[
https://issues.apache.org/jira/browse/IGNITE-11289?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Ivan Pavlukhin updated IGNITE-11289:
------------------------------------
Description:
There is a possibility that {{AbstractForwardCursor.nextPageId}} refers to a
page which was already excluded from the tree. The problem reproduces with a
scan query execution with put/remove load in background.
Linked PR contains a reproducer and some tricks allowing to reproduce the
problem more often (it is still possible to reproduce it without tricks but
likelihood is significantly lower). The problem becomes evident when a
problematic page is taken from REUSE_BUCKET. But there could be other hidden
problems which do not cause any runtime errors but lead to data inconsistency.
was:
There is a possibility that {{AbstractForwardCursor.nextPageId}} refers to a
page which was already excluded from the tree. The problem reproduces with a
scan query execution with put/remove load in background.
Linked PR contains reproducer and some tricks allowing to reproduce the problem
more often (it is still possible to reproduce it without tricks but likelihood
is significantly lower). Problem becomes evident when problematic page is taken
from REUSE_BUCKET. But there could be other hidden problems which do no cause
any runtime errors but lead to data inconsistency.
> BPlusTree.AbstractForwardCursor can use obsolete page
> -----------------------------------------------------
>
> Key: IGNITE-11289
> URL: https://issues.apache.org/jira/browse/IGNITE-11289
> Project: Ignite
> Issue Type: Bug
> Components: cache
> Reporter: Ivan Pavlukhin
> Priority: Major
>
> There is a possibility that {{AbstractForwardCursor.nextPageId}} refers to a
> page which was already excluded from the tree. The problem reproduces with a
> scan query execution with put/remove load in background.
> Linked PR contains a reproducer and some tricks allowing to reproduce the
> problem more often (it is still possible to reproduce it without tricks but
> likelihood is significantly lower). The problem becomes evident when a
> problematic page is taken from REUSE_BUCKET. But there could be other hidden
> problems which do not cause any runtime errors but lead to data inconsistency.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
