We have an established procedure to replace the CTA Frontend node but it requires a downtime during which the flag “mgmofs.protowfendpoint” on the EOS MGM node and
the flag “fstofs.protowfendpoint” on all EOS FST nodes will be updated with the new hostname.
My question: is this downtime really necessary?
Would it be possible to have a rolling kind of update of mgmofs.protowfendpoint and fstofs.protowfendpoint in which case some EOS nodes will point to the old Frontend and some others to the new Frontend node? Is there a danger for data integrity by having two Frontend nodes (even a for a few mins) sending requests to the CTA object store?
yes, in the current EOS implementation a restart of the MGM and FSTs are necessary to reload the new configuration.
In principle one may, indeed, think of implementing an EOS method which would dynamically reload
this configuration from the file without a restart, but this is not high on our priority list at the moment.
We may come back to designing a solution for this problem later in the development process.
Thanks for the reply. I am aware of the need to restart the MGM and FSTs after a config change,
What I asked was if - during a Frontend replacement - it is OK that some FSTs point to one Frontend node and some others point to another Frontend node?
In other words, is there any issue by having (temporarily) two Frontends sending requests to the object store?