in_forward plugin bind on both ipv4 and ipv6

[inistor]

Describe the bug
While using the in_forward plugin, I’m trying to have it bind on both IPv4 and IPv6 addresses (for dual-stack systems); typically, when using. “::” for bind config parameter, the socket api will bind to both (as long as AF_INET6 is used).
However, if I use “::” for bind config parameter, an IPv6 only socket will be created.
With the default value of bind (“0.0.0.0”), an IPv4 only socket is created.

To Reproduce
Using the configuration below:

<source>
        @type forward
        bind "::"
</source>

ipv6 works

[root@centos8-ion ~]# telnet -6 localhost 24224
Trying ::1...
Connected to localhost.
Escape character is '^]'.

ipv4 does not:

[root@centos8-ion ~]# telnet -4 localhost 24224
Trying 127.0.0.1...
telnet: connect to address 127.0.0.1: Connection refused

A workaround is to use two sections:

<source>
        @type forward
        bind "::"
</source>
<source>
        @type forward
        bind "0.0.0.0"
</source>

Expected behavior
When binding on "::" (IPv6 name for any address), IPv4 should also be bound on any address.
Your Environment

• Fluentd or td-agent version: fluentd --version or td-agent --version

[root@centos8-ion ~]# fluentd --version
fluentd 1.10.4

• Operating system: cat /etc/os-release

[root@centos8-ion ~]# cat /etc/os-release 
NAME="CentOS Linux"
VERSION="8 (Core)"
ID="centos"
ID_LIKE="rhel fedora"
VERSION_ID="8"
PLATFORM_ID="platform:el8"
PRETTY_NAME="CentOS Linux 8 (Core)"
ANSI_COLOR="0;31"
CPE_NAME="cpe:/o:centos:centos:8"
HOME_URL="https://www.centos.org/"
BUG_REPORT_URL="https://bugs.centos.org/"

CENTOS_MANTISBT_PROJECT="CentOS-8"
CENTOS_MANTISBT_PROJECT_VERSION="8"
REDHAT_SUPPORT_PRODUCT="centos"
REDHAT_SUPPORT_PRODUCT_VERSION="8"

• Kernel version: uname -r

[root@centos8-ion ~]# uname -r
4.18.0-147.8.1.el8_1.x86_64

Your Configuration

<source>
        @type forward
        bind "::"
</source>

Issue discussed on fluentd Google Group here.

[inistor]

Hi there,

Has anyone taken a look over the issue described above?

Thanks,
Ion

[github-actions]

This issue has been automatically marked as stale because it has been open 90 days with no activity. Remove stale label or comment or this issue will be closed in 30 days

[inistor]

@repeatedly hi there; any updates here?

[minusbat]

Just found this, as hitting the same issue. No updates I assume ?

[ashie]

FYI: From #2902

Describe the solution you'd like

• When specifying a hostname like localhost, all relevant addresses are listend/bound. For example:

irb(main):016:0> Addrinfo.getaddrinfo('localhost', 5140).select { |x| x.socktype == Socket::SOCK_DGRAM }           
=> [#<Addrinfo: [::1]:5140 UDP (localhost)>, #<Addrinfo: 127.0.0.1:5140 UDP (localhost)>]

[ashie]

We need to modify ServerEngine to realize it:

https://github.com/treasure-data/serverengine/blob/08193f12e8a36518a35244b94f2865d974538450/lib/serverengine/socket_manager_unix.rb#L52-L70

[jaymzh]

So #2902 got de-duped here, which is cool, but I wanna clarify there's 2 different use-cases being discussed:

1. :: should be all-families, all-addresses
2. should be all families that a name resolves to (ala comment in_forward plugin bind on both ipv4 and ipv6 #3001 (comment) where @ashie is quoting my other ticket)

The first is table-stakes, but the latter should also work.

[inistor]

Correct - the first item refers to how a numeric address is handled, while the second involves DNS resolution.

[soostdijck]

I just ran into this on a v6 only k8s cluster. It's been a while, is this going to make the roadmap any time soon?

[Athishpranav2003]

@daipom

i guess this issue got resolved in Webrick itself as per this comment
ruby/webrick#99

But switching out of Webrick is the correct direction i guess
#4648

I am not sure if this is 100% correct but i guess this is the right way

[daipom]

@Athishpranav2003 Thanks!
So, do you think we should enhance this feature after #4648?

[Athishpranav2003]

I guess this issue is simply solved by either #4648 or by simply updating webrick. But need to check it after the changes again

[daipom]

Fluentd uses WEBrick mainly for the RPC feature.
The server-helper does not use it.
So, WEBrick appears to have nothing to do with this issue.