feat(oracle-feeder): initialise core logic #1010
Conversation
feeder/feeder_test.go
Outdated
| }) | ||
| }) | ||
|
|
||
| t.Run("events stream invalid val set - has no in validators", func(t *testing.T) { |
There was a problem hiding this comment.
| t.Run("events stream invalid val set - has no in validators", func(t *testing.T) { | |
| t.Run("events stream invalid val set - has out validators", func(t *testing.T) { |
feeder/feeder.go
Outdated
|
|
||
| // init val set | ||
| select { | ||
| case initValidators := <-stream.ValidatorSetChanged(): |
There was a problem hiding this comment.
How does this work if the validator joins late, meaning after the chain has already started?
There was a problem hiding this comment.
there's a listener which listens to the current validator set changes, although we don't make use of this because of: https://github.com/NibiruChain/nibiru/blob/master/x/oracle/keeper/update_exchange_rates.go#L30
So the actual check is: "Validator.Bonded()", but can an unbonding validator be on the Active set?
Other implementations there's one in umee and one in terra, from what I've seen, did not listen to val set changes or to any change to validators at all.
I think I can refine this logic, but in a separate PR as it might complicate things a lot based on what do we need to listen for
feeder/feeder.go
Outdated
|
|
||
| func (f *Feeder) startNewVotingPeriod(vp types.VotingPeriod) { | ||
| /* | ||
| TODO(testinginprod): we need to refine this logic, other implementers did not handle this case as far as i can see. |
There was a problem hiding this comment.
Maybe it's not harmful for non-validators to send the tx, since on-chain we only grab the votes by the current validator set. I believe we can leverage the collections API to iterate through the current validator set and do point reads, so a flood of txs from random non-validator addresses won't slow down the iteration logic.
There was a problem hiding this comment.
It would just cost the non-validators sending votes gas, which is fine. The gas they pay gets accrued to the actual validators.
There was a problem hiding this comment.
Yeah Ithis was my plan, who runs this assumes they're a validator in the active set, worst case we could add a check which blocks txs from non validators which are bonded so all gets blocked at checkTX and feeder runners which are not part of the bonded set don't lose money.
There was a problem hiding this comment.
Actually,
nibiru/x/oracle/keeper/keeper.go
Line 110 in 8b2fb54
the check is already in place, so we can assume the tx from non validators (or validators which just unbonded) will be blocked at CheckTX level and not cause any loss of funds in failed txs.
Perhaps it might even make sense to just simplify the code further and get rid of the val set changes check.
There was a problem hiding this comment.
I don't think you can check if the tx is coming from an address that's part of the validator set because that's a stateful check and CheckTx is supposed to be stateless. Either the tx will fail in DeliverTx after it's included in the block, or we can add an AnteHandler to fail early and reduce the gas paid by the non-validator, though one could argue that taking away gas fees is important for preventing DoS attacks from this tx.
feeder/feeder_test.go
Outdated
| tf.validatorSetChanges <- expected | ||
| time.Sleep(10 * time.Millisecond) | ||
|
|
||
| require.Len(t, tf.f.validatorSet, len(expected.In)+1) |
There was a problem hiding this comment.
It's not necessary len(expected.In)+1, but rather the initial len(tf.validatorSet)+1.
…o mercilex/oracle-feeder-init
Description
Adds the base layer of logic for the feeder sidecar process. The PR programs the
feedertype against interfaces which will be implemented in follow-up PRs.Purpose
Helps fulfill the decentralized oracle.