Cosmos-sdk: Interaction of mandatory reward-withdraw and reward truncation

[cwgoes]

Ref https://github.com/cosmos/cosmos-sdk/issues/2764
Ref https://forum.cosmos.network/t/3-important-topics-to-discuss-on-fee-reward-distribution-logic/

Any delegator can force decimal truncation of a validators' or the validators' self-delegation rewards by delegating/undelegating a tiny amount (to cause a power change). As delegator rewards are truncated to integers and rewards are minted each block, this seems like a substantial attack vector, with which the reward of a validators' self-bond could be almost totally negated.

[cwgoes]

cc @rigelrozanski @alexanderbez

I wonder if all of these hooks are quite necessary for the lazy calculation, I think a few of the ones we have might not be.

[rigelrozanski]

We do not require withdrawing a validator self-delegation when a 3rd party delegates-to/undelegates-from that validator. We do however require that that validator withdraw all of it's unclaimed rewards from the global pool to it's local pools when a 3rd party delegates-to/undelegates-from it (affecting its power)

The reason the code currently does this is out of convenience when this was being added if I'm not mistaken

[alexanderbez]

Awesome. Mind if I tackle this?

[rigelrozanski]

no-t-at-tall :)

[alexanderbez]

Discussed with @cwgoes and we agreed that the staking spec should first be reconciled prior to this. This is mainly motivated by some inconsistencies/inefficiencies in the staking hooks.

[cwgoes]

As in, we should correct anything incorrect in the spec, and then update the code to match (which it presently does not).

[rigelrozanski]

I'm not entirely sure that the _spec_ is the best way for us to spend our documentation efforts at the moment - For instance, I think diagram creation will do better to both explain the logic flow, as well as clarify hook calls. Diagrams fall more under description documentation than specs - we have very little description documentation in general

[rigelrozanski]

sry I meant - _Explanation_ documentation https://www.divio.com/blog/documentation/

[cwgoes]

Reopening this as we have reintroduced the attack.

[alexanderbez]

I think we can close this due to F1, correct me if I'm wrong @cwgoes.

[cwgoes]

Let's leave it open for now; F1 should prevent forced-withdraws, but we haven't merged it yet.

[cwgoes]

Closed by #3099, which is about to be merged.