Is that really a problem though? IIRC, the change_attribute is supposed to be opaque. The client is only supposed to care if it's different from the last one it saw, not necessarily that it has gone forward.
Oh but...I suppose you could get bitten if you saw the change attribute transition from (for instance) 3->4 and then server reboots without committing that to disk. It then comes back again and does another 3->4 transition with a different set of data. Client then sees change attribute is "still" at 4 and doesn't purge the cache.
In that case...yeah -- maybe adding some sort of random offset to the change_attribute that's generated at boot time might make sense.
Copyright © 2018, Eklektix, Inc.
Comments and public postings are copyrighted by their creators.
Linux is a registered trademark of Linus Torvalds