Question

com.amazonaws.AmazonClientException: Unable to execute HTTP request: Connection reset
    at com.amazonaws.http.AmazonHttpClient.executeHelper(AmazonHttpClient.java:324)
    at com.amazonaws.http.AmazonHttpClient.execute(AmazonHttpClient.java:164)
    at com.amazonaws.services.dynamodb.AmazonDynamoDBClient.invoke(AmazonDynamoDBClient.java:985)
    at com.amazonaws.services.dynamodb.AmazonDynamoDBClient.batchWriteItem(AmazonDynamoDBClient.java:365)
    .
    .
    .

The above exception was caught when attempting to run the following code sample.

BatchWriteItemResult result;
BatchWriteItemRequest batchWriteItemRequest = new BatchWriteItemRequest();
do {
    System.out.println("Making the request.");                         
    batchWriteItemRequest.withRequestItems(requestItems);
    result = client.batchWriteItem(batchWriteItemRequest);

    // Print consumed capacity units
    for(Map.Entry<String, BatchWriteResponse> entry : result.getResponses().entrySet()) {
        String tableName = entry.getKey();
        Double consumedCapacityUnits = entry.getValue().getConsumedCapacityUnits();
        System.out.println("Consumed capacity units for table " + tableName + ": " + consumedCapacityUnits);
     }

  // Check for unprocessed keys which could happen if you exceed provisioned throughput
    System.out.println("Unprocessed Put and Delete requests: \n" + result.getUnprocessedItems());
    requestItems = result.getUnprocessedItems();
} while (result.getUnprocessedItems().size() > 0);

I have 1 table set up with 8 Write Units configured.

In my BatchWriteItemRequest I have 9 PutRequestItems.

When the first time in the while loop, 8 out of the 9 PutRequestItems were processed.
Therefore, the loop was entered again, trying to process the one remaining request.

However, the code would hang at
result = client.batchWriteItem(batchWriteItemRequest); for around 149 seconds. After then, the above mentioned exception is thrown.

It seems the only way to bypass this was to set a higher Write Provisioning Unit. However, isn’t the while-do loop a way to handle cases where we exceed our provisioned Write Units?

Answer 1

It seems to be a bug on the Java Driver v. 1.3.21 (or maybe including older versions).

1.3.21.1 was released just to take care of that:

Resolved Issues

SocketException During Request Retries

Fixes an issue where retries caused by service error responses (i.e.
not IO error retries) aren’t able to correctly resend the request
payload for services like Amazon DynamoDB that utilize request
payloads.

see http://aws.amazon.com/releasenotes/Java/5659251433242996