[
https://issues.apache.org/jira/browse/IGNITE-18960?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Denis Chudov updated IGNITE-18960:
----------------------------------
Description:
This is a responsibility of replication group to notify the placement driver
about becoming inoperable, for example, about majority loss in case of Raft.
The corresponding message should be sent from leaseholder to placement driver
so that the placement driver would not prolong the lease for this leaseholder
automatically.
As the replica may not know about the actual active actor of placement driver,
it should send the message to every node in placement driver group. On the
placement driver side, there may be no active actor at the moment. In this
case, the replica should wait for next lease prolongation (which means that
active actor has appeared) and after that repeat the messages to placement
driver group.
The active actor should send LeaseGrantMessage for this leaseholder, which
should, in the same way as while initial lease granting process, wait for group
operability to send response.
was:
This is a responsibility of replication group to notify the placement driver
about becoming inoperable, for example, about majority loss in case of Raft.
The corresponding message should be sent from leaseholder to placement driver
so that the placement driver would not prolong the lease for this leaseholder
automatically. The placement driver should send LeaseGrantMessage for this
leaseholder, which should, in the same way as while initial lease granting
process, wait for group operability to send response.
> Implement notification for placement driver about replication group
> inoperability
> ---------------------------------------------------------------------------------
>
> Key: IGNITE-18960
> URL: https://issues.apache.org/jira/browse/IGNITE-18960
> Project: Ignite
> Issue Type: Improvement
> Reporter: Denis Chudov
> Priority: Major
> Labels: ignite-3
>
> This is a responsibility of replication group to notify the placement driver
> about becoming inoperable, for example, about majority loss in case of Raft.
> The corresponding message should be sent from leaseholder to placement driver
> so that the placement driver would not prolong the lease for this leaseholder
> automatically.
> As the replica may not know about the actual active actor of placement
> driver, it should send the message to every node in placement driver group.
> On the placement driver side, there may be no active actor at the moment. In
> this case, the replica should wait for next lease prolongation (which means
> that active actor has appeared) and after that repeat the messages to
> placement driver group.
> The active actor should send LeaseGrantMessage for this leaseholder, which
> should, in the same way as while initial lease granting process, wait for
> group operability to send response.
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
