test: Error: getaddrinfo ENOTFOUND localhost

[AdamMajer]

Version

18.6.0

Platform

openSUSE Tumbleweed

Subsystem

No response

What steps will reproduce the bug?

Currently the failures are visible in the current build log,

https://build.opensuse.org/public/build/devel:languages:nodejs/openSUSE_Tumbleweed/x86_64/nodejs18/_log

How often does it reproduce? Is there a required condition?

No response

What is the expected behavior?

No response

What do you see instead?

not ok 1995 parallel/test-net-socket-connect-without-cb
  ---
  duration_ms: 0.506
  severity: fail
  exitcode: 1
  stack: |-
    node:events:491
          throw er; // Unhandled 'error' event
          ^
    
    Error: getaddrinfo ENOTFOUND localhost
        at GetAddrInfoReqWrap.onlookup [as oncomplete] (node:dns:111:26)
    Emitted 'error' event on Socket instance at:
        at emitErrorNT (node:internal/streams/destroy:151:8)
        at emitErrorCloseNT (node:internal/streams/destroy:116:3)
        at process.processTicksAndRejections (node:internal/process/task_queues:82:21) {
      errno: -3008,
      code: 'ENOTFOUND',
      syscall: 'getaddrinfo',
      hostname: 'localhost'
    }
    
    Node.js v18.6.0
  ...

...

not ok 2000 parallel/test-net-socket-ready-without-cb
  ---
  duration_ms: 0.507
  severity: fail
  exitcode: 1
  stack: |-
    node:events:491
          throw er; // Unhandled 'error' event
          ^
    
    Error: getaddrinfo ENOTFOUND localhost
        at GetAddrInfoReqWrap.onlookup [as oncomplete] (node:dns:111:26)
    Emitted 'error' event on Socket instance at:
        at emitErrorNT (node:internal/streams/destroy:151:8)
        at emitErrorCloseNT (node:internal/streams/destroy:116:3)
        at process.processTicksAndRejections (node:internal/process/task_queues:82:21) {
      errno: -3008,
      code: 'ENOTFOUND',
      syscall: 'getaddrinfo',
      hostname: 'localhost'
    }
    
    Node.js v18.6.0
  ...

not ok 2591 parallel/test-tcp-wrap-listen
  ---
  duration_ms: 0.818
  severity: fail
  exitcode: 1
  stack: |-
    (node:18408) internal/test/binding: These APIs are for internal testing only. Do not use them.
    (Use `node18 --trace-warnings ...` to show where the warning was created)
    node:events:491
          throw er; // Unhandled 'error' event
          ^
    
    Error: getaddrinfo ENOTFOUND localhost
        at GetAddrInfoReqWrap.onlookup [as oncomplete] (node:dns:111:26)
    Emitted 'error' event on Socket instance at:
        at emitErrorNT (node:internal/streams/destroy:151:8)
        at emitErrorCloseNT (node:internal/streams/destroy:116:3)
        at process.processTicksAndRejections (node:internal/process/task_queues:82:21) {
      errno: -3008,
      code: 'ENOTFOUND',
      syscall: 'getaddrinfo',
      hostname: 'localhost'
    }
    
    Node.js v18.6.0
  ...

Additional information

> cat /etc/hosts
127.0.0.1 localhost
#
# hosts         This file describes a number of hostname-to-address
#               mappings for the TCP/IP subsystem.  It is mostly
#               used at boot time, when no name servers are running.
#               On small systems, this file can be used instead of a
#               "named" name server.
# Syntax:
#    
# IP-Address  Full-Qualified-Hostname  Short-Hostname
#

127.0.0.1       localhost
# fallback hostname used by NetworkManager
127.0.0.1       localhost.localdomain

# special IPv6 addresses
::1             localhost ipv6-localhost ipv6-loopback

fe00::0         ipv6-localnet

ff00::0         ipv6-mcastprefix
ff02::1         ipv6-allnodes
ff02::2         ipv6-allrouters
ff02::3         ipv6-allhosts

and resolv.conf doesn't exist. All other test cases that use localhost are passing, as expected.

