[ 
https://issues.apache.org/jira/browse/IGNITE-26117?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Roman Puchkovskiy updated IGNITE-26117:
---------------------------------------
    Description: 
There are two ways to destroy a Raft group storages (namely, log storage and 
Raft meta storage): durable and non-durable. Durable version makes sure that, 
if the call returned, then the destruction will complete if it was not able to 
finish its job (for example, due to node crash). Durable destruction is more 
costly as it requires to write destruction invent to the Vault (and this does 
an fsync). Currently, both when destroying a table as a whole and when 
destroying just one partition of a table, we use durable Raft storages 
destruction.

But after IGNITE-25621, durability guarantee becomes redundant for Raft 
storages destruction done when destroying whole tables as the mechanism 
introduced in IGNITE-25621 provides its own guarantee on a higher level.

Hence, we should now have a separation:
 # When just a single partition is destroyed (due to eviction from node), its 
Raft storages should still be destroyed durably
 # When whole table is destroyed, Raft storages of its partitions should use 
non-durable destruction

> Use non-durable Raft storages destruction for table destruction
> ---------------------------------------------------------------
>
>                 Key: IGNITE-26117
>                 URL: https://issues.apache.org/jira/browse/IGNITE-26117
>             Project: Ignite
>          Issue Type: Improvement
>            Reporter: Roman Puchkovskiy
>            Assignee: Roman Puchkovskiy
>            Priority: Major
>              Labels: ignite-3
>          Time Spent: 10m
>  Remaining Estimate: 0h
>
> There are two ways to destroy a Raft group storages (namely, log storage and 
> Raft meta storage): durable and non-durable. Durable version makes sure that, 
> if the call returned, then the destruction will complete if it was not able 
> to finish its job (for example, due to node crash). Durable destruction is 
> more costly as it requires to write destruction invent to the Vault (and this 
> does an fsync). Currently, both when destroying a table as a whole and when 
> destroying just one partition of a table, we use durable Raft storages 
> destruction.
> But after IGNITE-25621, durability guarantee becomes redundant for Raft 
> storages destruction done when destroying whole tables as the mechanism 
> introduced in IGNITE-25621 provides its own guarantee on a higher level.
> Hence, we should now have a separation:
>  # When just a single partition is destroyed (due to eviction from node), its 
> Raft storages should still be destroyed durably
>  # When whole table is destroyed, Raft storages of its partitions should use 
> non-durable destruction



--
This message was sent by Atlassian Jira
(v8.20.10#820010)

Reply via email to