F5 iApp pointing at Varnish cache, but not serving fresh content from the Varnish cache after PURGE

Hi all. We're upgrading our old Drupal Scientific Linux VMs to newer Ubuntu VMs (yay PHP 5x deprecation!). To test this I've set up separate QA domains for Drupal and our live site, both of which point against the production systems. The QA version of the live site points at a new Varnish node. The PROD version of the live site points at the existing two Varnish nodes.

 

I've modified a custom module I wrote that purges updated content from Varnish, such that it sends purge requests to all three nodes. So, whenever anybody updates from drupal.domain.com or drupal-qa.domain.com, a purge request is sent to all three Varnish nodes.

 

When this happens the content at live.domain.com is purged as expected, but for live-qa.domain.com the F5 is still sending the previously cached content from Varnish.

 

If I go directly to the Varnish node I see the fresh content. I've even gone so far as to log on directly to the F5 VM itself and request live-qa.domain.com via Curl, and I see the correct updated content!

 

I've also added logging in the Drupal purge module I wrote such that all the Varnish node / domain pairs being purged, as well as the actual HTTP result from Varnish, are logged. The result from Varnish is 'HTTP/1.1 200 Purged', as it should be.

 

So there is no question that the F5 can access the new content, or that Drupal is correctly purging the content. It's just for some reason when the F5 sends back the content for www-qa.domain.com it's still the old cached content.

 

I've got caching disabled within the iApp.

 

Can anyone think of anything that might be causing this? To test this weekend I'm going to update our production Varnish nodes to use the new Ubuntu VMs as their backends, so I can verify that purging will work correctly from drupal.domain.com when Varnish is pointing at the new infrastructure, but I would really, REALLY, like to figure out why the purged content is not being retrieved by the F5.

 

If anyone has any ideas or insight into things I could look at to test I would greatly appreciate it.

 

Thanks, Pablo