[
https://issues.apache.org/jira/browse/IGNITE-21738?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Andrey Mashenkov updated IGNITE-21738:
--------------------------------------
Description:
In IGNITE-21585 there were fixed potential races:
1. ClientPrimaryReplicaTracker might update `primaryReplicas` after table
destroyed concurrently.
The fix idea was to forbid caching data for destroyed tables. However, cache
contains placeholders for all tables/partitions even if ther never be requested.
2. TableManager might register index (see `registerIndexesToTable`) in a table,
while LWM rising up concurrently.
The `registerIndexesToTable` execution was delegated into LowWatermark, to make
sure it runs under an lwm lock.
was:
After IGNITE-21585 there are potential races:
1. ClientPrimaryReplicaTracker may update `primaryReplicas` after table
destroyed concurrently.
2. TableManager may register index (see `registerIndexesToTable`) in a table,
while LWM rising up concurrently.
> Proper fix to linearize LWM and Metastorage event.
> --------------------------------------------------
>
> Key: IGNITE-21738
> URL: https://issues.apache.org/jira/browse/IGNITE-21738
> Project: Ignite
> Issue Type: Improvement
> Reporter: Andrey Mashenkov
> Priority: Major
> Labels: ignite-3
> Fix For: 3.0.0-beta2
>
>
> In IGNITE-21585 there were fixed potential races:
> 1. ClientPrimaryReplicaTracker might update `primaryReplicas` after table
> destroyed concurrently.
> The fix idea was to forbid caching data for destroyed tables. However, cache
> contains placeholders for all tables/partitions even if ther never be
> requested.
> 2. TableManager might register index (see `registerIndexesToTable`) in a
> table, while LWM rising up concurrently.
> The `registerIndexesToTable` execution was delegated into LowWatermark, to
> make sure it runs under an lwm lock.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
