Dear reporters (Sylvain Guittard, Sébastien Morel)
we are in need of more information, preferably within the next hour or so given the urgency of this issue, before we can effectively figure out what is wrong. Otherwise this will become a second guessing game, amplified into a ping-pong game by timezone differences.
1. Can you reproduce locally or not, using Symfony Proxy? If not then issue on platform.sh might be in regards to changes done on platform.sh config, ref: https://github.com/ezsystems/ezplatform/pull/276
2. The Varnish issue: With the exact same Varnish setup, was it working with 2.0? Or did the Varnish setup slightly change? Asking, because we are aware the Host header sent on purge might be wrong, so if Varnish is setup on a port other then 80 it might ignore the purges. Way to see this is by debugging purge requests and see that they all return info that 0 items where purged. But it's been like this since 1.12: Ref: https://github.com/ezsystems/ezplatform-http-cache/blame/master/src/PurgeClient/VarnishPurgeClient.php#L44-L47
If none of those are at relevant, then there might be another regression besides the one we just fixed in the kernel change which removes about 5000 lines of code, so this is where we will likely look further for possible clues to what might be wrong here: https://github.com/ezsystems/ezpublish-kernel/pull/2194 Unless we by chance gets QA to reproduce any of this.
EDIT: Fix below can potentially explain it, assuming both of you where testing with eZ Platform EE, as the problem below was not affecting eZ Platform. Please verify as soon as you can.