Closed lixiaopengy closed 3 years ago
@lixiaopengxiya fuse-client is not supported in cephcsi for now, closing this for the same reason. Please reopen this one if you have used kernel mounter not the fuse.
@Madhu-1 I saw the document that fuse is used as the client by default, and the kernel mode does not support quota. Is there any way to solve this problem?
Fuse is default but not production supported still under development in Cephcsi as it is having some issue. Using kernel client should fix the issue. Fuse is having a restart issue where we will get Transport endpoint not connected error. We are working on a new design to support the fuse client.
Thank you for your reply. The inability to use qouta in kernel mode may cause system risks. When is the new design expected to be available?
Currently, we don't have the exact date.
Fuse is default but not production supported still under development in Cephcsi as it is having some issue. Using kernel client should fix the issue. Fuse is having a restart issue where we will get Transport endpoint not connected error. We are working on a new design to support the fuse client.
Can you talk more about your new design on how fuse client avoid the restart problem, I've been stuck with the question these days, too. Thanks!
There is no design doc for now. Once we have something will create a design doc PR
There is no design doc for now. Once we have something will create a design doc PR
Thanks, hope everything goes smoothly.
@Madhu-1 I found that the 3.4.0 version contains rbd-nbd volume healer support. Has this problem been solved?
@cl51287 yes its is alpha support and it's only for RBD not for CephFS.
@Madhu-1 I found that the 3.4.0 version contains rbd-nbd volume healer support. Has this problem been solved?
We have introduced a method here to take care of the plugin service restart of user space mounted rbd volumes, so it is worth to experiment with nbd mounter if you are on RBD PVs and wanted to use user space mounter instead of krbd.
@humblec We just want to use rbd, and want to use user space mounted, but our version of kubernetes is still relatively low, and it is estimated that it will not be used for the time being. How is the problem solved?
@Madhu-1 I found that 3.6.0 already supports fuse remounting, has this problem been solved? In addition, does rbd-nbd also support this feature?
@Madhu-1 I found that 3.6.0 already supports fuse remounting, has this problem been solved? In addition, does rbd-nbd also support this feature?
Yes, we added support for nbd already. cc @pkalever
@Madhu-1 When we used 3.6.2, we found that the configuration in the document could not solve the problem. When the daemonset of csi was restarted, the corresponding fuse process still disappeared, and the mount also failed. I don't know if it is a problem. Our kubernetes version is 1.14, we have configured netNamespaceFilepath as the net file of the No. 1 process of the host (other pods of the host have also tried), and found that the command is indeed correct through dlv debugging, using the nsenter command to execute , but when the daemonset of the csi is restarted, the mount still fails.
@Madhu-1 When we used 3.6.2, we found that the configuration in the document could not solve the problem. When the daemonset of csi was restarted, the corresponding fuse process still disappeared, and the mount also failed. I don't know if it is a problem.
automatic recovery of fuse is still not supported if you restart the application pod or create one more pod on the same node to use same PVC the mount should get recover.
Our kubernetes version is 1.14, we have configured netNamespaceFilepath as the net file of the No. 1 process of the host (other pods of the host have also tried), and found that the command is indeed correct through dlv debugging, using the nsenter command to execute , but when the daemonset of the csi is restarted, the mount still fails.
nsenter is not related to fuse. it only for the pod networking
@Madhu-1 Will the next version solve this problem? What does the documentation say to solve this problem?
@Madhu-1 Will the next version solve this problem? What does the documentation say to solve this problem?
Auto recovery is not planned yet. you can use the existing mechanism https://github.com/ceph/ceph-csi/blob/devel/docs/ceph-fuse-corruption.md which is supported in 3.6.0 release.
@Madhu-1 Thank you very much
Environmental inventory: kubernetes kubeadm v1.17 docker: 19.03.4 ceph: 14.2.11 csi: v3.1.2
Problem phenomenon After the CSI cephfsplugin service is restarted, the client has the following problems. You need to restart the client service container to return to normal
csi-cephfsplugin:
If you need other service logs, please contact me. Thank you for your help