s3-sync-to-host super slow with lots of files

[joehoyle]

If I have a lot of files in the local S3 (from an import), starting the local-server takes an age (15+ minutes) due to s3-sync-to-host.

It seems every time we start the local-server, we are trying to pull all files over to the host, is that right? If so, I'm wondering why we would want to do that.

cc @roborourke as I think you know about this.

[roborourke]

Can you quantify what a lot of files is more precisely, and how you narrowed the slowness down to s3-sync-to-host?

s3-sync-to-host is a background service and isn't depended on by any other container so it shouldn't block the server coming up, it runs the Minio Client mirror command, which is like rsync. It shouldn't be copying anything that doesn't need to be copied so we're not trying to copy all files every time you start it up.

We have this in place to make the dev experience smoother when starting / destroying Local Server. Images uploaded would not appear on the host machine which caused some confusion, see #140

That said, between the import-uploads command and the fact that Minio has a built in S3 browser that could be a solution to original problem described, we would just need to document clearly why uploads don't appear on the host how to view files in S3 via the browser.

There's an annoying dependency on this sync to host functionality with the Elasticsearch packages feature as they have to be uploaded to S3 but then made available / mounted on the ES container - that requires syncing to the host.