Files
pezkuwi-subxt/polkadot/node/core/proposer/src/lib.rs
T
Peter Goodspeed-Niklaus 1a25c41277 start working on building the real overseer (#1795)
* start working on building the real overseer

Unfortunately, this fails to compile right now due to an upstream
failure to compile which is probably brought on by a recent upgrade
to rustc v1.47.

* fill in AllSubsystems internal constructors

* replace fn make_metrics with Metrics::attempt_to_register

* update to account for #1740

* remove Metrics::register, rename Metrics::attempt_to_register

* add 'static bounds to real_overseer type params

* pass authority_discovery and network_service to real_overseer

It's not straightforwardly obvious that this is the best way to handle
the case when there is no authority discovery service, but it seems
to be the best option available at the moment.

* select a proper database configuration for the availability store db

* use subdirectory for av-store database path

* apply Basti's patch which avoids needing to parameterize everything on Block

* simplify path extraction

* get all tests to compile

* Fix Prometheus double-registry error

for debugging purposes, added this to node/subsystem-util/src/lib.rs:472-476:

```rust
Some(registry) => Self::try_register(registry).map_err(|err| {
	eprintln!("PrometheusError calling {}::register: {:?}", std::any::type_name::<Self>(), err);
	err
}),
```

That pointed out where the registration was failing, which led to
this fix. The test still doesn't pass, but it now fails in a new
and different way!

* authorities must have authority discovery, but not necessarily overseer handlers

* fix broken SpawnedSubsystem impls

detailed logging determined that using the `Box::new` style of
future generation, the `self.run` method was never being called,
leading to dropped receivers / closed senders for those subsystems,
causing the overseer to shut down immediately.

This is not the final fix needed to get things working properly,
but it's a good start.

* use prometheus properly

Prometheus lets us register simple counters, which aren't very
interesting. It also allows us to register CounterVecs, which are.
With a CounterVec, you can provide a set of labels, which can
later be used to filter the counts.

We were using them wrong, though. This pattern was repeated in a
variety of places in the code:

```rust
// panics with an cardinality mismatch
let my_counter = register(CounterVec::new(opts, &["succeeded", "failed"])?, registry)?;
my_counter.with_label_values(&["succeeded"]).inc()
```

The problem is that the labels provided in the constructor are not
the set of legal values which can be annotated, but a set of individual
label names which can have individual, arbitrary values.

This commit fixes that.

* get av-store subsystem to actually run properly and not die on first signal

* typo fix: incomming -> incoming

* don't disable authority discovery in test nodes

* Fix rococo-v1 missing session keys

* Update node/core/av-store/Cargo.toml

* try dummying out av-store on non-full-nodes

* overseer and subsystems are required only for full nodes

* Reduce the amount of warnings on browser target

* Fix two more warnings

* InclusionInherent should actually have an Inherent module on rococo

* Ancestry: don't return genesis' parent hash

* Update Cargo.lock

* fix broken test

* update test script: specify chainspec as script argument

* Apply suggestions from code review

Co-authored-by: Bastian Köcher <bkchr@users.noreply.github.com>

* Update node/service/src/lib.rs

Co-authored-by: Bastian Köcher <bkchr@users.noreply.github.com>

* node/service/src/lib: Return error via ? operator

* post-merge blues

* add is_collator flag

* prevent occasional av-store test panic

* simplify fix; expand application

* run authority_discovery in Role::Discover when collating

* distinguish between proposer closed channel errors

* add IsCollator enum, remove is_collator CLI flag

* improve formatting

* remove nop loop

* Fix some stuff

Co-authored-by: Andronik Ordian <write@reusable.software>
Co-authored-by: Bastian Köcher <git@kchr.de>
Co-authored-by: Fedor Sakharov <fedor.sakharov@gmail.com>
Co-authored-by: Robert Habermeier <robert@Roberts-MBP.lan1>
Co-authored-by: Bastian Köcher <bkchr@users.noreply.github.com>
Co-authored-by: Max Inden <mail@max-inden.de>
2020-10-28 10:26:50 +00:00

