"""amazon_ads_client.py: the Amazon Ads API transport (S79-T12, contract C6). ⛔ THE CLIENT READS REPORTS AND SENDS ONLY NAMED WRITES. The Ads API has no read-only scope: `advertising::campaign_management` can change any campaign (primer F10). So the refusal lives here, in the request builder, before any credential read, token exchange or socket: reports Reporting v1: create, retrieve, the advertiser account query, and the report files streams create and list subscriptions for the dedicated V1 destination named writes `send_named_write(tenant_key, kind, payload, *, approval_id)`, kind in `bid`, `budget`, `negative_keyword`, `scheduled_change`, each operation one path constant Every other call raises `AdsWriteRefused`. A named write also needs an approval id that the Operator App stored as approved for this very item (`APPROVAL_SOURCE`, read lazily, because a harness module never imports `modules` at load). A missing approval source refuses. `AMAZON_ADS_FIXTURE=1` routes every call to the recording fixture below. No live Ads call is made until Royal's Direct Advertiser credential arrives (R7, R17). NA host `advertising-api.amazon.com`, one profile id per marketplace (F13). The profile id is stored with the credential. v1 names the account by `advertiserAccountId`, which the client looks up once per connection from `query/advertiserAccounts`. Every path, header and field comes from Amazon's saved contracts in `AutoWiki/research/amazon-ads-v1-schema/`. """ import gzip import importlib import json import logging import os import re import threading import time import urllib.error import urllib.parse import urllib.request from contextlib import contextmanager from datetime import date, datetime, timedelta, timezone from typing import NamedTuple from zoneinfo import ZoneInfo, ZoneInfoNotFoundError KEY = "amazon_ads" LABEL = "Amazon Ads" ENTRY_TYPE = "amazon_ads" FIXTURE_ENV = "AMAZON_ADS_FIXTURE" _log = logging.getLogger(__name__) HOSTS = {"na": "https://advertising-api.amazon.com", "eu": "https://advertising-api-eu.amazon.com", "fe": "https://advertising-api-fe.amazon.com"} DEFAULT_REGION = "na" LWA_TOKEN_URL = "https://api.amazon.com/auth/o2/token" CREDENTIAL_FIELDS = ("client_id", "client_secret", "refresh_token", "profile_id") #: Reporting v1 limits (limits.md, dates.md): one report in flight per client ID, about 100 a day, #: up to 3 hours per report, at most 120 days per request. REPORT_CONCURRENCY = 1 REPORTS_PER_DAY = 100 REPORT_MAX_WAIT_SECONDS = 3 * 3600 MAX_DAYS_PER_REQUEST = 120 REPORT_POLL_SECONDS = 60.0 #: Amazon's error text reaches the sync report verbatim, up to this many characters. ERROR_TEXT_LIMIT = 600 _TIMEOUT = 60 #: The only non-write calls: the Reporting v1 operations, each a POST to one literal path from the #: Reports and AdvertiserAccounts contracts. Every other `adsApi` path refuses, a GET included. REPORT_OPERATIONS = { "createReport": ("POST", "adsApi/v1/create/reports"), "retrieveReport": ("POST", "adsApi/v1/retrieve/reports"), "queryAdvertiserAccounts": ("POST", "adsApi/v1/query/advertiserAccounts"), } #: The contract names `Amazon-Ads-ClientId`. The limits page names the older header. v1 sends both. V1_CLIENT_HEADERS = ("Amazon-Ads-ClientId", "Amazon-Advertising-API-ClientId") #: A report file is a presigned S3 URL: https, on this host suffix, GET with no Ads header. REPORT_FILE_HOST_SUFFIX = ".amazonaws.com" # Subscription management never admits an advertising write. STREAM_OPERATIONS = { "createStreamSubscription": ("POST", "streams/subscriptions"), "listStreamSubscriptions": ("GET", "streams/subscriptions"), } STREAM_CONTENT_TYPE = "application/vnd.MarketingStreamSubscriptions.StreamSubscriptionResource.v1.0+json" def stream_operation(method, path): """Resolve only the two V1 subscription operations.""" return next((name for name, (verb, endpoint) in STREAM_OPERATIONS.items() if str(method).upper() == verb and str(path).strip("/") == endpoint), None) def _stream_payload(operation, body, query): """Reject unknown subscription fields before reading credentials.""" if operation == "createStreamSubscription": if query or not isinstance(body, dict) or set(body) - { "dataSetId", "destinationArn", "clientRequestToken", "notes"}: raise AdsWriteRefused("invalid stream subscription fields") if body.get("dataSetId") not in {"sp-traffic", "sp-conversion"}: raise AdsWriteRefused("the stream dataset is not allowed") if not re.fullmatch(r"arn:aws:sqs:us-east-1:023371594014:aios-amazon-marketing-stream", str(body.get("destinationArn") or "")): raise AdsWriteRefused("the stream destination is not the dedicated AIOS queue") if not re.fullmatch(r"[A-Za-z0-9_-]{1,128}", str(body.get("clientRequestToken") or "")): raise AdsWriteRefused("a stream create needs a stable retry token") if "notes" in body and (not isinstance(body["notes"], str) or len(body["notes"]) > 256): raise AdsWriteRefused("invalid stream notes") elif operation == "listStreamSubscriptions": if body is not None or set(query or {}) - {"startingToken", "maxResults"}: raise AdsWriteRefused("invalid stream listing fields") if "startingToken" in (query or {}) and not isinstance(query["startingToken"], str): raise AdsWriteRefused("invalid stream continuation token") if "maxResults" in (query or {}) and (type(query["maxResults"]) is not int or query["maxResults"] <= 0): raise AdsWriteRefused("invalid stream page size") else: raise AdsWriteRefused("the stream operation is not allowed") #: ⛔ C6: THE NAMED WRITES, AS DATA. `kind -> operation -> rule`. `endpoints` names the one path #: per ad type; an ad type absent there refuses. `item` is the body template: a "{field}" value #: is read from the validated payload, any other value is a constant. `each` repeats the item per #: id in that payload list. The body carries only template keys, never a caller's other fields. NAMED_WRITES = { "bid": { "set_bid": { "endpoints": {"sp": ("PUT", "sp/targets"), "sb": ("PUT", "sb/targets"), "sd": ("PUT", "sd/targets")}, "required": ("campaign_id", "ad_group_id", "target_id", "bid"), "collection": "targets", "item": {"targetId": "{target_id}", "adGroupId": "{ad_group_id}", "campaignId": "{campaign_id}", "bid": "{bid}"}, }, # QI3 "pause ads on a money loser": an ad state change, carried under the bid kind. "pause_ads": { "endpoints": {"sp": ("PUT", "sp/productAds"), "sb": ("PUT", "sb/v4/ads"), "sd": ("PUT", "sd/productAds")}, "required": ("campaign_id", "ad_group_id", "ad_ids", "state"), "collection": "ads", "each": "ad_ids", "state": "paused", "item": {"adId": "{each}", "adGroupId": "{ad_group_id}", "campaignId": "{campaign_id}", "state": "PAUSED"}, }, "enable_ads": { "endpoints": {"sp": ("PUT", "sp/productAds"), "sb": ("PUT", "sb/v4/ads"), "sd": ("PUT", "sd/productAds")}, "required": ("campaign_id", "ad_group_id", "ad_ids", "state"), "collection": "ads", "each": "ad_ids", "state": "enabled", "item": {"adId": "{each}", "adGroupId": "{ad_group_id}", "campaignId": "{campaign_id}", "state": "ENABLED"}, }, }, "budget": { "set_daily_budget": { "endpoints": {"sp": ("PUT", "sp/campaigns"), "sb": ("PUT", "sb/v4/campaigns"), "sd": ("PUT", "sd/campaigns")}, "required": ("campaign_id", "daily_budget"), "collection": "campaigns", "item": {"campaignId": "{campaign_id}", "budget": {"budget": "{daily_budget}", "budgetType": "DAILY"}}, }, }, "negative_keyword": { # Sponsored Display takes no negative keywords, so it has no endpoint and refuses. "add_negative_keyword": { "endpoints": {"sp": ("POST", "sp/negativeKeywords"), "sb": ("POST", "sb/negativeKeywords")}, "required": ("campaign_id", "ad_group_id", "keyword_text", "match_type"), "collection": "negativeKeywords", "item": {"campaignId": "{campaign_id}", "adGroupId": "{ad_group_id}", "keywordText": "{keyword_text}", "matchType": "{match_type}", "state": "ENABLED"}, }, }, } NAMED_WRITES["scheduled_change"] = { operation: NAMED_WRITES["bid"][operation] for operation in ("set_bid", "pause_ads", "enable_ads") } #: Payload field rules for the named writes. A tuple is the allowed values. FIELD_RULES = { "campaign_id": "id", "ad_group_id": "id", "target_id": "id", "ad_ids": "ids", "bid": "money", "daily_budget": "money", "keyword_text": "text", "match_type": ("negativeExact", "negativePhrase"), "state": ("paused", "enabled"), } #: C6: `fn(tenant_key, approval_id) -> record or None`, imported lazily. The record must be #: approved and name the payload's `item_id`. APPROVAL_SOURCE = "modules.amazon_operator:approval" #: schedule_approval(tenant_key, schedule_id) returns a fresh stored record, or None when the App is off. #: Required: id, approved_by, approved_at (aware ISO timestamp), enabled (bool), campaign_id, ad_type, #: timezone (IANA), weekday_convention="monday_zero" (Monday=0, Sunday=6), rules (list). #: Each rule binds weekday (int 0..6), hour (int 0..23), action (operation name), value, #: ad_group_id, and target_id for set_bid or exact ad_ids for pause_ads/enable_ads. #: value is the bid for set_bid, "paused" for pause_ads, or "enabled" for enable_ads. #: Optional profile_id binds the profile on the record and payload. Missing means the default profile. #: Payload adds schedule_id, item_id, timezone, weekday, hour to the operation's usual fields. #: approval_id equals schedule_id. The client checks its own clock in the stored timezone. #: The reader validates entity membership from trusted observations and invalidates approval after edits. #: The runner persists receipts and claims each local date/rule/hour once, including repeated DST hours. SCHEDULE_APPROVAL_SOURCE = "modules.amazon_operator:schedule_approval" SCHEDULE_WEEKDAY_CONVENTION = "monday_zero" #: sent_write(tenant_key, item_id) returns a stored proposal with id, kind, status, sent_payload, result. #: enable_ads under bid requires an exact approved payload with undo_of naming that sent pause. #: sent_payload must record previous_state="enabled" for every ad, plus the original pause fields. #: result is the original send_named_write receipt. Historical pauses without prior state refuse. SENT_WRITE_SOURCE = "modules.amazon_operator:sent_write" class AdsError(RuntimeError): """An Ads API failure, safe to show: no token or secret ever reaches it.""" class AdsWriteRefused(AdsError): """A call this connector does not make. Raised before any network I/O.""" def __init__(self, cause): self.cause = str(cause) super().__init__(f"Refused: {self.cause}. This connector reads reports and sends only " f"the named writes: bid, budget, negative keyword and scheduled change.") class AdsRejected(AdsError): """Amazon answered an error. `errors` keeps each entry's code, message and fieldLocation.""" def __init__(self, message, errors=()): super().__init__(message) self.errors = list(errors) class AdsReportUnavailable(AdsRejected): """Amazon does not offer this report to this advertiser.""" class AdsReportPending(AdsError): """A report did not finish inside the wait limit. `report_id` names it so a later pass can resume it.""" def __init__(self, message, report_id=None): super().__init__(message) self.report_id = report_id def fixture_mode(): """True when `AMAZON_ADS_FIXTURE=1` routes every call to the recording fixture.""" return os.environ.get(FIXTURE_ENV) == "1" def report_operation(method, path): """The Reporting v1 operation name for one call, or None. A path matches only its literal.""" return next((name for name, (verb, endpoint) in REPORT_OPERATIONS.items() if str(method).upper() == verb and str(path or "").strip("/") == endpoint), None) class _Prepared: """A request the builder accepted. Only `build_request` and `_build_named` make one.""" __slots__ = ("method", "path", "body", "query", "operation", "named", "_seal") def __init__(self, method, path, body, query, operation, named): self.method, self.path, self.body, self.query = method, path, body, query self.operation, self.named = operation, named self._seal = _SEAL _SEAL = object() def build_request(method, path, *, body=None, query=None): """Accept a report or scoped stream subscription call before any I/O.""" operation = report_operation(method, path) or stream_operation(method, path) if operation is None: raise AdsWriteRefused(f"{str(method).upper()} {_safe_path(path)} is not a report call") if operation in STREAM_OPERATIONS: _stream_payload(operation, body, query) return _Prepared(str(method).upper(), str(path).strip("/"), body, dict(query or {}), operation, None) def _safe_path(path): return re.sub(r"[^A-Za-z0-9/_{}.