> If the 'recv' member of struct remote_mp is a flag then I'd expect it > to have type bool, not long long.
Oops. Kruft leftover from recv_time. > Does this change the promptness of fault detection at all? That is, > will it wait longer in some cases to detect a fault? Since the previous patch, fault status only changes when fault_timer runs once every interval * 3.5 seconds. There are some things we could do to improve the responsiveness if necessary. First, we could have per-rmp fault timers. Second, we could unfault cfm when a heartbeat causes the last faulted rmp to suddenly come up. I'm not sure it's worth it at this point though. > While I was looking this over, I noticed what might be a bug or might > be just a documentation error. cfm_create() doesn't mention that the > caller must call cfm_configure() immediately, but I think that that is > necessary because otherwise nothing properly initializes the timer > than cfm_run() will check. That's a documentation error. I'll add a comment to cfm_create(). Ethan _______________________________________________ dev mailing list [email protected] http://openvswitch.org/mailman/listinfo/dev
