Occasionally, Batch will place a job on an instance that has more CPU cores than I requested. My task can take advantage of the extra threads easily, but the memory requirements go up if that happens, often leading to an OOM (out of memory) crash.
Is there a way to have a core-based memory limit? Or do I need to limit my job to only use the threads I requested?
Related question - if I request 16 vCPUs for a job but the container runs on an instance with 128 vCPUs, is Batch able to place other containers on the same instance? In which case, trying to use 128 threads would potentially be counterproductive anyway?
AWS 官方已更新 2 年前