package webhooks import ( "context" "crypto/hmac" "crypto/sha256" "database/sql" "encoding/base64" "encoding/hex" "encoding/json" "errors" "fmt" "io" "log" "net/http" "os" "strings" "sync" "time" "crussell/db" "crussell/handlers/payments" "github.com/jackc/pgx/v5" ) type SquareWebhookEvent struct { Type string `json:"type"` EventID string `json:"event_id"` CreatedAt string `json:"created_at"` Data json.RawMessage `json:"data"` LocationID string `json:"location_id"` } // squareWebhookDedup is a bounded, mutex-guarded set of recently handled // event IDs. It is a FAST-PATH cache only: the persistent source of truth is // the square_webhook_events table (see HandleSquareWebhook). It lets a replayed // delivery be dropped without a DB round-trip, but a restart clears it — the DB // row is what keeps delivery at-least-once across restarts (a replay after a // crash is re-dispatched and absorbed by the idempotent handlers, per the // dispatch-first ordering in HandleSquareWebhook). type squareWebhookDedup struct { mu sync.Mutex seen map[string]struct{} order []string max int } func newSquareWebhookDedup(max int) *squareWebhookDedup { return &squareWebhookDedup{ seen: make(map[string]struct{}), order: make([]string, 0, max), max: max, } } // has reports whether id is in the set WITHOUT recording it. Used as the // fast-path short-circuit before the DB dedup insert. func (d *squareWebhookDedup) has(id string) bool { d.mu.Lock() defer d.mu.Unlock() _, ok := d.seen[id] return ok } // register records id, reporting whether it was already present (set untouched // on a replay, preserving insertion order). Mutex-guarded — the handler may be // hit concurrently. Called only after the DB insert has confirmed the event's // fate, so a failed DB write never leaves a stale entry that would drop a retry. func (d *squareWebhookDedup) register(id string) bool { d.mu.Lock() defer d.mu.Unlock() if _, ok := d.seen[id]; ok { return true } d.seen[id] = struct{}{} d.order = append(d.order, id) if len(d.order) > d.max { oldest := d.order[0] d.order = d.order[1:] delete(d.seen, oldest) } return false } // 500 IDs far exceeds the latency payoff of the fast-path cache; the DB row // (square_webhook_events) is the unbounded, restart-safe source of truth. var squareWebhookEventsSeen = newSquareWebhookDedup(500) // errWebhookParseFailure marks a dispatch error caused by a KNOWN money event // whose payload could not be parsed or extracted (as opposed to a DB failure). // HandleSquareWebhook distinguishes it from other dispatch errors to log the // failure at ERROR level with the event_id and type — and, like every dispatch // error, it returns 5xx WITHOUT committing the dedup row, so Square re-delivers // the event instead of the money state being lost forever. var errWebhookParseFailure = errors.New("webhook payload parse failure") // isSquareMoneyFamily reports whether an event type belongs to a money-state // family (payment.*, refund.*, dispute.*, terminal.*, plus money-adjacent // cash_drawer.*, gift_card.*, transaction.* and invoice.*). invoice.* is // deliberately included even though this app does not use Square Invoices: an // invoice event is money-adjacent (Invoice carries amounts and payment state), // so if Square Invoices are ever used the funds must surface as unknown-money // and stay retried rather than being acked 200 + dedup'd permanently (M11). // Unknown events in these families MUST NOT be 200-acked — see the // default-branch split in HandleSquareWebhook. func isSquareMoneyFamily(eventType string) bool { for _, prefix := range []string{ "payment.", "refund.", "dispute.", "terminal.", "cash_drawer.", "gift_card.", "transaction.", "invoice.", } { if strings.HasPrefix(eventType, prefix) { return true } } return false } // isSquareNonMoneyFamily reports whether an event type belongs to a known // non-money family this app will never process (customer.*, card.*, order.*, // booking.* and the other Square families below, none of which carry money // state this app tracks). These are deliberately acked 200 WITH a dedup row so // Square stops retrying them — see the default-branch split in // HandleSquareWebhook. func isSquareNonMoneyFamily(eventType string) bool { for _, prefix := range []string{ "customer.", "card.", "order.", "booking.", "appointment.", "availability.", "loyalty.", "merchant.", "location.", "labor.", "inventory.", "site.", "device.", "team_member.", "subscription.", "webhook.", } { if strings.HasPrefix(eventType, prefix) { return true } } return false } // webhookDBTimeout bounds the DB work performed while dispatching a webhook. // The work runs on a Background-derived context — so a client disconnect cannot // cancel it (the at-least-once delivery contract must survive) — but is // timeout-bound so a hung DB call cannot hold a pgx pool connection forever; // repeated hangs would otherwise exhaust the pool. 30s is the same // post-request DB budget used elsewhere in the backend (handlers/user). const webhookDBTimeout = 30 * time.Second // webhookDBContext returns a timeout-bound, Background-derived context for // webhook dispatch DB work. func webhookDBContext() (context.Context, context.CancelFunc) { return context.WithTimeout(context.Background(), webhookDBTimeout) } // squareEnvironmentMismatch reports whether the webhook's square-environment // header conflicts with the deployment's configured SQUARE_ENVIRONMENT. // // The check is enforced for every non-dev/mock deployment: configured // production and sandbox are compared directly, and an EMPTY or UNKNOWN // configured SQUARE_ENVIRONMENT is treated as PRODUCTION (LOW-4) — matching how // the rest of the backend treats empty/unknown env fail-closed (main.go:214 // and payments twofa.go:40-42) — so a sandbox subscription mis-pointed at an // unconfigured production URL cannot process sandbox events against production // state. In dev/mock deployments the header is informational and a mismatch is // not rejectable; the dev/mock determination delegates to the shared // payments.IsExplicitDevOrMockEnv (handlers/payments/twofa.go) rather than // re-implementing the env-value list, so this check and the 2FA gate can never // diverge on what counts as the dev/mock stack. // // An absent header is allowed through: real Square deliveries always send it, // so a missing header in an enforced deployment is a non-Square client (which // already failed signature verification) or a local mock/dev poster — both // safely handled downstream. Rejection is 403. NOTE: Square retries ANY // non-2xx response (4xx included) with exponential backoff for up to ~24h, so // the 403 does not by itself stop the retry loop — but a square-environment // mismatch is a PERMANENT configuration error that no retry can resolve, the // handler is intentionally fail-closed (every replay is rejected identically // before any state change), and the DB-level dedup (square_webhook_events + // ON CONFLICT) makes the replayed deliveries replay-safe. The 403 forces the // operator to fix the subscription or the environment setting. func squareEnvironmentMismatch(headerEnv string) bool { headerEnv = strings.ToLower(strings.TrimSpace(headerEnv)) if headerEnv == "" { return false } // Dev/mock deployments (mock/dev/development/test) are never enforced — // the shared predicate is the single source of truth for that // classification, so the webhook's interpretation matches the 2FA gate // exactly (handlers/payments/twofa.go:IsExplicitDevOrMockEnv). if payments.IsExplicitDevOrMockEnv() { return false } configured := strings.ToLower(strings.TrimSpace(os.Getenv("SQUARE_ENVIRONMENT"))) if configured != "sandbox" { // Empty/unknown configured environment is treated as PRODUCTION for the // env check (LOW-4), matching the fail-closed production default the // rest of the backend applies to empty/unknown SQUARE_ENVIRONMENT. configured = "production" } return headerEnv != configured } // HandleSquareWebhook verifies and dispatches Square webhook events. // // Fail-closed chain: 503 when the signing key is unset, 403 on a missing/bad // signature, 400 on malformed JSON or an empty event_id (which cannot be // deduplicated — Square always sends one, so this is defensive). A correctly // signed, well-formed event is deduplicated by event_id before dispatch and // acknowledged 200. Unknown event types are split by family: money-state // families (payment.