[ 
https://issues.apache.org/jira/browse/IGNITE-4205?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15724582#comment-15724582
 ] 

Igor Rudyak commented on IGNITE-4205:
-------------------------------------

Valentin I can easily do this, but it looks like IgniteThread implementation 
doesn't have anything special related to marchalling/unmarshalling - just only 
a number of getters and setters. Thus it looks like it shouldn't work. Am I 
wrong?

> CassandraCacheStore should start IgniteThread threads in loadCache() method
> ---------------------------------------------------------------------------
>
>                 Key: IGNITE-4205
>                 URL: https://issues.apache.org/jira/browse/IGNITE-4205
>             Project: Ignite
>          Issue Type: Bug
>          Components: cache
>    Affects Versions: 1.7
>            Reporter: Valentin Kulichenko
>            Assignee: Igor Rudyak
>
> {{CassandraCacheStore.loadCache()}} method starts a generic thread pool for 
> parallel data load. Threads in this thread pool can't deserialize Ignite 
> internal objects (e.g. {{IgniteKernal}}) which can cause unexpected behavior. 
> Here is one of the scenarios:
> * There is column in Cassandra which stores an object as BLOB using 
> {{JavaSerializer}}.
> * {{CacheConfiguration.storeKeepBinary}} is {{true}}.
> * When an object is saved, it's passed to the store as an instance of 
> {{BinaryObject}} which is converted to a byte array and saved in Cassandra.
> * When the same object is loaded in {{loadCache}}, the store takes the byte 
> array and tries to convert it to {{BinaryObject}}. But it can't because this 
> implies calling {{IgnitionEx.localIgnite()}} from non-Ignite thread.
> To fix this we need to provide a thread factory that will create instances of 
> {{IgniteThread}} and use it in the pool that loads the data.
> Most likely the same issue exists in {{CacheAbstractJdbcStore}}.
> And in general, any threads created by Ignite internals should be 
> {{IgniteThread}}-s. This should be revisited.



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

Reply via email to