My cluster has 3 nodes, all the same version
Now I want to add a 4th node, I applied all updates to have the same version on all nodes.
Hosts is identical on all 4 nodes. Also DNS, passwords, they are in the same network an are able to reach each other.
node04 shall be added. On this node, I have /etc/pve/nodes/node04 folder existing, with all rights for root. Also I have all the usual files in /etc/pve/priv/
Then I start adding node04 by pasting the cluster join info. After a while, I see a message from node04 that creating /etc/pve/nodes/node04 failed due to missing permissions.
At this moment, I have to shutdown node04 quickly, otherwise my cluster breaks completely.
Investigation after that:
- /etc/pve/nodes/node04 exists on node01, node02, node03. On node04 even /etc/pve/nodes is missing and cannot be created manually, due to missing permissions.
- In journalctl I see a message that /etc/pve/priv/authorized_keys cannot be created. In fact, now this folder is empty with the only exception of a folder called "lock". the write permission is indeed removed.
Now I want to add a 4th node, I applied all updates to have the same version on all nodes.
Hosts is identical on all 4 nodes. Also DNS, passwords, they are in the same network an are able to reach each other.
node04 shall be added. On this node, I have /etc/pve/nodes/node04 folder existing, with all rights for root. Also I have all the usual files in /etc/pve/priv/
Then I start adding node04 by pasting the cluster join info. After a while, I see a message from node04 that creating /etc/pve/nodes/node04 failed due to missing permissions.
At this moment, I have to shutdown node04 quickly, otherwise my cluster breaks completely.
Investigation after that:
- /etc/pve/nodes/node04 exists on node01, node02, node03. On node04 even /etc/pve/nodes is missing and cannot be created manually, due to missing permissions.
- In journalctl I see a message that /etc/pve/priv/authorized_keys cannot be created. In fact, now this folder is empty with the only exception of a folder called "lock". the write permission is indeed removed.