*, refund.*, dispute.*, terminal.* and money-adjacent // prefixes including invoice.*) and truly unknown prefixes get 501 (Square // retries, no dedup row), while known non-money families (customer.*, card.*, // order.*, booking.*, ...) are acked 200 WITH the dedup row so the subscription // is never flooded into suspension. func HandleSquareWebhook(w http.ResponseWriter, r *http.Request) { r.Body = http.MaxBytesReader(w, r.Body, 512*1024) body, err := io.ReadAll(r.Body) if err != nil { log.Printf("Failed to read webhook body: %v", err) http.Error(w, "request body too large or unreadable", http.StatusRequestEntityTooLarge) return } defer r.Body.Close() // Verification logic per Square spec (HMAC-SHA256, base64, notificationURL + body). // Production setup: set SQUARE_WEBHOOK_SIGNATURE_KEY and SQUARE_WEBHOOK_NOTIFICATION_URL // in env vars (see Square Developer Console → Webhooks → Subscription). // Reference: https://developer.squareup.com/docs/webhooks/step3validate // Fail closed: a missing signing key means the webhook cannot be verified, // so reject rather than process unauthenticated events (S-4). Square // always sends the signature header, so an unset key in production is a // misconfiguration that must not silently accept forged events. signingKey := os.Getenv("SQUARE_WEBHOOK_SIGNATURE_KEY") notificationURL := os.Getenv("SQUARE_WEBHOOK_NOTIFICATION_URL") if notificationURL == "" { notificationURL = "http://localhost:8080/webhooks/square" } if signingKey == "" { log.Printf("SQUARE_WEBHOOK_SIGNATURE_KEY is not set — rejecting webhook (fail-closed)") http.Error(w, "webhook signature verification unavailable", http.StatusServiceUnavailable) return } signature := r.Header.Get("x-square-hmacsha256-signature") if signature == "" { log.Printf("Missing Square webhook signature header") http.Error(w, "Invalid signature", http.StatusForbidden) return } if !verifySquareSignature(body, signature, signingKey, notificationURL) { log.Printf("Invalid Square webhook signature") http.Error(w, "Invalid signature", http.StatusForbidden) return } // Fail-closed environment check: a sandbox subscription mis-pointed at the // production URL + key would otherwise process sandbox events against // production state. 403 is correct because a square-environment mismatch is // a permanent config error no retry can fix — note that Square retries ANY // non-2xx (4xx included) for up to ~24h, so the 403 does not stop the retry // loop by itself; the handler rejects every replay identically (fail-closed, // before any state change) and the DB-level dedup (square_webhook_events + // ON CONFLICT) makes the replayed deliveries replay-safe. See // squareEnvironmentMismatch for the exact enforcement conditions. if squareEnvironmentMismatch(r.Header.Get("square-environment")) { log.Printf("[SQUARE-WEBHOOK] Rejecting event: square-environment header %q does not match configured SQUARE_ENVIRONMENT %q (403)", r.Header.Get("square-environment"), os.Getenv("SQUARE_ENVIRONMENT")) http.Error(w, "square environment mismatch", http.StatusForbidden) return } var event SquareWebhookEvent if err := json.Unmarshal(body, &event); err != nil { log.Printf("Failed to parse webhook event: %v", err) http.Error(w, "Invalid event", http.StatusBadRequest) return } // An empty event_id cannot be deduplicated. Square always sends event_id, // so this is defensive — but once handlers mutate state, a duplicate // empty-ID event would double-apply. Reject with 400 (fail-safe, no // dispatch, no dedup row). NOTE: Square retries ANY non-2xx response (4xx // included) with exponential backoff for up to ~24h, so the 400 does NOT by // itself stop the retry loop — but the handler is fail-closed: every replay // is rejected identically BEFORE any state change or dedup insert, so no // side effect can ever be applied, and the DB-level dedup // (square_webhook_events + ON CONFLICT) keeps the repeated deliveries // replay-safe. if event.EventID == "" { log.Printf("[SQUARE-WEBHOOK] Rejecting event with empty event_id (400)") http.Error(w, "Invalid event", http.StatusBadRequest) return } // Fast-path dedup: a correctly signed replay of a recently handled event is // dropped in memory without a DB round-trip. The persistent source of truth // is the square_webhook_events row committed AFTER dispatch below, so this // cache never hides an event whose dedup row is not yet persisted — a crash // before that commit simply replays the event, which the idempotent handlers // absorb. if squareWebhookEventsSeen.has(event.EventID) { log.Printf("[SQUARE-WEBHOOK] Duplicate event_id %s; skipping (already processed)", event.EventID) w.WriteHeader(http.StatusOK) _, _ = w.Write([]byte("ok")) return } if db.Conn == nil { log.Printf("[SQUARE-WEBHOOK] DB unavailable — rejecting event_id %s (fail-closed)", event.EventID) http.Error(w, "webhook processing unavailable", http.StatusServiceUnavailable) return } log.Printf("[SQUARE-WEBHOOK] Received event: %s", event.Type) // Dispatch FIRST, then commit the dedup row. The handlers mutate state, so a // dedup row committed before dispatch would permanently drop the event on a // crash between the insert and the dispatch (Square's retry would be // 200-skipped, and dispute.created has no sweep fallback). Committing after // a successful dispatch keeps delivery at-least-once: on any dispatch error // NO dedup row is written and we return 5xx so Square retries. The handlers // are idempotent (status='pending'-guarded UPDATEs keyed on the Square id), // so a retry — or two retries dispatching concurrently — applies each state // change at most once and is otherwise a no-op. var dispatchErr error switch event.Type { // Square fires payment.completed and payment.canceled as SEPARATE event // types from payment.updated, but all of them carry the same full Payment // object in data.object.payment, and handlePaymentUpdated reconciles purely // from the payload's id/status (squarePaymentStatusToLocal maps COMPLETED/ // CANCELED/FAILED). Routing all four here closes the silent gap where a // completed or canceled charge was previously acked 200 and never // reconciled against the local pending row. case "payment.updated", "payment.created", "payment.completed", "payment.canceled": dispatchErr = handlePaymentUpdated(event.Data) // Same reasoning for refunds: refund.completed/refund.canceled are distinct // event types carrying the full PaymentRefund object in data.object.refund, // which handleRefundUpdated reconciles by status. case "refund.updated", "refund.created", "refund.completed", "refund.canceled": dispatchErr = handleRefundUpdated(event.Data) case "dispute.created": dispatchErr = handleDisputeCreated(event.Data) case "dispute.state.updated": dispatchErr = handleDisputeStateUpdated(event.Data) case "dispute.evidence.created", "dispute.evidence.deleted": dispatchErr = handleDisputeEvidence(event.Data) case "terminal.checkout.created", "terminal.checkout.updated": dispatchErr = handleTerminalCheckout(event.Data) default: // Unknown-type handling is SPLIT by event family so Square's retry // policy (it re-delivers 5xx responses) can never flood the // subscription into suspension — a suspended subscription silently // kills money-event delivery (payment.created/updated). // // • MONEY-STATE families (any payment.*, refund.*, dispute.*, // terminal.*, cash_drawer.*, gift_card.*, transaction.*, invoice.* // not explicitly handled above, plus future money-adjacent // prefixes): keep 501 so Square retries. A money event this app does // not yet handle is NEVER a safe 200-ack (the caller would commit the // dedup row and the event would be dropped forever); no dedup row is // written and a later retry re-dispatches idempotently if code // support for the type lands before the retry window closes. // invoice.