Files
zeekay e285e18128 feat: bootstrap notifyd service layer on hanzoai/base + hanzoai/tasks
Replaces the chi-based notifyd scaffold with the canonical
hanzoai/auto-style layout: hanzoai/base as the HTTP + SQLite + auth
substrate (via the platform plugin for IAM JWT validation and
X-Org-Id injection), hanzoai/tasks for durable async sends, and a
per-tenant Notifier resolver that pulls provider credentials from
KMS through platform.KMSClient.

Library at the module root stays untouched — all 33 provider impls
(plivo, twilio, ses, sendgrid, mail, whatsapp, discord, slack, …)
keep their existing public API. The service wraps them.

cmd/notifyd
  - base.New + platform.MustRegister
  - schema.MustRegister (6 collections: tenants, providers,
    templates, events, messages, meter)
  - tasks worker registered when TASKS_ADDR is set; sync path
    runs the same Deliver activity directly when not

internal/
  - boot/      thin platform.KMSClient wiring (env-driven)
  - schema/    6 base collections
  - routes/    /v1/notify/{send,messages,providers,templates,events,metering,health}
               + convenience /send/{sms,email,voice,whatsapp}
  - tenant/    KMS-backed resolver builds library Notifier per send
  - template/  draft → pending → approved → published → archived
  - event/     catalog routing (fills missing channel / template)
  - metering/  per-send ledger writer + aggregate query
  - tasks/     NotifySendWorkflow + Deliver activity (idempotent on
               replay; reloads message row before resending)
  - zaprpc/    procedure name constants (server impl follow-up)

pkg/
  - client/    public HTTP SDK (Send / Message / Providers / Templates)
  - types/     wire shapes shared by routes and client

cucloud: fmt.Errorf(msg) → errors.New(msg) for the go 1.26 vet pass
that landed with this dependency bump. Behaviour unchanged.

Boot smoke-tested locally:
  - SQLite migrations apply cleanly
  - /v1/notify/health returns {"service":"notify","status":"ok"}
  - missing X-Org-Id returns 401 on every authenticated route
  - send with stub tenant returns the expected base validation error
2026-06-01 16:06:27 -07:00

55 lines
1.9 KiB
Go

package tasks
import (
"time"
"github.com/hanzoai/tasks/pkg/sdk/workflow"
)
// SendInput is the workflow + activity input. One pair per recipient.
// MessageID is the notifyd-generated message row id the worker updates
// as the send progresses (queued → sending → sent/failed).
type SendInput struct {
MessageID string `json:"message_id"`
TenantSlug string `json:"tenant_slug"`
Channel string `json:"channel"`
Provider string `json:"provider,omitempty"`
To string `json:"to"`
Subject string `json:"subject,omitempty"`
Body string `json:"body"`
TemplateID string `json:"template_id,omitempty"`
Event string `json:"event,omitempty"`
Vars map[string]any `json:"vars,omitempty"`
}
// SendResult is the workflow return shape.
type SendResult struct {
MessageID string `json:"message_id"`
Status string `json:"status"`
Provider string `json:"provider"`
Error string `json:"error,omitempty"`
}
// NotifySendWorkflow is the durable-execution entry point for async
// sends. The workflow body must be deterministic on its inputs — all
// I/O happens inside the Deliver activity.
//
// Single activity: this is intentional. A "send" is a unit of work,
// not a DAG; if a provider fails, retry the whole thing rather than
// trying to split rendering from delivery (rendering against the same
// vars yields the same body, so re-rendering on retry is free).
func NotifySendWorkflow(ctx workflow.Context, in SendInput) (SendResult, error) {
actCtx := workflow.WithActivityOptions(ctx, workflow.ActivityOptions{
StartToCloseTimeout: 60 * time.Second,
})
var res SendResult
if err := workflow.ExecuteActivity(actCtx, "Deliver", in).Get(actCtx, &res); err != nil {
return SendResult{
MessageID: in.MessageID,
Status: "failed",
Error: err.Error(),
}, err
}
return res, nil
}