I have a RAID5 array with two aging HDDs that both need replacement. One drive has exceeded its SMART failure-prediction threshold, shows five grown defects, and reports many non-medium errors, but it has no uncorrectable read or write errors. The other still reports SMART as healthy, but its long self-test failed at a specific LBA. It has 60 grown defects, uncorrectable read, write, and verify errors, plus additional corrected errors. Both drives have roughly 44,800 hours of operation. If the array is still online, which drive is less likely to survive being used as the source during a rebuild, and which should be replaced first?
3 Answers
The safest plan is not to rebuild this array at all. Copy or restore the data to a new array or new storage, then retire both suspect disks. A RAID5 rebuild reads essentially the entire surviving array, which can expose more failures in drives with this much operating time. Also check the remaining members, since they are likely similarly aged.
If you absolutely must perform an in-place replacement, the disk with the failed self-test and uncorrectable errors is the one to remove first. Leaving it in place while replacing the other disk would require it to supply clean data during the rebuild, and its existing bad sectors make that especially dangerous. Even then, be prepared for the rebuild to fail and have a tested backup or recovery plan.
If the array is still functional, replace the drive with the failed long self-test and uncorrectable errors first. That drive has demonstrated actual media or mechanical problems, so it is the least trustworthy source for a rebuild. The other drive is also failing and should be treated as urgent, but it has not reported uncorrectable read errors yet. This is still a high-risk rebuild, so make and verify a backup before changing anything.
Understood. I’ll verify the backup first and prepare replacement storage rather than assuming the rebuild will succeed.