tarides / current-albatross-deployer Goto Github PK

from your README "It's been made with zero downtime in mind, meaning that when an unikernel is updated, a new instance is started while keeping the old one alive, and the switch to the new instance is managed using a port redirection to the new IP."

Now, I wonder: how does this work? Did you already solve the halting problem?

More seriously, how do you detect that a new unikernel is working nicely and switch over to the new IP address? Is there a generic path to do this?

I'm stumbling upon this due to the huge variance of unikernels and when they may be ready (some may need to clone a git repository, etc.) - plus interaction (the old may write to a git repo while the new cloned an old revision, is such a scenario handled?) of old and new.

Do you handle TCP connection to the old unikernel still working smoothly?

https://github.com/TheLortex/current-albatross-deployer/blob/7d7c9b35b4ca07542fae5ccba8850cee6959016e/lib/unikernel.ml#L61

Current_git should be used to track the commit and allow updating it.