[
https://issues.apache.org/jira/browse/CASSANDRA-396?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12851410#action_12851410
]
gabriele renzi commented on CASSANDRA-396:
------------------------------------------
instead of adding logic to avoid compaction while anticompacting, wouldn't it
be easier to create hard links, and remove them after send?
I recall Java7 has support for hard links in java.nio.file.Path, maybe there is
some backport?
> more-efficient anticompaction/bootstrap/load-balance
> ----------------------------------------------------
>
> Key: CASSANDRA-396
> URL: https://issues.apache.org/jira/browse/CASSANDRA-396
> Project: Cassandra
> Issue Type: Improvement
> Components: Core
> Reporter: Jonathan Ellis
> Priority: Minor
>
> From a comment in CASSANDRA-213: it's a lot less IO (50% less reads, to be
> exact) to have anticompact split the old file into new files, each of which
> are served live until bootstrap is complete, at which point the sent file is
> removed, than to copy the to-send files out to a staging dir, and later
> re-scan the original file for cleanup & writing what is left. plus we get to
> remove the cleanup methods in ColumnFamilyStore, since they will be
> unnecessary.
> one other detail: we need to keep compaction from merging the to-send file
> until boostrap is complete.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