* is treated as money-adjacent even though Square Invoices // are unused today: an invoice carries amounts and payment state, so // acked invoice funds would be invisible to the app (M11). // • KNOWN NON-MONEY families (customer.*, card.*, order.*, // booking.* — plus appointment.*, availability.*, loyalty.*, // merchant.*, location.*, labor.*, inventory.*, site.*, device.*, // team_member.*, subscription.*, webhook.* — and anything else // Square can emit that this app will never process): deliberately // acked 200 WITH the dedup row committed so Square stops retrying. // They carry no money state this app tracks, so acking loses // nothing — and retrying them at any regularity would fill the // subscription's retry queue until Square suspends it, silently // killing money-event delivery. Logged at WARN. // • TRULY unknown prefixes (matching neither list): conservative // 501 — the type could be a new money family Square just added. switch { case isSquareMoneyFamily(event.Type): log.Printf("[SQUARE-WEBHOOK] CRITICAL: unhandled money-family Square event type %q (event_id=%s) — not acknowledged; returning 501 so Square retries", event.Type, event.EventID) // A19: raise the operator-facing admin notification BEFORE the 501. // Without it the event is retried by Square for ~24h and then // silently dropped with only a log line. The notification is a // SEPARATE table — no square_webhook_events dedup row is written // here, so Square keeps retrying the event exactly as before. notifCtx, notifCancel := webhookDBContext() insertUnknownEventNotification(notifCtx, event.Type, event.EventID) notifCancel() http.Error(w, "unhandled webhook event type", http.StatusNotImplemented) return case isSquareNonMoneyFamily(event.Type): log.Printf("[SQUARE-WEBHOOK] WARNING: acknowledged unhandled non-money event %q (event_id=%s) — committed dedup row; Square stops retrying", event.Type, event.EventID) default: log.Printf("[SQUARE-WEBHOOK] CRITICAL: unhandled Square event type %q (event_id=%s) — not acknowledged; returning 501 so Square retries", event.Type, event.EventID) notifCtx, notifCancel := webhookDBContext() insertUnknownEventNotification(notifCtx, event.Type, event.EventID) notifCancel() http.Error(w, "unhandled webhook event type", http.StatusNotImplemented) return } } if dispatchErr != nil { // A parse failure on a known money event is logged at ERROR with the // event_id and type so operators can see exactly which delivery was // unparseable and is being retried (rather than silently lost). Every // dispatch error — parse failure or DB failure — returns 5xx WITHOUT // committing the dedup row, so Square re-delivers the event. if errors.Is(dispatchErr, errWebhookParseFailure) { log.Printf("[SQUARE-WEBHOOK] ERROR: known money event %s (event_id=%s) payload failed to parse: %v — NOT recording dedup row; Square will retry", event.Type, event.EventID, dispatchErr) } else { log.Printf("[SQUARE-WEBHOOK] Event %s (%s) dispatch failed: %v — NOT recording dedup row; Square will retry", event.Type, event.EventID, dispatchErr) } http.Error(w, "webhook processing failed", http.StatusServiceUnavailable) return } // Commit the dedup row AFTER successful dispatch. Fail closed on a write // error: without a persisted row we cannot prove the event was handled, so // reject and let Square retry (the retry re-dispatches idempotently and // retries the insert). event_id is not PII, so logging it is safe. // Commit the dedup row AFTER successful dispatch. Fail closed on a write // error: without a persisted row we cannot prove the event was handled, so // reject and let Square retry (the retry re-dispatches idempotently and // retries the insert). event_id is not PII, so logging it is safe. A // bounded Background context keeps this post-dispatch write alive across a // client disconnect without letting a hung insert hold the pool forever. dedupCtx, dedupCancel := webhookDBContext() defer dedupCancel() tag, err := db.Conn.Exec(dedupCtx, "INSERT INTO square_webhook_events (event_id) VALUES ($1) ON CONFLICT (event_id) DO NOTHING", event.EventID) if err != nil { log.Printf("[SQUARE-WEBHOOK] Failed to record event_id %s (dedup write failed): %v", event.EventID, err) http.Error(w, "webhook processing unavailable", http.StatusServiceUnavailable) return } // Record in the fast-path cache only after the DB write succeeds, so a // failed write never leaves a stale entry that would drop a retry. squareWebhookEventsSeen.register(event.EventID) if tag.RowsAffected() == 0 { // A concurrent duplicate delivery already committed this event's row. log.Printf("[SQUARE-WEBHOOK] Duplicate event_id %s; acknowledging (already processed)", event.EventID) w.WriteHeader(http.StatusOK) _, _ = w.Write([]byte("ok")) return } w.WriteHeader(http.StatusOK) _, _ = w.Write([]byte("ok")) } func verifySquareSignature(body []byte, signature, signingKey, notificationURL string) bool { mac := hmac.New(sha256.New, []byte(signingKey)) mac.Write([]byte(notificationURL)) mac.Write(body) expected := base64.StdEncoding.EncodeToString(mac.Sum(nil)) return hmac.Equal([]byte(signature), []byte(expected)) } // squareWebhookData is the `data` envelope of a Square webhook v1 event. The // affected object's id is at data.id; the full resource is nested at // data.object. (e.g. data.object.payment). Only the id is logged — the // nested object can contain PII and is never echoed to the log. type squareWebhookData struct { ID string `json:"id"` Type string `json:"type"` Object json.RawMessage `json:"object"` } // squareDisputePayload maps the Square Dispute fields this app records. // Reference: https://developer.squareup.com/reference/square/objects/Dispute type squareDisputePayload struct { ID string `json:"id"` State string `json:"state"` AmountMoney *squareMoneyPayload `json:"amount_money"` Reason string `json:"reason"` DisputedPayment *squareDisputedPaymentField `json:"disputed_payment"` } type squareMoneyPayload struct { Amount int64 `json:"amount"` // minor units (pence for GBP) Currency string `json:"currency"` } type squareDisputedPaymentField struct { PaymentID string `json:"payment_id"` } // squarePaymentPayload maps the Square Payment fields this app consumes. type squarePaymentPayload struct { ID string `json:"id"` Status string `json:"status"` // "APPROVED", "COMPLETED", "CANCELED", "FAILED", "PENDING" // ReferenceID / IdempotencyKey / AmountMoney feed the B1-support orphan // detection (detectOrphanedReplayCharge): Square's Payment object carries // them, and a sweep-minted duplicate charge preserves the origin row's // idempotency key / reference_id / amount (the sweep replays the stored // request verbatim). ReferenceID string `json:"reference_id"` IdempotencyKey string `json:"idempotency_key"` AmountMoney *squareMoneyPayload `json:"amount_money"` } // squareRefundPayload maps the Square Refund (PaymentRefund) fields this app // consumes. type squareRefundPayload struct { ID string `json:"id"` Status string `json:"status"` // "PENDING", "COMPLETED", "FAILED" } // parseSquareObject unmarshals data.object. into out. Returns false when // the nested resource is absent (legacy envelope carrying only data.id). func parseSquareObject(object json.RawMessage, key string, out any) bool { if len(object) == 0 { return false } var wrapper map[string]json.RawMessage if err := json.Unmarshal(object, &wrapper); err != nil { return false } raw, ok := wrapper[key] if !ok || len(raw) == 0 { return false } if err := json.Unmarshal(raw, out); err != nil { return false } return true } // squareMoneyToAmount converts a Square Money object (minor units) to an exact // two-decimal string for the NUMERIC(10,2) columns. String formatting avoids // float64 rounding artifacts for money. func squareMoneyToAmount(m *squareMoneyPayload) string { if m == nil || m.Amount <= 0 { return "0.00" } return fmt.Sprintf("%d.