Reading the recommendations related to the migration from the 0.14 to the 1.0 (https://discuss.konghq.com/t/kong-1-0rc2-available-for-testing/2142 for the 0.14), I noticed this sentence:
Start provisioning 1.0rc2 nodes, but do not use their Admin API yet. Prefer making Admin API requests to your 0.14 nodes instead
This looks to mean that during the period of time where Kong instances for both the old (0.14) and the new (1.0) versions are running, this is still possible to launch Admin commands, as soon as they target Kong nodes of the new version (1.0).
I’m worried about this exact use case (without, unfortunately having tested it concretely !): when an Admin command is sent to such a “new” node, I guess that the usual “kong data replication mechanism” (sorry, I don’t know the exact name for that, but let’s say “the mechanism that makes other nodes be aware that updates in the Service/Routes/Consumers/… have been done”) is triggered, leading to the clearing of internal caches of each node (including the “old” nodes) and the reloading of the entities from the database.
If so, then “old” nodes will also try to reload data from the database… but the database schema is no longer at the right version => failures may happen !
Is my reasoning correct ? Shouldn’t rather we say that Admin commands shall be forbidden until all “old” Kong instances are gone ?
Thanks for your help !