[
https://issues.apache.org/jira/browse/YUNIKORN-793?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Wilfred Spiegelenburg closed YUNIKORN-793.
------------------------------------------
Fix Version/s: 1.0.0
Target Version: 1.0.0
Resolution: Fixed
Change committed
Jira linked to the original Jira that introduced the new code.
> fix deadlock caused by listing queues with scheduling pending pods
> ------------------------------------------------------------------
>
> Key: YUNIKORN-793
> URL: https://issues.apache.org/jira/browse/YUNIKORN-793
> Project: Apache YuniKorn
> Issue Type: Bug
> Reporter: Chia-Ping Tsai
> Assignee: Chia-Ping Tsai
> Priority: Blocker
> Labels: pull-request-available
> Fix For: 1.0.0
>
>
> `GetPartitionQueues` calls read lock multiple times. If there is a thread
> which is waiting write lock, it can exclude all new read locks. In short, the
> following execution order can cause dead lock.
> 1. hold read lock ---> thread 0
> 2. wait write lock ---> thread 1 is locked by thread 0
> 3. acquire read lock ---> thread 0 is locked by thread 1
> see docs for more details ([https://pkg.go.dev/sync#RWMutex])
> The pprof is shown below.
> {noformat}
> 1 @ 0x43ada5 0x44ca85 0x44ca6e 0x46de27 0x47ce25 0x47e590 0x47e522 0x9c5753
> 0x9ad231 0x9fbcfa 0x9e4986 0x9e4851 0x9de655 0x9ff792 0x471c61
> # 0x46de26 sync.runtime_SemacquireMutex+0x46
>
> /Users/chia7712/Library/go/default/src/runtime/sema.go:71
> # 0x47ce24 sync.(*Mutex).lockSlow+0x104
>
> /Users/chia7712/Library/go/default/src/sync/mutex.go:138
> # 0x47e58f sync.(*Mutex).Lock+0x8f
>
> /Users/chia7712/Library/go/default/src/sync/mutex.go:81
> # 0x47e521 sync.(*RWMutex).Lock+0x21
>
> /Users/chia7712/Library/go/default/src/sync/rwmutex.go:111
> # 0x9c5752
> github.com/apache/incubator-yunikorn-core/pkg/scheduler/objects.(*Queue).incPendingResource+0x52
>
> /Users/chia7712/go/pkg/mod/github.com/chia7712/incubator-yunikorn-core@v0.0.0-20210811001640-eaa6afb10b62/pkg/scheduler/objects/queue.go:454
> 1 @ 0x43ada5 0x44ca85 0x44ca6e 0x46de27 0x9c7eae 0x9c7e34 0x9c51f8 0x9c54ab
> 0xa59e45 0xa59e94 0x711024 0xa5b310 0x711024 0xa011b3 0x7145e3 0x70fb0d
> 0x471c61
> # 0x46de26 sync.runtime_SemacquireMutex+0x46
>
> /Users/chia7712/Library/go/default/src/runtime/sema.go:71
> # 0x9c7ead sync.(*RWMutex).RLock+0xad
>
> /Users/chia7712/Library/go/default/src/sync/rwmutex.go:63
> # 0x9c7e33
> github.com/apache/incubator-yunikorn-core/pkg/scheduler/objects.(*Queue).IsLeafQueue+0x33
>
> /Users/chia7712/go/pkg/mod/github.com/chia7712/incubator-yunikorn-core@v0.0.0-20210811001640-eaa6afb10b62/pkg/scheduler/objects/queue.go:667
> # 0x9c51f7
> github.com/apache/incubator-yunikorn-core/pkg/scheduler/objects.(*Queue).GetPartitionQueues+0x1f7
>
> /Users/chia7712/go/pkg/mod/github.com/chia7712/incubator-yunikorn-core@v0.0.0-20210811001640-eaa6afb10b62/pkg/scheduler/objects/queue.go:426
> # 0x9c54aa
> github.com/apache/incubator-yunikorn-core/pkg/scheduler/objects.(*Queue).GetPartitionQueues+0x4aa
>
> /Users/chia7712/go/pkg/mod/github.com/chia7712/incubator-yunikorn-core@v0.0.0-20210811001640-eaa6afb10b62/pkg/scheduler/objects/queue.go:416
> {noformat}
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscr...@yunikorn.apache.org
For additional commands, e-mail: dev-h...@yunikorn.apache.org
