[
https://issues.apache.org/jira/browse/IGNITE-23252?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17916445#comment-17916445
]
Mikhail Efremov edited comment on IGNITE-23252 at 1/23/25 6:16 PM:
-------------------------------------------------------------------
The issue showed up a critical race for old colocation code base:
# There are still needed table replication group for some transaction related
requests (e.g. {{TxFinishReplicaRequestImpl}} that triggers
{{WriteIntentSwitchReplicaRequestImpl}} that requires table replica now).
# So we have to left table replicas related code yet.
# Then we have a race between table replicas assignments stable switch and zone
replica stable switch destruction:
* table replica stopping and partition storages destruction are triggered by
table assignmets switch;
* zone replica on zone stable assignemtns switch fires an event to destroy the
corresponding table replica's storages highly likely before table replica was
stopped;
* this race leads to "no such file" exception due to wrong order of a table
replica stopping process.
We want a correct solution that starts the table replica on
{{AFTER_REPLICA_STARTED}} event and {{TableManager}}'s assignments events are
blocked if colocation flag is enabled. End the corresponding table replication
group is stopped with storages in case {{AFTER_REPLICA_STOPPED}} event. The
first requires forced assignments for {{weakReplcaStart}} call.
was (Author: JIRAUSER303791):
The issue showed up a critical race for old colocation code base:
# There are still needed table replication group for some transaction related
request (e.g. {{TxFinishReplicaRequestImpl}} that triggers
{{WriteIntentSwitchReplicaRequestImpl}} that requires table replica now).
# So we have to left table replicas related code yet.
# Then we have a race between table replicas assignments stable switch and zone
replica stable switch destruction:
* table replica stopping and partition storages destruction are triggered by
table assignmets switch;
* zone replica on zone stable assignemtns switch fires an event to destroy the
corresponding table replica's storages highly likely before table replica was
stopped;
* this race leads to "no such file" exception due to wrong order of a table
replica stopping process.
We want a correct solution that starts the table replica on
{{AFTER_REPLICA_STARTED}} event and {{TableManager}}'s assignments events are
blocked if colocation flag is enabled. End the corresponding table replication
group is stopped with storages in case {{AFTER_REPLICA_STOPPED}} event. The
first requires forced assignments for {{weakReplcaStart}} call.
> ItReplicaLifecycleTest is unstable
> ----------------------------------
>
> Key: IGNITE-23252
> URL: https://issues.apache.org/jira/browse/IGNITE-23252
> Project: Ignite
> Issue Type: Bug
> Reporter: Alexander Lapin
> Assignee: Mikhail Efremov
> Priority: Major
> Labels: ignite-3
> Time Spent: 10m
> Remaining Estimate: 0h
>
> Various of different exceptions, e.g., [TC
> failure.|https://ci.ignite.apache.org/buildConfiguration/ApacheIgnite3xGradle_Test_RunAllTests/8490892?expandBuildDeploymentsSection=false&hideTestsFromDependencies=false&hideProblemsFromDependencies=false&expandBuildTestsSection=true&expandCode+Inspection=true]
> Stabilization required.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
