The node was low on resource: ephemeral-storage #3209
Replies: 2 comments 2 replies
-
Hi @simonmacklin. There are typically two disks attached to a Bottlerocket instance - the root disk for the system and a data disk for ephemeral/container storage. For that system disk, it really is not worth increasing the size of it as Bottlerocket has a read-only root filesystem that just contains two partitions to hold the Bottlerocket image. Any extra space allocated to the root disk is just wasted space, so you end up paying for extra space that can never be consumed. The second disk is the interesting one. When Bottlerocket first boots, it will create a filesystem on that to use all available disk space. So in your example above, |
Beta Was this translation helpful? Give feedback.
-
Hey @stmcginnis Thanks for the info. I think I figured out what was going on. The terraform we use to build the clusters mapped the root EBS volume to device /dev/xvda which was fine for amazon linux. It appears the bottle rocket AMIs use /dev/xvdb so the 100GB disk wasn't being used in the right place it appears. Changing the device mapping in terraform appeared to solve the ephemeral storage issue but something else has created another additional 2GB EBS disk. So we have one 100GB disk created in terraform used correctly and another 2GB disk. Does the bottle rocket software create this at boot time? |
Beta Was this translation helpful? Give feedback.
-
EKS version 1.2.1
I have been testing bottle rocket in EKS for a while now and since we rolled this out to one of our test clusters we have seen a lot of evictions that otherwise never happened with Amazon linux.
The eviction message is "The node was low on resource: ephemeral-storage". Usually this is due to low disk on the underlying host which may make sense looking at the details above as only 5GB is free on the root disk even though we have a 100GB disk EBS disk attached.
I was not able to find in the docs how we can adjust the values to increase this partitions. Does this mean we would need to extend using standard tools and rebuild a custom AMI or can this be done as startup time?
Beta Was this translation helpful? Give feedback.
All reactions