main.go 8.3 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297
  1. // Command deliverd consumes deliveries.<channel>.<company_id> subjects
  2. // and pushes the alert to the appropriate third-party sink
  3. // (FCM, Telegram, SMS, email, Slack, Teams, webhook).
  4. //
  5. // M0: per-channel worker binary that connects to NATS, /health, /metrics.
  6. // M1: deliverd-fcm — consumes deliveries.fcm.*, posts to fakefcmd,
  7. // records a row in Postgres per attempt.
  8. package main
  9. import (
  10. "bytes"
  11. "context"
  12. "encoding/json"
  13. "fmt"
  14. "io"
  15. "log/slog"
  16. "net/http"
  17. "os"
  18. "os/signal"
  19. "strings"
  20. "syscall"
  21. "time"
  22. "git3.techno-world.net/lrosales/broad-announce/internal/broker"
  23. "git3.techno-world.net/lrosales/broad-announce/internal/config"
  24. "git3.techno-world.net/lrosales/broad-announce/internal/httpserver"
  25. "git3.techno-world.net/lrosales/broad-announce/internal/observability"
  26. "git3.techno-world.net/lrosales/broad-announce/internal/postgres"
  27. "github.com/nats-io/nats.go/jetstream"
  28. )
  29. // M1: only the FCM channel. M3+ adds telegram, sms, etc.
  30. const fcmChannel = "fcm"
  31. type deliveryEnvelope struct {
  32. Alert json.RawMessage `json:"alert"`
  33. IndividualID string `json:"individual_id"`
  34. FCMToken string `json:"fcm_token"`
  35. Locale string `json:"locale,omitempty"`
  36. }
  37. func main() {
  38. cfg, err := config.LoadCommon("deliverd")
  39. if err != nil {
  40. os.Stderr.WriteString("config: " + err.Error() + "\n")
  41. os.Exit(1)
  42. }
  43. logger := observability.Init(cfg.Env, cfg.LogLevel, "deliverd")
  44. logger.Info("starting", "env", cfg.Env, "addr", cfg.HTTPAddr)
  45. ctx, stop := signal.NotifyContext(context.Background(), os.Interrupt, syscall.SIGTERM)
  46. defer stop()
  47. br, err := broker.Connect(ctx, cfg.NATSURL)
  48. if err != nil {
  49. logger.Error("nats connect", "err", err)
  50. os.Exit(1)
  51. }
  52. defer br.Close()
  53. pool, err := postgres.Connect(ctx, cfg.PostgresDSN)
  54. if err != nil {
  55. logger.Error("postgres connect", "err", err)
  56. os.Exit(1)
  57. }
  58. defer pool.Close()
  59. // M1 only: fakefcmd URL. M3+ uses the real FCM HTTP v1 endpoint.
