Description
What happened:
Blob CSI doesn't respect memory limits when invoking blobfuse2
process on the host
What you expected to happen:
Tbh I don't know if its expected behaviour or not but I was sure that blobfuse2
is somehow respecting limits of blob
pod in csi-blob-node
daemonset. If this is expected there should be a big warning sign for using block cache, since by default it can use up to 80% of host free memory calculated on process start (so in most cases when nothing else is already allocated on the machine). Or maybe there is another way to work around it?
How to reproduce it:
Create any mount definition that uses Blob CSI driver with blobfuse2
version > 2.3.0
using block cache. Detailed parameters are described in azure-storage-fuse
issue: Azure/azure-storage-fuse#1639
Anything else we need to know?:
None
Environment:
- CSI Driver version: 1.25.1
- Kubernetes version (use
kubectl version
): 1.30.7 - OS (e.g. from /etc/os-release): Ubuntu Linux (AKS)
- Kernel (e.g.
uname -a
): Ubuntu Linux (AKS) - Install tools: helm chart
- Others: N/A
Activity