292 lines
9.4 KiB
Rust

// Copyright 2020 Parity Technologies (UK) Ltd.
// This file is part of Polkadot.
// Polkadot is free software: you can redistribute it and/or modify
// it under the terms of the GNU General Public License as published by
// the Free Software Foundation, either version 3 of the License, or
// (at your option) any later version.
// Polkadot is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
// GNU General Public License for more details.
// You should have received a copy of the GNU General Public License
// along with Polkadot. If not, see <http://www.gnu.org/licenses/>.
//! The proposer proposes new blocks to include
#![deny(unused_crate_dependencies, unused_results)]
use futures::prelude::*;
use futures::select;
use polkadot_node_subsystem::{messages::{AllMessages, ProvisionerInherentData, ProvisionerMessage}, SubsystemError};
use polkadot_overseer::OverseerHandler;
use polkadot_primitives::v1::{
Block, Hash, Header,
};
use sc_block_builder::{BlockBuilderApi, BlockBuilderProvider};
use sp_core::traits::SpawnNamed;
use sp_api::{ApiExt, ProvideRuntimeApi};
use sp_blockchain::HeaderBackend;
use sp_consensus::{Proposal, RecordProof};
use sp_inherents::InherentData;
use sp_runtime::traits::{DigestFor, HashFor};
use sp_transaction_pool::TransactionPool;
use prometheus_endpoint::Registry as PrometheusRegistry;
use std::{fmt, pin::Pin, sync::Arc, time};
/// How long proposal can take before we give up and err out
const PROPOSE_TIMEOUT: core::time::Duration = core::time::Duration::from_secs(2);
/// Custom Proposer factory for Polkadot
pub struct ProposerFactory<TxPool, Backend, Client> {
inner: sc_basic_authorship::ProposerFactory<TxPool, Backend, Client>,
overseer: OverseerHandler,
}
impl<TxPool, Backend, Client> ProposerFactory<TxPool, Backend, Client> {
pub fn new(
spawn_handle: impl SpawnNamed + 'static,
client: Arc<Client>,
transaction_pool: Arc<TxPool>,
overseer: OverseerHandler,
prometheus: Option<&PrometheusRegistry>,
) -> Self {
ProposerFactory {
inner: sc_basic_authorship::ProposerFactory::new(
spawn_handle,
client,
transaction_pool,
prometheus,
),
overseer,
}
}
}
impl<TxPool, Backend, Client> sp_consensus::Environment<Block>
for ProposerFactory<TxPool, Backend, Client>
where
TxPool: 'static + TransactionPool<Block = Block>,
Client: 'static
+ BlockBuilderProvider<Backend, Block, Client>
+ ProvideRuntimeApi<Block>
+ HeaderBackend<Block>
+ Send
+ Sync,
Client::Api:
BlockBuilderApi<Block> + ApiExt<Block, Error = sp_blockchain::Error>,
Backend:
'static + sc_client_api::Backend<Block, State = sp_api::StateBackendFor<Client, Block>>,
// Rust bug: https://github.com/rust-lang/rust/issues/24159
sp_api::StateBackendFor<Client, Block>: sp_api::StateBackend<HashFor<Block>> + Send,
{
type CreateProposer = Pin<Box<
dyn Future<Output = Result<Self::Proposer, Self::Error>> + Send + 'static,
>>;
type Proposer = Proposer<TxPool, Backend, Client>;
type Error = Error;
fn init(&mut self, parent_header: &Header) -> Self::CreateProposer {
// create the inner proposer
let proposer = self.inner.init(parent_header).into_inner();
// data to be moved into the future
let overseer = self.overseer.clone();
let parent_header_hash = parent_header.hash();
async move {
Ok(Proposer {
inner: proposer?,
overseer,
parent_header_hash,
})
}.boxed()
}
}
/// Custom Proposer for Polkadot.
///
/// This proposer gets the ProvisionerInherentData and injects it into the wrapped
/// proposer's inherent data, then delegates the actual proposal generation.
pub struct Proposer<TxPool: TransactionPool<Block = Block>, Backend, Client> {
inner: sc_basic_authorship::Proposer<Backend, Block, Client, TxPool>,
overseer: OverseerHandler,
parent_header_hash: Hash,
}
// This impl has the same generic bounds as the Proposer impl.
impl<TxPool, Backend, Client> Proposer<TxPool, Backend, Client>
where
TxPool: 'static + TransactionPool<Block = Block>,
Client: 'static
+ BlockBuilderProvider<Backend, Block, Client>
+ ProvideRuntimeApi<Block>
+ HeaderBackend<Block>
+ Send
+ Sync,
Client::Api:
BlockBuilderApi<Block> + ApiExt<Block, Error = sp_blockchain::Error>,
Backend:
'static + sc_client_api::Backend<Block, State = sp_api::StateBackendFor<Client, Block>>,
// Rust bug: https://github.com/rust-lang/rust/issues/24159
sp_api::StateBackendFor<Client, Block>: sp_api::StateBackend<HashFor<Block>> + Send,
{
/// Get provisioner inherent data
///
/// This function has a constant timeout: `PROPOSE_TIMEOUT`.
fn get_provisioner_data(&self) -> impl Future<Output = Result<ProvisionerInherentData, Error>> {
