# analytics — overview

A stateless multi-tenant API proxy exposing four upstream services under one
FastAPI app:

1. **Google Analytics 4** — Data API (reporting) + Admin API (read).
2. **Google Search Console** — Search Analytics, Sites, Sitemaps, URL Inspection (read).
3. **Google Ads** — GAQL reporting (search / searchStream).
4. **Sklik** (Seznam) — Drak JSON-RPC API.

The three Google services share one OAuth mechanism (token wins over service
account) — they differ only in the OAuth *scope* and the header prefix
(`X-GA-*`, `X-GSC-*`, `X-GAds-*`). Google Ads additionally needs a developer
token.

The structure mirrors the sibling `idoklad` / `csob` services (config→env,
credentials→headers, client per upstream, routers, central exception handling,
Swagger at `/docs`), adapted to Python/FastAPI.

## Design principles

- **Stateless / no stored secrets.** Credentials arrive per request in `X-`
  headers and are used only to call the upstream. Nothing is persisted; the
  only in-memory state is a short-lived GA access-token cache (see below).
- **Thin passthrough.** GA request/response bodies and most Sklik calls are
  forwarded as-is, so callers keep the full upstream API surface. Only
  authentication, base URL and error mapping are added.
- **No silent failures.** Every error is logged (never the secret values) and
  surfaced as JSON. Upstream errors preserve the upstream status and body.

## Layout

```
app/
  config.py            env-driven config (base URLs, scopes, timeout) — no secrets
  logging_config.py    get_logger(); secrets are never logged
  errors.py            MissingCredentialsError, UpstreamError + handlers
  credentials.py       X- header dependencies (GA / GSC / Ads / Sklik)
  idempotency.py       in-memory idempotency store for writes (opt-in per request)
  sklik_guards.py      write guard rails: forced paused status, optional budget caps
  clients/
    google.py          shared Google client: Bearer/SA token minting + requests
    sklik_client.py    Sklik JSON-RPC client (login + session + list/report paging)
  routers/
    meta.py            /health, /version
    ga_data.py         /ga/data/...      (Google Analytics Data)
    ga_admin.py        /ga/admin/...     (Google Analytics Admin)
    gsc.py             /gsc/...          (Search Console)
    googleads.py       /googleads/...    (Google Ads)
    sklik.py           /sklik/...
  main.py              app factory, root_path, router + handler registration
```

## Reverse proxy

`ROOT_PATH` (e.g. `/apps/analytics`) is passed to FastAPI's `root_path`, so the
OpenAPI `servers` entry and Swagger "Try it out" use the public prefix. Internal
routes are unprefixed (Caddy `handle_path` strips the prefix).

## Authentication summary

| Upstream | Header(s) | Behaviour |
| --- | --- | --- |
| Google Analytics | `X-GA-Access-Token` **or** `X-GA-Credentials` (+ `X-GA-Quota-Project`) | Token used directly; else minted from base64 service-account JSON (scope `analytics.readonly`) and cached in memory until ~60 s before expiry. |
| Search Console | `X-GSC-Access-Token` **or** `X-GSC-Credentials` (+ `X-GSC-Quota-Project`) | Same as GA, scope `webmasters.readonly`. |
| Google Ads | `X-GAds-Developer-Token` (req) + `X-GAds-Access-Token` **or** `X-GAds-Credentials` (+ `X-GAds-Login-Customer-Id`, `X-GAds-Quota-Project`) | Same OAuth (scope `adwords`) plus `developer-token` / `login-customer-id` headers forwarded upstream. |
| Sklik | `X-Sklik-Token` (+ `X-Sklik-User-Id`) | `client.loginByToken` per request → session injected into the call. |

## Writes

**Sklik is the only upstream with write endpoints** — full CRUD over campaigns,
groups and ads (create / update / remove / restore). The pattern established
there — create paused, optional budget ceilings, optional idempotency key — is
the template for any future write support (Google Ads mutates, Meta campaign
management in the sibling `meta` service).

The one non-negotiable rule: **entities are always created paused**. Sklik
defaults `status` to `active`, so this must be forced server-side. Update does
*not* force a status (that is how you pause or resume a campaign), but an
operator can refuse activation entirely with `SKLIK_BLOCK_ACTIVATION`. Budget
ceilings and idempotency are opt-in, since the approval flow lives on the
caller's side. Removal is Sklik's own soft delete and is reversible.

See `sklik.md` for details.

## Deliberately not wired

- **Write operations** across the Google services: GA Admin (create/update
  properties, streams), Search Console (submit/delete sitemaps, add/remove
  sites), Google Ads mutates (create/update campaigns etc.). All requested
  scopes are read-only; add the read-write scope + endpoints if management is
  needed later. Google Ads exposes reporting (GAQL) only for now.
- **Sklik keyword / sitelink / product-set management.** Only campaigns, groups
  and ads have typed CRUD; the rest of the Sklik API remains reachable through
  `POST /sklik/rpc/{method}` without guard rails.
- **Sklik header-credential encryption.** Same deferral as `idoklad`/`csob`:
  header values are plaintext over TLS for now.
- **Sklik session reuse across requests** — the chosen model logs in per
  request; a future `X-Sklik-Session` passthrough could save the login call.

## Verification checklist (per AGENTS.md)

- `/health` returns 200.
- `/docs` loads; `/openapi.json` `servers` contains the proxy prefix.
- New endpoints appear in Swagger with their `X-` headers in "Try it out".
- Secrets never appear in logs or source.