ai-ops-agent/agent/agent.py

import os
import re
import yaml
import json
from fastapi import FastAPI, Request, HTTPException
from pydantic import BaseModel
import docker

# Basic AI Ops Agent for Docker Swarm
PORT = int(os.getenv("PORT", "8080"))
GUARD_DEFAULT_MAX = 25

# Load rules file if present
RULES_PATH = "rules.yaml"
if os.path.exists(RULES_PATH):
    with open(RULES_PATH, "r") as f:
        RULES = yaml.safe_load(f) or {}
else:
    RULES = {}

ALLOWED_REGEX = re.compile(RULES.get("guardrails", {}).get("allowed_services_regex", ".*"))
MAX_SCALE = int(RULES.get("guardrails", {}).get("max_scale_replicas", GUARD_DEFAULT_MAX))

# Docker client using the socket
client = docker.DockerClient(base_url='unix://var/run/docker.sock')

app = FastAPI(title="BurnServ AI Ops Agent")


def _guard_service(service_name: str):
    if not ALLOWED_REGEX.match(service_name):
        raise HTTPException(status_code=403, detail=f"Service '{service_name}' not allowed by guardrails.")


def _scale(service_name: str, replicas: int = None, step: int = None, min_replicas: int = None, max_replicas: int = None):
    _guard_service(service_name)
    try:
        svc = client.services.get(service_name)
    except docker.errors.NotFound:
        raise HTTPException(status_code=404, detail=f"Service '{service_name}' not found")
    spec = svc.attrs.get('Spec', {}).copy()
    mode = spec.get('Mode', {})
    replicas_current = mode.get('Replicated', {}).get('Replicas', 1)

    if replicas is None:
        # step-based scaling
        tgt = int(replicas_current) + int(step or 1)
        if min_replicas is not None:
            tgt = max(tgt, int(min_replicas))
        if max_replicas is not None:
            tgt = min(tgt, int(max_replicas))
        replicas = tgt

    replicas = max(1, min(int(replicas), MAX_SCALE))
    mode['Replicated'] = {'Replicas': replicas}
    spec['Mode'] = mode

    # svc.update expects keyword args matching the service spec shape; pass the full spec with update
    try:
        svc.update(task_template=spec.get('TaskTemplate'), **{k: v for k, v in spec.items() if k != 'TaskTemplate'})
    except Exception:
        # Fallback: use the update with the raw spec (works in many docker-py versions)
        svc.update(**spec)
    return {"service": service_name, "replicas": replicas}


def _restart_service(service_name: str):
    _guard_service(service_name)
    try:
        svc = client.services.get(service_name)
    except docker.errors.NotFound:
        raise HTTPException(status_code=404, detail=f"Service '{service_name}' not found")
    spec = svc.attrs.get('Spec', {}).copy()
    # Force a rolling update by bumping ForceUpdate / Version index
    try:
        current_index = svc.attrs.get('Version', {}).get('Index', 0)
        svc.update(**spec, force_update=current_index + 1)
    except Exception:
        # If update signature differs, try a simple update
        svc.update(**spec)
    return {"service": service_name, "status": "rolling-restart-issued"}


class Command(BaseModel):
    text: str = ""
    action: str | None = None
    params: dict | None = None


@app.get("/health")
def health():
    return {"ok": True}


@app.post("/command")
def command(cmd: Command):
    \"\"\"Accepts either structured commands or simple text instructions.

    Structured example:
      {"action":"scale","params":{"service":"weblabs_php","replicas":3}}

    Text example:
      {"text":"scale weblabs_php to 3"}
    \"\"\"
    # Structured commands first
    if cmd.action:
        if cmd.action == "scale":
            p = cmd.params or {}
            if "service" not in p or "replicas" not in p:
                raise HTTPException(status_code=400, detail="Missing params for scale: service, replicas")
            return _scale(p["service"], replicas=int(p["replicas"]))
        elif cmd.action == "restart_service":
            p = cmd.params or {}
            if "service" not in p:
                raise HTTPException(status_code=400, detail="Missing service param for restart_service")
            return _restart_service(p["service"])
        else:
            raise HTTPException(status_code=400, detail="Unknown action")

    # Free-text parsing (simple)
    t = (cmd.text or "").strip().lower()
    if not t:
        raise HTTPException(status_code=400, detail="Empty command")

    if t.startswith("scale "):
        # "scale weblabs_php to 3"
        parts = t.split()
        try:
            svc = parts[1]
            if "to" in parts:
                idx_to = parts.index("to")
                reps = int(parts[idx_to + 1])
                return _scale(svc, replicas=reps)
            else:
                # if no explicit amount, treat as step +1
                return _scale(svc, step=1)
        except Exception:
            raise HTTPException(status_code=400, detail="Format: 'scale <service> to <n>'")
    elif t.startswith("restart "):
        # "restart weblabs_php"
        try:
            svc = t.split()[1]
            return _restart_service(svc)
        except Exception:
            raise HTTPException(status_code=400, detail="Format: 'restart <service>'")

    raise HTTPException(status_code=400, detail="Unrecognized command")


@app.post("/alert")
async def alert(request: Request):
    \"\"\"Receive Alertmanager webhooks and execute configured actions based on rules.yaml.\"\"\"
    payload = await request.json()
    alerts = payload.get("alerts", [])
    executed = []

    for a in alerts:
        labels = a.get("labels", {}) or {}
        # For each rule in RULES, check if the match conditions apply
        for rule in RULES.get("alerts", []):
            match = rule.get("match", {})
            if all(labels.get(k) == v for k, v in match.items()):
                for act in rule.get("actions", []):
                    if "scale" in act:
                        cfg = act["scale"].copy()
                        svc = cfg.get("service", "").replace("{{ $labels.service_name }}", labels.get("service_name", ""))
                        res = _scale(
                            service_name=svc,
                            replicas=cfg.get("replicas"),
                            step=cfg.get("step"),
                            min_replicas=cfg.get("min_replicas"),
                            max_replicas=cfg.get("max_replicas", MAX_SCALE),
                        )
                        executed.append({"alert": labels.get("alertname"), "action": "scale", "result": res})
                    elif "restart_service" in act:
                        cfg = act["restart_service"].copy()
                        svc = cfg.get("service", "").replace("{{ $labels.service_name }}", labels.get("service_name", ""))
                        res = _restart_service(svc)
                        executed.append({"alert": labels.get("alertname"), "action": "restart_service", "result": res})

    return {"executed": executed}