apache/druid · error · RuntimeException

RuntimeException

Error message

RuntimeException

What it means

fetchRecords catches SdkException and classifies it via AWSClientUtil.isClientExceptionRecoverable. Recoverable ones are retried after EXCEPTION_RETRY_DELAY_MS; unrecoverable ones are logged as 'will not retry' and rethrown as a RuntimeException wrapping the SDK exception, which fails the fetch loop.

Source

Thrown at extensions-core/kinesis-indexing-service/src/main/java/org/apache/druid/indexing/kinesis/KinesisRecordSupplier.java:366

          if (recordsResult != null) {
            shardIterator = recordsResult.nextShardIterator(); // will be null if the shard has been closed
            scheduleBackgroundFetch(fetchDelayMillis);
          } else {
            throw new ISE("can't reschedule fetch records runnable, recordsResult is null??");
          }
        }
        catch (ResourceNotFoundException | InvalidArgumentException e) {
          // aws errors
          log.error(e, "encounted AWS error while attempting to fetch records, will not retry");
          throw e;
        }
        catch (SdkException e) {
          if (AWSClientUtil.isClientExceptionRecoverable(e)) {
            log.warn(e, "encounted unknown recoverable AWS exception, retrying in [%,dms]", EXCEPTION_RETRY_DELAY_MS);
            scheduleBackgroundFetch(EXCEPTION_RETRY_DELAY_MS);
          } else {
            log.warn(e, "encounted unknown unrecoverable AWS exception, will not retry");
            throw new RuntimeException(e);
          }
        }
        catch (Throwable e) {
          // non transient errors
          log.error(e, "unknown fetchRecords exception, will not retry");
          throw new RuntimeException(e);
        }

      };
    }

    private void seek(ShardIteratorType iteratorEnum, String sequenceNumber)
    {
      log.debug(
          "Seeking partition [%s] to [%s]",
          streamPartition.getPartitionId(),
          sequenceNumber != null ? sequenceNumber : iteratorEnum.toString()
      );

View on GitHub (pinned to 9b90983fd2)

Solutions

1. Read the wrapped SdkException cause and its error code; fix the underlying AWS-side problem (credentials, permissions, stream existence)
2. Verify druid kinesis consumer AWS accessKey/secretKey/region config
3. Restart the task after fixing AWS configuration — the error is deliberately non-retried
4. Add the error code to isClientExceptionRecoverable handling if it is actually transient

Example fix

// before
throw new RuntimeException(e);
// after
if (e.errorCode().equals("ThrottlingException")) {
  scheduleBackgroundFetch(EXCEPTION_RETRY_DELAY_MS);
} else {
  throw new RuntimeException(e);
}

// verify credentials and stream exist before consuming
GetRecordsPermissionCheck: kinesisClient.describeStreamSummary(DescribeStreamRequest.builder().streamName(stream).build());

static boolean unrecoverableSdkError(Throwable t) {
  return t instanceof RuntimeException && t.getCause() instanceof SdkException
      && !AWSClientUtil.isClientExceptionRecoverable((SdkException) t.getCause());
}

try {
  startIngestion();
} catch (RuntimeException e) {
  if (e.getCause() instanceof SdkException) {
    LOG.error("AWS SDK error code=%s", ((SdkException) e.getCause()).errorCode());
  }
  throw e;
}

When it happens

Trigger: scheduleBackgroundFetch -> fetchRecords calls getRecords and the AWS SDK throws a non-recoverable SdkClientException/SdkException (e.g. credentials failure, permission denied, malformed request).

Common situations: Expired or missing AWS credentials; IAM policy revocation mid-task; stream deleted while task is running; proxy/endpoint misconfiguration making every request fail deterministically.

Related errors

• StreamException
• can't reschedule fetch records runnable, recordsResult is nu
• getPosition() is not supported in Kinesis
• Partition [%s] has not been assigned
• Invalid operation - KinesisRecordSupplier has already been c

AI-assisted analysis of apache/druid@9b90983fd2 (2026-09-07). Data as JSON: /api/errors/0f336edeaa8c2e2f. Report an issue: GitHub.