cloudstack
cloudstack copied to clipboard
apache
Restarting management server causes multiple entries for the same node
Restart management server several times, even if there's only one mgmt server, it causes several entries in the cloud.mshost table and in the API:
ISSUE TYPE
- Bug Report
COMPONENT NAME
Mgmt server
CLOUDSTACK VERSION
4.18.1.0
CONFIGURATION
Adv zone KVM env with Ubuntu 22.04
Tried to reproduce on ubuntu22 and centos7 on 4.18-head, no go. I'll give it one more try with stock 4.18.1 on ubuntu22
@rohityadavcloud
can you share the entries in database table mshost ?
If a new record each time when management server is restarted, it might be because the management server id is calculated from a random mac address.
https://github.com/apache/cloudstack/blob/78213da82860761e174d5876afe857977f3889b3/utils/src/main/java/com/cloud/utils/net/MacAddress.java#L145
may need to verify all the commands before the line in same class
If a new record each time when management server is restarted, it might be because the management server id is calculated from a random mac address.
https://github.com/apache/cloudstack/blob/78213da82860761e174d5876afe857977f3889b3/utils/src/main/java/com/cloud/utils/net/MacAddress.java#L145
may need to verify all the commands before the line in same class
What is the use case for this logic? Why not just use the servers FQDN?
If a new record each time when management server is restarted, it might be because the management server id is calculated from a random mac address. https://github.com/apache/cloudstack/blob/78213da82860761e174d5876afe857977f3889b3/utils/src/main/java/com/cloud/utils/net/MacAddress.java#L145
may need to verify all the commands before the line in same class
What is the use case for this logic? Why not just use the servers FQDN?
@poddm Actually it does. It gets the local address at first. If it does not work, then fall back to the process above. If the issue still exists, please share more information like
- hostname
- hostname -f
- /etc/hosts
This happens on a server which is used both as management server and KVM hosts (and docker host), the number of interfaces keep on changing over time and the current code is flaky to be deterministic about which interface to pick. I've addressed this at: https://github.com/apache/cloudstack/pull/8968/commits/fb82e352bf4cd9306421effdd91bdc6e9387df5f
The solution to the issue is to only pick non-virtual, possibly actual physical Nic (last in the list seems to be correct) and use it to convert/use as the (long) ID of the mgmt server. I'm also thinking this value should be configurable via server.properties.