Hi all, If you are not using the new fpc scheduler in a cluster that cannot handle zero downtime, you can stop reading here.
I have two PR’s linked below that I will be merging in later this week. This is to allow getting the max instances per action within a namespace into master and to allow future changes to the CreateQueue message without another breaking change like this. With that said, it does not prevent an in place rolling cluster upgrade but it does require taking a certain order. So you have two options to continue on the latest master commit before we do an official release of the fpc scheduler: 1. Perform a blue / green deployment if capable which the new scheduler makes simple with the etcd cluster prefix config. 2. - To perform an in-place cluster change you should first update half of the schedulers. At this point the controller will not be able to send a CreateQueue message for actions without a queue running onto a scheduler that has been upgraded. However the controller should gracefully handle this and try to create queue on all schedules before failing the activation. - Once half of the schedulers have been upgraded, upgrade all of the controllers. Any action that already has a queue on a scheduler will work fine, any new action at this point will now be able to be created on any scheduler that has been upgraded but not the yet to be upgraded. - Finally upgrade the remaining schedulers. So to succinctly summarize the rolling upgrade order: upgrade half the schedulers, then all of the controllers, then finish the remaining schedulers. If you object to these changes, please respond by Thursday of this week which is when I plan to merge. https://github.com/apache/openwhisk/pull/5389 https://github.com/apache/openwhisk/pull/5287 Thanks, Brendan
