j4ckstraw
j4ckstraw
/remove-lifecycle stale
> memory evict is just like oom killer, which dose not restricted by pdb. _by default_, memory evict should always kill containers even the eviction is failed for rpc timeout...
> > > memory evict is just like oom killer, which dose not restricted by pdb. _by default_, memory evict should always kill containers even the eviction is failed for...
> > want > > Actually, after the eviction api returns ok, there is no need to kill container if you want to solve the grace termination or PDB problem....
@zwzhang0107 PTAL
@zwzhang0107 rebased
We observed a steep drop of batch-cpu allocatable. metric `koordlet_node_resource_allocatable{resource="kubernetes.io/batch-cpu",node=~"$node"}/1000`  One pod with 10cores normal cpu requested scheduled on the node at the time of problem, and batch-cpu usage...
Here's my question: why add up HPUsed with pod request if no metric found, how about just skip it?