Fix P-chain validator set lookup race condition #2672
Conversation
StephenButtolph
requested review from
abi87,
danlaine,
dhrubabasu and
joshua-kim
as code owners
| validators.NewLockedState( | ||
| &chainCtx.Lock, | ||
| validatorManager, | ||
| ), |
There was a problem hiding this comment.
This is the actual fix. chainCtx.ValidatorState was replaced with validatorManager because chainCtx.ValidatorState is actually equal to vm here and I found that very confusing. Additionally, this made testing much better because we don't need to rely on mocking the ValidatorState.
| @@ -2218,6 +2225,61 @@ func TestSubnetValidatorSetAfterPrimaryNetworkValidatorRemoval(t *testing.T) { | |||
| require.NoError(err) | |||
| } | |||
|
|
|||
| func TestValidatorSetRaceCondition(t *testing.T) { | |||
There was a problem hiding this comment.
I couldn't find a cleaner way to write this test... Very open to suggestions.
| func MarshalAppRequest(filter, salt []byte) ([]byte, error) { | ||
| request := &sdk.PullGossipRequest{ | ||
| Filter: filter, | ||
| Salt: salt, | ||
| } | ||
| return proto.Marshal(request) | ||
| } | ||
|
|
||
| func ParseAppRequest(bytes []byte) (*bloom.ReadFilter, ids.ID, error) { | ||
| request := &sdk.PullGossipRequest{} | ||
| if err := proto.Unmarshal(bytes, request); err != nil { | ||
| return nil, ids.Empty, err | ||
| } | ||
|
|
||
| salt, err := ids.ToID(request.Salt) | ||
| if err != nil { | ||
| return nil, ids.Empty, err | ||
| } | ||
|
|
||
| filter, err := bloom.Parse(request.Filter) | ||
| return filter, salt, err | ||
| } | ||
|
|
||
| func MarshalAppResponse(gossip [][]byte) ([]byte, error) { | ||
| return proto.Marshal(&sdk.PullGossipResponse{ | ||
| Gossip: gossip, | ||
| }) | ||
| } | ||
|
|
||
| func ParseAppResponse(bytes []byte) ([][]byte, error) { | ||
| response := &sdk.PullGossipResponse{} | ||
| err := proto.Unmarshal(bytes, response) | ||
| return response.Gossip, err | ||
| } | ||
|
|
||
| func MarshalAppGossip(gossip [][]byte) ([]byte, error) { | ||
| return proto.Marshal(&sdk.PushGossip{ | ||
| Gossip: gossip, | ||
| }) | ||
| } | ||
|
|
||
| func ParseAppGossip(bytes []byte) ([][]byte, error) { | ||
| msg := &sdk.PushGossip{} | ||
| err := proto.Unmarshal(bytes, msg) | ||
| return msg.Gossip, err | ||
| } |
There was a problem hiding this comment.
Technically we don't need to export all of these utilities... But I felt like it made the code easier to understand... If we'd prefer to just directly marshal the proto messages I'll revert these. @joshua-kim thoughts?
There was a problem hiding this comment.
Strongly prefer this
There was a problem hiding this comment.
This is better, this is also good since we frequently re-write this during testing as well
| handlerStr := strconv.FormatUint(handlerID, 10) | ||
|
|
There was a problem hiding this comment.
We could reduce the diff here, but I don't think there is any good reason to do this with the lock held.
![github-advanced-security[bot]](https://avatars.githubusercontent.com/in/57789?s=60&v=4){kind=small}
| @@ -277,13 +277,14 @@ func defaultVM(t *testing.T, fork activeFork) (*VM, database.Database, *mutableS | |||
| return nil | |||
| } | |||
|
|
|||
| dynamicConfigBytes := []byte(`{"network":{"max-validator-set-staleness":0}}`) | |||
There was a problem hiding this comment.
There might be other configs that would be good to default to. But this disables the optimization so that we re-calculate the validator set aggressively.
| @@ -205,7 +205,10 @@ func (vm *VM) Initialize( | |||
| chainCtx.Log, | |||
| chainCtx.NodeID, | |||
| chainCtx.SubnetID, | |||
| chainCtx.ValidatorState, | |||
| validators.NewLockedState( | |||
There was a problem hiding this comment.
Are there other VMs that provide the raw state here?
Probably makes sense to double-check C/X-chain initialization.
There was a problem hiding this comment.
This is a P-chain specific bug. In the C-chain and the X-chain the ctx.ValidatorState is already locked.
Co-authored-by: Stephen Buttolph <[email protected]>
Why this should be merged
The P-chain p2p network currently fetches the primary network validator set without first grabbing the P-chain context lock. This can result in racy (incorrect) calculations of the validator set that are cached. This can cause persistently incorrect validator set lookups at P-chain heights where this race occurred.
How this works
validators.Statepassed into the p2p network with the context lock.How this was tested