You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Worker container is able to use HugePage resource from Host.
Creating docker container with the argument --device=/dev/hugepages:/dev/hugepages may help.
Why is this needed:
To test applications that need HugePage to run, for example SDPK and some Databases.
Problem:
This may cause K8S Node resource consistency problem, if there are multiple nodes, which are enabled to use HugePage, on one machine. Therefore, this feature should be only used for testing purpose.
This issue comes with another question, it seems Memory resource has the same issue. If multiple nodes are created on one machine, each one of them can see all the memory from Host, but they cannot use all the memory.
The text was updated successfully, but these errors were encountered:
This may cause K8S Node resource consistency problem, if there are multiple nodes, which are enabled to use HugePage, on one machine. Therefore, this feature should be only used for testing purpose.
When we test Kubernetes we use multiple nodes. What's the use case for this?
This issue comes with another question, it seems Memory resource has the same issue. If multiple nodes are created on one machine, each one of them can see all the memory from Host, but they cannot use all the memory.
Yes, in general resource limits are not best tested with nested containers and should use a VM or similar platform instead. See: #1963
What would you like to be added:
Worker container is able to use HugePage resource from Host.
Creating docker container with the argument
--device=/dev/hugepages:/dev/hugepages
may help.Why is this needed:
To test applications that need HugePage to run, for example SDPK and some Databases.
Problem:
This may cause K8S Node resource consistency problem, if there are multiple nodes, which are enabled to use HugePage, on one machine. Therefore, this feature should be only used for testing purpose.
This issue comes with another question, it seems Memory resource has the same issue. If multiple nodes are created on one machine, each one of them can see all the memory from Host, but they cannot use all the memory.
The text was updated successfully, but these errors were encountered: