We like the simplicity of longhorn and would like to use consumer hardware for the storage. We know Ceph is using some automatic healing and detection of disk failures, but ceph is quite more complicated than longhorn. We also saw different other situation where the data is completely corrupted at the end of live of the ssd. We already saw disks which start to add 0-Bytes somewhere on the disks. We also considered if we could use the snapshot data for this.Ĭonsumer SSDs are dying in a very bad way. This would lead to a drastically performance decrease if we set them read only for the hash time. To work like this, the disk have to be without changes for some time.
If a corrupt volume is detected with the python interface remove the replica. This wouldn't lower the performance drastically but increase the snapshot, backup and replica integrity significantly.Ī cron job which automatically calculates a hash of the mounted devices and compare these.
These hashes don't have to be cryptography correctly(sha). This would help to identify common storage problems.
It would be possible to identify corrupted blocks automatically. If you add some meta data to the snapshot in form of hashes of the changed blocks. You already described a way of identifying corrupted replicas.
0 kommentar(er)
