Skip to content

Latest commit

 

History

History
170 lines (136 loc) · 7.2 KB

debugging.md

File metadata and controls

170 lines (136 loc) · 7.2 KB
Raw

Debug aggregation runs with encrypted payloads

This document describes the debugging support for the Aggregation Service running in a TEE using encrypted payloads of aggregatable reports. This allows you to debug your production setup and understand how the encrypted payloads of aggregatable reports are processed. Reports with debug_cleartext_payload can be used with the local testing tool and are helpful for understanding the content of reports and validating that registrations on the browser client or device are configured properly.

To test the Aggregation Service, you can enable debug aggregation runs which use encrypted payloads of aggregatable reports to generate debug summary reports. When executing a debug run, no noise is added to the debug summary report, and annotations are added to indicate whether keys are present in domain input and/or reports. This allows developers to:

  • Analyze the reports
  • Determine if the aggregation was completed correctly, per the adtech's specifications
  • Understand the impact of noise in summary reports
  • Determine how to set the proper domain keys

Additionally, debug runs do not enforce the No-Duplicates rule across batches. The No-Duplicates rule is still enforced within a batch. This allows adtech to try different batches without worrying about making them disjoint during testing or debugging.

Once third-party cookies are deprecated, the client (browser or operating system) will no longer generate aggregatable reports that are enabled for debugging. At that time, debug runs with encrypted payloads will no longer be supported for reports from real user devices or browsers.

In this document, you'll find code snippets and instructions for how to debug the Aggregation Service and create debug summary reports.

Create a debug job

To create an aggregation debug job, add the debug_run parameter to the job_parameters object of the createJob API request.

AWS: POST https://<frontend_api_id>.execute-api.us-east-1.amazonaws.com/stage/v1alpha/createJob GCP: POST https://<cloud-run-endpoint>/v1alpha/createJob

{
    "input_data_blob_prefix": "input/reports.avro",
    "input_data_bucket_name": "<data_bucket>",
    "output_data_blob_prefix": "output/summary_report.avro",
    "output_data_bucket_name": "<data_bucket>",
    "job_parameters": {
        "attribution_report_to": "<your_attribution_domain>",
        "output_domain_blob_prefix": "domain/domain.avro",
        "output_domain_bucket_name": "<data_bucket>",
        "debug_run": "true"
    },
    "job_request_id": "test01"
}

If debug_run is not present injob_parameters or it's set to false, a normal noised aggregation run is created. More details about createJob API can be found in detailed API spec.

Debuggable aggregatable reports

A debug run only considers reports that have the flag "debug_mode": "enabled" in the report shared_info (aggregatable report sample). Reports with the debug_mode flag missing or the debug_mode value isn't set to enabled aren't included in the results generated by a debug run.

The count of reports that were not processed during a debug run is returned in the job response, which can be previewed in the detailed API spec. In the error_counts field, the category NUM_REPORTS_DEBUG_NOT_ENABLED shows the numbers of reports not processed during a debug run.

Results

Two summary reports are generated from a debug run: a regular summary report and a debug summary report. The regular summary report format generated from the debug run is consistent with that of a regular aggregation run. The debug summary report has a different format. The path of the summary report is set in the createJob API. The debug summary report will be stored in the "debug" folder under the summary report's path with the same object name.

Considering the following createJob parameters for output_data_bucket_name and output_data_blob_prefix:

{
    "output_data_blob_prefix": "output/summary_report.avro",
    "output_data_bucket_name": "<data_bucket>"
}

the following objects are created by a debug run:

s3://<data_bucket>/output/summary_report.avro and

s3://<data_bucket>/output/debug/summary_report.avro.

Note that the regular summary report generated during a debug run will only include reports which have the flag "debug_mode": "enabled" in the reports shared_info.

Debug summary report

The debug summary report includes the following data:

  • bucket: The aggregation key
  • unnoised_metric: The aggregation value without noise
  • noise: The approximate noise applied to the aggregated results in the regular summary report
  • annotations: The annotations associated with the bucket

The keys in the debug summary report will include all the keys from the output domain.

If the key is only present in the output domain (not in any of the processed aggregatable reports), the key will be included in the debug report with unnoised_metric=0 and annotations=["in_domain"].

The keys that are only present in aggregatable reports (not in output domain) will also be included in the debug report with unnoised_metric=<unnoised aggregated value> and annotations=["in_reports"].

Keys that are present in domain and aggregatable reports will have annotations for both ["in_domain", "in_reports"].

The schema of debug summary reports is in the following Avro format:

{
"type": "record",
"name": "DebugAggregatedFact",
"fields": [
    {
    "name": "bucket",
    "type": "bytes",
    "doc": "Histogram bucket used in aggregation. 128-bit integer encoded as a 16-byte big-endian bytestring. Leading 0-bits will be left out."
    },
    {
    "name": "unnoised_metric",
    "type": "long",
    "doc": "Unnoised metric associated with the bucket."
    },
    {
    "name": "noise",
    "type": "long",
    "doc": "The noise applied to the metric in the regular result."
    }
    {
    "name":"annotations",
    "type":
       {
       "type": "array",
       "items": {
         "type":"enum",
         "name":"bucket_tags",
         "symbols":["in_domain","in_reports"]
       }
    }
  ]
}

Privacy Budget Service Errors

When an aggregation job is performed with debug_run enabled, the privacy budget is not consumed. Furthermore, budget availability will not be checked while debug_run is enabled -- even if budget is not available, the summary report will still be generated successfully.

This change was made in version 2.10. For versions 2.9 and lower, debug runs would consume privacy budget. Budget that was consumed with previous debug runs will not be recovered automatically. If budget recovery is necessary for previously consumed budget, please see the budget recovery process