data-manager-api-troubleshooting
Diagnoses and fixes errors, warnings, and missing data for Data Manager API event and audience ingestion requests. Use when an IngestEvents, IngestAudienceMembers, RemoveAudienceMembers, RemoveAllAudienceMembers, or RetrieveRequestStatus call fails (INVALID_ARGUMENT, REQUIRED_FIELD_MISSING, INVALID_HEX_ENCODING, PERMISSION_DENIED, UNAUTHENTICATED, RESOURCE_EXHAUSTED, INVALID_REQUEST_ID), when a response has fieldWarnings, when checking request status or diagnostics, when request status shows FAILED or PARTIAL_SUCCESS or processing errors, when finding which records failed, or when a request re
Try your first task
A source-grounded starter task has not been verified for this skill yet.
SKILL.md
name: data-manager-api-troubleshootingdescription: >- Diagnoses and fixes errors, warnings, and missing data for Data Manager API event and audience ingestion requests. Use when an IngestEvents, IngestAudienceMembers, RemoveAudienceMembers, RemoveAllAudienceMembers, or RetrieveRequestStatus call fails (INVALID_ARGUMENT, REQUIRED_FIELD_MISSING, INVALID_HEX_ENCODING, PERMISSION_DENIED, UNAUTHENTICATED, RESOURCE_EXHAUSTED, INVALID_REQUEST_ID), when a response has fieldWarnings, when checking request status or diagnostics, when request status shows FAILED or PARTIAL_SUCCESS or processing errors, when finding which records failed, or when a request returned 200 but conversions or audience members are missing. Don't use for writing new ingestion code (use the data-manager-api-event-ingestion or data-manager-api-audience-ingestion skills) or first-time credential setup (use the data-manager-api-setup skill).metadata: version: "1.0.0" category: GoogleAdsData Manager API Troubleshooting
The Data Manager API reports problems in three places, at three different times.
Step 1: Classify the problem
Ask for the exact error payload, response, or diagnostics output if the user hasn't shared it.
| Symptom | When | What it means | Go to |
|---|---|---|---|
| Request throws an exception or returns 4xx/5xx | Immediately | The entire request was rejected. No records were ingested. | Step 2 |
Request succeeds with field_warnings | Immediately | The request was accepted, but the flagged field had an issue (for example, an invalid or missing optional value). Don't resend. Fix the source data for future requests. | None |
| Request succeeded, but data is missing or lower than expected | Check retrieve_request_status 30 minutes to 24 hours later | Records can still fail during asynchronous processing. Diagnostics show per-destination results, including which records failed. | Step 3 |
Step 2: Fix request errors
The API uses a fast-fail model: if any record fails synchronous validation, the
whole request fails and no records are processed. Fix the problem and resend the
entire request. To confirm a fix first, send the ingestion request with
validate_only set to true.
Read the full error
For INVALID_ARGUMENT, the top-level message is always generic ("There was a
problem with the request."). Read every entry in the google.rpc.BadRequest
detail's field_violations, including field (a path with list indexes, such
as events.events[1].user_data.user_identifiers[2]), reason (an
ErrorReason), and description. For any error, also log request_id from
google.rpc.RequestInfo and include it in support requests.
In Python (google.api_core.exceptions.GoogleAPICallError), read err.details:
from google.api_core import exceptions
from google.rpc import error_details_pb2
try:
...
except exceptions.GoogleAPICallError as err:
for detail in err.details:
if isinstance(detail, error_details_pb2.BadRequest):
for v in detail.field_violations:
print(v.field, v.reason, v.description)
elif isinstance(detail, error_details_pb2.RequestInfo):
print("request_id:", detail.request_id)
Fix by error code
| Error code | What to do |
|---|---|
INVALID_ARGUMENT (400) | Follow Fix INVALID_ARGUMENT errors. |
UNAUTHENTICATED (401), PERMISSION_DENIED (403) | Read ErrorInfo.reason, then follow Permission errors. |
RESOURCE_EXHAUSTED (429) | Check ErrorInfo.metadata.quota_limit. For a per-minute limit, resend with exponential backoff and jitter. For a daily limit, don't resend until the quota resets at midnight Pacific Time. In both cases, reduce the request count by batching more records per request (up to 2,000 events or 10,000 audience members) and sending at most 10 requests concurrently. |
RetrieveRequestStatus fails with INVALID_REQUEST_ID or REQUEST_TOO_OLD | Status exists only for recent requests that succeeded and didn't set validate_only. For INVALID_REQUEST_ID, check that the ID is the request_id from a successful, non-validation ingestion response (if the original request failed, fix it using its own error). For REQUEST_TOO_OLD, diagnostics are no longer available. |
For other error codes, check whether the error can be retried before resending. See Best practices for error handling.
Step 3: Check request status and diagnostics
Each destination in the response has an overall request_status, error and
warning counts, and a request-type status with record and identifier counts.
Some audience data also has a match rate. Field names differ by request type
and data type. Use the fields in
Request status and diagnostics.
Requirements and timing
- Diagnostics exist only for requests that succeeded and didn't set
validate_only. - Wait at least 30 minutes after ingestion, then poll with exponential backoff
until every destination reaches
SUCCESS,PARTIAL_SUCCESS, orFAILED. Processing can take up to 24 hours. When writing code that reads request status, handle destinations still inPROCESSINGby polling again later instead of reporting their results.
Map records to their errors and warnings
record_indices are 0-based positions in the original request's events or
audience_members list. When a user asks which records failed or wants to fix
them, invert error_info.error_counts and warning_info.warning_counts for
each destination into a per-record list of errors and warnings. Resend only the
corrected records that had errors, and only to the destinations where they
failed. For warning-only records, fix the issue for future uploads.
To find what a processing reason means and how to fix it, see Processing reasons.