broad-announce/cmd/ingestd/process.go

// process.go is the shared alert-processing pipeline used by
// both the HTTP handler (cmd/ingestd/http.go) and the MQTT
// subscriber (cmd/ingestd/mqtt.go). It does the M0→M4 SPEC §22
// work in one place: parse → validate → look up source → HMAC
// verify → rate-limit (per-source, per-company) → dedupe → stamp
// → publish to NATS → return a result struct the caller maps
// to its own transport-level response.
//
// The function is pure (no global state, no transport types).
// It returns Result{Accepted/Rejected/Error + reason + alert_id
// + dedupe_count} so the HTTP handler can map to a status code
// and the MQTT handler can map to a log line.
package main

import (
	"context"
	"encoding/json"
	"fmt"
	"log/slog"
	"strconv"
	"time"

	"git3.techno-world.net/lrosales/broad-announce/internal/alert"
	"git3.techno-world.net/lrosales/broad-announce/internal/broker"
	"git3.techno-world.net/lrosales/broad-announce/internal/dedupe"
	"git3.techno-world.net/lrosales/broad-announce/internal/observability"
	"git3.techno-world.net/lrosales/broad-announce/internal/ratelimit"
	"github.com/nats-io/nats.go"
)

// Result is the outcome of a ProcessAlert call. The caller
// (HTTP/MQTT) maps it to its transport's response shape.
type Result struct {
	// Accepted is true if the alert passed all checks and was
	// published to NATS.
	Accepted bool
	// AlertID is the server-assigned id (empty on error paths).
	AlertID string
	// DedupeCount is the count returned by the dedupe layer.
	DedupeCount uint32
	// IsNew is true for the first alert in a dedupe window.
	IsNew bool
	// RejectReason is the SPEC §22 / auth reason; one of:
	//   "payload_too_large", "bad_request", "invalid_json",
	//   "invalid", "unknown_source", "bad_signature",
	//   "rate_limited_source", "rate_limited_company",
	//   "marshal_failed", "broker_unavailable"
	RejectReason string
	// HTTPStatus is the suggested HTTP status code (0 for
	// accepted / 202).
	HTTPStatus int
	// Detail is the free-form string the caller can show in
	// a response body or a log message.
	Detail string
}

// Accept is the canonical "ok" result.
func Accept(id string, count uint32, isNew bool) Result {
	return Result{Accepted: true, AlertID: id, DedupeCount: count, IsNew: isNew, HTTPStatus: 202}
}

// Reject is the canonical "no" result.
func Reject(reason string, status int, detail string) Result {
	return Result{RejectReason: reason, HTTPStatus: status, Detail: detail}
}

// processDeps is the process-pipeline dependency set. Smaller
// than httpDeps — no HTTP-specific fields. Both httpDeps and
// the MQTT subscriber construct one and call ProcessAlert.
type processDeps struct {
	Logger    *slog.Logger
	Metrics   *observability.IngestdMetrics
	Limiter   *ratelimit.Limiter
	Deduper   *dedupe.Deduper
	JetStream natsPublisher
	// Sources is the (company_id, source_id) → SourceConfig map.
	// M0 reads it from env; M2 from Postgres. The MQTT subscriber
	// uses the same map keyed on the topic-parsed (co, src).
	Sources map[string]SourceConfig
	// Per-company default rate cap. HTTP and MQTT use the same
	// constant; once the per-company cap lives in DB, both
	// transports read it.
	CompanyRatePerSec int
	// Now is overridable in tests.
	Now func() time.Time
}

// ProcessAlert runs the full SPEC §22 protection chain on one
// alert body. It is the single source of truth for the ingest
// pipeline; both the HTTP POST handler and the MQTT subscriber
// call it.
//
// Layer order (matches SPEC §22):
//   1. payload-size cap   (caller does this — http.go via
//                          MaxBytesReader; mqtt.go via the
//                          broker-side max-inflight setting)
//   3. per-source rate limit
//   4. per-company rate limit
//   5. schema validate
//   5b. parse JSON
//   (auth) HMAC verify — see verifyHMAC
//   5c. dedupe
//   publish to NATS
func (d *processDeps) ProcessAlert(ctx context.Context, body []byte, sigHeader string) Result {
	now := d.now()

	// 5. Parse + validate. We treat any parse failure as invalid.
	var a alert.Alert
	if err := json.Unmarshal(body, &a); err != nil {
		d.Metrics.AlertsReceived.WithLabelValues("invalid").Inc()
		return Reject("invalid_json", 400, err.Error())
	}
	if err := a.Validate(); err != nil {
		d.Metrics.AlertsReceived.WithLabelValues("invalid").Inc()
		return Reject("invalid", 400, err.Error())
	}

