This repository was archived by the owner on Nov 6, 2020. It is now read-only.
Fix unstable peers and slowness in sync#9967
Merged
Merged
Conversation
sorpaas
approved these changes
Nov 26, 2018
sorpaas
reviewed
Nov 27, 2018
Contributor
|
Should we backport this as well? |
Co-Authored-By: ngotchac <ngotchac@gmail.com>
5chdn
pushed a commit
that referenced
this pull request
Nov 28, 2018
* 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>
Contributor
|
@ngotchac just out of curiosity, how did you generate the graphs? |
5chdn
added a commit
that referenced
this pull request
Nov 29, 2018
* 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
Contributor
Author
|
@folsen I wrote a small Rust script that runs the given binary with a given config basically, then prints the graphs using |
Contributor
|
@ngotchac definitely interested, seems to cross over with what some of the QA guys are working on |
Contributor
Author
|
The script used to get the graphs : https://github.com/ngotchac/eth-metrics |
niklasad1
pushed a commit
that referenced
this pull request
Dec 16, 2018
* 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>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
This PR fixes some issues with the stability of peer counts, and overall slowness in block sync.
Here are graphs that compares block height against time for the current




stableand this branch:Here for the peer counts:
The issue was that
sync.continue_syncwas called every time a packet was received, so it had to go through the whole list of peers, randomize it, etc.. Which could be quite expensive when connected to a lot of peers.Now it is called on a timer, every 2.5 seconds.