FLINK-35815, titled "KinesisProxySyncV2 doesn't always retry throttling exceptions," describes a problem where the KinesisProxySyncV2 class fails to retry certain throttling exceptions, specifically LimitExceededException, during DescribeStreamSummary calls to AWS Kinesis.

The root cause is identified in the AwsV2Util.isRecoverableException method, which only checks the cause of an exception (exception.getCause()) and not the exception itself. As a result, when a LimitExceededException is thrown directly (without being wrapped in another exception), the retry mechanism does not recognize it as recoverable, leading to job failures instead of retries.

The bug report provides a stack trace showing a LimitExceededException causing a FlinkRuntimeException, indicating that the exception was not retried as expected. This issue affects versions like aws-connector-4.2.0, aws-connector-4.3.0, and 1.19.1, with fixes released in aws-connector-4.4.0 and aws-connector-5.0.0.