-
Notifications
You must be signed in to change notification settings - Fork 1.7k
Improve block and transaction propagation #9954
Conversation
@@ -583,6 +583,13 @@ impl<K: Kind> VerificationQueue<K> { | |||
result | |||
} | |||
|
|||
/// Returns true if there is nothing currently in the queue. | |||
/// TODO [ToDr] Optimize to avoid locking |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Would you mind to create a new issue and include an issue number for this TODO item?
@@ -67,14 +65,9 @@ pub struct SyncHandler; | |||
impl SyncHandler { | |||
/// Handle incoming packet from peer | |||
pub fn on_packet(sync: &mut ChainSync, io: &mut SyncIo, peer: PeerId, packet_id: u8, data: &[u8]) { | |||
if packet_id != STATUS_PACKET && !sync.peers.contains_key(&peer) { |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Note to reviewers: those logic are moved to https://github.com/paritytech/parity-ethereum/pull/9954/files#diff-6bb6c36e06ede3bf5eca86a14dfe62c8R94
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM. Awesome results :) 🥇
@@ -228,6 +228,37 @@ impl AttachedProtocol { | |||
} | |||
} | |||
|
|||
/// A prioritized tasks run in a specialised timer. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
/// A prioritized tasks run in a specialised timer. | |
/// A prioritized task runs in a specialised timer. |
} | ||
} | ||
|
||
const PEERS_TIMER: TimerToken = 0; | ||
const SYNC_TIMER: TimerToken = 1; | ||
const TX_TIMER: TimerToken = 2; | ||
const PRIORITY_TIMER: TimerToken = 3; |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
We should be careful when merging this PR with #9967 because if it doesn't conflict we end up with duplicate tokens.
* Refactor sync to add priority tasks. * Send priority tasks notifications. * Propagate blocks, optimize transactions. * Implement transaction propagation. Use sync_channel. * Tone down info. * Prevent deadlock by not waiting forever for sync lock. * Fix lock order. * Don't use sync_channel to prevent deadlocks. * Fix tests.
* version: bump beta to 2.2.2 * Add experimental RPCs flag (#9928) * WiP * Enable experimental RPCs. * Keep existing blocks when restoring a Snapshot (#8643) * Rename db_restore => client * First step: make it compile! * Second step: working implementation! * Refactoring * Fix tests * PR Grumbles * PR Grumbles WIP * Migrate ancient blocks interating backward * Early return in block migration if snapshot is aborted * Remove RwLock getter (PR Grumble I) * Remove dependency on `Client`: only used Traits * Add test for recovering aborted snapshot recovery * Add test for migrating old blocks * Fix build * PR Grumble I * PR Grumble II * PR Grumble III * PR Grumble IV * PR Grumble V * PR Grumble VI * Fix one test * Fix test * PR Grumble * PR Grumbles * PR Grumbles II * Fix tests * Release RwLock earlier * Revert Cargo.lock * Update _update ancient block_ logic: set local in `commit` * Update typo in ethcore/src/snapshot/service.rs Co-Authored-By: ngotchac <ngotchac@gmail.com> * Adjust requests costs for light client (#9925) * PIP Table Cost relative to average peers instead of max peers * Add tracing in PIP new_cost_table * Update stat peer_count * Use number of leeching peers for Light serve costs * Fix test::light_params_load_share_depends_on_max_peers (wrong type) * Remove (now) useless test * Remove `load_share` from LightParams.Config Prevent div. by 0 * Add LEECHER_COUNT_FACTOR * PR Grumble: u64 to u32 for f64 casting * Prevent u32 overflow for avg_peer_count * Add tests for LightSync::Statistics * Fix empty steps (#9939) * Don't send empty step twice or empty step then block. * Perform basic validation of locally sealed blocks. * Don't include empty step twice. * prevent silent errors in daemon mode, closes #9367 (#9946) * Fix a deadlock (#9952) * Update informant: - decimal in Mgas/s - print every 5s (not randomly between 5s and 10s) * Fix dead-lock in `blockchain.rs` * Update locks ordering * Fix light client informant while syncing (#9932) * Add `is_idle` to LightSync to check importing status * Use SyncStateWrapper to make sure is_idle gets updates * Update is_major_import to use verified queue size as well * Add comment for `is_idle` * Add Debug to `SyncStateWrapper` * `fn get` -> `fn into_inner` * ci: rearrange pipeline by logic (#9970) * ci: rearrange pipeline by logic * ci: rename docs script * fix docker build (#9971) * Deny unknown fields for chainspec (#9972) * Add deny_unknown_fields to chainspec * Add tests and fix existing one * Remove serde_ignored dependency for chainspec * Fix rpc test eth chain spec * Fix starting_nonce_test spec * Improve block and transaction propagation (#9954) * Refactor sync to add priority tasks. * Send priority tasks notifications. * Propagate blocks, optimize transactions. * Implement transaction propagation. Use sync_channel. * Tone down info. * Prevent deadlock by not waiting forever for sync lock. * Fix lock order. * Don't use sync_channel to prevent deadlocks. * Fix tests. * Fix unstable peers and slowness in sync (#9967) * Don't sync all peers after each response * Update formating * Fix tests: add `continue_sync` to `Sync_step` * Update ethcore/sync/src/chain/mod.rs Co-Authored-By: ngotchac <ngotchac@gmail.com> * fix rpc middlewares * fix Cargo.lock * json: resolve merge in spec * rpc: fix starting_nonce_test * ci: allow nightl job to fail
* Refactor sync to add priority tasks. * Send priority tasks notifications. * Propagate blocks, optimize transactions. * Implement transaction propagation. Use sync_channel. * Tone down info. * Prevent deadlock by not waiting forever for sync lock. * Fix lock order. * Don't use sync_channel to prevent deadlocks. * Fix tests.
Closes #9576
This addresses the issue identified by @AlexeyAkhunov regarding block propagation. Before the PR blocks were propagated after they were successfully imported to the DB (which means we had to execute all transactions).
This behaviour leads to a situation that I replicated in tests by running a PoA network with one validator (Node 5) and 4 regular nodes that are connected into a chain:
5 -- 4 -- 3 -- 2 --1
:To address this issue we propagate blocks right after initial verification (
engine::verify_block_basic
), which means we verify PoW or signatures in case of PoA network. Similar mechanism for priority propagation is also applied for transactions to improve their propagation as well.The results looks like this:
(the fluctation between 1.2s and 1.6s per block import is just caused by my machine).
This PR also improves receiving of transactions from other peers, as we don't require
sync.write()
lock.The fix should lower the uncle rate.