Gerrit is down

While running the switchover process, we faced a race condition on the final steps of the process; where the puppet change required to switch Gerrit instances over was not visible to all Gerrit instances. That issue came from the fact that Gerrit was seen as unavailable for the puppetserver instances that tried to git fetch the puppet change, despite the DNS cache wipe. Once the Puppet change had been merged on the primary instance where it was initially fetched at the start of the process, it was impossible to re-apply that change anywhere. At this stage, we chose to revert to the previous state and use the emergency DNS procedure.