  60. fakefcmdURL := os.Getenv("BA_FAKECMD_URL")
  61. if fakefcmdURL == "" {
  62. fakefcmdURL = "http://fakefcmd:8820"
  63. }
  64. logger.Info("fcm target", "url", fakefcmdURL)
  65. httpClient := &http.Client{Timeout: 10 * time.Second}
  66. js := br.JS()
  67. stream, err := js.Stream(ctx, "DELIVERIES")
  68. if err != nil {
  69. logger.Error("nats stream DELIVERIES", "err", err)
  70. os.Exit(1)
  71. }
  72. consumer, err := stream.CreateOrUpdateConsumer(ctx, jetstream.ConsumerConfig{
  73. Name: "deliverd-fcm",
  74. Durable: "deliverd-fcm",
  75. FilterSubjects: []string{"deliveries.fcm.>"},
  76. AckPolicy: jetstream.AckExplicitPolicy,
  77. })
  78. if err != nil {
  79. logger.Error("nats consumer", "err", err)
  80. os.Exit(1)
  81. }
  82. runCtx, runCancel := context.WithCancel(ctx)
  83. defer runCancel()
  84. go consume(runCtx, logger, consumer, pool, httpClient, fakefcmdURL)
  85. reg, _ := observability.NewRegistry("deliverd")
  86. srv := httpserver.New(httpserver.Config{
  87. Addr: cfg.HTTPAddr,
  88. ServiceName: "deliverd",
  89. ShutdownGrace: cfg.ShutdownGrace,
  90. }, logger, observability.MetricsHandler(reg))
  91. errCh := make(chan error, 1)
  92. go func() { errCh <- srv.Start() }()
  93. select {
  94. case <-ctx.Done():
  95. logger.Info("shutdown signal received")
  96. case err := <-errCh:
  97. if err != nil {
  98. logger.Error("http server", "err", err)
  99. os.Exit(1)
  100. }
  101. }
  102. runCancel()
  103. time.Sleep(500 * time.Millisecond)
  104. if err := srv.Shutdown(ctx); err != nil {
  105. logger.Warn("graceful shutdown", "err", err)
  106. }
  107. logger.Info("bye")
  108. }
  109. func consume(ctx context.Context, logger *slog.Logger, c jetstream.Consumer, pool *postgres.Pool, httpClient *http.Client, fakefcmdURL string) {
  110. for {
  111. if ctx.Err() != nil {
  112. return
  113. }
  114. batch, err := c.Fetch(16, jetstream.FetchMaxWait(2*time.Second))
  115. if err != nil {
  116. if ctx.Err() != nil {
  117. return
  118. }
  119. logger.Warn("nats fetch", "err", err)
  120. time.Sleep(500 * time.Millisecond)
  121. continue
  122. }
  123. for m := range batch.Messages() {
  124. handleOne(ctx, logger, m, pool, httpClient, fakefcmdURL)
  125. if batch.Error() != nil {
  126. logger.Warn("batch error", "err", batch.Error())
  127. break
  128. }
  129. }
  130. }
  131. }
  132. func handleOne(ctx context.Context, logger *slog.Logger, m jetstream.Msg, pool *postgres.Pool, httpClient *http.Client, fakefcmdURL string) {
  133. var env deliveryEnvelope
  134. if err := json.Unmarshal(m.Data(), &env); err != nil {
  135. logger.Warn("malformed delivery envelope", "err", err, "subject", m.Subject())
  136. _ = m.Ack() // poison message; we don't have a DLQ for malformed yet
  137. return
  138. }
  139. // Parse out the alert_id and company_id from the inner alert JSON.
  140. var alertHeader struct {
  141. ID string `json:"id"`
  142. CompanyID string `json:"company_id"`
  143. Title string `json:"title"`
  144. Body string `json:"body"`
  145. Data map[string]string `json:"data"`
  146. Category string `json:"category"`
  147. Severity string `json:"severity"`
  148. }
  149. _ = json.Unmarshal(env.Alert, &alertHeader)
  150. companyID := alertHeader.CompanyID
  151. if companyID == "" {
  152. // subject is "deliveries.fcm.<company_id>"
  153. parts := strings.SplitN(m.Subject(), ".", 3)
  154. if len(parts) == 3 {
  155. companyID = parts[2]
  156. }
  157. }
  158. if companyID == "" || env.FCMToken == "" || alertHeader.ID == "" {
  159. logger.Warn("delivery envelope missing fields", "subject", m.Subject(), "company", companyID, "token", env.FCMToken != "", "alert_id", alertHeader.ID)
  160. _ = m.Ack()
  161. return
  162. }
  163. // Build the FCM HTTP v1 message body. The shape matches what
  164. // real FCM expects, so the M3 swap is a no-op at this layer.