// clone this (lightweight) data because we're going to move it into the future
let mut overseer = self.overseer.clone();
let parent_header_hash = self.parent_header_hash.clone();
let mut provisioner_inherent_data = async move {
let (sender, receiver) = futures::channel::oneshot::channel();
overseer.wait_for_activation(parent_header_hash, sender).await?;
receiver.await.map_err(|_| Error::ClosedChannelAwaitingActivation)??;
let (sender, receiver) = futures::channel::oneshot::channel();
// strictly speaking, we don't _have_ to .await this send_msg before opening the
// receiver; it's possible that the response there would be ready slightly before
// this call completes. IMO it's not worth the hassle or overhead of spawning a
// distinct task for that kind of miniscule efficiency improvement.
overseer.send_msg(AllMessages::Provisioner(
ProvisionerMessage::RequestInherentData(parent_header_hash, sender),
)).await?;
receiver.await.map_err(|_| Error::ClosedChannelAwaitingInherentData)
}
.boxed()
.fuse();
let mut timeout = wasm_timer::Delay::new(PROPOSE_TIMEOUT).fuse();
async move {
select! {
pid = provisioner_inherent_data => pid,
_ = timeout => Err(Error::Timeout),
}
}
}
}
impl<TxPool, Backend, Client> sp_consensus::Proposer<Block> for Proposer<TxPool, Backend, Client>
where
TxPool: 'static + TransactionPool<Block = Block>,
Client: 'static
+ BlockBuilderProvider<Backend, Block, Client>
+ ProvideRuntimeApi<Block>
+ HeaderBackend<Block>
+ Send
+ Sync,
Client::Api:
BlockBuilderApi<Block> + ApiExt<Block, Error = sp_blockchain::Error>,
Backend:
'static + sc_client_api::Backend<Block, State = sp_api::StateBackendFor<Client, Block>>,
// Rust bug: https://github.com/rust-lang/rust/issues/24159
sp_api::StateBackendFor<Client, Block>: sp_api::StateBackend<HashFor<Block>> + Send,
{
type Transaction = sc_client_api::TransactionFor<Backend, Block>;
type Proposal = Pin<Box<
dyn Future<Output = Result<Proposal<Block, sp_api::TransactionFor<Client, Block>>, Error>> + Send,
>>;
type Error = Error;
fn propose(
self,
mut inherent_data: InherentData,
inherent_digests: DigestFor<Block>,
max_duration: time::Duration,
record_proof: RecordProof,
) -> Self::Proposal {
let provisioner_data = self.get_provisioner_data();
async move {
let provisioner_data = match provisioner_data.await {
Ok(pd) => pd,
Err(err) => {
log::warn!("could not get provisioner inherent data; injecting default data: {}", err);
Default::default()
}
};
inherent_data.put_data(
polkadot_primitives::v1::INCLUSION_INHERENT_IDENTIFIER,
&provisioner_data,
)?;
self.inner
.propose(inherent_data, inherent_digests, max_duration, record_proof)
.await
.map_err(Into::into)
}
.boxed()
}
}
// It would have been more ergonomic to use thiserror to derive the
// From implementations, Display, and std::error::Error, but unfortunately
// one of the wrapped errors (sp_inherents::Error) also
// don't impl std::error::Error, which breaks the thiserror derive.
#[derive(Debug)]
pub enum Error {
Consensus(sp_consensus::Error),
Blockchain(sp_blockchain::Error),
Inherent(sp_inherents::Error),
Timeout,
ClosedChannelAwaitingActivation,
ClosedChannelAwaitingInherentData,
Subsystem(SubsystemError)
}
impl From<sp_consensus::Error> for Error {
fn from(e: sp_consensus::Error) -> Error {
Error::Consensus(e)
}
}
impl From<sp_blockchain::Error> for Error {
fn from(e: sp_blockchain::Error) -> Error {
Error::Blockchain(e)
}
}
impl From<sp_inherents::Error> for Error {
fn from(e: sp_inherents::Error) -> Error {
Error::Inherent(e)
}
}
impl From<SubsystemError> for Error {
fn from(e: SubsystemError) -> Error {
Error::Subsystem(e)
}
}
impl fmt::Display for Error {
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
match self {
Self::Consensus(err) => write!(f, "consensus error: {}", err),
Self::Blockchain(err) => write!(f, "blockchain error: {}", err),
Self::Inherent(err) => write!(f, "inherent error: {:?}", err),
Self::Timeout => write!(f, "timeout: provisioner did not return inherent data after {:?}", PROPOSE_TIMEOUT),
Self::ClosedChannelAwaitingActivation => write!(f, "closed channel from overseer when awaiting activation"),
Self::ClosedChannelAwaitingInherentData => write!(f, "closed channel from provisioner when awaiting inherent data"),
Self::Subsystem(err) => write!(f, "subsystem error: {:?}", err),
}
}
}
impl std::error::Error for Error {
fn source(&self) -> Option<&(dyn std::error::Error + 'static)> {
match self {
Self::Consensus(err) => Some(err),
Self::Blockchain(err) => Some(err),
Self::Subsystem(err) => Some(err),
_ => None
}
}
}