tysonnorris commented on pull request #4978: URL: https://github.com/apache/openwhisk/pull/4978#issuecomment-693681340
Thanks for this! I think there is already some handling of the `runBuffer` to coerce it to "preserve order until the activation can be run". see https://github.com/apache/openwhisk/blob/39131e2cde5f30ecb331f06c47e2b70129b2aa3c/core/invoker/src/main/scala/org/apache/openwhisk/core/containerpool/ContainerPool.scala#L124 So you can check `isResentFromBuffer` and skip (re)enqueuing if there is no ability to run the activation immediately. Separately, there will probably need to be additional special handling for concurrent actions - where actions can support concurrency within the container. If ContainerPool receives 2000 activations for an action that supports concurrency=200, a maximum of 10 containers (maybe less, if some were already running) should be launched. Since an activation shows up in the pool processing multiple times, each would have to be tracked as "already affected a container launch" to prevent a container launch on second (and subsequent) passes. If you enable concurrency in your containers, this is simpler, but has some caveats (not all runtimes/actions can support this), and is currently buffered inside the ContainerProxy instance for that container (with the waiting problem that you mention). ---------------------------------------------------------------- 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]