-]", "", str(path or ""))[:120] or "an empty path" def _valid(field, value): rule = FIELD_RULES.get(field) if rule == "id": return isinstance(value, (str, int)) and not isinstance(value, bool) \ and bool(re.fullmatch(r"[0-9]{1,24}", str(value))) if rule == "ids": return isinstance(value, (list, tuple)) and bool(value) and all(_valid("campaign_id", v) for v in value) if rule == "money": try: number = float(value) except (TypeError, ValueError): return False return not isinstance(value, bool) and number > 0 and number == number and number != float("inf") if rule == "text": return isinstance(value, str) and 0 < len(value.strip()) <= 80 if isinstance(rule, tuple): return value in rule return False def _render(value, payload, each=None): if isinstance(value, dict): return {k: _render(v, payload, each) for k, v in value.items()} if isinstance(value, str) and value.startswith("{") and value.endswith("}"): field = value[1:-1] raw = each if field == "each" else payload[field] if FIELD_RULES.get(field) == "money": return round(float(raw), 2) return str(raw) if field == "each" or FIELD_RULES.get(field) == "id" else raw return value def _build_named(kind, payload): """Accept one of C6's named writes, or raise `AdsWriteRefused` before any I/O.""" if kind not in NAMED_WRITES: raise AdsWriteRefused(f"write kind {str(kind)[:40]!r} is not a named write") if not isinstance(payload, dict): raise AdsWriteRefused("the write payload is not an object") operation = str(payload.get("operation") or "") rule = NAMED_WRITES[kind].get(operation) if rule is None: raise AdsWriteRefused(f"operation {operation[:40]!r} is not named under {kind}") ad_type = str(payload.get("ad_type") or "") endpoint = rule["endpoints"].get(ad_type) if endpoint is None: raise AdsWriteRefused(f"{operation} is not named for ad type {ad_type[:8]!r}") for field in rule["required"]: if not _valid(field, payload.get(field)): raise AdsWriteRefused(f"{operation} needs a valid {field}") if "state" in rule and payload.get("state") != rule["state"]: raise AdsWriteRefused(f"{operation} needs state {rule['state']}") each = rule.get("each") items = ([_render(rule["item"], payload, v) for v in payload[each]] if each else [_render(rule["item"], payload)]) method, path = endpoint return _Prepared(method, path, {rule["collection"]: items}, {}, operation, {"kind": kind, "operation": operation, "ad_type": ad_type}) def _read_record(source, tenant_key, record_id, label): """Read one tenant-bound approval or outcome through its lazy source.""" module_name, _, name = source.partition(":") try: reader = getattr(importlib.import_module(module_name), name) except Exception: # noqa: BLE001 raise AdsWriteRefused(f"the {label} source is not installed") from None try: return reader(tenant_key, record_id) except Exception: # noqa: BLE001 raise AdsWriteRefused(f"the {label} could not be read") from None def _utcnow(): return datetime.now(timezone.utc) def _approved_schedule(tenant_key, approval_id, payload): """Refuse unless the stored schedule permits these entities and this action at the current hour.""" if payload.get("schedule_id") != approval_id: raise AdsWriteRefused("the schedule id must be its approval id") record = _read_record(SCHEDULE_APPROVAL_SOURCE, tenant_key, approval_id, "schedule approval") if not isinstance(record, dict) or record.get("id") != approval_id: raise AdsWriteRefused("the approved schedule does not exist") if record.get("enabled") is not True: raise AdsWriteRefused("the schedule is switched off") if not isinstance(record.get("approved_by"), str) or not record["approved_by"].strip(): raise AdsWriteRefused("the schedule needs its approver") now = _utcnow() try: approved_at = datetime.fromisoformat(record["approved_at"].replace("Z", "+00:00")) if approved_at.utcoffset() is None or approved_at > now: raise ValueError("invalid approval time") except (KeyError, AttributeError, TypeError, ValueError): raise AdsWriteRefused("the schedule needs a valid approval time") from None if record.get("decision", "approved") != "approved": raise AdsWriteRefused("the schedule is not approved") if record.get("weekday_convention") != SCHEDULE_WEEKDAY_CONVENTION: raise AdsWriteRefused("the schedule must use Monday-zero weekdays") try: local = now.astimezone(ZoneInfo(record["timezone"])) except (KeyError, TypeError, ValueError, ZoneInfoNotFoundError): raise AdsWriteRefused("the schedule needs an IANA timezone") from None if (type(payload.get("weekday")) is not int or type(payload.get("hour")) is not int or payload["weekday"] != local.weekday() or payload["hour"] != local.hour or payload.get("timezone") != record["timezone"]): raise AdsWriteRefused("the schedule is not due in its stored timezone") if any(record.get(field) != payload.get(field) for field in ("campaign_id", "ad_type", "profile_id")): raise AdsWriteRefused("the schedule does not name this campaign and profile") operation = payload["operation"] value = payload["bid"] if operation == "set_bid" else payload["state"] entity_fields = ("ad_group_id", "target_id") if operation == "set_bid" else ("ad_group_id", "ad_ids") rules = record.get("rules") if not isinstance(rules, list) or not any( isinstance(rule, dict) and type(rule.get("weekday")) is int and type(rule.get("hour")) is int and rule.get("weekday") == payload["weekday"] and rule.get("hour") == payload["hour"] and rule.get("action") == operation and not isinstance(rule.get("value"), bool) and rule.get("value") == value and all(rule.get(field) == payload.get(field) for field in entity_fields) for rule in rules): raise AdsWriteRefused("the schedule rule does not approve this action and these entities") return record def _approved_pause_undo(tenant_key, record, payload): """Require