Slow S3 blobstore docker image ingest rate

Nexus 3.12.1

When loading an S3 blobstore with docker images in an Amazon VPC, we experience a bottleneck, as though there was only one connection to S3 for all the bits to go through.

Poring over the documentation, I see:

Pushing large images can result in failures due to network interruptions and other issues. These partial uploads result in temporary storage for these transfers in the repository manager filling up.

Is this related? Are there ways to configure Nexus to speed things up?


This may be due to this issue:

If you can get a thread dump in Nexus while the uploads are running you should be able to compare the threads to see if they match the above issue. The issue above is fixed in the upcoming 3.13.0 release.