%02d", m.Amount/100, m.Amount%100) } // squarePaymentStatusToLocal maps Square's payment state machine to the local // payment_status enum. APPROVED/PENDING are NON-terminal (Square may still // complete or void them), so they map to a zero local status and the caller // leaves the row untouched — the same classification the stale-pending sweeps // use (handlers/payments/sweep.go). func squarePaymentStatusToLocal(status string) (string, bool) { switch status { case "COMPLETED": return "completed", true case "CANCELED", "FAILED": return "failed", true case "APPROVED", "PENDING": return "", false default: return "", false } } // squareRefundStatusToLocal maps Square's refund status to the local // payment_status enum. Square's PaymentRefund states are PENDING, APPROVED, // COMPLETED, CANCELED, FAILED and REJECTED (developer.squareup.com/reference/ // square/objects/PaymentRefund). COMPLETED/FAILED/REJECTED are TERMINAL — // REJECTED (Square declined the refund) is a definitive failure and must be // surfaced as local 'failed' instead of leaving the row pending until the slow // sweep; PENDING and APPROVED are NON-terminal (the refund may still complete // or be rejected) and map to a zero local status so the caller leaves the row // untouched. func squareRefundStatusToLocal(status string) (string, bool) { switch status { case "COMPLETED": return "completed", true case "FAILED", "REJECTED": return "failed", true default: return "", false } } // squareDisputeStateToLocal maps Square's dispute state to the local // disputes.status. Only the terminal resolutions move the row to won/lost; // ACCEPTED (seller accepted the dispute) is a loss — the money is gone. // Everything else (inquiries, evidence required, processing) stays open. func squareDisputeStateToLocal(state string) string { switch state { case "WON": return "won" case "LOST", "ACCEPTED": return "lost" default: return "open" } } // findPaymentBySquareID resolves the local payment id and booking id for a // Square payment id. Multiple local rows can share one Square charge id (e.g. // a deposit + balance split); the most recent is used. The caller supplies a // bounded context (webhookDBContext) so this post-dispatch DB work survives a // client disconnect without holding a pool connection forever. func findPaymentBySquareID(ctx context.Context, squarePaymentID string) (paymentID, bookingID string, ok bool) { if squarePaymentID == "" { return "", "", false } var pid string var bid *string err := db.Conn.QueryRow(ctx, ` SELECT id, booking_id FROM payments WHERE square_payment_id = $1 ORDER BY created_at DESC, id DESC LIMIT 1 `, squarePaymentID).Scan(&pid, &bid) if err != nil { return "", "", false } if bid != nil { bookingID = *bid } return pid, bookingID, true } // findPaymentByDisputeID resolves the local payment (and its booking) recorded // for a dispute row. Used by dispute.state.updated when the dispute row already // exists but the webhook payload carries no resolvable Square payment id. The // caller supplies a bounded context (webhookDBContext). func findPaymentByDisputeID(ctx context.Context, squareDisputeID string) (paymentID, bookingID string) { var pid string var bid *string err := db.Conn.QueryRow(ctx, ` SELECT d.payment_id, p.booking_id FROM disputes d JOIN payments p ON p.id = d.payment_id WHERE d.square_dispute_id = $1 `, squareDisputeID).Scan(&pid, &bid) if err != nil { return "", "" } if bid != nil { bookingID = *bid } return pid, bookingID } // disputeNotificationID derives the deterministic admin_notifications id for an // untracked dispute's critical_payment_log notification: 'D' + 11 lowercase hex // chars of a SHA-256 over 'dispute-'. generate_short_id // (init-script.sql) only ever emits 12 lowercase hex chars // (substr(encode(gen_random_bytes(6),'hex'),1,12)), so the uppercase 'D' prefix // guarantees this can never collide with a DB-generated id. The id is stable // per dispute, giving ON CONFLICT (id) DO NOTHING per-dispute idempotency. func disputeNotificationID(squareDisputeID string) string { sum := sha256.Sum256([]byte("dispute-" + squareDisputeID)) return "D" + hex.EncodeToString(sum[:])[:11] } // insertCriticalPaymentNotification surfaces a money event in the admin // notification centre (reason='critical_payment_log'), the DB-backed stand-in // for un-watched CRITICAL log lines (see ScanCriticalPaymentLogs in // internal/jobs/cleanup.go). Dedup: one unacknowledged row per (reason, // booking_id) — acknowledging re-arms it. // // NOTE: this is the WEBHOOK-specific variant of a same-named helper in the // payments package (handlers/payments/sweep.go, insertCriticalPaymentNotification) // with DIFFERENT semantics: that one has the signature (ctx, bookingID, // userID *string), writes the user_id column, and dedups on (reason, // booking_id, user_id); this one takes (ctx, bookingID, disputeID string), // never writes user_id, and dedups on (reason, booking_id) or the // deterministic per-dispute id. They share the admin_notifications table and // the 'critical_payment_log' reason but serve different call paths (sweep vs // webhook) — do not merge them. // // AUTHORITATIVE NOT EXISTS GUARD: the booking-scoped branch below is the // canonical form of this dedup — `an.reason = 'critical_payment_log' AND // an.booking_id IS NOT DISTINCT FROM $1 AND an.acknowledged_at IS NULL` (a // notification blocks a re-notify until the admin acknowledges it, then a NEW // event re-arms). sweep.go's insertCriticalPaymentNotification mirrors this // predicate in shape (its copy applies the same NOT EXISTS/acknowledged_at // IS NULL guard over (reason, booking_id, user_id) — an intentional // user-scoping addition for sweep-originated events, documented on the sweep // copy itself); if the guard ever changes, sweep.go must be updated to match. // // Untracked disputes (no local payment row, booking_id NULL) pass disputeID // instead: each DISTINCT square dispute gets its OWN notification under the // deterministic id (disputeNotificationID), so a second distinct chargeback is // never suppressed by the first's (reason, NULL booking) row — and re-delivery // of the same dispute is a no-op (ON CONFLICT (id) DO NOTHING). The // booking-scoped NOT EXISTS guard does NOT apply to this path: it would // collapse every untracked dispute onto one unacknowledged NULL-booking row. // The caller supplies a bounded context (webhookDBContext). func insertCriticalPaymentNotification(ctx context.Context, bookingID, disputeID string) { if disputeID != "" { id := disputeNotificationID(disputeID) tag, err := db.Conn.Exec(ctx, ` INSERT INTO admin_notifications (id, reason, booking_id, created_at) VALUES ($1, 'critical_payment_log'::admin_notification_reason, NULL, NOW()) ON CONFLICT (id) DO NOTHING `, id) if err != nil { log.Printf("[SQUARE-WEBHOOK] Failed to insert critical_payment_log admin notification: %v", err) return } if tag.RowsAffected() > 0 { log.Printf("[SQUARE-WEBHOOK] Inserted critical_payment_log admin notification (dispute_id=%s, booking_id=NULL)", disputeID) } return } var bid any if bookingID != "" { bid = bookingID } tag, err := db.Conn.Exec(ctx, ` INSERT INTO admin_notifications (reason, booking_id, created_at) SELECT 'critical_payment_log'::admin_notification_reason, $1, NOW() WHERE NOT EXISTS ( SELECT 1 FROM admin_notifications an WHERE an.reason = 'critical_payment_log' AND an.booking_id IS NOT DISTINCT FROM $1 AND an.acknowledged_at IS NULL ) `, bid) if err != nil { log.Printf("[SQUARE-WEBHOOK] Failed to insert critical_payment_log admin notification: %v", err) return } if tag.RowsAffected() > 0 { log.Printf("[SQUARE-WEBHOOK] Inserted critical_payment_log admin notification (booking_id=%s)", bookingID) } } // unknownEventNotificationID derives the deterministic admin_notifications id // for an unhandled money-family/truly-unknown webhook event's // critical_payment_log notification: 'U' + 11 lowercase hex chars of a SHA-256 // over 'unknown-event-'. Mirrors disputeNotificationID (which uses an // uppercase 'D' prefix): the two are intentionally parallel — same // prefix + hex(sha256(input))[:11] scheme (44 bits), DIFFERENT prefixes, so a // dispute id and an unknown-event id can never collide even on identical hash // input. generate_*_id (init-script.sql) only ever emits 12 // lowercase hex chars, so the uppercase 'U' prefix guarantees no collision with // a DB-generated id. The id is stable per