I’m having the same issue, mentioned in another thread with a less appropriate title.
My (perhaps incorrect) impression is that this relates to the (mis)handling of compressed files, based on the introduction of support for brotli compression in 0.21.0 and a post I saw somewhere (but can’t find now) about the protocol getting caught in a loop while trying to validate a chunk.
The release notes for 0.21.0 suggest that there was some update to error handling, where “URLs that contain invalid encodings now return 400 Bad Request instead of 500 Internal Server Error”, which I take as a clue.
This cropped up for me while I was updating the header encodings, so it led me to think it is related. My .ic-assets.json file is in another post but @Severin believes this is not related, as this is on the application level and the error suggests it has deeper roots.
I am open to suggestions as this is currently blocking development. What would be the steps for a rigourous cleaning of all cache, both locally and existing on the canister (frontend) that could rule out this being triggered by artefacts from previous builds? I have tried deleting dist and .dfx directories, but this again seems to be at the application level, so might be the wrong tack.