SignerUtils#moveContentLength Opens a Stream Unnecessarily

[RealTYPICAL]

Describe the bug

SignerUtils#moveContentLength conditionally has to read the entire contents of an input stream in case a content length header isn't there. This is problematic because:

• Some streams can only be opened once.
• It's a performance penalty if the stream takes time to be opened.

Expected Behavior

The stream should only be opened if the content length needs to be computed.

Current Behavior

It will open a stream and then close it for no purpose.

Reproduction Steps

AtomicBoolean opened = new AtomicBoolean(false);
s3Client.putObject(b -> b.bucket("bucket").key("key"), RequestBody.fromContentProvider(() -> {
    if (opened.getAndSet(true)) {
        throw new RuntimeException("Could only be opened once.");
    }
    return new ByteArrayInputStream("a".repeat(10).getBytes());
}, 10, "application/octet-stream"));

Possible Solution

SignerUtils#moveContentLength should take a ContentStreamProvider and only open the stream if it's necessary for computing the content length.

Additional Information/Context

No response

AWS Java SDK version used

2.26.22

JDK version used

openjdk version "21.0.1"

Operating System and version

Windows 10 and WSL2

[debora-ito]

@RealTYPICAL

Content-length is required for S3 PutObject, that's why the SDK will read the entire stream when no content-length is provided.

and only open the stream if it's necessary for computing the content length.

How do you know if it's necessary? Are you using a third-party solution that integrates with s3? Can you tell us more about your use case?

[RealTYPICAL]

Hi debora-tio,

Thanks for responding.

In the snippet I provided in the description, the content-length is already given as the second argument to fromContentProvider. There is a second factory method on RequestBody with the signature RequestBody#fromContentProvider(ContentStreamProvider, String) which would then need to compute the content-length from the InputStream returned from the ContentStreamProvider. This issue revolves around the fact that the stream is opened even though the content-length is already there.

This is the relevant code from SignerUtils.moveContentLength with some changes demonstrating what I mean:

public static long moveContentLength(SdkHttpRequest.Builder request, ContentStreamProvider payload /*Used to be `InputStream payload`*/) {
        Optional<String> decodedContentLength = request.firstMatchingHeader(X_AMZ_DECODED_CONTENT_LENGTH);
        if (!decodedContentLength.isPresent()) {
            String contentLength = request.firstMatchingHeader(Header.CONTENT_LENGTH).orElseGet(
                () -> String.valueOf(readAll(payload)) //<--- now the stream will only be opened if the content length header is missing
            );

            request.putHeader(X_AMZ_DECODED_CONTENT_LENGTH, contentLength)
                .removeHeader(Header.CONTENT_LENGTH);
            return Long.parseLong(contentLength);
        }

        request.removeHeader(Header.CONTENT_LENGTH);
        return Long.parseLong(decodedContentLength.get());
    }

    private static int readAll(ContentStreamProvider provider) {
        //stream only opened here
        try (InputStream inputStream = provider.newStream()) {
            byte[] buffer = new byte[4096];
            int read = 0;
            int offset = 0;
            while (read >= 0) {
                read = inputStream.read(buffer);
                if (read >= 0) {
                    offset += read;
                }
            }
            return offset;
        } catch (Exception e) {
            throw new RuntimeException("Could not finish reading stream: ", e);
        }
    }