[
https://issues.apache.org/jira/browse/CASSANDRA-14408?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Blake Eggleston updated CASSANDRA-14408:
----------------------------------------
Status: Patch Available (was: Open)
[trunk|https://github.com/bdeggleston/cassandra/tree/14408]
[dtest|https://github.com/bdeggleston/cassandra-dtest/tree/14408]
First pass ready for review. The main change is discarding transiently
replicated data at the successful completion of the incremental repair session.
Since a repair session can include both full and transiently replicated data
for a given node, we need to keep the transient data separate from the
non-transient data for the life of the repair session so we don’t have to do
another anti-compaction at the end of the session. This is done by creating
separate sstable buckets for full and transiently replicated data (basically
adding another dimension to the sstable bucketing by repair session added in
CASSANDRA-9143).
An {{isTransient}} flag was added to sstable metadata, and anti-compaction will
write transient data to sstables with that flag. The full and transient
sstables are then managed by separate compaction strategy groups, preventing
them from being compacted together. Once the session is finalized, the
compaction task that promotes sstables obsoletes the transient sstables without
adding a replacement, deleting them.
This patch also prevents streaming data to transient replicas, and considers
transiently replicated ranges when grouping token ranges for repair.
Note: this patch depends on CASSANDRA-14621
> Transient Replication: Incremental & Validation repair handling of transient
> replicas
> -------------------------------------------------------------------------------------
>
> Key: CASSANDRA-14408
> URL: https://issues.apache.org/jira/browse/CASSANDRA-14408
> Project: Cassandra
> Issue Type: Sub-task
> Components: Repair
> Reporter: Ariel Weisberg
> Assignee: Blake Eggleston
> Priority: Major
> Fix For: 4.0
>
>
> At transient replicas anti-compaction shouldn't output any data for transient
> ranges as the data will be dropped after repair.
> Transient replicas should also never have data streamed to them.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