event_id, giving // ON CONFLICT (id) DO NOTHING per-event idempotency across redeliveries. func unknownEventNotificationID(eventID string) string { sum := sha256.Sum256([]byte("unknown-event-" + eventID)) return "U" + hex.EncodeToString(sum[:])[:11] } // insertUnknownEventNotification raises a critical_payment_log admin // notification for a money-family/truly-unknown webhook event the handler // refuses to ack (501). Without it the event is retried by Square for ~24h and // then silently dropped with only a log line. The notification is deduped by // its deterministic id (unknownEventNotificationID) and lives in a SEPARATE // table — no square_webhook_events dedup row is written on the 501 path, so // Square keeps retrying the event exactly as before. Best-effort: an insert // failure is logged, never a dispatch error (the 501 is already the response). // The caller supplies a bounded context (webhookDBContext). func insertUnknownEventNotification(ctx context.Context, eventType, eventID string) { id := unknownEventNotificationID(eventID) tag, err := db.Conn.Exec(ctx, ` INSERT INTO admin_notifications (id, reason, booking_id, created_at) VALUES ($1, 'critical_payment_log'::admin_notification_reason, NULL, NOW()) ON CONFLICT (id) DO NOTHING `, id) if err != nil { log.Printf("[SQUARE-WEBHOOK] Failed to insert critical_payment_log admin notification for unhandled event %q (event_id=%s): %v", eventType, eventID, err) return } if tag.RowsAffected() > 0 { log.Printf("[SQUARE-WEBHOOK] Inserted critical_payment_log admin notification for unhandled event type %q (event_id=%s)", eventType, eventID) } } // markPaymentFailed flips a payment to 'failed' after a lost dispute — the // money was charged back, so the row must not read as collected. 'refunded' // rows are left alone (the money was returned by refund, not charged back). // The caller supplies a bounded context (webhookDBContext). func markPaymentFailed(ctx context.Context, paymentID string) error { if paymentID == "" { return nil } _, err := db.Conn.Exec(ctx, "UPDATE payments SET status = 'failed', updated_at = NOW() WHERE id = $1 AND status IN ('pending', 'completed')", paymentID) if err != nil { log.Printf("[SQUARE-WEBHOOK] Failed to mark payment %s failed after lost dispute: %v", paymentID, err) return err } return nil } // squarePaymentKnown reports whether ANY local payments row carries the Square // payment id, whatever its status. The pending-only UPDATE in // handlePaymentUpdated matches zero rows both when no row exists at all and // when the row already settled (e.g. a completed webhook replay of an // already-'completed' charge). This existence check distinguishes those two: // a row existing means the event is a plain no-op replay of a known charge, // while no row at all is the signature of an ORPHANED sweep-minted duplicate. // Unlike findPaymentBySquareID (which swallows errors), a DB failure here // propagates so the caller rejects the event and Square retries. func squarePaymentKnown(ctx context.Context, squarePaymentID string) (bool, error) { var one int err := db.Conn.QueryRow(ctx, `SELECT 1 FROM payments WHERE square_payment_id = $1 LIMIT 1`, squarePaymentID).Scan(&one) if errors.Is(err, pgx.ErrNoRows) { return false, nil } if err != nil { return false, err } return true, nil } // findPendingByOrphanKeys locates the pending ORIGIN row of a likely // sweep-minted duplicate charge: the pending row that shares the replayed // charge's idempotency key (primary — exact, because Square returns the key on // the Payment object and payments.idempotency_key is UNIQUE) or, failing that, // the pending row whose booking/gift-card reference AND amount match the // payload's reference_id/amount_money (the sweep replays the stored snapshot // verbatim, preserving both). Only 'pending' rows WITHOUT a square_payment_id // are candidates — that is exactly the population the keyed stale-pending sweep // replays (sweep.go), so a match is the sweep-minted duplicate's origin. func findPendingByOrphanKeys(ctx context.Context, payment squarePaymentPayload) (paymentID, bookingID string, found bool, err error) { if payment.IdempotencyKey != "" { var pid string var bid *string err := db.Conn.QueryRow(ctx, ` SELECT id, booking_id FROM payments WHERE status = 'pending' AND idempotency_key = $1 AND square_payment_id IS NULL ORDER BY created_at DESC, id DESC LIMIT 1 `, payment.IdempotencyKey).Scan(&pid, &bid) if err == nil { if bid != nil { bookingID = *bid } return pid, bookingID, true, nil } if !errors.Is(err, pgx.ErrNoRows) { return "", "", false, err } } // reference_id fallback: booking charges carry the booking id as // reference_id; a gift-card-linked payment carries the card code. The // amount guard stops a coincidental reference match (a pending row for the // same booking at a different amount) from being mistaken for the origin. if payment.ReferenceID != "" && payment.AmountMoney != nil && payment.AmountMoney.Amount > 0 { amount := float64(payment.AmountMoney.Amount) / 100.0 var pid string var bid *string err := db.Conn.QueryRow(ctx, ` SELECT id, booking_id FROM payments WHERE status = 'pending' AND square_payment_id IS NULL AND ABS(amount - $2) < 0.005 AND (booking_id = $1 OR gift_card_id = $1) ORDER BY created_at DESC, id DESC LIMIT 1 `, payment.ReferenceID, amount).Scan(&pid, &bid) if err == nil { if bid != nil { bookingID = *bid } return pid, bookingID, true, nil } if !errors.Is(err, pgx.ErrNoRows) { return "", "", false, err } } return "", "", false, nil } // orphanReplayChargeNotificationID derives the deterministic admin_notifications // id for an orphaned sweep-minted duplicate charge's critical_payment_log // notification: 'O' + 11 lowercase hex chars of a SHA-256 over // 'orphan-replay-'. Mirrors disputeNotificationID ('D') and // unknownEventNotificationID ('U'): same prefix + hex(sha256(input))[:11] // scheme with a distinct uppercase prefix, so the three id spaces can never // collide and the uppercase prefix guarantees no collision with a DB-generated // id (generate_*_id emits 12 lowercase hex chars). The id is stable per ORIGIN // payment row, so ON CONFLICT (id) DO NOTHING keeps re-deliveries and // re-notifications of the same orphan charge to one row. func orphanReplayChargeNotificationID(paymentID string) string { sum := sha256.Sum256([]byte("orphan-replay-" + paymentID)) return "O" + hex.EncodeToString(sum[:])[:11] } // insertOrphanReplayChargeNotification surfaces an orphaned sweep-minted // duplicate charge in the admin notification centre (reason // 'critical_payment_log'), deduped by the deterministic per-origin-row id so // repeated deliveries of the same orphan charge's event never add a second // row. booking_id is set when the origin payment row has one, giving the owner // a booking to act from. Best-effort: an insert failure is logged, never a // dispatch error. func insertOrphanReplayChargeNotification(ctx context.Context, paymentID, bookingID string) { if paymentID == "" { return } id := orphanReplayChargeNotificationID(paymentID) var bid any if bookingID != "" { bid = bookingID } tag, err := db.Conn.Exec(ctx, ` INSERT INTO admin_notifications (id, reason, booking_id, created_at) VALUES ($1, 'critical_payment_log'::admin_notification_reason, $2, NOW()) ON CONFLICT (id) DO NOTHING `, id, bid) if err != nil { log.Printf("[SQUARE-WEBHOOK] Failed to insert orphaned-replay admin notification: %v", err) return } if tag.RowsAffected() > 0 { log.Printf("[SQUARE-WEBHOOK] Inserted orphaned-replay-charge admin notification (origin payment=%s, booking=%q)", paymentID, bookingID) } } // detectOrphanedReplayCharge handles a COMPLETED payment.updated event whose // square_payment_id matches NO local payments row — the signature of an // ORPHANED SWEEP-MINTED duplicate charge (B1-support). When the stale-pending // sweep replays a pending row's stored idempotency key against Square and the // key has expired, Square creates a NEW charge under that same key; the new // charge's payment.updated event arrives here with no local row of its own, // while the pending ORIGIN row (the one the sweep replayed) still