Hacker News new | ask | show | jobs
by sean13013 4309 days ago
What does "scrub repaired 1.58M in 201h47m with 0 errors on Fri Feb 21 13:37:24 2014" mean?

I have a much smaller zfs setup at home, as well as a few at work, and I would be super concerned if the scrubs were repairing data. Am I overreacting?
1 comments
jspiros 4309 days ago
It means I ran a scrub that completed on February 21, 2014 after almost 202 hours, and there were no errors, but there were some corrupt data blocks that were repaired from mirrors, adding up to 1.58MB of data that was repaired.

See http://docs.oracle.com/cd/E23823_01/html/819-5461/gbbwa.html... for more information.

link
sean13013 4308 days ago
Sorry, I should have worded that better. I know what the message means, but what do you make of the fact that it found & repaired corrupt data? If I understand right (and correct me if I'm wrong), corrupt data should only occur if some bit of hardware is failing or possibly after improper shutdown. Am I wrong? Is it no big deal to see scrubs repairing data?
link
jspiros 4308 days ago
My understanding is that some significant (thousands of bytes) silent corruption is inevitable when you start reaching huge capacities over long-ish periods of time; even cosmic radiation has the potential to flip a few bits here and there every once in a while.

http://en.wikipedia.org/wiki/Data_corruption#Silent_data_cor...

So, yes, I think even with the best hardware, and proper maintenance, seeing some data repaired in a scrub is to be expected.

(That said, I did spend way too long using technology [SATA PM] that often failed and made it impossible for me to run a scrub. It's very possible that normal error rates are more like what I'm seeing now, a single byte every month or so, and that the megabyte figure is representative of errors from the days of my arrays dropping out unexpectedly.)

link