A guy use Claude Code recovered nearing 90TB of corrupted ZFS pool that rejected by data recovery companies
Originally posted by @ shujueitsao in Threads
The following content is the post content translated by ChatGPT
Note the "dollar" is TWD.
Last May, the company NAS crashed. 90TB of animation project files — three years of hard work. The hard drives weren’t broken, but the ZFS metadata was corrupted. The system couldn’t recognize the data. I contacted several data recovery companies. I was ready to pay the price of a domestic car. In the end, no one was willing to take the job — all said it was unrecoverable. It sat there for almost a year. One day, while playing with Claude, I suddenly thought: Could I just try to fix it myself? I spent less than 5,000 dollars to buy an HBA card. Connected 12 hard drives to the computer. Installed Claude Code and let the AI issue commands on its own. It tried ZFS historical restore points one by one. Found a point in time that wasn’t corrupted. Mounted the entire data pool back. Recovered everything. Directory structure intact, filenames intact — three years of projects. I had absolutely no Linux experience. AI guided me through the entire process. Today, I saved the cost of a domestic car 😮💨
The following content is the summaries for the chat message that done by ChatGPT
Recovery is difficult; ECC errors cannot currently be corrected.
They will return the hard drives first, but backups are retained.
No recommended contacts are available locally; recovery in Taiwan is not feasible.
Consultation was done with overseas experts. They connected remotely and wrote several test programs; additional tests were also written locally.
Conclusion: recovery is unlikely.
The following content is the Chinese content that in the Ubuntu's Claude Code screenshot
All background tasks have completed; data integrity is confirmed.
Main data is located in zpool1/zfs19 (88.6TB). How would you like to proceed?
Option 1: Extract selectively (using zdb)
Suitable for recovering a small number of important files
Slow; impractical for 88TB
Option 2: VM approach (recommended)
Run a QuTS hero VM via KVM/QEMU on this machine
Pass through all 12 drives to the VM
Inside the VM, use zpool import -o readonly=on to mount the pool fully
Then export data via SMB/NFS
Which option do you want to use? If you choose option 2, do you have another set of drives or enough space to store the recovered data?