exists and // shares the replayed idempotency key — and, for booking/gift-card charges, // the same reference_id and amount. // // Action (the webhook half of B1): mark the origin row 'failed' — it is NOT // the charge that completed, so rescuing it to 'completed' would hide the // duplicate behind the original and rescuing by square_payment_id is // impossible (the orphan has no row) — and surface a deduped 'orphaned replay // charge detected' admin notification. The AUTO-REFUND of the orphan charge is // the SWEEP's job (sweep.go, the money agent's B1 fix): this path NEVER issues // a refund and must never be turned into one; it detects + notifies + settles // the origin row so the sweep does not blind-fail or double-rescue it later. func detectOrphanedReplayCharge(ctx context.Context, payment squarePaymentPayload) error { originID, bookingID, found, err := findPendingByOrphanKeys(ctx, payment) if err != nil { log.Printf("[SQUARE-WEBHOOK] Orphan-replay origin lookup failed for square payment %s: %v", payment.ID, err) return err } if !found { log.Printf("[SQUARE-WEBHOOK] payment.updated: COMPLETED square payment %s matches no local row and no pending origin row by idempotency key/reference_id — acknowledging (not a sweep-minted duplicate)", payment.ID) return nil } tag, err := db.Conn.Exec(ctx, `UPDATE payments SET status = 'failed', updated_at = NOW() WHERE id = $1 AND status = 'pending'`, originID) if err != nil { log.Printf("[SQUARE-WEBHOOK] Failed to mark orphaned-replay origin payment %s failed: %v", originID, err) return err } if tag.RowsAffected() > 0 { log.Printf("[SQUARE-WEBHOOK] CRITICAL: COMPLETED square payment %s (orphaned sweep-minted duplicate) — origin pending payment %s marked failed; admin notified", payment.ID, originID) } else { log.Printf("[SQUARE-WEBHOOK] CRITICAL: COMPLETED square payment %s (orphaned sweep-minted duplicate) — origin payment %s already resolved; admin notified", payment.ID, originID) } insertOrphanReplayChargeNotification(ctx, originID, bookingID) return nil } // handlePaymentUpdated reconciles a Square Payment state change against the // local payments row (real-time counterpart to the stale-pending sweep). The // Square id is logged, never the payload (PII). Idempotent: the UPDATE is a // no-op when the local status already matches, and event_id dedup prevents // re-entry at the handler level. A non-nil error means dispatch failed and the // caller must NOT commit the dedup row (Square retries). All DB work runs on a // bounded Background context (webhookDBContext): a client disconnect must not // cancel the state mutation, and a hung DB call must not hold the pool forever. func handlePaymentUpdated(data json.RawMessage) error { ctx, cancel := webhookDBContext() defer cancel() var env squareWebhookData if err := json.Unmarshal(data, &env); err != nil { // Parse failure on a known money event: the caller returns 5xx without // committing the dedup row (errWebhookParseFailure), so Square // re-delivers and the payment state change is not permanently lost. return fmt.Errorf("payment event data envelope: %v: %w", err, errWebhookParseFailure) } if len(env.Object) == 0 { // Legacy envelope carrying only data.id — no nested payment object to // reconcile, so there is nothing to apply; acknowledge as received. log.Printf("[SQUARE-WEBHOOK] payment.updated received (data.id=%s)", env.ID) return nil } var payment squarePaymentPayload if !parseSquareObject(env.Object, "payment", &payment) || payment.ID == "" || payment.Status == "" { // A payment object is present but unusable (malformed, or missing the // id/status reconciliation needs): cannot apply money state — retry. return fmt.Errorf("payment.updated payload for data.id=%q missing/invalid payment object (id=%q status=%q): %w", env.ID, payment.ID, payment.Status, errWebhookParseFailure) } localStatus, terminal := squarePaymentStatusToLocal(payment.Status) if !terminal { log.Printf("[SQUARE-WEBHOOK] payment.updated: square payment %s status %q is non-terminal — no local state change", payment.ID, payment.Status) return nil } // Only 'pending' rows are candidates for a terminal transition — the same // conservative rule the stale-pending sweeps use. A webhook for an already // settled row (Square fires payment.updated for ANY field change, e.g. fee // recalculation on a fully-refunded charge) must never revert a terminal // status like 'refunded' back to 'completed'. tag, err := db.Conn.Exec(ctx, `UPDATE payments SET status = $1, updated_at = NOW() WHERE square_payment_id = $2 AND status = 'pending'`, localStatus, payment.ID) if err != nil { log.Printf("[SQUARE-WEBHOOK] Failed to update payment %s to status %s: %v", payment.ID, localStatus, err) return err } if tag.RowsAffected() > 0 { log.Printf("[SQUARE-WEBHOOK] payment.updated: square payment %s → local status %s", payment.ID, localStatus) } else if localStatus == "completed" { // B1-support: a COMPLETED payment.updated that matched NO pending // 'payments' row may be an ORPHANED SWEEP-MINTED duplicate — the // stale-pending sweep replayed a stored idempotency key against an // expired key, Square landed a NEW charge whose id matches nothing // locally, and this is that new charge's completion event. If NO local // row carries this square_payment_id at all (not even a settled one), // hunt for the pending origin row and settle it so the sweep never // blind-fails or double-rescues it. A settled row existing means this // is a plain no-op replay of a known charge and is left untouched. known, kErr := squarePaymentKnown(ctx, payment.ID) if kErr != nil { log.Printf("[SQUARE-WEBHOOK] Failed to check whether square payment %s is known: %v", payment.ID, kErr) return kErr } if !known { if dErr := detectOrphanedReplayCharge(ctx, payment); dErr != nil { return dErr } } } // A Square charge can also map to a till_sales row (online gift-card // purchase, retail at the till) — reconcile those too. Same pending-only // guard: never revert a terminal till-sale status. A definitively failed // charge (Square FAILED/CANCELED) claws back the gift-card funding those // pending sales added, exactly like the stale-pending sweep // (handlers/payments/sweep.go); an ambiguous status never reaches here. if localStatus == "failed" { return clawbackFailedTillSales(ctx, payment.ID) } tsTag, err := db.Conn.Exec(ctx, `UPDATE till_sales SET status = $1, updated_at = NOW() WHERE square_payment_id = $2 AND status = 'pending'`, localStatus, payment.ID) if err != nil { log.Printf("[SQUARE-WEBHOOK] Failed to reconcile till_sales for square payment %s: %v", payment.ID, err) return err } if tsTag.RowsAffected() > 0 { log.Printf("[SQUARE-WEBHOOK] payment.updated: reconciled %d till_sale(s) for square payment %s → status %s", tsTag.RowsAffected(), payment.ID, localStatus) } return nil } // clawbackFailedTillSales reverts the gift-card funding of every still-pending // till sale funded by a Square charge that is DEFINITIVELY failed (Square // FAILED/CANCELED — never ambiguous). It mirrors the stale-pending sweep's // clawbackTillSaleFunding + revertGiftCardFunding (handlers/payments/sweep.go, // till.go): each sale is claimed with a status='pending' guard so an // already-resolved row is skipped without error, and the failed mark + funding // revert commit atomically. A non-nil error means a DB failure left a pending // sale's funding unreverted — the caller rejects the webhook so Square retries // the clawback (the sweep is the eventual backstop). func clawbackFailedTillSales(ctx context.Context, squarePaymentID string) error { rows, err := db.Conn.Query(ctx, ` SELECT ts.id, ts.item_type, ts.item_id, ts.total_amount, gc.redeemed_by, (ts.created_at = gc.created_at) AS is_create FROM till_sales ts LEFT JOIN gift_cards gc ON gc.id = ts.item_id WHERE ts.square_payment_id = $1 AND ts.status = 'pending' `, squarePaymentID) if err != nil { log.Printf("[SQUARE-WEBHOOK] Failed to read pending till_sales for funding clawback (square payment %s): %v", squarePaymentID, err) return err } defer rows.Close() for rows.Next() { var ( saleID string itemType string itemID sql.NullString