisors keep things alive

Pick a supervisor (launchd on macOS, systemd on Linux, supervisord, k8s) and let it own restart-on-crash. Your service should never be the thing that decides 'I should restart myself' — that is what supervisors are for. cwkPippa runs under launchd; launchctl kickstart -k is the operator's restart command.

Health endpoints that mean something

A liveness endpoint returns 200 if the process is up. A readiness endpoint returns 200 only if dependencies are healthy (Anthropic reachable, DB writable, MCP servers responsive). Use the right one for the right consumer — supervisors want liveness; load balancers want readiness.

Self-editing services need cold restart

cwkPippa runs Uvicorn without --reload because the agent edits its own backend code routinely. Auto-reload during a self-edit causes partial-file restarts and crashes. The pattern: edit, save, run a controlled launchctl kickstart -k — predictable, debuggable, and visible in the StatusBar.

Principle: Supervisors restart, services do not. Health endpoints distinguish liveness from readiness. Self-editing systems need cold restart, not auto-reload.

Code

Liveness vs readiness·python

from fastapi import FastAPI, Response
import httpx

app = FastAPI()

@app.get("/healthz")
async def liveness():
    # Process is up. That is all this means.
    return {"status": "alive"}

@app.get("/readyz")
async def readiness(response: Response):
    # Are we ready to serve traffic? Check real dependencies.
    checks = {}
    try:
        r = await httpx.AsyncClient(timeout=2).get("https://api.anthropic.com/v1/health")
        checks["anthropic"] = r.status_code == 200
    except Exception as e:
        checks["anthropic"] = False
    checks["db"] = await db_check()
    if not all(checks.values()):
        response.status_code = 503
    return {"checks": checks}

launchd plist (sketch) for an always-on service·yaml

# ~/Library/LaunchAgents/com.example.pippa.plist (XML in reality)
# Key fields:
ProgramArguments:
  - /usr/local/bin/uvicorn
  - backend.main:app
  - --host=0.0.0.0
  - --port=8000
KeepAlive:
  SuccessfulExit: false
RunAtLoad: true
StandardOutPath: /var/log/pippa.out.log
StandardErrorPath: /var/log/pippa.err.log
# Restart with: launchctl kickstart -k gui/$(id -u)/com.example.pippa

Exercise

For one service you operate, audit: which supervisor, what is the restart command, what is the health endpoint surface (liveness + readiness), how does the operator know when to restart. Document the gaps.

Hint

If you do not have a readiness check beyond '200 if process up', you are conflating two questions. Add the dependency probes.

Deployment, Monitoring, and Restart Discipline

Process supervisors keep things alive

Health endpoints that mean something

Self-editing services need cold restart

Code

External links

Exercise

Progress

Comments 0