I worked on the complete migration of the LibreOffice extensions and templates website repository to Python 3 and Plone 5.2.x during the last days. I was able to upgrade the staging (development / single instance) buildout first. I worked on some Plone add-ons to reach this goal in my spare time during the last weeks.
But for the success of the migration to Python 3 and Plone 5.2.x is necessary a working production buildout. Thus I worked on the live.cfg buildout script for the site and created a new version of it: live2.cfg. I used this new script for the setup of a pure zeoserver with four instances. I could also add the supervisor controller to start and stop the zeoserver and its instances.
Then I worked on adding the support of Varnish to it. I used the add-on plone.recipe.varnish from the Github repository for that and changed Varnish to the a current LTS version 6.0.6. I could start varnishd with supervisor although I got an error message with supervisorctl status for it. The Varnish log stated that it could not get the socket of the port, it is configured to run on.
I looked with netstat -pln into the processes running on my PC and found varnishd running and listening. Thus Varnish had been started, but supervisor lost control over it. I was not able to stop Varnish with a supervisorctl command.
I worked further on the migration of HAProxy to the new Python 3 setup. I got a hint from a member of the Plone community that he already created a working branch of plone.recipe.haproxy for Python 3. I changed my buildout scripts to use this branch and I could build and run HAProxy too. But HAProxy didn’t get in contact with Varnish and the zeoserver instances in my setting. It stated that this instances were not running / available, although they are.
I got another hint from this community member and changed my buildout further. I changed the ‘bind’ Varnish configuration line to point to the Haproxy port of my buildout setting. The error message from Varnish disappeared. I was under control of supervisor yet.
Then I read in a blog about an issue of the HAProxy configuration, if it checks via http for the availability of Plone instances. I removed this option (httpchk) from the haproxy.conf file and everything went green. I could finish the migration to Python 3 and Plone 5.2.x.
The next step would be a test with a copy of the database and content of the current LibreOffice extensions website. I asked for this some days ago, but got no answer yet.