Puma leaking Temp Files for large payloads

[virgoproz]

Using a fairly standard Rails 4.2.X, Grape API, Puma stack, Ruby 2.3.1

In production we are seeing a leakage of large temp files containing the request body in clear (POST with large bodies). After a few days we are forced to delete these files manually (or write a hack to delete periodically). Puma should call .unlink or .delete on whatever temp file it decides to create.

[janko]

That's weird, because Puma does unlink created tempfiles, even in case of an error.

However, it seems there are two places where Puma creates tempfiles, in Puma::Server#read_body and in Puma::Client#setup_body, and only the one created by Puma::Client seems to get unlinked. I don't know when does Puma::Server#read_body get called.

[nateberkopec]

Please reopen if you can reporduce this.

[smcgivern]

Because the unlink call happens in an ensure, it's possible for the tempfile to stick around if the process receives SIGKILL (or the ensure doesn't run for any other reason; this is just the easiest way to explain it) before that happens - say, the request is still in progress.

I created #2613 to cover that case, even if it wasn't the root cause of this issue.

[dentarg]

With #2613 merged, I think we can close this again.