We are using Java 8 and using AWS SDK to programmatically upload files to AWS S3. For uploading large file (>100MB), we read that the preferred method to use is Multipart Upload. We tried that but it seems it does not speed it up, upload time remains almost the same as not using multipart upload. Worse is, we even encountered out of memory errors saying heap space is not sufficient.
Questions:
Is using multipart upload really supposed to speed up the upload? if not, then why use it?
How come using multi part upload eats up memory faster than not using? does it concurrently upload all the parts?
See below for the code we used:
private static void uploadFileToS3UsingBase64(String bucketName, String region, String accessKey, String secretKey,
String fileBase64String, String s3ObjectKeyName) {
byte[] bI = org.apache.commons.codec.binary.Base64.decodeBase64((fileBase64String.substring(fileBase64String.indexOf(",")+1)).getBytes());
InputStream fis = new ByteArrayInputStream(bI);
long start = System.currentTimeMillis();
AmazonS3 s3Client = null;
TransferManager tm = null;
try {
s3Client = AmazonS3ClientBuilder.standard().withRegion(region)
.withCredentials(new AWSStaticCredentialsProvider(new BasicAWSCredentials(accessKey, secretKey)))
.build();
tm = TransferManagerBuilder.standard()
.withS3Client(s3Client)
.withMultipartUploadThreshold((long) (50* 1024 * 1025))
.build();
ObjectMetadata metadata = new ObjectMetadata();
metadata.setHeader(Headers.STORAGE_CLASS, StorageClass.Standard);
PutObjectRequest putObjectRequest = new PutObjectRequest(bucketName, s3ObjectKeyName,
fis, metadata).withSSEAwsKeyManagementParams(new SSEAwsKeyManagementParams());
Upload upload = tm.upload(putObjectRequest);
// Optionally, wait for the upload to finish before continuing.
upload.waitForCompletion();
long end = System.currentTimeMillis();
long duration = (end - start)/1000;
// Log status
System.out.println("Successul upload in S3 multipart. Duration = " + duration);
} catch (Exception e) {
e.printStackTrace();
} finally {
if (s3Client != null)
s3Client.shutdown();
if (tm != null)
tm.shutdownNow();
}
}
Using multipart will only speed up the upload if you upload multiple parts at the same time.
In your code you're setting withMultipartUploadThreshold. If your upload size is larger than that threshold, then you should observe concurrent upload of separate parts. If it is not, then only one upload connection should be used. You're saying that you have >100 MB file, and in your code you have 50 * 1024 * 1025 = 52 480 000 bytes as the multipart upload threshold, so concurrent upload of parts of that file should have been happening.
However, if your upload throughput is anyway capped by your network speed, there would not be any increase in throughput. This might be the reason you're not observing any speed increase.
There are other reasons to use multipart too, as it is recommended for fault tolerance reasons as well. Also, it has a larger maximum size than single upload.
For more details see documentation:
Multipart upload allows you to upload a single object as a set of
parts. Each part is a contiguous portion of the object's data. You can
upload these object parts independently and in any order. If
transmission of any part fails, you can retransmit that part without
affecting other parts. After all parts of your object are uploaded,
Amazon S3 assembles these parts and creates the object. In general,
when your object size reaches 100 MB, you should consider using
multipart uploads instead of uploading the object in a single
operation.
Using multipart upload provides the following advantages:
Improved throughput - You can upload parts in parallel to improve throughput.
Quick recovery from any network issues - Smaller part size minimizes the impact of restarting a failed upload due to a network
error.
Pause and resume object uploads - You can upload object parts over time. After you initiate a multipart upload, there is no expiry; you
must explicitly complete or stop the multipart upload.
Begin an upload before you know the final object size - You can upload an object as you are creating it.
We recommend that you use multipart upload in the following ways:
If you're uploading large objects over a stable high-bandwidth network, use multipart upload to maximize the use of your available
bandwidth by uploading object parts in parallel for multi-threaded
performance.
If you're uploading over a spotty network, use multipart upload to increase resiliency to network errors by avoiding upload restarts.
When using multipart upload, you need to retry uploading only parts
that are interrupted during the upload. You don't need to restart
uploading your object from the beginning.
The answer from eis is very fine. Though you still should take some action:
String.getBytes(StandardCharsets.US_ASCII) or ISO_8859_1 prevents using a more costly encoding, like UTF-8. If the platform encoding would be UTF-16LE the data would even be corrupt (0x00 bytes).
The standard java Base64 has some de-/encoders that might work. It can work on a String. However check the correct handling (line endings).
try-with-resources closes also in case of exceptions/internal returns.
