We have identified that one particular node was getting errors. The requests that were going to the unhealthy node were receiving a 503 error. We completely disconnect the node to not have more invalid requests. We are still investigating what was happening.
