> From: Chris Ferebee > Sent: Friday, January 19, 2018 11:23 AM > > The concern I’ve read about is that you may be running smoothly with an SSD > slog, have a power failure, and find, when you reboot, that the SSD was > actually dead, it just hadn’t stopped moving yet – so it never comes back on > line. I have no idea whether that is a common failure mode. > [...] > them. Maybe they’re just very reliable, but my understanding was always > that Illumos would recover transparently from a slog device failure.
If a log device fails while the pool is in active use, the ZIL transparently swaps to be on the pool itself rather than the separate log device. If after a clean shutdown a log device fails to initialize and is unusable, the ZIL also transparently swaps to be on the pool itself, and given the clean shutdown, there is no unprocessed data on it so nothing is lost. There is only one scenario where you would lose data due to the failure of a dedicated log device; if you have an unclean shutdown with unprocessed data on the log device, and it fails and is unreadable the next time the pool is imported. In that case the ZIL still transparently swaps to the pool itself, but any unprocessed transactions on the log device are lost. The cost/benefit comparison is simply as you say to consider the likelihood of that specific failure mode occurring in your deployment. I'm not sure if there are any statistics regarding, but in general the major players (such as Joyent) don't consider it necessary to mirror a log device, and they certainly have a lot more experience than me :). ------------------------------------------- smartos-discuss Archives: https://www.listbox.com/member/archive/184463/=now RSS Feed: https://www.listbox.com/member/archive/rss/184463/25769125-55cfbc00 Modify Your Subscription: https://www.listbox.com/member/?member_id=25769125&id_secret=25769125-7688e9fb Powered by Listbox: http://www.listbox.com