How to Fix an Erlang Fail to Resolve IPv6 Address Using Parameters from RabbitMQ

When I first deployed a RabbitMQ cluster on Kubernetes (K8s) with only pure IPv6 addresses, I ran into an unexpected problem: RabbitMQ couldn’t resolve hostnames over IPv6, even though my cluster was fully IPv6-enabled.

The error I saw was:

{failed_connect,
  [{to_address,{"kubernetes.default.svc.cluster.local",443}},
   {inet,[inet],nxdomain}]}

At first glance, this looked like a simple DNS issue. But after digging deeper, I realized the problem was that Erlang’s HTTP client (httpc) was trying to resolve the hostname using IPv4 (inet) only, while my services only had IPv6 (inet6) available.

First Code Setup (with the Problem)

My initial RabbitMQ setup included the following environment variables:

RABBITMQ_SERVER_ADDITIONAL_ERL_ARGS="+A 128 -kernel inetrc '/etc/rabbitmq/erl_inetrc' -proto_dist inet6_tcp"
RABBITMQ_CTL_ERL_ARGS="-proto_dist inet6_tcp"

And my erl_inetrc file looked like this:

{inet6, true}.

Despite this, the rabbit_peer_discovery_k8s plugin failed when invoking the Kubernetes API. Erlang’s httpc:request still attempted IPv4 resolution, which led to:

{inet,[inet],nxdomain}

Error Explanation

This happens because:

• Erlang’s resolver (inet) defaults to IPv4 unless explicitly told to use IPv6.
• httpc does not automatically switch to IPv6 mode unless the environment forces it.
• RabbitMQ itself doesn’t control DNS resolution—the Erlang runtime does.

So, even though I had {inet6, true} in erl_inetrc, Erlang modules like httpc still defaulted to IPv4-first resolution.

How I Fix It

To fix this, I needed to force Erlang to prefer IPv6 resolution everywhere.

Update erl_inetrc

I expanded the file to look like this:

{inet6, true}.
{hosts_file, "/etc/hosts"}.
{lookup, [dns, file]}.

This explicitly prioritizes IPv6 and ensures Erlang uses DNS lookups.

Verify Container DNS Setup

I made sure that /etc/resolv.conf pointed to an IPv6-capable DNS server and that /etc/hosts included the proper entries.

Start RabbitMQ with IPv6 Parameters

I restarted RabbitMQ with:

RABBITMQ_SERVER_ADDITIONAL_ERL_ARGS="-kernel inetrc '/etc/rabbitmq/erl_inetrc' -proto_dist inet6_tcp -ssl_dist inet6_tcp"

This guaranteed that both Erlang distribution and SSL traffic used IPv6.

Test Resolver Inside the Pod

I ran:

kubectl exec -ti <rabbit-pod> -- \
  rabbitmqctl eval 'inet:gethostbyname("kubernetes.default.svc.cluster.local").'

And I finally got the expected result:

{ok,{hostent,"kubernetes.default.svc.cluster.local",[],inet6,16,
    [{64769,43981,0,0,0,0,0,1}]}}

Success! Erlang could now resolve IPv6 addresses correctly.

Extra Debugging & Practice Functionality

To better understand the gap between inet:gethostbyname/1 and httpc:request/1, I played around with Erlang directly:

% Test with inet (IPv4)
inet:gethostbyname("kubernetes.default.svc.cluster.local").

% Test with inet6 (IPv6)
inet:gethostbyname("kubernetes.default.svc.cluster.local", inet6).

% Simulate httpc request (forcing IPv6 transport)
httpc:request(get, {"https://[fd01:abcd::1]:443/api", []}, [], []).

Why this helps:

• inet:gethostbyname/1 → shows what Erlang’s DNS resolver does by default.
• inet:gethostbyname/2 → lets me force and confirm IPv6 resolution.
• httpc:request/4 → verifies how Erlang’s HTTP client behaves with IPv6 addresses.

Best Practices for RabbitMQ + IPv6 on K8s

Here are the practices I now follow whenever I deploy RabbitMQ in an IPv6-only cluster:

1. Force IPv6 in

{inet6, true}.

1. Verify DNS Config:
   Make sure /etc/resolv.conf points to an IPv6-capable DNS server.
2. Test Resolution Before Plugins:
   Run rabbitmqctl eval 'inet:gethostbyname(...).' to confirm hostnames resolve.
3. Use Bracketed IPv6 Addresses in Configs:
   For example: [fd01:abcd::1].
4. Enable Verbose Debugging:

RABBITMQ_SERVER_ADDITIONAL_ERL_ARGS="-kernel inetrc '/etc/rabbitmq/erl_inetrc' -proto_dist inet6_tcp -sasl sasl_error_logger tty"

Final Thought

At first, I thought my issue was with RabbitMQ itself. But in reality, it was about how Erlang handles DNS resolution under IPv6. Once I understood that httpc and inet behave differently, the fix was straightforward: configure erl_inetrc properly, verify the container DNS, and always test with rabbitmqctl eval. Now, my RabbitMQ cluster can resolve IPv6-only Kubernetes services without crashing, and the peer discovery plugin works seamlessly.