Hi,

Many systems still generate v1 uuids (
http://en.wikipedia.org/wiki/Universally_unique_identifier#Version_1_.28MAC_address.29
-
for example still used by MySQL by default ) and seeing how most OSes are
handling the leap second is kind of worrying me, since it greatly increases
the risk of seeing the same UUID generate twice (since the same unix
timestamp happens twice).

While this might be seen as a bug in UUID implementations or the way time
is represented in unix timestamps, it's more of an issue in the way leap
seconds are handled in the kernel (repeating :59), causing applications
which rely on the fact that the same time shouldn't happen twice to fail
(on top of getting some kernels to crash during the previous day, see
http://serverfault.com/questions/403732/anyone-else-experiencing-high-rates-of-linux-server-crashes-during-a-leap-second).


Anyway before we go and modify the code in Linux implementation of the leap
second, I'd like to open a discussion to find the best way to handle leap
seconds, and here might be a good idea to start with. One thing I could
think about and which shouldn't be too complex is to have the microtime
part (tv_usec) go twice as slow, which would ensure the same time does not
happen twice.


Thanks,
Mark
_______________________________________________
pool mailing list
[email protected]
http://lists.ntp.org/listinfo/pool

Reply via email to