an exact approved Undo of a sent pause whose recorded prior state was enabled.""" if not isinstance(record.get("writes"), list): raise AdsWriteRefused("an enable Undo needs its exact approved write") original_id = payload.get("undo_of") if not isinstance(original_id, str) or not original_id or original_id == payload["item_id"]: raise AdsWriteRefused("an enable Undo must name the original sent pause") original = _read_record(SENT_WRITE_SOURCE, tenant_key, original_id, "sent pause") if (not isinstance(original, dict) or original.get("id") != original_id or original.get("status") != "sent" or original.get("kind") not in ("bid", "scheduled_change")): raise AdsWriteRefused("the original pause was not sent") sent = original.get("sent_payload") result = original.get("result") if (not isinstance(sent, dict) or sent.get("item_id") != original_id or sent.get("operation") != "pause_ads" or sent.get("state") != "paused" or not isinstance(result, dict) or result.get("sent") is not True or result.get("operation") != "pause_ads" or result.get("item_id") != original_id or result.get("kind") != original["kind"] or not result.get("approval_id")): raise AdsWriteRefused("the original pause needs its sent receipt") if sent.get("previous_state") != "enabled": raise AdsWriteRefused("the original pause has no recorded enabled prior state") fields = ("campaign_id", "ad_type", "profile_id", "ad_group_id", "ad_ids") if any(sent.get(field) != payload.get(field) for field in fields): raise AdsWriteRefused("the enable Undo does not restore the original paused entities") def _approved(tenant_key, approval_id, kind, payload): """Refuse unless the Operator App stored this approval for this item and this payload.""" if not isinstance(approval_id, str) or not approval_id.strip(): raise AdsWriteRefused("a named write needs an approval id") item_id = payload.get("item_id") if not isinstance(item_id, str) or not item_id: raise AdsWriteRefused("a named write needs the item id it was approved for") if kind == "scheduled_change": return _approved_schedule(tenant_key, approval_id, payload) module_name, _, name = APPROVAL_SOURCE.partition(":") try: reader = getattr(importlib.import_module(module_name), name) except Exception: # noqa: BLE001 raise AdsWriteRefused("the approval source is not installed") from None try: record = reader(tenant_key, approval_id) except Exception: # noqa: BLE001 raise AdsWriteRefused("the approval could not be read") from None if not isinstance(record, dict) or record.get("decision") != "approved": raise AdsWriteRefused("the approval id is not an approved decision") if item_id not in (record.get("item_ids") or ()): raise AdsWriteRefused("the approval does not name this item") writes = record.get("writes") if writes is not None: stored = [w for w in writes if isinstance(w, dict) and w.get("item_id") == item_id] if not stored or stored[0].get("kind") != kind or stored[0].get("payload") != payload: raise AdsWriteRefused("the payload is not the one that was approved") if payload.get("operation") == "enable_ads": _approved_pause_undo(tenant_key, record, payload) return record def send_named_write(tenant_key, kind, payload, *, approval_id, rt=None): """Send one approved named write and return its request, approver and reply.""" prepared = _build_named(kind, payload) key = str(tenant_key or "").strip().lower() if not key: raise AdsWriteRefused("a named write needs a tenant") approval = _approved(key, approval_id, kind, payload) approver = approval.get("approved_by") if kind == "scheduled_change" else approval.get("by") _log.info("Amazon Ads write approved tenant=%s kind=%s operation=%s approval=%s approver=%s item=%s", key, kind, prepared.operation, approval_id, approver, payload["item_id"]) reply = _send(prepared, key, rt=rt, profile_id=payload.get("profile_id")) return {"sent": True, "kind": kind, "operation": prepared.operation, "ad_type": prepared.named["ad_type"], "method": prepared.method, "path": prepared.path, "approval_id": approval_id, "item_id": payload["item_id"], "approved_by": approver, "approved_at": approval.get("approved_at", approval.get("at")), "sent_at": _utcnow().isoformat(), "response": reply, **({field: payload[field] for field in ("schedule_id", "timezone", "weekday", "hour")} if kind == "scheduled_change" else {})} def request(tenant_key, method, path, *, body=None, query=None, rt=None, profile_id=None, connection_id=None): """A report call by method and path. Anything else raises `AdsWriteRefused` before any I/O.""" return _send(build_request(method, path, body=body, query=query), str(tenant_key or "").strip().lower(), rt=rt, profile_id=profile_id, connection_id=connection_id) def _admitted(prepared): """Re-check at the transport door, so no hand-made object reaches a socket.""" if not isinstance(prepared, _Prepared) or prepared._seal is not _SEAL: return False if prepared.named is None: if prepared.operation is not None and report_operation(prepared.method, prepared.path) == prepared.operation: return True if stream_operation(prepared.method, prepared.path) != prepared.operation: return False try: _stream_payload(prepared.operation, prepared.body, prepared.query) except AdsWriteRefused: return False return True rule = NAMED_WRITES.get(prepared.named["kind"], {}).get(prepared.named["operation"]) return bool(rule) and rule["endpoints"].get(prepared.named["ad_type"]) == (prepared.method, prepared.path) def _send(prepared, tenant_key, *, rt=None, profile_id=None, connection_id=None): if not _admitted(prepared): raise AdsWriteRefused("the request did not come from the request builder") if fixture_mode(): headers = _headers(prepared, FIXTURE.credential(tenant_key), FIXTURE_TOKEN, profile_id) return FIXTURE.answer(prepared, tenant_key, profile_id, headers) return _live(prepared, tenant_key, rt=rt, profile_id=profile_id, connection_id=connection_id) # --------------------------------------------------------------------------- live transport _NO_CREDENTIAL = ("No Amazon Ads credential is stored for this tenant. " "Connect Amazon Ads under