[ruby-core:103016] [Ruby master Bug#17748] Ruby 3.0 takes a long time to resolv DNS of nonexistent domains

[jean.boussier@gmail.com]

Issue #17748 has been updated by byroot (Jean Boussier).


I ran a quick profiler:

All the time is spent waiting on the socket to be readable:
```
==================================
  Mode: wall(1000)
  Samples: 75017 (0.00% miss rate)
  GC: 1 (0.00%)
==================================
     TOTAL    (pct)     SAMPLES    (pct)     FRAME
     75016 (100.0%)       75016 (100.0%)     IO#wait_readable
...
```

The long time it takes it likely the default socket timeout of the system.

That wait_readable is in `Resolv::DNS::Requester#request`

```
IO#wait_readable (<cfunc>:1)
  samples:  75016 self (100.0%)  /   75016 total (100.0%)
  callers:
    75016  (  100.0%)  Resolv::DNS::Requester#request
```

Likely this one: https://github.com/ruby/ruby/blob/ad8842c06d26ee634f90008efecf1cd4d76342df/lib/resolv.rb#L677

----------------------------------------
Bug #17748: Ruby 3.0 takes a long time to resolv DNS of nonexistent domains
https://bugs.ruby-lang.org/issues/17748#change-91081

* Author: xdmx (Eric Bloom)
* Status: Open
* Priority: Normal
* ruby -v: ruby 3.0.0p0 (2020-12-25 revision 95aff21468) [x86_64-linux]
* Backport: 2.5: UNKNOWN, 2.6: UNKNOWN, 2.7: UNKNOWN, 3.0: UNKNOWN
----------------------------------------
I'm running into a subtle bug when trying to resolv a nonexistent domain.
```
$ ruby -v          
ruby 2.7.2p137 (2020-10-01 revision 5445e04352) [x86_64-linux]
$ irb
require 'resolv'
=> true
t = Time.now ; Resolv::DNS.open { |dns| dns.getresources('thisisaninvaliddomain.com', Resolv::DNS::Resource::IN::A) } ; Time.now - t
=> 0.051897333
```

This works fine and it's fast, but as soon as I try to run this on 3.0:
```
$ rbenv local 3.0.0
$ ruby -v
ruby 3.0.0p0 (2020-12-25 revision 95aff21468) [x86_64-linux]
$ irb
require 'resolv'
=> true
t = Time.now ; Resolv::DNS.open { |dns| dns.getresources('thisisaninvaliddomain.com', Resolv::DNS::Resource::IN::A) } ; Time.now - t
=> 76.314624548
t = Time.now ; Resolv::DNS.open { |dns| dns.getresources('domain.com', Resolv::DNS::Resource::IN::A) } ; Time.now - t
=> 0.081165397
```

In this case the resolver takes a very long time to return a result, and it only happens with nonexistent domains. For those that exist it's running fast.

What's weird is that the only change I do is switching from 2.7 to 3.0, and then I suddenly have this problem.

The above code was run on my local machine (Arch Linux) with Ruby installed through rbenv. I've also double checked it on a completely different machine (server running Ubuntu 18.04) that is connected from a different country and I'm having the exact same problem. In that case Ruby was compiled directly from the source and installed on the server.

So this should potentially exclude any connection issue (as locally with 2.7 is fast, and it was tried in 2 different places), the way it was installed (rbenv vs source) and the distro (Arch vs Ubuntu). I asked someone running macos and ruby 3.0 and it seems that it was fast there, so maybe it's only a linux related bug, I wasn't able to get other people to try.

Resolving the domain with `resolvctl` is also running fast:
```
$ time resolvectl query thisisaninvaliddomain.com
thisisaninvaliddomain.com: resolve call failed: 'thisisaninvaliddomain.com' not found
resolvectl query thisisaninvaliddomain.com  0.00s user 0.01s system 71% cpu 0.009 total
```



-- 
https://bugs.ruby-lang.org/