How does tape backup handle intermediate failed verification?

[wbk]

Hi all,

I just completed backup and verification of a snapshot and started a task to write it to tape.

• There are six snapshots of this container, the fifth of which failed verification.
• Only the first snapshot has been written to tape previously

My intention was to write only the last snapshot, which I perceived to be the delta between snapshot 1 (on tape) and snapshot 6 (not yet on tape).

Not totally expected, but just as well:

• PBS starts writing the second, then the third snapshot to tape, and just started the fourth snapshot
• In this fashion, it will start writing the fifth snapshot later tonight

Assuming the data in the container has not been changed but only has been extended, will the sixth backup have overwritten/corrected the corrupt chunks of the fifth backup? I would expect the verification process of 'Nr 6' to verify all chunks that are part of that snapshot, also if they are also part of another snapshot.

Datastore screenshot: I highlighted the fifht snapshot, that has verification failed

 

[dcsapak]

the tape backup code has no idea of the verification state of the snapshots and will start backing up "corrupt" snapshots
(this make sense, as the verification state is only ever valid at that point it in time)

it will cancel however if it encounteres a chunk that has an invalid crc checksum (only that is checked not the full hash, for performance reasons)

i admit that it might make sense to filter out snapshots with failed verification tasks, maybe as an option, but in general i'd
either try to fix such a snapshot, or to delete it (you won't be able to properly restore from a snapshot with invalid chunks anyway)

as for

I would expect the verification process of 'Nr 6' to verify all chunks that are part of that snapshot, also if they are also part of another snapshot.

yes, but it won't modify the state of other snapshots (since we don't know beforehand which chunks belong to which snapshots)
if the subsequent backup fixed the corrupt chunks from the previous one, you have to verify that failed one again to get a valid status

 

[wbk]

Hi Dominik,

Thank you once more for your patient and in-depth explanation, it makes total sense!