geode-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Jason Huynh (JIRA)" <j...@apache.org>
Subject [jira] [Updated] (GEODE-2900) BucketRegionQueue transitions from primary/secondary/primary can lead to events lingering in queue
Date Tue, 09 May 2017 22:18:04 GMT

     [ https://issues.apache.org/jira/browse/GEODE-2900?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]

Jason Huynh updated GEODE-2900:
-------------------------------
    Description: 
In this scenario when peek() is called from BucketRegionQueue, a side effect is that the key
is removed from the eventSeqNumQueue and an event is placed into the peekedEvents list.

If there is failure dispatching the event, we add the peekedEvents list to a new peekedEvents
list.  Before doing so, we remove any events that we are not primary for.  Now the event is
not in the eventSeqNumQueue or the peekedEvents list

If we now become primary (before the other node could dispatch this event), and because we
do not have that event in the eventSeqNumQueue or the peekedEvents, it gets "stuck."

This also affects the Lucene implementation.  An stuck event can mean incorrectly indexed
data or data inconsistencies


  was:
In this scenario when peek() is called from BucketRegionQueue, a side effect is that the key
is removed from the eventSeqNumQueue and an event is placed into the peekedEvents list.

If there is failure dispatching the event, we add the peekedEvents list to a new peekedEvents
list.  Before doing so, we remove any events that we are not primary for.  Now the event is
not in the eventSeqNumQueue or the peekedEvents list

If we now become primary (before the other node could dispatch this event), and because we
do not have that event in the eventSeqNumQueue or the peekedEvents, it gets "stuck."


    Component/s: lucene

> BucketRegionQueue transitions from primary/secondary/primary can lead to events lingering
in queue
> --------------------------------------------------------------------------------------------------
>
>                 Key: GEODE-2900
>                 URL: https://issues.apache.org/jira/browse/GEODE-2900
>             Project: Geode
>          Issue Type: Bug
>          Components: lucene, wan
>            Reporter: Jason Huynh
>            Assignee: Jason Huynh
>
> In this scenario when peek() is called from BucketRegionQueue, a side effect is that
the key is removed from the eventSeqNumQueue and an event is placed into the peekedEvents
list.
> If there is failure dispatching the event, we add the peekedEvents list to a new peekedEvents
list.  Before doing so, we remove any events that we are not primary for.  Now the event is
not in the eventSeqNumQueue or the peekedEvents list
> If we now become primary (before the other node could dispatch this event), and because
we do not have that event in the eventSeqNumQueue or the peekedEvents, it gets "stuck."
> This also affects the Lucene implementation.  An stuck event can mean incorrectly indexed
data or data inconsistencies



--
This message was sent by Atlassian JIRA
(v6.3.15#6346)

Mime
View raw message