chore(sales): log error when datasets size is different then expected #1155
Conversation
AuHau
added
the
Marketplace
AuHau
force-pushed
the
chore/download-size-mismatch
branch
from
08a7bb4 to
125cf9b
Compare
There was a problem hiding this comment.
If the data is smaller than StorageRequest.ask.slotSize then unused capacity in a Reservation gets returned to the Availability on clean up. If the data is greater than the Reservation capacity, we should get a BytesOutOfBoundsError exception in the result. So from an accounting perspective, I believe we're already ok. However, from a data integrity perspective, wouldn't it be better to recreate the Cid from the dataset and see if it matches? Slower, yes, but more accurate.
Also, maybe node.onStore might be a better place to implement that.
This is already the case, we always check the integrity of the downloaded data. |
| error "Couldn't get reservation key", id = $reservation.id, error = error.msg | ||
|
|
||
| if updatedReservation =? await reservations.get(keyId, Reservation): | ||
| if updatedReservation.size != 0: |
There was a problem hiding this comment.
This should not fail the contract tho, if this is whats going on. We should either return that extra capacity back, or just ignore it, since the user is allegedly already paying for it.
This is a safeguard for when the downloaded Slot's size does not match what was expected. I want to add this, because back when I was working on IPFS I discovered a security bug where you could forge the manifest in a way that it would report a different size than the actual data's size. You could have then stored an "infinite" amount of data with the pinning services because they relied on the manifest info and did not double-check actual data size.
Originally part of #798 but that PR got reworked by Arnoud at #1099 so just pulling this part out of it.