WIP: [TENTATIVE] Feature/caddy frontend root direct configuration
WARNING This MR is very tentative, as it's not proven yet that such change would improve the situation.
Outcome is to provide shared instance list configuration to the nodes (kedifa, exposed, etc) directly via HTTP protocol instead with SlapOS Master.
Generally request shall have only information required to setup the node and make it functioning. All other information, like details about slaves, shall be exposed on the root partition. Some kind of checksum/timestamp can be added to request of the nodes, so that they'll reconfigure themselves. It's important, that download of the information by the node can happen asynchronously, but processing shall happen after being initiated due to root node information change.
It's still open question if the exposed and kedifa nodes shall use publish to provide all information, or also use HTTP to avoid data duplication. Note: This generally has to be rethinked, maybe the exposed/kedifa nodes does not have a lot to publish, as a lot could be calculated by the root node.
Keys to be moved away:
- kedifa node:
-
slave-list
-
- exposed node:
-
extra_slave_instance_list -
slave-kedifa-information
-
Requirements:
- transfer massive information to the nodes w/o using SlapOS Master
- cleanup the data
- bonus: improve speed a lot
- bonus: provide slave connection information ASAP instead of relying on all exposed and kedifa nodes to process
Depends on: