[batch] Prevent the worker container from being killed on OOM event (h…

…ail-is#13977) This PR fixes that crun has a different way to specify memory requirements when using cgroups v2 instead of cgroups v1. Should fix hail-is#13902.
populationgenomics · Nov 9, 2023 · 3b38d0b · 3b38d0b
1 parent ebfbc24
commit 3b38d0b
Showing 1 changed file with 4 additions and 3 deletions.
diff --git a/batch/batch/worker/worker.py b/batch/batch/worker/worker.py
@@ -1192,9 +1192,10 @@ async def container_config(self):
                 'resources': {
                     "devices": [{"allow": False, "access": "rwm"}],
                     'cpu': {'shares': weight},
-                    'memory': {
-                        'limit': self.memory_in_bytes,
-                        'reservation': self.memory_in_bytes,
+                    'unified': {  # https://github.com/opencontainers/runtime-spec/blob/main/config-linux.md
+                        'memory.max': str(int(0.99 * self.memory_in_bytes)),
+                        'memory.high': str(int(0.95 * self.memory_in_bytes)),
+                        'memory.swap.max': '0',
                     },
                     # 'blockIO': {'weight': min(weight, 1000)}, FIXME blkio.weight not supported
                 },