totalAmount float64 redeemedBy sql.NullString isCreate *bool ) if err := rows.Scan(&saleID, &itemType, &itemID, &totalAmount, &redeemedBy, &isCreate); err != nil { log.Printf("[SQUARE-WEBHOOK] Failed to scan pending till_sale for funding clawback (square payment %s): %v", squarePaymentID, err) return err } if err := clawbackOneTillSale(ctx, saleID, itemType, itemID, totalAmount, redeemedBy, isCreate); err != nil { return err } } return rows.Err() } // clawbackOneTillSale resolves one pending till sale of a definitively failed // charge. A gift-card sale has its funding reverted atomically with the failed // mark; a sale with no gift card (future retail product / orphaned item) is // only marked failed. An already-resolved sale is skipped, not an error. func clawbackOneTillSale(ctx context.Context, saleID, itemType string, itemID sql.NullString, totalAmount float64, redeemedBy sql.NullString, isCreate *bool) error { if itemType != "gift_card" || !itemID.Valid || itemID.String == "" || isCreate == nil { // No gift card to claw back — mark the sale failed without touching // any card (mirrors the sweep's non-gift-card branch). tag, err := db.Conn.Exec(ctx, ` UPDATE till_sales SET status = 'failed', updated_at = NOW() WHERE id = $1 AND status = 'pending' `, saleID) if err != nil { log.Printf("[SQUARE-WEBHOOK] Failed to mark till sale %s failed: %v", saleID, err) return err } if tag.RowsAffected() > 0 { log.Printf("[SQUARE-WEBHOOK] payment.updated: marked till sale %s failed (no gift card to claw back)", saleID) } return nil } action := "topup" if *isCreate { action = "create" } var redeem *string if redeemedBy.Valid && redeemedBy.String != "" { redeem = &redeemedBy.String } if err := revertTillSaleGiftCardFunding(ctx, action, itemID.String, totalAmount, redeem, saleID); err != nil { if payments.IsTillSaleNotPending(err) { log.Printf("[SQUARE-WEBHOOK] Till sale %s was already resolved (not pending) — skipping funding clawback", saleID) return nil } log.Printf("CRITICAL: [SQUARE-WEBHOOK] failed to claw back gift card %s funding for failed till sale %s: %v — MANUAL RECONCILIATION REQUIRED: gift card may still be funded", itemID.String, saleID, err) return err } return nil } // revertTillSaleGiftCardFunding undoes the gift-card funding of a till sale // whose charge definitively failed. It delegates to the single shared clawback // implementation, payments.RevertGiftCardFunding (handlers/payments/ // giftcard_clawback.go) — the same helper the till handler and the stale-pending // sweep use — so the webhook's money reversal can never drift from theirs. The // claim-first status='pending' guard, the create/top-up branches, the // redeem-to-user reversal, the CRITICAL reconciliation log lines and the // errTillSaleNotPending sentinel all live in that one place. func revertTillSaleGiftCardFunding(ctx context.Context, action, giftCardID string, amount float64, redeemToUserID *string, tillSaleID string) error { return payments.RevertGiftCardFunding(ctx, action, giftCardID, amount, redeemToUserID, tillSaleID) } // handleRefundUpdated reconciles a Square Refund state change against the local // refunds row. Idempotent (status-guarded UPDATE + event_id dedup). A non-nil // error means dispatch failed and the caller must NOT commit the dedup row. DB // work runs on a bounded Background context (webhookDBContext). func handleRefundUpdated(data json.RawMessage) error { ctx, cancel := webhookDBContext() defer cancel() var env squareWebhookData if err := json.Unmarshal(data, &env); err != nil { // Parse failure on a known money event: retry via 5xx, no dedup row. return fmt.Errorf("refund event data envelope: %v: %w", err, errWebhookParseFailure) } if len(env.Object) == 0 { // Legacy envelope carrying only data.id — no nested refund object to // reconcile; acknowledge as received. log.Printf("[SQUARE-WEBHOOK] refund.updated received (data.id=%s)", env.ID) return nil } var refund squareRefundPayload if !parseSquareObject(env.Object, "refund", &refund) || refund.ID == "" || refund.Status == "" { // A refund object is present but unusable: cannot apply money state — retry. return fmt.Errorf("refund.updated payload for data.id=%q missing/invalid refund object (id=%q status=%q): %w", env.ID, refund.ID, refund.Status, errWebhookParseFailure) } localStatus, terminal := squareRefundStatusToLocal(refund.Status) if !terminal { log.Printf("[SQUARE-WEBHOOK] refund.updated: square refund %s status %q is non-terminal — no local state change", refund.ID, refund.Status) return nil } // COMPLETED may promote any non-completed row (incl. a sweep-failed refund // Square later shows complete) — the over-refund guard counts completed // refunds, so this only tightens it. FAILED only demotes a 'pending' row: // demoting 'completed' would let the guard exclude money that already moved // (the exact risk refunds.go documents for failed refunds). switch localStatus { case "completed": tag, err := db.Conn.Exec(ctx, `UPDATE refunds SET status = 'completed' WHERE square_refund_id = $1 AND status <> 'completed'`, refund.ID) if err != nil { log.Printf("[SQUARE-WEBHOOK] Failed to update refund %s to status %s: %v", refund.ID, localStatus, err) return err } if tag.RowsAffected() > 0 { log.Printf("[SQUARE-WEBHOOK] refund.updated: square refund %s → local status %s", refund.ID, localStatus) } return nil case "failed": // A5d: this webhook demotes a 'pending' row to 'failed' BEFORE the sweep // ever sees it — the sweep only processes 'pending' rows, so its // failed-refund admin notification would be permanently lost. Raise the // same 'refund_failed' notification here (via the shared exported // payments.InsertRefundFailedNotifications — the single source for the // SQL and its (reason='refund_failed', booking_id) dedup guard), guarded // to the actually-demoted row. var rowID string err := db.Conn.QueryRow(ctx, `UPDATE refunds SET status = 'failed' WHERE square_refund_id = $1 AND status = 'pending' RETURNING id`, refund.ID).Scan(&rowID) if errors.Is(err, pgx.ErrNoRows) { // No pending row matched — the refund is already resolved; a late // FAILED/REJECTED replay must not demote or notify anything. return nil } if err != nil { log.Printf("[SQUARE-WEBHOOK] Failed to update refund %s to status %s: %v", refund.ID, localStatus, err) return err } log.Printf("[SQUARE-WEBHOOK] refund.updated: square refund %s → local status failed (row %s)", refund.ID, rowID) payments.InsertRefundFailedNotifications(ctx, []string{rowID}) return nil } return nil } // truncateDisputeReason caps a Square dispute reason at the disputes.reason // VARCHAR(192) column width. An over-long reason would fail the disputes // INSERT; the handler treats that as a dispatch error (no dedup row, 5xx), so // Square would retry forever — truncating lets the event succeed instead. // VARCHAR(192) counts CHARACTERS, not bytes, so truncation must slice on a rune // boundary: byte-slicing (reason[:192]) can split a multi-byte UTF-8 rune and // store invalid UTF-8 (which Postgres rejects), failing the INSERT the same way // an over-long reason would. Slicing the []rune form keeps the stored reason a // valid, at-most-192-character UTF-8 string. func truncateDisputeReason(reason string) string { runes := []rune(reason) if len(runes) > 192 { return string(runes[:192]) } return reason } // handleDisputeCreated records a newly opened dispute: inserts the disputes row // and surfaces a critical_payment_log admin notification so the owner sees the // chargeback in-app. A disputed Square payment with NO local payments row (a // Dashboard-initiated charge, a mismatched Square id, or a deleted/erased row) // cannot be reconciled to a booking — no disputes row is written — but the // admin notification is STILL raised with booking_id NULL, because there is no // sweep fallback for disputes and a chargeback the app cannot see is a silent // money-loss path the owner must always be told about. Idempotent via // ON CONFLICT (square_dispute_id) DO NOTHING plus the event_id dedup. A non-nil // error means dispatch failed (no dedup row committed — Square retries). DB // work runs on a