January 2, 200917 yr There have been several people that have encountered subtle drive corruption by running parity checks while a failing drive is present in the array. It seems that in these situations, the failing drive can somehow return bad data and cause unRAID to update parity based on the bad information. So even after a drive reconstruction, the data is not 100% perfect. This is insidious as it is impossible to determine where the problem occurred without having a backup. Here is a recent example ... Link One possible solution would be to have the parity check work differently. It could "remember" where the parity problems are but NOT correct them on the fly. After the parity check is complete, it could prompt the user through emhttp as to whether to correct the errors or not. If the answer is yes, unRAID could then recheck just the few sync error sectors and update parity to fix them. With this model, you'd be able to safely run parity checks even when a drive is suspect, and not risk corrupting parity in the process.
January 2, 200917 yr I hope Tom will read this & consider it. It would make running parity checks safer.
January 7, 200917 yr I hope Tom will read this & consider it. It would make running parity checks safer. What we can do is add a "Fix errors" checkbox next to the "Check" button on the Main page. We'll try to get this in the next beta release (4.5-beta2).
January 7, 200917 yr I hope Tom will read this & consider it. It would make running parity checks safer. What we can do is add a "Fix errors" checkbox next to the "Check" button on the Main page. We'll try to get this in the next beta release (4.5-beta2). That would be very cool. If you were to print in the syslog the first hundred or so blocks where the parity check found differences it might help us to figure out the involved/affected files. (I don't think printing every block with a parity error is good, as it might run you out of RAM if something odd were to occur. The first few hundred would let most errors affecting a single block or two to be found and analyzed. ) Joe L.
January 7, 200917 yr Author What we can do is add a "Fix errors" checkbox next to the "Check" button on the Main page. We'll try to get this in the next beta release (4.5-beta2). Excellent!!! Thank you!
Archived
This topic is now archived and is closed to further replies.