allocrunner: ensure alloc state updates are handled correctly. #22
Conversation
Implementing the GetUpdatePriority function is not enough to ensure updates are handling in a similar fashion to real clients. The alloc runner needs to be careful in how and when it updates its own state. Specifically, we do not want to directly force the client status information and instead have this populated once updates have been sent to the servers. The new functionality helps ensure updates are sent at the correct pacing, and handled in a manner more similar to real clients.
There was a problem hiding this comment.
Yeah it seems absolutely necessary to simulate staggered hook events which cause task state updates (and therefore rpcs to the server if the updates don't fall within a batch period)...
...but at what point do we just switch to using real alloc and task runners with the mock driver (or similar)?
Yeah good point and I was thinking about this last night. It might even be nice in the future to add a config flag that flips between real alloc/task runner and the bare simulated alloc runner. |
Implementing the GetUpdatePriority function is not enough to ensure updates are handling in a similar fashion to real clients.
The alloc runner needs to be careful in how and when it updates its own state. Specifically, we do not want to directly force the client status information and instead have this populated once updates have been sent to the servers. The new functionality helps ensure updates are sent at the correct pacing, and handled in a manner more similar to real clients.