The ByteArrayInputStream was not closed, which would have been better style (easier garbage collection?).
You could set the ExecutorFactory to a thread pool factory limiting the number of threads globally.
So
byte[] bI = Base64.getDecoder().decode(
fileBase64String.substring(fileBase64String.indexOf(',') + 1));
try (InputStream fis = new ByteArrayInputStream(bI)) {
...
}
Related
As part of my web service, I have a picture repository which retrieves an image from Amazon S3 (a datastore) then returns it. This is how the method that does this looks:
File getPicture(String path) throws IOException {
File file = File.createTempFile(path, ".png");
S3Object object = s3Client.getObject(new GetObjectRequest(bucketName, path));
IOUtils.copy(object.getObjectContent(), new FileOutputStream(file));
return file;
}
The problem is that it takes way too long to get a response from the service - (a 3MB image took 7.5 seconds to download). I notice that if I comment out the IOUtils.copy() line, the response time is significantly faster so it must be that particular method that's causing this delay.
I've seen this method used in almost all modern examples of converting an S3Object to a file but I seem to be a unique case. Am I missing a trick here?
Appreciate any help!
From the AWS documentation:
public S3Object getObject(GetObjectRequest getObjectRequest)
the returned Amazon S3 object contains a direct stream of data from the HTTP connection. The underlying HTTP connection cannot be reused until the user finishes reading the data and closes the stream.
public S3ObjectInputStream getObjectContent()
Note: The method is a simple getter and does not actually create a stream. If you retrieve an S3Object, you should close this input stream as soon as possible, because the object contents aren't buffered in memory and stream directly from Amazon S3.
If you remove the IOUtils.copy line, then method exits quickly because you don't actually process the stream. If the file is large it will take time to download. You can't do much about that unless you can get a better connection to the AWS services.
I'm quite new in AWS. I have designed an architecture that uses Api Gateway to call a lambda function written in java. Since I have some configuration I decided to create an S3 file to store a standard Java configuration file there and load it when needed. This took a lot of time, about 15 sec, for a very small file.
To read the file I'm using AmazonS3Client client class, Do I have other options?
long ms = System.nanoTime();
AmazonS3Client client = new AmazonS3Client(new DefaultAWSCredentialsProviderChain());
GetObjectRequest request = new GetObjectRequest("bucket","filepath");
InputStream inputStream = client.getObject(request).getObjectContent();
try {
PropertiesConfiguration p = new PropertiesConfiguration();
p.load(inputStream);
composite.addConfiguration(p);
log.debug(String.format("Configuration read in %f mS",(System.nanoTime()-ms)/1000000f));
}catch (ConfigurationException e) {
logger.error("error reading configuration on S3:"+e);
}
So the questions: if storing the config file in an s3 bucket is a bad idea, where is supposed to be stored a configuration?
Is that performance normal? I'm thinking in using s3 a lot in my architecture for something else, but having a 15 sec handshake for a file is, of course, unacceptable.
In this case I think you should try to store it with EBS. But it will cost you more because EBS is optimized I/O and EBS storage is organized into volumes and once an EBS volume is attached to a server it is treated like a local disk drive.
We have an environment where each user may get different html/js/css resources. I'm using the following code to compress and transfer a java script resource:
public static byte[] compress(String str) throws IOException {
if (str == null || str.length() == 0) {
return null;
}
ByteArrayOutputStream obj=new ByteArrayOutputStream();
GZIPOutputStream gzip = new GZIPOutputStream(obj);
gzip.write(str.getBytes("UTF-8"));
gzip.close();
return obj.toByteArray();
}
...
HttpServletResponse raw = response.raw();
raw.setBufferSize(file.length().intValue());
ServletOutputStream servletOutputStream = raw.getOutputStream();
servletOutputStream.write(compress(FileUtils.readFileToString(file)));
servletOutputStream.flush();
servletOutputStream.close();
...
Inspecting the problem using chrome network tab, the download time is 2 seconds for 300KB of compressed data - this seems unreasonable.
The problem is not the bandwidth or jetty itself, because static resources transfer time is fast.
Don't know if that is the source of your bottleneck, but I wouldn't do:
raw.setBufferSize(file.length().intValue());
If the gzipped file is around 300KB then you could create response buffers which are a magnitude greater than that. And you don't need a big response buffer at all when you are streaming static content.
From the servlet javadoc:
Sets the preferred buffer size for the body of the response. The
servlet container will use a buffer at least as large as the size
requested. The actual buffer size used can be found using
getBufferSize.
A larger buffer allows more content to be written before anything is
actually sent, thus providing the servlet with more time to set
appropriate status codes and headers. A smaller buffer decreases
server memory load and allows the client to start receiving data more
quickly.
I have the next code for uploading files to an Amazon S3:
AmazonS3Client client = new AmazonS3Client(credentials,
new ClientConfiguration().withMaxConnections(100)
.withConnectionTimeout(120 * 1000)
.withMaxErrorRetry(15));
TransferManager tm = new TransferManager(client);
TransferManagerConfiguration configuration = new TransferManagerConfiguration();
configuration.setMultipartUploadThreshold(MULTIPART_THRESHOLD);
tm.setConfiguration(configuration);
Upload upload = tm.upload(bucket, key, file);
try {
upload.waitForCompletion();
} catch(InterruptedException ex) {
logger.error(ex.getMessage());
} finally {
tm.shutdownNow(false);
}
It works, but some uploads(1GB) produce the next log message:
INFO AmazonHttpClient:Unable to execute HTTP request: bucket-name.s3.amazonaws.com failed to respond
org.apache.http.NoHttpResponseException: bucket-name.s3.amazonaws.com failed to respond
I have tried to create TransferManager without AmazonS3Client, but it doesn't help.
Is there any way to fix it?
The log message is telling you that there was a transient error sending data to S3. You've configured .withMaxErrorRetry(15), so the AmazonS3Client is transparently retrying the request that failed and the overall upload is succeeding.
There isn't necessarily anything to fix here - sometimes packets get lost on the network, especially if you're trying to push through a lot of packets at once. Waiting a little while and retrying is usually the right way to deal with this, and that's what's already happening.
If you wanted, you could try turning down the MaxConnections setting to limit how many chunks of the file will be uploaded at a time - there's probably a sweet spot where you're still getting reasonable throughput, but not overloading the network.
I'm trying to insert large file into Google's drive using google-api-services-drive version v2-rev93-1.16.0-rc
I've set setChunkSize() for minimum in order to have my own ProgressListener notified more frequent. The following code is used to insert file:
File body = new File();
body.setTitle(filetobeuploaded.getName());
body.setMimeType("application/zip");
body.setFileSize(filetobeuploaded.length());
InputStreamContent mediaContent =
new InputStreamContent("application/zip",
new BufferedInputStream(new FileInputStream(filetobeuploaded)));
mediaContent.setLength(filetobeuploaded.length());
Insert insert = drive.files().insert(body, mediaContent);
MediaHttpUploader uploader = insert.getMediaHttpUploader();
uploader.setChunkSize(MediaHttpUploader.MINIMUM_CHUNK_SIZE);
uploader.setProgressListener(new CustomProgressListener(filetobeuploaded));
insert.execute();
After 'a while' (sometimes 200 MB sometimes 300 MB ) I got IOException :
Exception in thread "main" java.io.IOException: insufficient data written
at sun.net.www.protocol.http.HttpURLConnection$StreamingOutputStream.close(HttpURLConnection.java:3213)
at com.google.api.client.http.javanet.NetHttpRequest.execute(NetHttpRequest.java:81)
at com.google.api.client.http.HttpRequest.execute(HttpRequest.java:960)
at com.google.api.client.googleapis.media.MediaHttpUploader.executeCurrentRequest(MediaHttpUploader.java:482)
at com.google.api.client.googleapis.media.MediaHttpUploader.upload(MediaHttpUploader.java:390)
at com.google.api.client.googleapis.services.AbstractGoogleClientRequest.executeUnparsed(AbstractGoogleClientRequest.java:418)
at com.google.api.client.googleapis.services.AbstractGoogleClientRequest.executeUnparsed(AbstractGoogleClientRequest.java:343)
at com.google.api.client.googleapis.services.AbstractGoogleClientRequest.execute(AbstractGoogleClientRequest.java:460)
Any ideas how to get this code working?
You wont be able to get it working from a frontend because of time constrains. The only reliable way (but a pain) is to do it from a backend using resumable upload since the backend/task queue may also be shut down while processing chunks.
Does 'a while' happen to mean 1 hour? In this case you are probably experiencing the following bug:
http://code.google.com/p/gdata-issues/issues/detail?id=5124
This issue is only for Drive Resumable Media upload. check this reply ..
https://stackoverflow.com/a/30796105/4576135
For me this meant "you are doing a POST and specified a content length, but then the stream you uploaded wasn't long enough to match that content length" (a closed bytearray stream, in my case, closed because previously used to basically "exhausted" already, as it were).