	// Look up source. M0: in-memory map. M2: DB.
	src, ok := d.Sources[a.CompanyID+":"+a.SourceID]
	if !ok {
		d.Metrics.AlertsReceived.WithLabelValues("invalid").Inc()
		return Reject("unknown_source", 401,
			fmt.Sprintf("no such source %s/%s", a.CompanyID, a.SourceID))
	}

	// Auth. Stripe-style: X-BA-Signature: t=<unix>,v1=<hex>.
	// For HTTP it's a header; for MQTT it's a top-level field
	// on the envelope — both call sites pass the same string.
	if !verifyHMAC(sigHeader, src.HMACSecret, body, now) {
		d.Metrics.AlertsReceived.WithLabelValues("invalid").Inc()
		return Reject("bad_signature", 401, "")
	}

	// 3. Per-source rate limit.
	if ok, ttl, err := d.Limiter.Allow(ctx, "source:"+a.CompanyID+":"+a.SourceID, src.RateLimitPerSec); err != nil {
		d.Logger.Warn("ratelimit redis error (failing open)", "err", err, "scope", "source")
	} else if !ok {
		d.Metrics.AlertsReceived.WithLabelValues("rate_limited").Inc()
		d.Metrics.RateLimitHits.WithLabelValues("source").Inc()
		_ = ttl
		return Reject("rate_limited_source", 429, strconv.Itoa(int(ttl.Seconds())))
	}

	// 4. Per-company rate limit.
	if ok, ttl, _ := d.Limiter.Allow(ctx, "company:"+a.CompanyID, d.CompanyRatePerSec); !ok {
		d.Metrics.AlertsReceived.WithLabelValues("rate_limited").Inc()
		d.Metrics.RateLimitHits.WithLabelValues("company").Inc()
		_ = ttl
		return Reject("rate_limited_company", 429, strconv.Itoa(int(ttl.Seconds())))
	}

	// 5c. Dedupe.
	isNew, count, err := d.Deduper.Check(ctx, a.SourceID, a.DedupeKey)
	if err != nil {
		d.Logger.Warn("dedupe redis error (failing open)", "err", err)
		isNew, count = true, 1
	}

	// Stamp server-side fields.
	a.ID = alert.NewID()
	a.ReceivedAt = now.UTC()
	a.DedupeCount = count

	// Publish to NATS.
	subject := broker.AlertsSubject(a.CompanyID)
	payload, err := json.Marshal(a)
	if err != nil {
		d.Metrics.AlertsReceived.WithLabelValues("invalid").Inc()
		return Reject("marshal_failed", 500, err.Error())
	}
	start := time.Now()
	if err := d.JetStream.PublishAsync(subject, payload); err != nil {
		// Circuit breaker (M9) wraps this. For M0 we just fail loud.
		d.Metrics.AlertsReceived.WithLabelValues("circuit_open").Inc()
		return Reject("broker_unavailable", 503, err.Error())
	}
	d.Metrics.PublishLatency.Observe(time.Since(start).Seconds())
	d.Metrics.PayloadBytes.Observe(float64(len(payload)))

	if isNew {
		d.Metrics.AlertsReceived.WithLabelValues("accepted").Inc()
	} else {
		d.Metrics.AlertsReceived.WithLabelValues("deduped").Inc()
	}

	d.Logger.Info("alert accepted",
		"alert_id", a.ID,
		"company_id", a.CompanyID,
		"source_id", a.SourceID,
		"severity", string(a.Severity),
		"dedupe_count", count,
	)
	return Accept(a.ID, count, isNew)
}

func (d *processDeps) now() time.Time {
	if d.Now != nil {
		return d.Now()
	}
	return time.Now()
}

// natsPublisher is the minimal NATS interface. The HTTP and
// MQTT paths share it; tests can swap in a fake.
type natsPublisher interface {
	PublishAsync(subj string, data []byte) error
}

// jsPublisher adapts a nats.JetStreamContext to the natsPublisher interface.
type jsPublisher struct {
	js nats.JetStreamContext
}

func (j *jsPublisher) PublishAsync(subj string, data []byte) error {
	_, err := j.js.PublishAsync(subj, data)
	return err
}

// newNatsPublisher is the constructor used by main.
func newNatsPublisher(js nats.JetStreamContext) natsPublisher {
	return &jsPublisher{js: js}
}