bounded Background context (webhookDBContext). func handleDisputeCreated(data json.RawMessage) error { ctx, cancel := webhookDBContext() defer cancel() var env squareWebhookData if err := json.Unmarshal(data, &env); err != nil { // Parse failure on a known money event (chargeback): retry via 5xx, no // dedup row — disputes have no sweep fallback, so a lost event is a // silent money-loss path. return fmt.Errorf("dispute event data envelope: %v: %w", err, errWebhookParseFailure) } if len(env.Object) == 0 { // Legacy envelope carrying only data.id — no nested dispute object to // record; acknowledge as received. log.Printf("[SQUARE-WEBHOOK] dispute.created received (data.id=%s)", env.ID) return nil } var dispute squareDisputePayload if !parseSquareObject(env.Object, "dispute", &dispute) || dispute.ID == "" { // A dispute object is present but unusable: cannot record the // chargeback — retry. return fmt.Errorf("dispute.created payload for data.id=%q missing/invalid dispute object (id=%q): %w", env.ID, dispute.ID, errWebhookParseFailure) } squarePaymentID := "" if dispute.DisputedPayment != nil { squarePaymentID = dispute.DisputedPayment.PaymentID } paymentID, bookingID, paymentFound := findPaymentBySquareID(ctx, squarePaymentID) if !paymentFound { // Untracked chargeback: no local payments row for this Square charge // (Dashboard-initiated, mismatched Square payment id, or a deleted/erased // row). There is NO sweep fallback for disputes — this notification is // the only in-app trace the owner gets that Square is clawing back funds, // so it must never be skipped. booking_id stays NULL; each DISTINCT // dispute gets its OWN deterministic-id notification (the booking-scoped // dedup would collapse separate chargebacks into one suppressed row). // Still return nil so the dedup row commits and Square's retry is // acknowledged. log.Printf("[SQUARE-WEBHOOK] CRITICAL: dispute %s created for square payment %q with NO local payment row — chargeback cannot be reconciled in-app — admin notified (booking_id NULL)", dispute.ID, squarePaymentID) insertCriticalPaymentNotification(ctx, "", dispute.ID) return nil } amount := squareMoneyToAmount(dispute.AmountMoney) tag, err := db.Conn.Exec(ctx, ` INSERT INTO disputes (square_dispute_id, payment_id, status, amount, reason, created_at, updated_at) VALUES ($1, $2, 'open', $3, NULLIF($4, ''), NOW(), NOW()) ON CONFLICT (square_dispute_id) DO NOTHING `, dispute.ID, paymentID, amount, truncateDisputeReason(dispute.Reason)) if err != nil { log.Printf("[SQUARE-WEBHOOK] Failed to insert dispute %s: %v", dispute.ID, err) return err } _ = tag insertCriticalPaymentNotification(ctx, bookingID, "") log.Printf("[SQUARE-WEBHOOK] CRITICAL: dispute %s created (amount %s, reason %q) for square payment %s — admin notified", dispute.ID, amount, truncateDisputeReason(dispute.Reason), squarePaymentID) return nil } // handleDisputeStateUpdated applies a Square dispute state change to the local // disputes row (upsert — a state.updated may arrive before the created event), // and on a terminal loss marks the payment failed + raises CRITICAL. Won is // logged only. Idempotent: the upsert converges to the same row. A non-nil // error means dispatch failed (no dedup row committed — Square retries). DB // work runs on a bounded Background context (webhookDBContext). func handleDisputeStateUpdated(data json.RawMessage) error { ctx, cancel := webhookDBContext() defer cancel() var env squareWebhookData if err := json.Unmarshal(data, &env); err != nil { // Parse failure on a known money event (chargeback): retry via 5xx, no // dedup row — disputes have no sweep fallback. return fmt.Errorf("dispute event data envelope: %v: %w", err, errWebhookParseFailure) } if len(env.Object) == 0 { // Legacy envelope carrying only data.id — no nested dispute object; // acknowledge as received. log.Printf("[SQUARE-WEBHOOK] dispute.state.updated received (data.id=%s)", env.ID) return nil } var dispute squareDisputePayload if !parseSquareObject(env.Object, "dispute", &dispute) || dispute.ID == "" { // A dispute object is present but unusable: cannot apply the state // change — retry. return fmt.Errorf("dispute.state.updated payload for data.id=%q missing/invalid dispute object (id=%q): %w", env.ID, dispute.ID, errWebhookParseFailure) } localStatus := squareDisputeStateToLocal(dispute.State) amount := squareMoneyToAmount(dispute.AmountMoney) squarePaymentID := "" if dispute.DisputedPayment != nil { squarePaymentID = dispute.DisputedPayment.PaymentID } paymentID, bookingID, paymentFound := findPaymentBySquareID(ctx, squarePaymentID) if !paymentFound { // Row may already exist from dispute.created — recover its payment. paymentID, bookingID = findPaymentByDisputeID(ctx, dispute.ID) if paymentID == "" { // Untracked chargeback: no local payments row for this Square // charge AND no dispute row to recover one from. Mirror the // dispute.created untracked branch — there is NO sweep fallback for // disputes, so this critical notification is the only in-app trace // the owner gets that Square is clawing back funds; a state.updated // arriving without a prior created event must surface it too, never // drop it silently. booking_id stays NULL; disputeNotificationID // gives each DISTINCT dispute its own deterministic-id notification // and makes this insert a no-op if dispute.created already raised // it. Still return nil so the dedup row commits and Square's retry // is acknowledged. log.Printf("[SQUARE-WEBHOOK] CRITICAL: dispute %s state change (state %s) for square payment %q with NO local payment row — chargeback cannot be reconciled in-app — admin notified (booking_id NULL)", dispute.ID, dispute.State, squarePaymentID) insertCriticalPaymentNotification(ctx, "", dispute.ID) return nil } } _, err := db.Conn.Exec(ctx, ` INSERT INTO disputes (square_dispute_id, payment_id, status, amount, reason, created_at, updated_at) VALUES ($1, $2, $3, $4, NULLIF($5, ''), NOW(), NOW()) ON CONFLICT (square_dispute_id) DO UPDATE SET status = EXCLUDED.status, amount = EXCLUDED.amount, reason = EXCLUDED.reason, updated_at = NOW() `, dispute.ID, paymentID, localStatus, amount, truncateDisputeReason(dispute.Reason)) if err != nil { log.Printf("[SQUARE-WEBHOOK] Failed to update dispute %s to state %s: %v", dispute.ID, dispute.State, err) return err } switch localStatus { case "lost": if err := markPaymentFailed(ctx, paymentID); err != nil { return err } insertCriticalPaymentNotification(ctx, bookingID, "") log.Printf("[SQUARE-WEBHOOK] CRITICAL: dispute %s LOST — payment %s marked failed; admin notified", dispute.ID, paymentID) case "won": log.Printf("[SQUARE-WEBHOOK] dispute %s WON — resolved in seller's favour; no action", dispute.ID) default: log.Printf("[SQUARE-WEBHOOK] dispute %s state → %s (status %s)", dispute.ID, dispute.State, localStatus) } return nil } // handleDisputeEvidence logs evidence submissions/removals. Evidence does not // change the dispute's local status, so it is informational only. func handleDisputeEvidence(data json.RawMessage) error { var env squareWebhookData if err := json.Unmarshal(data, &env); err != nil { log.Printf("[SQUARE-WEBHOOK] dispute evidence event received (payload length=%d)", len(data)) return nil } var dispute squareDisputePayload if !parseSquareObject(env.Object, "dispute", &dispute) || dispute.ID == "" { log.Printf("[SQUARE-WEBHOOK] dispute evidence event received (data.id=%s)", env.ID) return nil } log.Printf("[SQUARE-WEBHOOK] dispute evidence event for dispute %s (state %s)", dispute.ID, dispute.State) return nil } // handleTerminalCheckout logs terminal checkout lifecycle events. Terminal // checkout state is owned by the poll/sweep handlers (handlers/payments/), // which fetch the authoritative status from Square — no state mutation here. func handleTerminalCheckout(data json.RawMessage) error { var env squareWebhookData if err := json.Unmarshal(data, &env); err != nil { log.Printf("[SQUARE-WEBHOOK] terminal.checkout event received (payload length=%d)", len(data)) return nil } log.Printf("[SQUARE-WEBHOOK] terminal.checkout event received (data.id=%s)", env.ID) return nil }