Bug #14713: <internal:prelude>:132:in `__write_nonblock': Protocol wrong type for socket (Errno::EPROTOTYPE)

• Status: Closed
• Assignee: -
• Target version: -
• Backport: 2.3: REQUIRED, 2.4: REQUIRED, 2.5: DONE

[ioquatix (Samuel Williams)]

Very occasionally on macOS, using Ruby 2.5.0, I get the above error.

Traceback (most recent call last):
	23: from /Users/samuel/.rvm/gems/ruby-2.5.0/gems/async-1.6.0/lib/async/task.rb:74:in `block in initialize'
	22: from /Users/samuel/.rvm/gems/ruby-2.5.0/gems/async-io-1.7.0/lib/async/io/socket.rb:79:in `block in accept'
	21: from /Users/samuel/.rvm/gems/ruby-2.5.0/gems/async-io-1.7.0/lib/async/io/socket.rb:46:in `block in accept_each'
	20: from /Users/samuel/.rvm/gems/ruby-2.5.0/gems/async-http-0.20.0/lib/async/http/server.rb:51:in `accept'
	19: from /Users/samuel/.rvm/gems/ruby-2.5.0/gems/async-http-0.20.0/lib/async/http/server.rb:51:in `catch'
	18: from /Users/samuel/.rvm/gems/ruby-2.5.0/gems/async-http-0.20.0/lib/async/http/server.rb:52:in `block in accept'
	17: from /Users/samuel/.rvm/gems/ruby-2.5.0/gems/async-http-0.20.0/lib/async/http/protocol/http1.rb:59:in `receive_requests'
	16: from /Users/samuel/.rvm/gems/ruby-2.5.0/gems/async-http-0.20.0/lib/async/http/protocol/http11.rb:75:in `receive_requests'
	15: from /Users/samuel/.rvm/gems/ruby-2.5.0/gems/async-http-0.20.0/lib/async/http/server.rb:54:in `block (2 levels) in accept'
	14: from /Users/samuel/Documents/socketry/falcon/lib/falcon/server.rb:33:in `handle_request'
	13: from /Users/samuel/Documents/socketry/falcon/lib/falcon/verbose.rb:46:in `call'
	12: from /Users/samuel/.rvm/gems/ruby-2.5.0/gems/async-http-0.20.0/lib/async/http/middleware.rb:45:in `call'
	11: from /Users/samuel/Documents/socketry/falcon/lib/falcon/proxy.rb:71:in `call'
	10: from /Users/samuel/.rvm/gems/ruby-2.5.0/gems/async-http-0.20.0/lib/async/http/client.rb:65:in `call'
	 9: from /Users/samuel/.rvm/gems/ruby-2.5.0/gems/async-http-0.20.0/lib/async/http/protocol/http11.rb:98:in `call'
	 8: from /Users/samuel/.rvm/gems/ruby-2.5.0/gems/async-http-0.20.0/lib/async/http/protocol/http11.rb:112:in `write_request'
	 7: from /Users/samuel/.rvm/gems/ruby-2.5.0/gems/async-io-1.7.0/lib/async/io/stream.rb:76:in `flush'
	 6: from /Users/samuel/.rvm/gems/ruby-2.5.0/gems/async-io-1.7.0/lib/async/io/stream.rb:175:in `syswrite'
	 5: from /Users/samuel/.rvm/gems/ruby-2.5.0/gems/async-io-1.7.0/lib/async/io/generic.rb:47:in `block in wrap_blocking_method'
	 4: from /Users/samuel/.rvm/gems/ruby-2.5.0/gems/async-io-1.7.0/lib/async/io/generic.rb:124:in `async_send'
	 3: from /Users/samuel/.rvm/gems/ruby-2.5.0/gems/async-io-1.7.0/lib/async/io/generic.rb:131:in `async'
	 2: from /Users/samuel/.rvm/gems/ruby-2.5.0/gems/async-io-1.7.0/lib/async/io/generic.rb:125:in `block in async_send'
	 1: from <internal:prelude>:132:in `write_nonblock'

I'm not even sure how it's possible to receive this message in write_nonblock since by that point the socket is well and truely setup. Maybe a bug in the OS?

I'm not sure how to reproduce, I've only seen it once a month and I don't have any strategy to reproduce yet. I could try a busy loop.

[nobu (Nobuyoshi Nakada)]

ioquatix (Samuel Williams) wrote:

Very occasionally on macOS, using Ruby 2.5.0, I get the above error.

It happens often on macOS.

I'm not even sure how it's possible to receive this message in write_nonblock since by that point the socket is well and truely setup. Maybe a bug in the OS?

Probably, but not sure.

[ioquatix (Samuel Williams)]

It happens often on macOS.

I don't have this experience. I can't systematically reproduce the issue. It only happens perhaps once a month even thought I'm running networking code almost daily on macOS, and these are code paths that do occasionally show the issue..

It would be nice to figure out a busy loop that can reproduce the issue on demand, even if it takes a few minutes.

I'm running on macOS 10.13.4

[ioquatix (Samuel Williams)]

I tried digging into this issue.

I found some other people with similar problems: https://github.com/nodejs/node/issues/2382

The most interesting article digging into the issue: http://erickt.github.io/blog/2014/11/19/adventures-in-debugging-a-potential-osx-kernel-bug/

I'm still digesting this information.

[ioquatix (Samuel Williams)]

It seems that in this specific case, EPROTOTYPE is the same as EPIPE. It's a bug in the kernel code. I wonder if we should compensate for it by detecting EPROTOTYPE by write, and raising EPIPE.

[nobu (Nobuyoshi Nakada)]

• Related to Bug #12914: FTPTest#test_list_read_timeout_exceeded causes EPROTOTYPE on Mac OS X 10.10 added

[ioquatix (Samuel Williams)]

I've made a PR which fixes the issue.

https://github.com/ruby/ruby/pull/1868

I believe it would make sense to back-port it.

Additionally the following change could be reverted: https://bugs.ruby-lang.org/issues/12914

[ioquatix (Samuel Williams)]

I made a patch for write_nonblock. But this probably equally applies to send, sendmsg, etc. Perhaps we can discuss the right course of action to cover all these cases.

[nobu (Nobuyoshi Nakada)]

• Status changed from Open to Closed

Applied in changeset trunk|r63304.

---

io.c: workaround for EPROTOTYPE

• io.c (internal_write_func, internal_writev_func): retry at
  unexpected EPROTOTYPE on macOS, to get rid of a kernel bug.
  [ruby-core:86690] [Bug #14713]

• ext/socket/init.c (rsock_{sendto,send,write}_blocking): ditto.

[ioquatix (Samuel Williams)]

Thanks for solving this problem. Your patch was much more thorough.

Any change to back port this?

[nobu (Nobuyoshi Nakada)]

• Backport changed from 2.3: UNKNOWN, 2.4: UNKNOWN, 2.5: UNKNOWN to 2.3: REQUIRED, 2.4: REQUIRED, 2.5: REQUIRED

[nagachika (Tomoyuki Chikanaga)]

• Backport changed from 2.3: REQUIRED, 2.4: REQUIRED, 2.5: REQUIRED to 2.3: REQUIRED, 2.4: REQUIRED, 2.5: DONE

ruby_2_5 r63826 merged revision(s) 63304.