Issues related to applications and software problems and general support
- Posts: 3263
- Joined: 2007/12/11 08:17:33
- Location: Finland
TrevorH wrote: ↑
always do a fresh install. With everything in configuration management it really isn't that much effort and you end up with a supported machine with no cruft and no legacy crap on it.
This. One can put their effort on setting up configuration management script that is likely to work time and again, or one can spend the time on creating upgrade magic that might possess undefined behaviour.
Everyone is naturally free to choose ... as long as they assign blame truthfully.
The configuration management implies that configuration is not stored in
a machine; it is applied to
a machine. Management is hosted outside a machine.
This also implies that user data is "not part of the system". User data could be on separate partition (that needs not to be overwritten due to OS install). For example, on network share.
- Posts: 1
- Joined: 2020/06/17 11:01:40
_"isn't that much effort"_
At my place of work we currently have ~32k hosts (physical and virtual) running CentOS of which ~29k are currently running CentOS 7. Replacing them with new builds of CentOS 8, even with heavily automated processes and not disrupting the business is quite a considerable effort. A supported in-place upgrade would speed that up and save many months of work.
- Forum Moderator
- Posts: 30173
- Joined: 2009/09/24 10:40:56
- Location: Brighton, UK
It will never happen. Start designing your deployment practices to make it easier to replace them not upgrade them.
- Posts: 414
- Joined: 2013/12/18 15:04:03
- Location: xfs file system
I have a large number of centos 7 servers with linode, I've created a semi-automated method with ansible. My playbook creates a new centos 8 server, generates ssh keys, migrates data between the two servers (via the local interface), switches the public IP addresses between the two servers, thus the new server becomes visible to the outside world. Finally, the old server is destroyed.
In most cases its easy to migrate the data, but in some limited cases the migration is rather difficult and can only be done with a live cluster that allows servers to shut down without affecting performance.