Amazon Ads Setup.") def credential(tenant_key, rt=None, connection_id=None): """The tenant's stored Amazon Ads credential fields, or None. The keychain is the only source.""" try: import core.keychain as kc bound = getattr(rt, "key", None) or getattr(getattr(rt, "tenant", None), "key", None) if bound != tenant_key: return None fields = kc.amazon_ads_creds(rt, connection_id=connection_id) except Exception: # noqa: BLE001 return None if not isinstance(fields, dict) or not all(fields.get(f) for f in CREDENTIAL_FIELDS): return None return fields def _token(fields): data = urllib.parse.urlencode({"grant_type": "refresh_token", "refresh_token": fields["refresh_token"], "client_id": fields["client_id"], "client_secret": fields["client_secret"]}).encode() req = urllib.request.Request(LWA_TOKEN_URL, data=data, method="POST", headers={"Content-Type": "application/x-www-form-urlencoded"}) try: with urllib.request.urlopen(req, timeout=_TIMEOUT) as r: return json.loads(r.read().decode("utf-8", "replace"))["access_token"] except Exception as exc: # noqa: BLE001 raise AdsError(f"Amazon Ads sign in failed ({type(exc).__name__})") from None def _headers(prepared, fields, token, profile_id=None): """The headers one admitted call sends. A v1 call names its account in the body, not a scope.""" if prepared.operation in REPORT_OPERATIONS: return {"Authorization": f"Bearer {token}", **dict.fromkeys(V1_CLIENT_HEADERS, fields["client_id"]), "Content-Type": "application/json", "Accept": "application/json"} headers = {"Authorization": f"Bearer {token}", "Amazon-Advertising-API-ClientId": fields["client_id"], "Amazon-Advertising-API-Scope": str(profile_id or fields["profile_id"]), "Content-Type": "application/json", "Accept": "application/json"} if prepared.operation in STREAM_OPERATIONS: headers["Content-Type"] = STREAM_CONTENT_TYPE headers["Accept"] = STREAM_CONTENT_TYPE return headers def _errors(detail): """Amazon's error entries from an error body, each with its code, message and fieldLocation.""" if not isinstance(detail, dict): return [] entries = detail["errors"] if isinstance(detail.get("errors"), list) else [detail] return [{"code": str(e.get("code") or ""), "message": str(e.get("details") or e.get("message") or ""), "fieldLocation": str(e.get("fieldLocation") or "")} for e in entries if isinstance(e, dict) and (e.get("code") or e.get("details") or e.get("message"))] def _error_text(errors): """Amazon's errors as one line a person reads, verbatim, cut to `ERROR_TEXT_LIMIT`.""" return " ".join(" ".join(filter(None, (e["code"] + ":" if e["code"] else "", e["message"], f"(field {e['fieldLocation']})" if e["fieldLocation"] else ""))) for e in errors)[:ERROR_TEXT_LIMIT] def _rejection(prefix, errors, *, may_be_unoffered=True): """The exception for Amazon's errors: not offered when Amazon says so, else rejected.""" text = _error_text(errors) if may_be_unoffered and re.search(r"not (?:supported|offered|available)", text, re.I): return AdsReportUnavailable(f"Amazon does not offer this report: {text}", errors) return AdsRejected(f"{prefix}: {text}" if text else prefix, errors) def _live(prepared, tenant_key, *, rt=None, profile_id=None, connection_id=None): fields = credential(tenant_key, rt, connection_id=connection_id) if fields is None: raise AdsError(_NO_CREDENTIAL) if prepared.operation in STREAM_OPERATIONS and ( str(profile_id or fields["profile_id"]) != str(fields["profile_id"]) or str(fields.get("region") or DEFAULT_REGION).lower() != "na"): raise AdsWriteRefused("the stream scope must match the stored NA business profile") host = HOSTS.get(str(fields.get("region") or DEFAULT_REGION).lower(), HOSTS[DEFAULT_REGION]) url = f"{host}/{prepared.path}" if prepared.query: url += "?" + urllib.parse.urlencode(prepared.query) headers = _headers(prepared, fields, _token(fields), profile_id) data = json.dumps(prepared.body).encode() if prepared.body is not None else None req = urllib.request.Request(url, data=data, method=prepared.method, headers=headers) try: with urllib.request.urlopen(req, timeout=_TIMEOUT) as r: return json.loads(r.read().decode("utf-8", "replace") or "{}") except urllib.error.HTTPError as exc: try: detail = json.loads(exc.read().decode("utf-8", "replace") or "{}") except Exception: # noqa: BLE001 an unreadable body still raises its status detail = {} raise _rejection(f"Amazon Ads answered HTTP {exc.code} on {prepared.operation}", _errors(detail), may_be_unoffered=exc.code == 400) from None except Exception as exc: # noqa: BLE001 raise AdsError(f"Amazon Ads could not be reached ({type(exc).__name__})") from None # --------------------------------------------------------------------------- reports _REPORT_LOCKS = {} _REPORT_LOCKS_GUARD = threading.Lock() #: Process caches: the advertiser account per connection, the optional fields Amazon refused per #: (client id, group) with Amazon's text, and the report file URLs a retrieve returned. _ACCOUNTS = {} _REFUSED = {} _ISSUED_FILES = set() class GroupReport(NamedTuple): """One finished group report: its id, flat rows per member table, and notes a person reads.""" report_id: str rows: dict notes: tuple def _report_lock(client_id): """One lock per client ID: Amazon runs one report at a time for each app login.""" with _REPORT_LOCKS_GUARD: return _REPORT_LOCKS.setdefault(str(client_id), threading.Lock()) def _report_credential(tenant_key, rt, connection_id): """The credential a report runs under: its client id keys the lock, its profile names the account.""" if fixture_mode(): return FIXTURE.credential(tenant_key) fields = credential(tenant_key, rt, connection_id=connection_id) if fields is None: raise AdsError(_NO_CREDENTIAL) return fields def _lists_profile(account, profile_id): return (isinstance(account, dict) and bool(account.get("advertiserAccountId")) and any(isinstance(alt, dict) and str(alt.get("profileId")) == str(profile_id) for alt in account.get("alternateIds") or ())) def _advertiser_account(tenant_key, *, rt, profile_id, connection_id): """The v1 `advertiserAccountId` whose alternate ids carry the stored profile id, cached per connection.""" key, token = (tenant_key, connection_id or "", str(profile_id)), None while key not in _ACCOUNTS: page = request(tenant_key, "POST", REPORT_OPERATIONS["queryAdvertiserAccounts"][1], body={"maxResults": 100, **({"nextToken": token} if token else {})}, rt=rt, profile_id=profile_id, connection_id=connection_id) match = next((a for a in page.get("advertiserAccounts") or () if _lists_profile(a, profile_id)), None) token = page.get("nextToken") if match is not None: _ACCOUNTS[key] = str(match["advertiserAccountId"]) elif not token: raise AdsError(f"No Amazon Ads advertiser account lists profile {profile_id}. " f"Check the profile ID on this Amazon Ads connection.") return _ACCOUNTS[key] def _create_body(group, start, end, account, fields): """A CreateReportRequest for one group and window, filtered to the group's ad products.""" from harness import amazon_ads_schema as schema return {"accessRequestedAccounts": [{"advertiserAccountId": account}], "reports": [{"format": "GZIP_JSON", "periods": [{"datePeriod": {"startDate": start.isoformat(), "endDate": end.isoformat()}}], "query": {"fields": list(fields), "filter": {"on": {"field": schema.AD_PRODUCT_FIELD, "comparisonOperator": "IN", "not": False, "values": list(group.tables)}}}}]} def _first_report(reply, operation): """The one report a 207 multi-status answer carries, or its errors raised verbatim.""" errors = [e for entry in reply.get("error") or () if isinstance(entry, dict) for e in _errors({"errors": entry.get("errors") or []})] if errors or reply.get("error"): raise _rejection(f"Amazon Ads refused {operation}", errors) for entry in reply.get("success") or (): report = entry.get("report") if isinstance(entry, dict) else None if isinstance(report, dict) and re.fullmatch(r"[A-Za-z0-9_.:-]{1,128}", str(report.get("reportId") or "")): return report raise AdsError(f"Amazon Ads returned no report on {operation}") def _named_fields(exc, fields): """The requested fields an error names, by id in its text or by index in its field location.""" text = " ".join([str(exc)] + [e["fieldLocation"] for e in exc.errors]) named = {f for f in fields if re.search(rf"(? its report file URLs.""" started = now() while True: report = _first_report(request(tenant_key, "POST", REPORT_OPERATIONS["retrieveReport"][1], body={"reportIds": [report_id]}, **opts), "retrieveReport") state = str(report.get("status") or "").upper() if state == "COMPLETED": urls = [str(part.get("url") or "") for part in report.get("completedReportParts") or () if isinstance(part, dict)] _ISSUED_FILES.update(urls) return urls if state == "FAILED": reason = " ".join(filter(None, (str(report.get("failureCode") or ""), str(report.get("failureReason") or ""))))[:ERROR_TEXT_LIMIT] or state if re.search(r"not (?:supported|offered|available)", reason, re.I): raise AdsReportUnavailable(f"Amazon does not offer this report: {reason}") raise AdsError(f"Amazon Ads report {report_id} ended FAILED: {reason}") elapsed = now() - started if elapsed > REPORT_MAX_WAIT_SECONDS: raise AdsReportPending(f"Amazon Ads report {report_id} did not finish in 3 hours", report_id) if limit is not None and elapsed >= limit: raise AdsReportPending(f"Amazon Ads report {report_id} is still being prepared", report_id) pause = 0 if fixture_mode() else REPORT_POLL_SECONDS sleep(pause if limit is None else min(pause, limit - elapsed)) def _flat(row, prefix=""): """A nested v1 row -> flat dotted field ids. A flat row stays as it is.""" out = {} for key, value in row.items(): if isinstance(value, dict): out.update(_flat(value, f"{prefix}{key}.")) else: out[f"{prefix}{key}"] = value return out def _part_rows(raw): """One report file -> flat rows. GZIP or plain, a JSON array or JSON lines.""" try: text = (gzip.decompress(raw) if raw[:2] == b"\x1f\x8b" else raw).decode("utf-8", "replace").strip() if not text: return [] try: data = json.loads(text) except ValueError: data = [json.loads(line) for line in text.splitlines() if line.strip()] except (OSError, EOFError, ValueError) as exc: raise AdsError(f"Amazon Ads sent a report file Loopable cannot read ({type(exc).__name__})") from None return [_flat(row) for row in (data if isinstance(data, list) else [data]) if isinstance(row, dict)] def _read_file(url): """GET one report file with no Ads header. Only an https S3 URL a retrieve returned is admitted.""" parts = urllib.parse.urlsplit(str(url or "")) host = parts.hostname or "" if parts.scheme != "https" or not host.endswith(REPORT_FILE_HOST_SUFFIX) or url not in _ISSUED_FILES: raise AdsWriteRefused(f"{parts.scheme or 'no scheme'} {_safe_path(host)} is not a report file Amazon returned") _ISSUED_FILES.discard(url) headers = {} if fixture_mode(): return _part_rows(FIXTURE.download(url, headers)) try: with urllib.request.urlopen(urllib.request.Request(url, method="GET", headers=headers), timeout=_TIMEOUT) as r: raw = r.read() except Exception as exc: # noqa: BLE001 the URL carries a signature: never echo it raise AdsError(f"Amazon Ads report file could not be read ({type(exc).__name__})") from None return _part_rows(raw) def _split(group, rows): """Flat rows -> {member table: rows} by `adProduct.value`, and a note for rows no member takes.""" from harness import amazon_ads_schema as schema out, tables, stray = {table: [] for table in group.members}, group.tables, 0 for row in rows: table = tables.get(str(row.get(schema.AD_PRODUCT_FIELD) or "").upper()) if table is None or not row.get(schema.vendor_name("date")): stray += 1 else: out[table].append(row) note = (f"Amazon Ads sent {stray} rows with no date or with another ad product. Loopable left them out.",) return out, note if stray else () def run_group_report(tenant_key, group, start, end, *, rt=None, profile_id=None, connection_id=None, sleep=time.sleep, now=time.monotonic, max_wait=None, report_id=None): """Create one Reporting v1 group report, wait for it, and read every file. -> GroupReport. `max_wait` caps this call's poll loop in seconds. `report_id` resumes that report without a create. """ from