-
Notifications
You must be signed in to change notification settings - Fork 1.7k
Fixed AuthorityRound deadlock on shutdown #8803
Conversation
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.
Could confirm that the issue is fixed. Code changes also LGTM (no more engine.upgrade()
calls from TransitionHandler
).
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.
Can not reproduce #8088 on Kovan with this changes.
// Make sure to advance up to the actual step. | ||
while self.step.inner.duration_remaining().as_millis() == 0 { | ||
self.step.inner.increment(); | ||
self.step.can_propose.store(true, AtomicOrdering::SeqCst); |
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.
Maybe it's worth extracting into a function:
fn step(step: &PermissionedStep, client: &RwLock<Option<Weak<EngineClient>>>) {
step.inner.increment();
step.can_propose.store(true, AtomicOrdering::SeqCst);
if let Some(ref weak) = *client.read() {
if let Some(c) = weak.upgrade() {
c.update_sealing();
}
}
}
to avoid code duplication?
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.
this code would also leak, cause it upgrades the reference to EngineClient
from worker thread
|
* master: Fix subcrate test compile (#8862) network-devp2p: downgrade logging to debug, add target (#8784) Clearing up a comment about the prefix for signing (#8828) Disable parallel verification and skip verifiying already imported txs. (#8834) devp2p: Move UDP socket handling from Discovery to Host. (#8790) Fixed AuthorityRound deadlock on shutdown, closes #8088 (#8803) Specify critical release flag per network (#8821) Fix `deadlock_detection` feature branch compilation (#8824) Use system allocator when profiling memory (#8831) added from and to to Receipt (#8756)
* Fixed AuthorityRound deadlock on shutdown, closes #8088 (#8803) * CI: Fix docker tags (#8822) * scripts: enable docker builds for beta and stable * scripts: docker latest should be beta not master * scripts: docker latest is master * Fix concurrent access to signer queue (#8854) * Fix concurrent access to signer queue * Put request back to the queue if confirmation failed * typo: fix docs and rename functions to be more specific `request_notify` does not need to be public, and it's renamed to `notify_result`. `notify` is renamed to `notify_message`. * Change trace info "Transaction" -> "Request" * Add new ovh bootnodes and fix port for foundation bootnode 3.2 (#8886) * Add new ovh bootnodes and fix port for foundation bootnode 3.2 * Remove old bootnodes. * Remove duplicate 1118980bf48b0a3640bdba04e0fe78b1add18e1cd99bf22d53daac1fd9972ad650df52176e7c7d89d1114cfef2bc23a2959aa54998a46afcf7d91809f0855082 * Block 0 is valid in queries (#8891) Early exit for block nr 0 leads to spurious error about pruning: `…your node is running with state pruning…`. Fixes #7547, #8762 * update jsonrpc libs, fixed ipc leak, closes #8774 (#8876) Instead of cherrypicking 8b78141, just ran cargo update -p jsonrpc-core * Add ETC Cooperative-run load balanced parity node (#8892) * Minor fix in chain supplier and light provider (#8906) * fix chain supplier increment * fix light provider block_headers
* `duration_ns: u64 -> duration: Duration` (#8457) * duration_ns: u64 -> duration: Duration * format on millis {:.2} -> {} * Keep all enacted blocks notify in order (#8524) * Keep all enacted blocks notify in order * Collect is unnecessary * Update ChainNotify to use ChainRouteType * Fix all ethcore fn defs * Wrap the type within ChainRoute * Fix private-tx and sync api * Fix secret_store API * Fix updater API * Fix rpc api * Fix informant api * Eagerly cache enacted/retracted and remove contain_enacted/retracted * Fix indent * tests: should use full expr form for struct constructor * Use into_enacted_retracted to further avoid copy * typo: not a function * rpc/tests: ChainRoute -> ChainRoute::new * Handle removed logs in filter changes and add geth compatibility field (#8796) * Add removed geth compatibility field in log * Fix mocked tests * Add field block hash in PollFilter * Store last block hash info for log filters * Implement canon route * Use canon logs for fetching reorg logs Light client removed logs fetching is disabled. It looks expensive. * Make sure removed flag is set * Address grumbles * Fixed AuthorityRound deadlock on shutdown, closes #8088 (#8803) * CI: Fix docker tags (#8822) * scripts: enable docker builds for beta and stable * scripts: docker latest should be beta not master * scripts: docker latest is master * ethcore: fix ancient block error msg handling (#8832) * Disable parallel verification and skip verifiying already imported txs. (#8834) * Reject transactions that are already in pool without verifying them. * Avoid verifying already imported transactions. * Fix concurrent access to signer queue (#8854) * Fix concurrent access to signer queue * Put request back to the queue if confirmation failed * typo: fix docs and rename functions to be more specific `request_notify` does not need to be public, and it's renamed to `notify_result`. `notify` is renamed to `notify_message`. * Change trace info "Transaction" -> "Request" * Don't allocate in expect_valid_rlp unless necessary (#8867) * don't allocate via format! in case there's no error * fix test? * fixed ipc leak, closes #8774 (#8876) * Add new ovh bootnodes and fix port for foundation bootnode 3.2 (#8886) * Add new ovh bootnodes and fix port for foundation bootnode 3.2 * Remove old bootnodes. * Remove duplicate 1118980bf48b0a3640bdba04e0fe78b1add18e1cd99bf22d53daac1fd9972ad650df52176e7c7d89d1114cfef2bc23a2959aa54998a46afcf7d91809f0855082 * Block 0 is valid in queries (#8891) Early exit for block nr 0 leads to spurious error about pruning: `…your node is running with state pruning…`. Fixes #7547, #8762 * Add ETC Cooperative-run load balanced parity node (#8892) * Minor fix in chain supplier and light provider (#8906) * fix chain supplier increment * fix light provider block_headers * Check whether we need resealing in miner and unwrap has_account in account_provider (#8853) * Remove unused Result wrap in has_account * Check whether we need to reseal for external transactions * Fix reference to has_account interface * typo: missing ) * Refactor duplicates to prepare_and_update_sealing * Fix build * Allow disabling local-by-default for transactions with new config entry (#8882) * Add tx_queue_allow_unknown_local config option - Previous commit messages: dispatcher checks if we have the sender account Add `tx_queue_allow_unknown_local` to MinerOptions Add `tx_queue_allow_unknown_local` to config fix order in MinerOptions to match Configuration add cli flag for tx_queue_allow_unknown_local Update refs to `tx_queue_allow_unknown_local` Add tx_queue_allow_unknown_local to config test revert changes to dispatcher Move tx_queue_allow_unknown_local to `import_own_transaction` Fix var name if statement should return the values derp de derp derp derp semicolons Reset dispatch file to how it was before fix compile issues + change from FLAG to ARG add test and use `into` import MinerOptions, clone the secret Fix tests? Compiler/linter issues fixed Fix linter msg - case of constants IT LIVES refactor to omit yucky explict return update comments Fix based on diff AccountProvider.has_account method * Refactor flag name + don't change import_own_tx behaviour fix arg name Note: force commit to try and get gitlab tests working again 😠 * Add fn to TestMinerService * Avoid race condition from trusted sources - refactor the miner tests a bit to cut down on code reuse - add `trusted` param to dispatch_transaction and import_claimed_local_transaction Add param to `import_claimed_local_transaction` Fix fn sig in tests
closes #8088
follow up for @svyatonik explanation
memory in our event loop looks as follows:
IoService
->manager_thread
->IoManager
->Workers
->workers_threads
and in case of
AuthorityRound
it was additionally:workers_threads
->TransitionHandler
->Weak<AuthorityRound>
->IoService
everything was fine, as long as we were not trying to shutdown one of the workers when he owned the strong reference of
IoService
as a solution, I refactored the code so
TransitionHandler
thread never has ownership ofIoService