[
https://issues.apache.org/jira/browse/IGNITE-23880?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Kirill Sizov updated IGNITE-23880:
-----------------------------------
Description:
h3. Motivation
For the purpose of the partition graceful start
https://issues.apache.org/jira/browse/IGNITE-23874
we need to implement the functionality to check if raft group is operable for
the specified partition.
In the flow of the partition graceful start, when a node sends a request to a
node from the next chain,
this node must immediately respond that raft group for the partition is
opereable, so the node that sent this request could
decide that it should not start a raft group.
h3. Implementation notes
Granular node aliveness check is already implemented, see
{{DisasterRecoveryManager.localPartitionStatesInternal}}, partition aliveness
check is also implemented in
{{DisasterRecoveryManager.assembleGlobalStateFromLocal}} for non-HA mode.
What needs to be done in the scope of this task is to check partition state for
HA mode.
The difference between HA and non-HA is that we want to check partition
aliveness based on the last chain link size, not on the zone's replicas number.
*For example:*
Imagine, we have a group of 7 nodes, A-G. Four nodes (that form the majority)
die: A-D die, EFG stay online.
Automatic reset triggers with E as pending, EFG as planned.
Once it is completely executed, the new stable will be EFG. The chain will be
A-G -> EFG.
If node G restarts, it should check whether the group EFG is up and running,
not A-G.
h3. Definition of done
* Node must be able to check if raft group for the provided partition is alive
and operable.
was:
h3. Motivation
For the purpose of the partition graceful start
https://issues.apache.org/jira/browse/IGNITE-23874
we need to implement the functionality to check if raft group is operable for
the specified partition.
In the flow of the partition graceful start, when a node sends a request to a
node from the next chain,
this node must immediately respond that raft group for the partition is
opereable, so the node that sent this request could
decide that it should not start a raft group.
h3. Implementation notes
h3. Definition of done
* Node must be able to check if raft group for the provided partition is alive
and operable.
> Implement node aliveness check
> ------------------------------
>
> Key: IGNITE-23880
> URL: https://issues.apache.org/jira/browse/IGNITE-23880
> Project: Ignite
> Issue Type: New Feature
> Reporter: Alexander Lapin
> Priority: Major
> Labels: ignite-3
>
> h3. Motivation
> For the purpose of the partition graceful start
> https://issues.apache.org/jira/browse/IGNITE-23874
> we need to implement the functionality to check if raft group is operable for
> the specified partition.
> In the flow of the partition graceful start, when a node sends a request to a
> node from the next chain,
> this node must immediately respond that raft group for the partition is
> opereable, so the node that sent this request could
> decide that it should not start a raft group.
> h3. Implementation notes
> Granular node aliveness check is already implemented, see
> {{DisasterRecoveryManager.localPartitionStatesInternal}}, partition aliveness
> check is also implemented in
> {{DisasterRecoveryManager.assembleGlobalStateFromLocal}} for non-HA mode.
> What needs to be done in the scope of this task is to check partition state
> for HA mode.
> The difference between HA and non-HA is that we want to check partition
> aliveness based on the last chain link size, not on the zone's replicas
> number.
> *For example:*
> Imagine, we have a group of 7 nodes, A-G. Four nodes (that form the majority)
> die: A-D die, EFG stay online.
> Automatic reset triggers with E as pending, EFG as planned.
> Once it is completely executed, the new stable will be EFG. The chain will be
> A-G -> EFG.
> If node G restarts, it should check whether the group EFG is up and running,
> not A-G.
>
> h3. Definition of done
> * Node must be able to check if raft group for the provided partition is
> alive and operable.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)