  165. fcmBody := map[string]any{
  166. "message": map[string]any{
  167. "token": env.FCMToken,
  168. "notification": map[string]any{
  169. "title": alertHeader.Title,
  170. "body": alertHeader.Body,
  171. },
  172. "data": mergeData(alertHeader.Data, map[string]string{
  173. "company_id": companyID,
  174. "alert_id": alertHeader.ID,
  175. "severity": alertHeader.Severity,
  176. "category": alertHeader.Category,
  177. "individual_id": env.IndividualID,
  178. "locale": env.Locale,
  179. "deep_link": fmt.Sprintf("broadannounce://alert/%s", alertHeader.ID),
  180. }),
  181. "android": map[string]any{
  182. "priority": androidPriority(alertHeader.Severity),
  183. "notification": map[string]any{
  184. "sound": androidSound(alertHeader.Category, alertHeader.Severity),
  185. "channel_id": "alerts." + channelForSeverity(alertHeader.Severity),
  186. },
  187. },
  188. },
  189. }
  190. body, _ := json.Marshal(fcmBody)
  191. // POST to fakefcmd. M3: real FCM endpoint.
  192. url := fakefcmdURL + "/v1/projects/fakefcmd/messages:send"
  193. req, _ := http.NewRequestWithContext(ctx, "POST", url, bytes.NewReader(body))
  194. req.Header.Set("Content-Type", "application/json")
  195. resp, err := httpClient.Do(req)
  196. status := "failed"
  197. lastErr := ""
  198. if err != nil {
  199. lastErr = err.Error()
  200. } else {
  201. defer resp.Body.Close()
  202. respBody, _ := io.ReadAll(resp.Body)
  203. if resp.StatusCode/100 == 2 {
  204. status = "sent"
  205. } else {
  206. lastErr = fmt.Sprintf("status %d: %s", resp.StatusCode, string(respBody))
  207. }
  208. }
  209. // Persist delivery row. M1: just insert. M3+: per-channel
  210. // retry with exp backoff and DLQ on terminal failure.
  211. _, dbErr := pool.Exec(ctx, `
  212. INSERT INTO deliveries
  213. (alert_id, company_id, individual_id, channel, target, status, attempts, last_error, payload, sent_at)
  214. VALUES ($1, $2, $3, $4, $5, $6, 1, NULLIF($7, ''), $8,
  215. CASE WHEN $6 = 'sent' THEN now() ELSE NULL END)
  216. `, alertHeader.ID, companyID, env.IndividualID, fcmChannel, env.FCMToken, status, lastErr, json.RawMessage(m.Data()))
  217. if dbErr != nil {
  218. logger.Warn("delivery row insert", "err", dbErr)
  219. }
  220. logger.Info("delivery",
  221. "alert_id", alertHeader.ID,
  222. "company", companyID,
  223. "individual", env.IndividualID,
  224. "channel", fcmChannel,
  225. "status", status,
  226. "err", lastErr,
  227. )
  228. _ = m.Ack()
  229. }
  230. func mergeData(base, add map[string]string) map[string]string {
  231. if base == nil {
  232. base = map[string]string{}
  233. }
  234. for k, v := range add {
  235. if _, ok := base[k]; !ok {
  236. base[k] = v
  237. }
  238. }
  239. return base
  240. }
  241. func androidPriority(sev string) string {
  242. if sev == "critical" || sev == "inminent_colapse" {
  243. return "HIGH"
  244. }
  245. return "NORMAL"
  246. }
  247. // androidSound picks a sound file. SPEC §4 says the Android app
  248. // uses data.category to pick siren_<category>.ogg; the per-severity
  249. // override for inminent_colapse wins.
  250. func androidSound(category, sev string) string {
  251. if sev == "inminent_colapse" {
  252. return "klaxon"
  253. }
  254. if category == "" {
  255. return "default"
  256. }
  257. return "siren_" + category
  258. }
  259. func channelForSeverity(sev string) string {
  260. switch sev {
  261. case "critical":
  262. return "critical"
  263. case "inminent_colapse":
  264. return "imminent"
  265. case "warning":
  266. return "warning"
  267. default:
  268. return "info"
  269. }
  270. }