[OmniOS-discuss] ZFS data set corruption

[Sebastian Gabler]

Hi,

I recently migrated from OI 151a7 to Omnios April 2015 LTS with my 
backup ZFS file server. Probably because of vibration-related issues a 
pool was corrupted during the first nightly backup. Several attempts to 
resilver the pool ended with checksum errors in one of the raidz vdevs, 
so I decided to try a recovery. The affetced files were the file that 
was being written when the first error occurred, and two blocks in 
pool/datasets/corrupted_dataset itself. A further salience was that 
copying some files under corrupted_dataset could not be read (bad 
exchange descriptor error). From other reports on that error message, I 
concluded that deleting the files would not work either, so I skipped that.
Exporting pool did not work anymore (pool busy error, even with -f)
Now, when trying to destroy 'pool/datasets/corrupted_dataset', I got a 
kernel panic. The same happened when trying to import -nfF the pool, 
after forcing an export with the pool not mounted on reboot. So, I 
decided to cut my losses and re-create all datasets from scratch, 
without further analyzing the reasons.
Any ideas how I could have done better with an additional work of 2-3 h 
(5.7 TB dataset)?

BR

Sebastian