Closed stereoscott closed 10 years ago
TL;DR - go back to 0.5.2 :) I don't quite understand this bug myself. For the best of my understanding, this is not needed: https://github.com/romanbsd/heroku-deflater/commit/10ca119b59ea3f51fc004a0b86012bad95ae98a9 , however, many people were complaining again and again, so I just did it to make them happy. I still don't know what was/is the core issue here, and unfortunately I don't have time to invest into it. Having said that, I'm still very curious to know what is going on there, and pull requests are always welcomed.
Very interesting indeed. Thanks for the reply and for the great plugin.
TL;DR After updating to 0.5.3, nearly all of my asset requests were with a
closed stream (IOError)
Stack trace:
I read up on this a bit and found a thread here (http://rubyforge.org/pipermail/mongrel-unicorn/2013-September/001874.html) that mentions:
I dug into unicorn and rack/lint a bit and both seem to call
body.close
on their own, which apparently is something that all rack middleware should do, but now I'm getting complaints that the steam is already closed.I then disabled unicorn and just went through webrick. Same issue:
I think that
Rack::Deflater
is the middleware that is complaining because it's the middleware that runs immediately afterHerokuDeflater::ServeZippedAssets
:I'm happy to go back to 0.5.2 but I'm curious what the cause of this issue is, and how we might solve it in a way that doesn't cause requests to fail. Happy to help in whatever way I can.