[ 
https://issues.apache.org/jira/browse/HBASE-20597?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Andrew Purtell updated HBASE-20597:
-----------------------------------
    Description: The code that closes down a ZKW that fails to initialize when 
attempting to connect to the remote cluster is not MT safe and can in theory 
leak ZooKeeperWatcher instances. The allocation of a new ZKW and store to the 
reference is not atomic. Might have concurrent allocations with only one 
winning store, leading to leaked ZKW instances.   (was: The code that closes 
down a ZKW that fails to initialize when attempting to connect to the remote 
cluster is not MT safe and can in theory leak ZooKeeperWatcher instances. The 
allocation of a new ZKW and store to the reference is not atomic. Might have 
concurrent allocations with only one winning store, leading to leaked ZKW 
instances. If the connection problem is persistent, like loss of shared trust 
between the clusters, we may accumulate unclosed ZKW instances over time, with 
a ZK send thread and event thread each, and eventually have enough leaked 
threads to cause OOME (cannot allocate native thread). )

> Use a lock to serialize access to a shared reference to ZooKeeperWatcher in 
> HBaseReplicationEndpoint
> ----------------------------------------------------------------------------------------------------
>
>                 Key: HBASE-20597
>                 URL: https://issues.apache.org/jira/browse/HBASE-20597
>             Project: HBase
>          Issue Type: Bug
>    Affects Versions: 1.3.2, 1.4.4
>            Reporter: Andrew Purtell
>            Assignee: Andrew Purtell
>            Priority: Minor
>             Fix For: 3.0.0, 2.1.0, 1.5.0, 1.3.3, 2.0.1, 1.4.5
>
>
> The code that closes down a ZKW that fails to initialize when attempting to 
> connect to the remote cluster is not MT safe and can in theory leak 
> ZooKeeperWatcher instances. The allocation of a new ZKW and store to the 
> reference is not atomic. Might have concurrent allocations with only one 
> winning store, leading to leaked ZKW instances. 



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

Reply via email to