tysonnorris commented on issue #4326: Invoker backpressure URL: https://github.com/apache/incubator-openwhisk/pull/4326#issuecomment-470992379 @style95 the state of resources is not real-time, which is why this change also tracks "reservations", e.g. resources that ContainerPool has decided to use, but are not actually in use yet. This change includes no adjustments to controller, but the controller behavior is affected by: - whisk.container-pool.user-memory setting should be artificially high, which causes all actions to be routed to their "home invoker" - controller will only stop routing to the "home invoker" when that invoker becomes Unresponsive (which is based on health check activation timeouts). This will leave the invoker in a queued/rescheduling state (since there will be a delay between reaching capacity, and the health action timeout), but once the Unresponsive state is reached, it won't be used again until it is Healthy again (has capacity to schedule containers) - This should be improved with such ideas like allowing invoker to declare itself Unavailable at will (as opposed to indirectly making this state change from timeout), and allowing controller to shard based on different data (activation memory affect at invoker, with respect to /init and /run invoker heap usage), etc.
---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected] With regards, Apache Git Services
