* sc-consensus-beefy: improve beefy gossip validator
Old gossip validator was pretty dumb, being very permissive with
incoming votes - only condition it had was to be newer than best
finalized.
New filter conditions:
- voter rounds are initialized (discarding votes until voter is
actually active),
- only votes for current active set id are accepted,
- only votes for rounds in the current voting session are accepted,
- only votes for GRANDPA finalized blocks are accepted,
- when BEEFY voter reaches mandatory round, only votes for said
mandatory round are accepted.
New validator uses the VoterOracle to easily implement above conditions
and only allow through votes that are immediately useful to the voter.
After every GRANDPA or BEEFY finality, the gossip validator filter is
updated.
* sc-consensus-beefy: remove votes enqueueing
Since gossip validator will simply disallow votes for future rounds,
and only allow votes that the voter can immediately process, there
is no need for the voter to enqueue votes.
It will see these "future" votes later in rebroadcasts, when voter
will also be able to process them. Only at that point does gossip
accept and consume them.
* sc-consensus-beefy: refactor persistent state
Move best-beefy and best-grandpa into VoterOracle instead
of passing them around as params.
VoterOracle ultimately needs to know best-beefy and/or best-grandpa
for most of its functions.
* sc-consensus-beefy: further restrict gossip validator
Assuming mandatory done in current session:
Instead of allowing votes for any round in the current session, only
accept votes for rounds equal or better than best BEEFY finalized.
* sc-consensus-beefy: add a couple of comments
* sc-consensus-beefy: fix tests involving multiple tasks
Finalize blocks one a time in tests where we want gossip to happen
in a certain round. Otherwise, some tasks may be left behind in
terms of gossip round numbers because once "scheduled" a task will
greedily process as much as possible.
This change should be in line with the real-world scenario where
voters run "in parallel" across nodes, the only points of
synchronization being the finality notifications.
* sc-consensus-beefy: address review comments
---------
Signed-off-by: acatangiu <adrian@parity.io>
`futures_util::pending!()` macro only yields to the event loop once,
resulting in many calls to the `OnDemandJustificationsEngine::next()`
made by the tokio reactor even though the on-demand-engine is idle.
Replace it with the correct `futures::future::pending::<()>()` function
which forever yields to the event loop, which is what we want when
the engine is idle.
Signed-off-by: Adrian Catangiu <adrian@parity.io>
* Move service tests to `client/network/tests`
These tests depend on `sc-network` and `sc-network-sync` so they should
live outside the crate.
* Move some configs from `sc-network-common` to `sc-network`
* Move `NetworkService` traits to `sc-network`
* Move request-responses to `sc-network`
* Remove more stuff
* Remove rest of configs from `sc-network-common` to `sc-network`
* Remove more stuff
* Fix warnings
* Update client/network/src/request_responses.rs
Co-authored-by: Dmitry Markin <dmitry@markin.tech>
* Fix cargo doc
---------
Co-authored-by: Dmitry Markin <dmitry@markin.tech>
Adrian Catangiu