natalie-lang.org server.rb doesn't work with latest Natalie

[seven1m]

Starting in 2ef5771 the payload returned from https://natalie-lang.org/stats/stats.json is truncated. Need to investigate...

[seven1m]

Also, 9 out of 10 of the child processes see the following error for every socket connection:

["Errno::EAGAIN", "Resource temporarily unavailable", ["server.rb:41:in `accept'", "server.rb:41:in `block in block in block in block'", "server.rb:40:in `loop'", "server.rb:40:in `block in block in block'", "server.rb:39:in `fork'", "server.rb:39:in `block in block'", "src/enumerable.rb:624:in `block in collect'", "src/enumerable.rb:624:in `each'", "src/enumerable.rb:624:in `collect'", "server.rb:38:in `<main>'"]]["Errno::EAGAIN", "Resource temporarily unavailable", ["server.rb:41:in `accept'", "server.rb:41:in `block in block in block in block'", "server.rb:40:in `loop'", "server.rb:40:in `block in block in block'", "server.rb:39:in `fork'", "server.rb:39:in `block in block'", "src/enumerable.rb:624:in `block in collect'", "src/enumerable.rb:624:in `each'", "src/enumerable.rb:624:in `collect'", "server.rb:38:in `<main>'"]]

It seems that server.accept returns one good connection and raises an error for all the other children waiting.

[funny-falcon]

It is well known issue with non-blocking accept from many threads/processes: kernel wakes all epool/poll/select waiters of socket, but only one succeed.

Nginx works around by using cross-process lock around waiting for accept. Some other servers use blocking listen socket and do accept in separate threads (iirc, Rainbow did it).