[
https://issues.apache.org/jira/browse/CASSANDRA-21653?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Aleksey Yeschenko updated CASSANDRA-21653:
------------------------------------------
Bug Category: Parent values: Correctness(12982)Level 1 values:
Unrecoverable Corruption / Loss(13161)
Complexity: Normal
Component/s: Cluster/Schema
Discovered By: Code Inspection
Fix Version/s: NA
Reviewers: Blake Eggleston
Severity: Normal
Status: Open (was: Triage Needed)
> CEP-45: Fixes and improvements to MT migration logic
> ----------------------------------------------------
>
> Key: CASSANDRA-21653
> URL: https://issues.apache.org/jira/browse/CASSANDRA-21653
> Project: Apache Cassandra
> Issue Type: Bug
> Components: Cluster/Schema
> Reporter: Aleksey Yeschenko
> Assignee: Aleksey Yeschenko
> Priority: Normal
> Fix For: NA
>
>
> Fixes the following TCM-related issues in MT migration logic:
> 1. Repeat repairs of already migrated ranges no longer issue redundant
> transformations
> 2. Incomplete repair jobs (Paxos-only, or not involving all replicas) no
> longer advance ranges to migrated status
> 3. A virtual table added for exposing migrated and migrating ranges so
> operators can have visibility into migration process
> 4. Error handling and logging has been improved
> 5. [Optional] (more on this point below) Batch repair jobs at parent repair
> session level so that one transformation per set of ranges - and not one
> transformation per table per set of ranges - gets committed, to reduce
> potential TCM contention issues
> The original motivation for this patch set was to deal with the storm of
> transformations issued on repair jobs' completions contending and
> overwhelming TCM.
> The build that exposed this problem, however, did not include the following
> two patch sets that are intended to help with just this sort of problem:
> - CASSANDRA-21453 (Move long running cluster metadata operations to a longer
> timeout)
> - CASSANDRA-21456 (Add policy for selecting CMS host when submitting commit
> request)
> So I suspect that commit (5) in the submitted PR is no longer necessary. I'm
> also not a massive fan of having two inconsistent approaches (per-job
> callback for Accord, per-session callback for MT) co-existing. So my
> preference is to include all the fixes and improvements except the last one,
> and revisit committing the optional one only if a build with those trunk TCM
> improvements still exhibits contention problems.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
