Limitations of AWS Clean Rooms Differential Privacy
AWS Clean Rooms Differential Privacy doesn't address the following situations:
-
AWS Clean Rooms Diļ¬erential Privacy only supports Amazon S3-backed AWS Glue tables. It does not support queries with Snowflake or Amazon Athena tables.
-
AWS Clean Rooms Differential Privacy doesn't address timing attacks. For example, these attacks are possible in scenarios where an individual user contributes a large number of rows and adding or removing this user significantly changes the query computation time.
-
AWS Clean Rooms Differential Privacy doesn't guarantee differential privacy when a SQL query can result in overflow or invalid cast errors at run time due to the use of certain SQL constructs. The following table is a list of some, but not all, SQL constructs that may produce run-time errors and should be verified in analysis templates. We recommend that you approve analysis templates that minimize the chances of such run-time errors and periodically review query logs to determine if the queries align with the collaboration agreement.
The following SQL constructs are vulnerable to overflow errors:
-
Aggregate functions - AVG, LISTAVG, PERCENTILE_COUNT, PERCENTILE_DISC, SUM/SUM_DISTINCT
-
Data type formatting functions - TO_TIMESTAMP, TO_DATE
-
Date and time functions - ADD_MONTHS, DATEADD, DATEDIFF
-
Math functions - +, -, *, /, POWER
-
String functions - ||, CONCAT, REPEAT, REPLICATE
-
Window functions - AVG, LISTAGG, PERCENTILE_COUNT, PERCENTILE_DISC, RATIO_TO_REPORT, SUM
The CAST data type formatting function is vulnerable to invalid cast errors.
You can configure CloudWatch to create a metric filter for a log group and then create a CloudWatch alarm on that metric filter to receive alerts if a potential overflow or cast error was encountered. Specifically, you should monitor for the error codes
CastError
,OverflowError
,ConversionError
. The presence of these error codes indicates a potential side-channel attack, but might indicate an erroneous SQL query.For more information, see Query logging in AWS Clean Rooms.
-