Folks can I ask everyone to help improve reliability of Flyte plugins. As we work on agents it will become easier to test. Let’s help to make tests run on various cloud providers and fix them.

We run test executions for pytorch/elastic tasks in our gke staging cluster automatically before we bump flyte helm chart or flytekit versions (we usually quickly upgrade to the newest versions). There was one bug introdued a while ago about resources not being set correctly any more that we caught with this but apart from that, it’s been stable …