[JCLOUDS-1521] Automatic computation of content length for input streams

• Type: Improvement
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: 2.1.1
• Component/s: jclouds-blobstore
• Assignee: Unassigned
• Reporter: Alexander Tsvetkov
• Created: 25/Oct/19 06:49
• Updated: 11/Apr/22 12:44

Description

I have a REST API that allows upload of potentially large files (up to 4GBs). Due to the size, I cannot load these files in-memory, as that could quickly crash my application. I also don't want to store them as temporary files, since that could fill up my disk if a lot of people decide to upload at the same time.

Instead, I want to process the incoming files as InputStreams and forward them to the S3 object store. I understand that this is not possible directly, since S3 requires the content length to be known before the upload. However, I saw on StackOverflow (https://stackoverflow.com/questions/8653146/can-i-stream-a-file-upload-to-s3-without-a-content-length-header) that it's possible to workaround this problem by reading the InputStream in memory in chunks of 5 (or more) MBs and uploading these chunks via the S3 multipart upload API. As a result, I assume that I'll be able to upload a 4GB file, by having no more than 5 MBs of its content stored in-memory at any given time.

I tried to do so with JClouds (version 2.1.1), but I've hit a problem. I have the following code:

Blob blob = blobStore.blobBuilder(name)
    .payload(inputStream)
    ...
    .build();
blobStore.putBlob(container, blob, PutOptions.Builder.multipart());

If I run it like this, I get a NullPointerException, because I didn't specify the content's length:

java.lang.NullPointerException: while trying to invoke the method java.lang.Long.longValue() of a null object returned from org.jclouds.io.MutableContentMetadata.getContentLength()
   at org.jclouds.blobstore.internal.BaseBlobStore.putMultipartBlob(BaseBlobStore.java:356)
   at org.jclouds.blobstore.internal.BaseBlobStore.putMultipartBlob(BaseBlobStore.java:347)
   at org.jclouds.aws.s3.blobstore.AWSS3BlobStore.putBlob(AWSS3BlobStore.java:79)

I think it would be possible for JClouds to compute the size of the InputStream dynamically:

1. Slice the stream into chunks of X MBs and store the chunks in-memory (where X has a default value but is also configurable).
2. Upload the chunks sequentially - the content length header can be set to X MBs.
3. Finalize the multipart upload.

That way, no more than X MBs will be stored in memory for any given upload.

Would you accept a pull request for this?

PS: I've set the priority to blocker, because we really can't use JClouds for our upload right now, because of the memory and disk space concerns listed above.