>> >> How about .5 seconds? We sync our machines to a local ntp server, so >> the clocks get really close (sub-ms). I forget that NTP isn't so >> accurate over a WAN. >> >> The issue is that the monitor leases and timeouts are something like 2 >> seconds (so that failures/unavailability) are detected quickly, so the >> tolerable clock drift can't get too close to that. >> >> sage > > ok, if this is the problem, then I would also recommend running an ntp > daemon (not ntpdate) on every node. A running ntpd will manage the > timesource to be in sync. > > If anyone likes to run an mds/mon on a VM I expect problems - > virtualized clocksources are far less stable (my experience). > started the ntpd on the nodes. now i'm able to run it with "mon lease wiggle room = 0.1". as Wido is using ntpdate and he's successfull with 0.5 I would say 0.5 is possible a good default. - Thomas -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
