openresty
could not be resolved (3: Host not found)
I deployed an OpenResty Pod in kubernetes, and I use "ngx.socket.tcp" to request another application of the current kubernetes cluster in the format of "serviceName.namespace", but could not be resolved (3: Host not found) error, why is this? My sample code is as follows:
local ok, err = ngx.sokcet.connect("echo.eplat", 80)
if not ok then
ngx.log(err) -- could not be resolved (3: Host not found)
end
The resolver directive has been set in my Nginx configuration file, and the configuration is as follows:
http{
...
resolver ipv6=off local=on;
...
}
When I use the proxy_pass directive of Nginx as a proxy, I can access the target service. and the configuration is as follows:
...
location =/test {
proxy_pass http://echo.eplat;
}
...
bash-5.1$ curl -X GET http://localhost:9080/test
Hostname: echo-7cc69d9cd8-27xhd
Pod Information:
node name: minikube
pod name: echo-7cc69d9cd8-27xhd
pod namespace: eplat
pod IP: 172.17.0.9
Server values:
server_version=nginx: 1.13.3 - lua: 10008
Request Information:
client_address=::ffff:172.17.0.1
method=GET
real path=/test
query=
request_version=1
request_scheme=http
request_uri=http://echo.eplat:80/test
Request Headers:
accept=*/*
connection=close
host=echo.eplat
user-agent=curl/7.80.0
Request Body:
-no body in request-
bash-5.1$
I can ping this address directly inside the container, and it can also be pinged：
bash-5.1$ ping echo.eplat
PING echo.eplat (10.102.19.162): 56 data bytes
ping: permission denied (are you root?)
I checked OpenResty's documentation (https://github.com/openresty/lua-nginx-module#tcpsockconnect), which clearly states that the first positional parameter can be in domain name format. If it is in domain name format, this method will use Nginx The core dynamic resolver resolves the domain name without blocking, and the resolver nginx.conf directive needs to be configured in the file; at this point, I am very confused, why the proxy_pass directive of Nginx can proxy this address, while the ngx.socket. Connect can't resolve this address? Don't they all use Nginx's dns resolver?
It sounds wired, Can you try to Tcpdump to capture your container traffic to analyze DNS requests? I am sure openresty cosocket will use the nginx resolver to dynamic resolver DNS name.
I also encountered this problem, please help me solve it。
│ 2023/10/25 07:56:31 [error] 10#10: *42 [lua] gfe.lua:256: pass(): connect error: engine-master.app-xx.svc.cluster.local:2202 could not be resolved (3: Host │
same here, please if you fixed the issue, comment the solution. I will greatly appreciate your time
@GitDemis, hi
In a Kubernetes environment, you'll need to use the FQDN echo.eplat.svc.cluster.local in the ngx.socket.connect function.
The root cause here is that Nginx's resolver doesn't support ndots or search lists —it rigidly asks CoreDNS to resolve just echo.eplat, which predictably fails.
On the other hand, proxy_pass works because Nginx uses getaddrinfo during startup to resolve echo.eplat, and that function's underlying implementation does support ndots and search lists. It will try resolving echo.eplat.svc.cluster.local via the CoreDNS server, leading to a successful result.
This issue is pretty stale at this point, but I hope this explanation still helps!
