Config reset after upgrade from 8.1 to 8.2

[frsc-ku]

We observe an issue that breaks [rolling] updates in a bad way. A yum update does not act consistently on the beegfs config files, causing services to fail to start up. In our case, this affected the MGMTD.

The problem is that yum update replaces the current beegfs-mgmtd.toml with a new one and moves the current config out to beegfs-mgmtd.toml.rpmsave. This breaks restart since the DB file location is reset to an incorrect path.

The update of the client config behaves correctly, the user-modified config is not replaced, but left untouched while the new config is deployed as an rpmnew file. Here the relevant file listing after upgrade:

# correct:
beegfs-client.conf
beegfs-client.conf.rpmnew

# oh boy:
beegfs-mgmtd.toml
beegfs-mgmtd.toml.rpmsave

All changes in these files are modifications to comments and do not even require touching the running config. Could you please change the deployment procedure to use the rpmnew method throughout?

If you would like to merge configs, please consider adopting the conf.d-paradigm or any other method that has been established to avoid config update clashes.

Thanks,
Frank

[iamjoemccormick]

Hi @frsc-ku ,

This is definitely not the intended behavior. I see the beegfs-mgmtd.toml is correctly marked as a config file, but RPM packages also need a NOREPLACE flag to ensure the existing file is kept and the new file saved as beegfs-mgmtd.rpmnew. This will be fixed with ThinkParQ/beegfs-rust#37.

Thanks for the report!

~Joe