According to the instance listing for deployment-prep at https://wikitech.wikimedia.org/wiki/Special:NovaInstance, deployment-cache-text04 should be the Varnish instance we are interested in.

deployment-cache-text04 does have the vhtcpd purger daemon up and running, but it's mutlicast config points to the production one:

root@deployment-cache-text04:/# ps -ef | grep htcp
vhtcpd    1662     1  0 Aug18 ?        00:01:39 /usr/sbin/vhtcpd -m 239.128.0.112 -c 127.0.0.1:80 -c 127.0.0.1:3128 start

The IP is hard-coded in the Puppet manifest.

I have tried to issue some requests to it, but couldn't see anything in the varnish logs.

I've built vhtcpd locally, and started it with the command:

./src/vhtcpd -m 224.0.0.0 -c 127.0.0.1:8000 startfg

Then, I've used our command-line utility to send HTCP packets like this:

node ./maintenance/purge.js 127.0.0.1:4827 http://test.com/test.html

Finally, I've made up a very simple HTTP server on 8000 port, that just logs the contents of a received request, so, for the previous HTCP request I get the following HTTP request from daemon:

{
    method: 'purge',
    uri: '/test.html'
    headers: {
        host: 'test.com',
        'user-agent': 'vhtcpd'
    }
}

This appears to be correct, and vhtcpd daemon doesn't crash, so at least this part of the puzzle seem to work.

To test that sending HTCP to a multicast address locally works, I've had to change the HTCP purger code a little bit to enable an IP_MULTICAST_LOOP option. After enabling it, I can join vhtcpd to any multicast group, and use it to send HTCP packets, everything works correctly.

As I understand, in production we don't need to loopback, right?

Thnx for the feedback @Pchelolo! I forgot to mention in my previous comment that in spite of the production multicast IP in labs, I could verify that packets are successfully received and correctly interpreted by vhtcpd.

The HTTP requests produced are indeed correct @Pchelolo.

In T113235#1666437, @Pchelolo wrote:

As I understand, in production we don't need to loopback, right?

Nope, that's already set up there.

I've also deployed PR 317 in the Beta-Cluster-Infrastructure and the logic works as advertised.