Skip to content
This repository has been archived by the owner on Nov 6, 2020. It is now read-only.
This repository has been archived by the owner on Nov 6, 2020. It is now read-only.

Panic (deadlock) on Parity shutdown when using AuthorityRound engine #8088

Closed
svyatonik opened this issue Mar 12, 2018 · 1 comment · Fixed by #8803
Closed

Panic (deadlock) on Parity shutdown when using AuthorityRound engine #8088

svyatonik opened this issue Mar 12, 2018 · 1 comment · Fixed by #8803
Assignees
Labels
F1-panic 🔨 The client panics and exits without proper error handling. M4-core ⛓ Core client code / Rust.
Milestone

Comments

@svyatonik
Copy link
Collaborator

Environment: linux + branch with up-to-date master merged-in (origin/secretstore_doc_key_via_contract) + PoA of three nodes (configs are generated using this script).

When trying to stop parity instances, I often see:

thread '<unnamed>' panicked at 'failed to join thread: Resource deadlock avoided (os error 35)', /checkout/src/libstd/sys/unix/thread.rs:178:12

The reason is that we're trying to drop IO service from IO worker thread. Here are stacks of two threads that are trying to join each other: https://gist.github.com/svyatonik/5f1251da6990df46b7033ff843b7a9ee . In this case, IO worker thread was the last thread holding reference to the Arc<AuthorityRound> in this block. Arc<AuthorityRound>, in turn was the last holder of IO service reference => when Arc<AuthorityRound> goes out of scope, we're sending IoMessage::Shutdown message to service thread and start joining this thread. The reaction to this message is clearing and joining all the IO workers (including the one who's currently dropping the service) => deadlock, which is detected by pthread_join.

@svyatonik svyatonik added M4-core ⛓ Core client code / Rust. F1-panic 🔨 The client panics and exits without proper error handling. labels Mar 12, 2018
@5chdn 5chdn added this to the 1.11 milestone Mar 28, 2018
@5chdn 5chdn modified the milestones: 1.11, 1.12 Apr 24, 2018
@saulbein
Copy link

Having same issue on linux environment, with Parity 1.10.4, on Kovan, with large number of addresses

@debris debris self-assigned this Jun 5, 2018
dvdplm added a commit that referenced this issue Jun 11, 2018
* 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)
5chdn pushed a commit that referenced this issue Jun 19, 2018
* 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
5chdn pushed a commit that referenced this issue Jun 19, 2018
* `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
Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Labels
F1-panic 🔨 The client panics and exits without proper error handling. M4-core ⛓ Core client code / Rust.
Projects
None yet
Development

Successfully merging a pull request may close this issue.

4 participants