[ 
https://issues.apache.org/jira/browse/IGNITE-6695?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Alexey Goncharuk updated IGNITE-6695:
-------------------------------------
    Labels: IEP-4 Phase-2  (was: IEP-4)

> Validation of joining node data consistency WRT the same data in grid
> ---------------------------------------------------------------------
>
>                 Key: IGNITE-6695
>                 URL: https://issues.apache.org/jira/browse/IGNITE-6695
>             Project: Ignite
>          Issue Type: New Feature
>          Components: persistence
>            Reporter: Sergey Chugunov
>            Priority: Major
>              Labels: IEP-4, Phase-2
>
> h2. Scenario
> Consider the following simple scenario (persistence is active):
> # Start nodes A and B, activate, add (K1, V1) to cache.
> # Stop A; update K1 to (K1, V2) (only B is aware of update). Stop B.
> # Start A, activate, update K1 to (K1, V3).
> After that B joining the cluster will lead to ambiguity of K1 value.
> Also even having BaselineTopology tracking history of cluster nodes 
> activations won't help here as after #3 node B's history is compatible with 
> node A's history.
> h2. Description
> When there is load of data updates and user turns off nodes one by one, it is 
> important to start nodes back in the opposite order. Node turned off the last 
> must be started first and so one.
> If it is not the case, situations like described above may happen.
> A mechanism to detect this scenarios and refuse to join nodes with 
> potentially conflicting data is needed.



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

Reply via email to