[
https://issues.apache.org/jira/browse/CASSANDRA-14089?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16275175#comment-16275175
]
Jeff Jirsa commented on CASSANDRA-14089:
----------------------------------------
You've marked this as invalid, but there are two ways to do this without
changing TWCS:
1) You can write data with {{USING TIMESTAMP}}, which will go into the old
windows. You just need care to only write one window at a time, and then
{{flush}} between windows.
2) You can load the data from the old CF into the new by creating new sstables
offline (see: CQLSSTableWriter), and separate them into time windows offline as
you create them. Then, load into new table with {{sstableloader}} .
> Time Window Compaction Strategy: Determine window by column value
> -----------------------------------------------------------------
>
> Key: CASSANDRA-14089
> URL: https://issues.apache.org/jira/browse/CASSANDRA-14089
> Project: Cassandra
> Issue Type: New Feature
> Reporter: Taylor Cressy
>
> Recently, we came across an issue where we needed to repartition a column
> family and alter it's compaction strategy to TWCS.
> We wanted to ensure that the time series data was bucketized properly into
> it's appropriate window when we migrate data from the old CF to the new one.
> However, there appears to be no simple way to do this unless we:
> 1) Set the time window to some arbitrary length
> 2) Backload a chunk of the data within that time window
> 3) After all chunks backload, reset the time window to the desired size
> I propose that we allow TWCS to have the ability to key off a column value
> for placing it in a window.
> Also, maybe we can adjust the TTL accordingly - based on that column value,
> upon insert - considering TTLs and TWCS are generally based on some form of
> business logic. Of course, the user could calculate the TTL on the fly, but
> it could be a nice feature.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