harness import amazon_ads_schema as schema group = schema.GROUPS[group] if isinstance(group, str) else group if (end - start).days + 1 > MAX_DAYS_PER_REQUEST: raise ValueError(f"a report window is at most {MAX_DAYS_PER_REQUEST} days") fields = _report_credential(tenant_key, rt, connection_id) opts = {"rt": rt, "profile_id": str(profile_id or fields["profile_id"]), "connection_id": connection_id} refused = _REFUSED.setdefault((str(fields["client_id"]), group.name), {}) limit = None if max_wait is None else max(0.0, min(float(max_wait), REPORT_MAX_WAIT_SECONDS)) with _report_lock(fields["client_id"]): if report_id is None: account = _advertiser_account(tenant_key, **opts) report_id = _create(tenant_key, group, start, end, account, refused, opts)["reportId"] urls = _await(tenant_key, str(report_id), opts, sleep=sleep, now=now, limit=limit) rows = [row for url in urls for row in _read_file(url)] split, stray = _split(group, rows) notes = tuple(f"Amazon Ads refused the field {field} on this report, so Loopable reads it without that " f"field. Amazon said: {text}" for field, text in refused.items()) return GroupReport(str(report_id), split, notes + stray) def run_report(tenant_key, table, start, end, **options): """One table's rows from its group report. -> (report_id, rows). The store reads `run_group_report`.""" from harness import amazon_ads_schema as schema group = schema.group_of(table) if group is None: raise AdsReportUnavailable(f"Amazon Reporting v1 offers no report for {schema.short_name(table)}") result = run_group_report(tenant_key, group, start, end, **options) return result.report_id, result.rows[schema.short_name(table)] # --------------------------------------------------------------------------- recording fixture FIXTURE_TOKEN = "fixture-access-token" FIXTURE_CREDENTIAL = {"client_id": "amzn1.application-oa2-client.fixture", "client_secret": "fixture", "refresh_token": "fixture", "profile_id": "1000000001", "region": "na"} FIXTURE_FILE_HOST = "https://fixture-ads-reports.s3.amazonaws.com" FIXTURE_CREATED = "2026-10-20T09:00:00Z" _FIXTURE_PREFIX = {"SPONSORED_PRODUCTS": ("71", "SP"), "SPONSORED_BRANDS": ("72", "SB"), "SPONSORED_DISPLAY": ("73", "SD")} #: Per campaign and day: the metric totals every finer row splits evenly. _FIXTURE_TOTALS = {"metric.impressions": 400, "metric.clicks": 12, "metric.totalCost": 5.0, "metric.sales": 38.0, "metric.purchases": 4, "metric.unitsSold": 4} def _multi_status(report=None, errors=None): """A ReportMultiStatusResponse with one success or one error entry at index 0.""" return {"success": [{"index": 0, "report": report}] if report else [], "error": [{"index": 0, "errors": errors}] if errors else []} class _Fixture: """The recording fixture `AMAZON_ADS_FIXTURE=1` routes to, in the saved contracts' v1 shapes. It records method, path, operation, headers and body of every call, report file downloads included. Toggles: `drop_days`, `unsupported` (group names that end FAILED), `refuse_fields` (a create naming one answers 400), `nested` (files carry nested JSON lines), `responder`. """ FILES = 2 def __init__(self): self.reset() def reset(self): self.calls, self.reports, self.subscriptions, self.credentials = [], {}, {}, {} self.in_flight = self.max_in_flight = 0 self.drop_days, self.unsupported, self.refuse_fields = set(), set(), set() self.nested = False self.responder = None def credential(self, tenant_key): """The fixture credential for one tenant. `credentials[tenant]` overrides its fields.""" return {**FIXTURE_CREDENTIAL, **self.credentials.get(tenant_key, {})} def answer(self, prepared, tenant_key, profile_id, headers=None): self.calls.append({"method": prepared.method, "path": prepared.path, "operation": prepared.operation, "body": prepared.body, "query": dict(prepared.query), "tenant": tenant_key, "headers": dict(headers or {}), "named": dict(prepared.named) if prepared.named else None}) if self.responder is not None: return self.responder(prepared) if prepared.operation == "listStreamSubscriptions": return {"subscriptions": list(self.subscriptions.get((tenant_key, profile_id), {}).values())} if prepared.operation == "createStreamSubscription": subscriptions = self.subscriptions.setdefault((tenant_key, profile_id), {}) token = prepared.body["clientRequestToken"] if token not in subscriptions: subscriptions[token] = {**prepared.body, "subscriptionId": f"stream{len(subscriptions) + 1}", "status": "PENDING_CONFIRMATION"} return dict(subscriptions[token]) if prepared.named is not None: items = next(iter(prepared.body.values())) return {"success": [{"index": i} for i in range(len(items))], "error": []} if prepared.operation == "queryAdvertiserAccounts": return self._accounts(prepared.body, profile_id) if prepared.operation == "createReport": return self._create(prepared.body) return self._retrieve(prepared.body["reportIds"][0]) def _accounts(self, body, profile_id): """Two pages: an unrelated account, then the account that lists the caller's profile.""" if not body.get("nextToken"): return {"advertiserAccounts": [{"advertiserAccountId": "amzn1.ads-account.g.fxother", "alternateIds": [{"profileId": "999", "countryCode": "CA"}]}], "nextToken": "fixture-page-2"} return {"advertiserAccounts": [{"advertiserAccountId": f"amzn1.ads-account.g.fx{profile_id}", "alternateIds": [{"profileId": str(profile_id), "countryCode": "US", "entityId": "ENTITYFX"}]}]} def _create(self, body): if not (body.get("accessRequestedAccounts") or [{}])[0].get("advertiserAccountId"): return _multi_status(errors=[{"code": "400001", "message": "An advertiser account is required.", "fieldLocation": "accessRequestedAccounts"}]) refused = [f for f in body["reports"][0]["query"]["fields"] if f in self.refuse_fields] if refused: raise _rejection("Amazon Ads answered HTTP 400 on createReport", [{"code": "400004", "message": f"Field {refused[0]} is invalid for this request.", "fieldLocation": ""}], may_be_unoffered=False) report_id = f"fx{len(self.reports) + 1:05d}" self.in_flight += 1 self.max_in_flight = max(self.max_in_flight, self.in_flight) self.reports[report_id] = {"body": body, "polls": 0, "files": None, "done": False} return _multi_status(self._record(report_id, "PENDING")) def _record(self, report_id, status, **extra): """A contract `Report`: every required key, the request's format, periods and query echoed.""" request = self.reports[report_id]["body"]["reports"][0] return {"reportId": report_id, "status": status, "format": request["format"], "periods": request["periods"], "query": request["query"], "creationDateTime": FIXTURE_CREATED, "lastUpdatedDateTime": FIXTURE_CREATED, **extra} def _retrieve(self, report_id): from harness import amazon_ads_schema as schema report = self.reports[report_id] report["polls"] += 1 query = report["body"]["reports"][0]["query"] group = schema.group_for(query["fields"], query["filter"]["on"]["values"]) if group is not None and group.name in self.unsupported: self._finish(report) return _multi_status(self._record(report_id, "FAILED", failureCode="REPORT_NOT_SUPPORTED", failureReason="This report is not supported for this advertiser.")) if report["polls"] < 2: return _multi_status(self._record(report_id, "PROCESSING")) if report["files"] is None: rows = fixture_rows(report["body"], drop_days=self.drop_days) report["files"] = {f"{FIXTURE_FILE_HOST}/{report_id}/part-{i}.json.gz?X-Amz-Signature=fx{report_id}{i}": _fixture_file(rows[i::self.FILES], self.nested) for i in range(self.FILES)} return _multi_status(self._record(report_id, "COMPLETED", completedDateTime=FIXTURE_CREATED, completedReportParts=[ {"url": url, "sizeInBytes": len(raw), "urlExpirationDateTime": "2026-12-31T00:00:00Z"} for url, raw in report["files"].items()])) def _finish(self, report): if not report["done"]: report["done"] = True self.in_flight -= 1 def download(self, url, headers): """Serve one report file and record the GET with the headers the client sent.""" self.calls.append({"method": "GET", "path": url, "operation": "downloadReportPart", "body": None, "query": {}, "tenant": None, "headers": dict(headers), "named": None}) report = self.reports[urllib.parse.urlsplit(url).path.split("/")[1]] raw = report["files"][url] report.setdefault("read", set()).add(url) if report["read"] == set(report["files"]): self._finish(report) return raw def _fixture_file(rows, nested): """GZIP JSON: an array of flat rows, or nested objects one per line.""" def nest(row): out = {} for field, value in row.items(): head, _, tail = field.partition(".") out.setdefault(head, {})[tail] = value return out text = "\n".join(json.dumps(nest(r)) for r in rows) if nested else json.dumps(rows) return gzip.compress(text.encode("utf-8")) def _fixture_query(body): report = body["reports"][0] period = report["periods"][0]["datePeriod"] return (report["query"]["fields"], report["query"]["filter"]["on"]["values"], date.fromisoformat(period["startDate"]), date.fromisoformat(period["endDate"])) def _fixture_days(start, end, drop_days): return [start + timedelta(days=i) for i in range((end - start).days + 1) if (start + timedelta(days=i)).isoformat() not in drop_days] def fixture_totals(body, *, drop_days=()): """The fixture's own totals for one create body. -> {(ad product, date, campaign id): {field: total}}.""" _fields, products, start, end = _fixture_query(body) out = {} for day in _fixture_days(start, end, drop_days): for index, product in enumerate(products): for c in (1, 2): n = day.toordinal() % 7 + c + index out[(product, day.isoformat(), f"{_FIXTURE_PREFIX[product][0]}10{c}")] = { field: total * n for field, total in _FIXTURE_TOTALS.items()} return out def fixture_rows(body, *, drop_days=()): """Deterministic v1 rows for one create body: one row per distinct value of every requested dimension and day, only the requested fields and ad products. Two ad groups split a campaign.""" fields, _products, start, end = _fixture_query(body) groups = (1, 2) if "adGroup.id" in fields else (0,) out = [] for (product, day, campaign), totals in fixture_totals(body, drop_days=drop_days).items(): prefix, short = _FIXTURE_PREFIX[product] c = campaign[-1] for g in groups: row = {"date.value": day, "adProduct.value": product, "campaign.id": campaign, "campaign.name": f"{short} campaign {c}", "campaign.deliveryStatus": "DELIVERING", "campaign.budgetAmount": 25 * int(c), "budgetCurrency.value": "USD", "adGroup.id": f"{prefix}20{c}{g}", "adGroup.name": f"Ad group {c}.{g}", "adGroupTarget.id": f"{prefix}30{c}{g}", "target.value": f"term {c}{g}", "target.matchType": "EXACT", "adGroupTarget.bid": str(0.5 + g / 10), "ad.id": f"{prefix}40{c}{g}", "advertisedProduct.id": f"B0FX{prefix}{c}{g}00", "advertisedProductMarketplace.value": "ATVPDKIKX0DER", "advertisedProductSku.value": f"SKU-{c}{g}", "searchTerm.value": f"search {c}{g}", **{f: (t // len(groups) if isinstance(t, int) else t / len(groups)) for f, t in totals.items()}} out.append({f: row[f] for f in fields}) return out FIXTURE = _Fixture() def _reset_caches(): _ACCOUNTS.clear() _REFUSED.clear() _ISSUED_FILES.clear() @contextmanager def recording_fixture(): """Route every call to a fresh recording fixture and fresh report caches for the block. -> the fixture.""" previous = os.environ.get(FIXTURE_ENV) os.environ[FIXTURE_ENV] = "1" FIXTURE.reset() _reset_caches() try: yield FIXTURE finally: FIXTURE.reset() _reset_caches() if previous is None: os.environ.pop(FIXTURE_ENV, None) else: os.environ[FIXTURE_ENV] = previous __all__ = ["KEY", "LABEL", "ENTRY_TYPE", "FIXTURE_ENV", "HOSTS", "REPORT_OPERATIONS", "V1_CLIENT_HEADERS", "NAMED_WRITES", "STREAM_OPERATIONS", "STREAM_CONTENT_TYPE", "stream_operation", "FIELD_RULES", "APPROVAL_SOURCE", "SCHEDULE_APPROVAL_SOURCE", "SENT_WRITE_SOURCE", "SCHEDULE_WEEKDAY_CONVENTION", "REPORT_CONCURRENCY", "REPORTS_PER_DAY", "MAX_DAYS_PER_REQUEST", "AdsError", "AdsWriteRefused", "AdsRejected", "AdsReportUnavailable", "AdsReportPending", "GroupReport", "fixture_mode", "build_request", "report_operation", "send_named_write", "request", "credential", "run_group_report", "run_report", "FIXTURE", "recording_fixture", "fixture_rows", "fixture_totals"]