[
https://issues.apache.org/jira/browse/IGNITE-28077?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Roman Puchkovskiy updated IGNITE-28077:
---------------------------------------
Description:
Both leases and MS safe time need to be distributed over the cluster in a
timely manner to make it possible for cluster nodes to process transactions.
Currently, they are distributed using a Raft group (Metastorage group, MG). The
problem is that, when a system is overloaded (for example, the disk I/O is
saturated), Raft cannot provide any guarantee on how much time a command will
take to be delivered and executed (even on the leader).
We need to invent another mechanism that is less dependent on the system
working without bursts.
> Reliance on Raft replication for lease and MS safe time distribution
> --------------------------------------------------------------------
>
> Key: IGNITE-28077
> URL: https://issues.apache.org/jira/browse/IGNITE-28077
> Project: Ignite
> Issue Type: Task
> Reporter: Roman Puchkovskiy
> Priority: Major
> Labels: ignite-3
>
> Both leases and MS safe time need to be distributed over the cluster in a
> timely manner to make it possible for cluster nodes to process transactions.
> Currently, they are distributed using a Raft group (Metastorage group, MG).
> The problem is that, when a system is overloaded (for example, the disk I/O
> is saturated), Raft cannot provide any guarantee on how much time a command
> will take to be delivered and executed (even on the leader).
> We need to invent another mechanism that is less dependent on the system
> working without bursts.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
