Should we fail a retrieval with an identical CID+Selector?

[hannahhoward]

Looking through state code, I found this line: https://github.com/filecoin-project/lassie/blob/main/pkg/session/state.go#L132

It looks like if we have two retrievals in progress for the same cid + selector, we fail the second one. BUT does this make sense? I feel like this line comes from our autoretrieve days when it was all going ot the same blockstore. But now it's not. Should we reject these requests just cause another is in progress?

[rvagg]

Seems wasteful to let these through, the argument here would be to make this the user's problem I suppose. From an API perspective I guess it is a bit weird "sorry, can't fulfil this request because someone is already doing the same thing". No technical reason anymore that it shouldn't be duplicable I think.

From a technical standpoint I like rejecting duplicates. From a user perspective it probably is a bit weird. So I guess that means we should remove it.

Do we know if it's causing problems in practice?

[hannahhoward]

that's what I'm curious about -- theoretically it shouldn't cause same cid + selector should get responded to from cache if hit twice.... but who knows.