# Jobs, queues, and integrations

## Processing queue

ASSUMPTION in code: full River client is deferred. Production MVP uses **Postgres processing_jobs** with FOR UPDATE SKIP LOCKED + pg_notify('processing_jobs').

- Enqueue: internal/jobs.Queue.EnqueueProcessingJob
- Workers: internal/processing.JobSlots.Fill → ClaimNext (count from config / ClampProcessingWorkers)
- Process starts also hit HTTP rate limits (RPM per company)

## Worker loop (what runs)

From apps/api/cmd/worker:

1. Fill processing job slots  
2. ProcessPendingAutoJobs (support AI fallback)  
3. WooCommerce / Shopify ClaimNextPendingJob + sync  
4. Periodic: EnqueueDueScheduled (stores), RunDueBillingCycles

API also runs a light RunAutoJobsLoop for support AI — keep **worker** in production.

```mermaid
flowchart TB
  subgraph ingest [Ingest]
    FeedURL[Feed URL / CSV]
    Woo[WooCommerce]
    Shop[Shopify]
  end
  subgraph core [Core]
    Jobs[(processing_jobs)]
    Worker[cmd/worker]
    Catalog[(products)]
  end
  subgraph out [Outbound]
    Export[export feeds CSV/XML]
    StorePush[Woo/Shopify push]
  end
  FeedURL --> Catalog
  Woo --> Catalog
  Shop --> Catalog
  Catalog --> Jobs
  Jobs --> Worker
  Worker --> Catalog
  Catalog --> Export
  Worker --> StorePush
```

## Integrations (where configured)

| Integration | Preferred config | Notes |
|-------------|------------------|-------|
| AI (BYOK / OpenAI-compatible) | Tenant /integrations/ai | Encrypted with APP_ENCRYPTION_KEY; optional process OPENAI_* fallback |
| Marketing email | /integrations/email | Separate from platform invite SMTP |
| Woo / Shopify / feeds | /stores | Woo secrets at rest; Shopify Admin domain SSRF-hardened |
| Stripe / EPREL / feed private-URL allowlist | /admin/settings | Env fallbacks exist; prefer UI |
| Platform invite SMTP | Process SMTP_* | See docs/ops-runtime.md |

Support AI auto-reply jobs: table support_auto_jobs → TryAutoReplyLLM after FAQ miss.

Sources: apps/api/internal/jobs/river.go, apps/api/cmd/worker/main.go, docs/ops-runtime.md.