fleet engine constantly running #742
Conversation
|
Per discussion, we should revert to the earlier behaviour of only triggering the engine on specific events (e.g. dd6636c) |
|
@jonboulle this is reasonable for fixing the engine side of things, but we've got the same problem with the agent. All of the job-state events are causing a lot of unnecessary reconciliation for the Agents. Could we replace JobEvent with JobTargetChangeEvent and have the agents listen to both JobTargetChangeEvent and JobTargetStateChangeEvent? That feels like the 80% solution here. |
|
Original ticket makes no mention of the agent 😞 Will add after lunch. |
|
@jonboulle yeah, well, I was testing you |
|
updated |
|
One nit, but LGTM if you don't care. |
|
I'm getting wary of using pointers unless really necessary, and I think the |
fleet engine constantly running
|
@jonboulle I'd like to be on the same page w.r.t. this sort of stuff in fleet. Maybe we should discuss offline. |
The fleet engine is triggered at a 2s interval, but also by any events generated by an etcd watch on the /_coreos.com/fleet/ namespace. Given the amount of heartbeating going on (machines, jobs, states, etc) that doesn't represent actual changes, this causes the engine to be triggered constantly.
We should figure out a way to only trigger the engine when an event represents an actual change that the to-which the engine could respond.