[ 
https://issues.apache.org/jira/browse/IGNITE-8719?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16667382#comment-16667382
 ] 

Sergey Antonov edited comment on IGNITE-8719 at 10/29/18 4:05 PM:
------------------------------------------------------------------

[~agoncharuk] I wrote test by your scenario. But the problem wasn't reproduced. 
What I did wrong? [^IndexRebuildAfterNodeCrashTest.java] 


was (Author: antonovsergey93):
[~agoncharuk] I wrote test by your scena [^IndexRebuildAfterNodeCrashTest.java] 
rio. But the problem wasn't reproduced. What I did wrong? 
[^IndexRebuildAfterNodeCrashTest.java] 

> Index left partially built if a node crashes during index create or rebuild
> ---------------------------------------------------------------------------
>
>                 Key: IGNITE-8719
>                 URL: https://issues.apache.org/jira/browse/IGNITE-8719
>             Project: Ignite
>          Issue Type: Bug
>            Reporter: Alexey Goncharuk
>            Assignee: Sergey Antonov
>            Priority: Critical
>             Fix For: 2.8
>
>         Attachments: IndexRebuildAfterNodeCrashTest.java, 
> IndexRebuildingTest.java
>
>
> Currently, we do not have any state associated with the index tree. Consider 
> the following scenario:
> 1) Start node, put some data
> 2) start CREATE INDEX operation
> 3) Wait for a checkpoint and stop node before index create finished
> 4) Restart node
> Since the checkpoint finished, the new index tree will be persisted to the 
> disk, but not all data will be present in the index.
> We should somehow store information about initializing index tree and mark it 
> valid only after all data is indexed. The state should be persisted as well.



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

Reply via email to