Checksum validation is a slow blocking operation inside a future

[jdisanti]

The current client implementation of @httpChecksum calls into a sync Checksum trait to compute and validate a checksum for requests/responses as part of the streaming body. If a chunk is large enough, calculating this checksum can take longer than a millisecond, which can cause issues for the async runtime.

How often this happens in practice is hard to say. This issue needs some investigation. If it happens often, then the size of the chunk should be examined, and if large enough, the checksum computation should be split off into a blocking thread.

[rcoh]

Chunks in practice are usually about 16KB—how large does the chunk have to be to be slow?

[jdisanti]

It's something we should measure.

[aajtodd]

IF this is in fact an issue (haven't looked into it) could be a fit for annotating with something like https://github.com/jlizen/vacation