I'm running Varnish behind an Nginx reverse proxy with an apache 2 backend serving a Drupal 7 site.
I've recently moved to a new server, this time running Ubuntu 22.04, meaning that I've upgraded the Varnish edition I have been using from 4 to 6. The config is otherwise identical.
On the previous server the cache hit rate, off a 12G malloc setting, was 96% and the site ran predictably fast.
On the new server I'm seeing at roughly 50% hit rate. On investigating I've found that many pages are expiring in just an hour or two (as opposed to over a day I specify). They are not actually being served from GRACE, because when I request them I get a cache miss and a complete new BE request, slowing things down enormously.
The pages are not being nuked owing to a full cache. The cache is only about 4G full, with 8G spare when this is happening.
The fact that there is no LRU_nuked log suggests that something is purging these pages, but I have no idea what or how, nor why this should be happening now but not on the previous server.
I'm wondering how to investigate further?