[
https://issues.apache.org/jira/browse/IGNITE-8356?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Alexey Goncharuk updated IGNITE-8356:
-------------------------------------
Description:
The problem was discovered on the IGNITE-8355.
But *root cause* is the race on the start of the node discovery.
The race happens because a joining node may start processing NodeAddMessage
before processing local node's NodeAddFinishedMessage. Because of this, the
local node will not have any constructed DiscoCache yet and NPE happens.
We need to take a look at the workaround suggested in IGNITE-8355 and
1) Check if any public API changes are needed on DiscoverySpi interface
2) Verify it works for ZookeeperDiscoverySpi.
was:
The problem was discovered on the IGNITE-8355.
But *root cause* is the race on the start of the node discovery.
> Possible race at the discovery on the start node
> ------------------------------------------------
>
> Key: IGNITE-8356
> URL: https://issues.apache.org/jira/browse/IGNITE-8356
> Project: Ignite
> Issue Type: Bug
> Affects Versions: 2.4
> Reporter: Taras Ledkov
> Priority: Major
> Fix For: 2.6
>
>
> The problem was discovered on the IGNITE-8355.
> But *root cause* is the race on the start of the node discovery.
> The race happens because a joining node may start processing NodeAddMessage
> before processing local node's NodeAddFinishedMessage. Because of this, the
> local node will not have any constructed DiscoCache yet and NPE happens.
> We need to take a look at the workaround suggested in IGNITE-8355 and
> 1) Check if any public API changes are needed on DiscoverySpi interface
> 2) Verify it works for ZookeeperDiscoverySpi.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