[kvakil]

curious what you see when doing strace. On Ubuntu I see the following:

$ strace -ff ./node test/parallel/test-tcp-wrap-listen.js |& grep localhost
[pid  4385] read(19, "127.0.0.1 localhost\n\n# The follo"..., 4096) = 221
[pid  4385] sendto(19, "\362*\1 \0\1\0\0\0\0\0\1\tlocalhost\3ec2\10inter"..., 51, MSG_NOSIGNAL, NULL, 0) = 51
[pid  4385] recvfrom(19, "\362*\201\203\0\1\0\0\0\0\0\1\tlocalhost\3ec2\10inter"..., 1024, 0, {sa_family=AF_INET, sin_port=htons(53), sin_addr=inet_addr("127.0.0.53")}, [28 => 16]) = 51
[pid  4385] sendto(19, "Uk\1 \0\1\0\0\0\0\0\1\tlocalhost\0\0\34\0\1\0\0)\4\260"..., 38, MSG_NOSIGNAL, NULL, 0) = 38
[pid  4385] recvfrom(19, "Uk\205\240\0\1\0\1\0\0\0\1\tlocalhost\0\0\34\0\1\300\f\0\34\0"..., 1024, 0, {sa_family=AF_INET, sin_port=htons(53), sin_addr=inet_addr("127.0.0.53")}, [28 => 16]) = 66

[theanarkh]

I get this error sometimes on my pc, but not always.

[bnoordhuis]

ENOTFOUND is a synthetic error that node fabricates when it doesn't get an answer back from getaddrinfo(3), it's not a real operating system-level error. Reasons include running out of file descriptors in libc (not visible to node), not getting back a response from the DNS server, etc. It's basically a catch-all.

[AdamMajer]

In this case, there is no DNS server and it works with other tests...

strace shows nothing. This is only a problem since about 18.2.0 and not before and not in any other released line. And I only see it in a VM build environment and not in a chroot (at least it disappeared there) which is very strange.

I will try to bisect the issue.

[bnoordhuis]

You mean v18.1.0 works? There are no changes between the two that seem like likely culprits. I'm interested to see what bisect turns up.

[AdamMajer]

So, the bisect ended up on 70b516e which is #43054 . Looking at the pull request, it kind of makes sense this is the one as the tests in question are explicitly skipped on some platforms.

https://github.com/nodejs/node/pull/43054/files#diff-5ba63fa1f910a3ec440b5174993719278ddca69569c0159e2c782173af190602

The test is reproducible every time on openSUSE:Tumbleweed build VM. I can't reproduce it outside of it on a workstation, but in the VM with no networking connections and only a hosts file to resolve localhost to ipv6 and ipv4, it reproduces every time.

I will try to look more closely what is actually failing here.

[AdamMajer]

This was introduced in 18.4.0. I mistyped earlier the versions. 18.2.0 still passed

[kapouer]

Is there no network interface at all, no loopback ?

[AdamMajer]

So, I've learned something about debugging node over last few days :-) But I did find the problem.

> NODE_DEBUG=net ./node18 test/parallel/test-net-socket-ready-without-cb.js
NET 1834: setupListenHandle ::1 0 6 false undefined
NET 1834: setupListenHandle: create a handle
NET 1834: bind to ::1
{ address: '::1', family: 'IPv6', port: 32841 }
NET 1834: pipe false undefined
NET 1834: connect: find host localhost
NET 1834: connect: dns options { family: 'IPv6', hints: 32 }

The problem is that family is set to a string, which causes hints to be set to 32 which means AI_ADDRCONFIG and then address resolution fails.

On a network connected device, this passes.

Why? We can look at the manpage for getaddrinfo

If hints.ai_flags includes the AI_ADDRCONFIG flag, then IPv4 addresses are returned in the list pointed to by res only if the local system has at least one IPv4 address configured, and IPv6 addresses are returned only if the local system has at least one IPv6 address configured. The loopback address is not considered for this case as valid as a configured address. This flag is useful on, for example, IPv4-only systems, to ensure that getaddrinfo() does not return IPv6 socket addresses that would always fail in connect(2) or bind(2).

This is why these tests fail every time when you have a loopback interface as the only network interface active. They will pass if you have other network interfaces active.