[jira] [Commented] (IGNITE-2528) Deadlocks caused by Ignite.close()

2016-05-19 Thread Alexei Scherbakov (JIRA)

[ 
https://issues.apache.org/jira/browse/IGNITE-2528?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15290631#comment-15290631
 ] 

Alexei Scherbakov commented on IGNITE-2528:
---

Pull request

https://github.com/apache/ignite/pull/730

> Deadlocks caused by Ignite.close()
> --
>
> Key: IGNITE-2528
> URL: https://issues.apache.org/jira/browse/IGNITE-2528
> Project: Ignite
>  Issue Type: Bug
>  Components: general
>Affects Versions: 1.5.0.final
>Reporter: Denis Magda
>Assignee: Alexei Scherbakov
> Fix For: 1.7
>
>
> If to start stopping an Ignite instance and execute {{cluster.nodes()}} from 
> an {{EntryProcessor}} or some other place of the code that holds a lock on 
> cache's gateway then this can lead to the deadlock:
> Ignite.close:
> - holds kernel.gateway lock;
> - tries to get a gateway lock on cache A;
> Entry.processor is called for cache A:
> - a gateway lock is acquired for cache A;
> - calling {{cluster.nodes()}};
> - trying to acquire kernel's gateway lock.
> To fix this deadlock we can do the following:
> - introduce a volatile variable that has to be set to 'true' when a node is 
> being stopped;
> - check this variable before acquiring kernel's gateway.
> Also probably it makes sense to try to use try lock here.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)


[jira] [Commented] (IGNITE-2528) Deadlocks caused by Ignite.close()

2016-05-18 Thread Alexei Scherbakov (JIRA)

[ 
https://issues.apache.org/jira/browse/IGNITE-2528?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15289098#comment-15289098
 ] 

Alexei Scherbakov commented on IGNITE-2528:
---

Patch available.
Awaiting TC results.

> Deadlocks caused by Ignite.close()
> --
>
> Key: IGNITE-2528
> URL: https://issues.apache.org/jira/browse/IGNITE-2528
> Project: Ignite
>  Issue Type: Bug
>  Components: general
>Affects Versions: 1.5.0.final
>Reporter: Denis Magda
>Assignee: Alexei Scherbakov
> Fix For: 1.7
>
>
> If to start stopping an Ignite instance and execute {{cluster.nodes()}} from 
> an {{EntryProcessor}} or some other place of the code that holds a lock on 
> cache's gateway then this can lead to the deadlock:
> Ignite.close:
> - holds kernel.gateway lock;
> - tries to get a gateway lock on cache A;
> Entry.processor is called for cache A:
> - a gateway lock is acquired for cache A;
> - calling {{cluster.nodes()}};
> - trying to acquire kernel's gateway lock.
> To fix this deadlock we can do the following:
> - introduce a volatile variable that has to be set to 'true' when a node is 
> being stopped;
> - check this variable before acquiring kernel's gateway.
> Also probably it makes sense to try to use try lock here.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)


[jira] [Commented] (IGNITE-2528) Deadlocks caused by Ignite.close()

2016-05-16 Thread Alexei Scherbakov (JIRA)

[ 
https://issues.apache.org/jira/browse/IGNITE-2528?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15284665#comment-15284665
 ] 

Alexei Scherbakov commented on IGNITE-2528:
---

Created a reproducer.

> Deadlocks caused by Ignite.close()
> --
>
> Key: IGNITE-2528
> URL: https://issues.apache.org/jira/browse/IGNITE-2528
> Project: Ignite
>  Issue Type: Bug
>  Components: general
>Affects Versions: 1.5.0.final
>Reporter: Denis Magda
>Assignee: Alexei Scherbakov
> Fix For: 1.7
>
>
> If to start stopping an Ignite instance and execute {{cluster.nodes()}} from 
> an {{EntryProcessor}} or some other place of the code that holds a lock on 
> cache's gateway then this can lead to the deadlock:
> Ignite.close:
> - holds kernel.gateway lock;
> - tries to get a gateway lock on cache A;
> Entry.processor is called for cache A:
> - a gateway lock is acquired for cache A;
> - calling {{cluster.nodes()}};
> - trying to acquire kernel's gateway lock.
> To fix this deadlock we can do the following:
> - introduce a volatile variable that has to be set to 'true' when a node is 
> being stopped;
> - check this variable before acquiring kernel's gateway.
> Also probably it makes sense to try to use try lock here.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)