[ 
https://issues.apache.org/jira/browse/CASSANDRA-17372?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17490503#comment-17490503
 ] 

Benedict Elliott Smith commented on CASSANDRA-17372:
----------------------------------------------------

If we did this purely based on insert timestamp and repair status this would be 
very easy, and likely much easier to administer for users. It is likely to lead 
to easier optimisations on our side as well.

I'm not convinced by some of the arguments against presented in the linked 
comment, personally. I think it is fine to simply document that TTL-expiry is 
non-deterministic, and that without a major compaction you can expect data 
resurrection if you set the TTL up again later.

Additionally, I don't really see the issue with assuming a timestamp is a valid 
server-side timestamp. At worst I think we might want to document that we 
assume unix epoch microsecond timestamps for TTL-expiry if the user ever 
provides them (which most won't). If a user ever wants it, we can no doubt make 
the expiration epoch and granularity configurable.


> Dynamic Table TTL
> -----------------
>
>                 Key: CASSANDRA-17372
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-17372
>             Project: Cassandra
>          Issue Type: Improvement
>          Components: CQL/Semantics
>            Reporter: Paulo Motta
>            Priority: Normal
>
> One limitation of the {{default_time_to_live}} option is that it only applies 
> to newly inserted data so an expensive migration is required when altering 
> the table-level TTL, which is not an uncommon request due to changes in 
> retention policies.
> This seems to have been a deliberate design decision when adding the Table 
> TTL feature on CASSANDRA-3974, due to the reasons stated [on this 
> comment|https://issues.apache.org/jira/browse/CASSANDRA-3974?focusedCommentId=13427314&page=com.atlassian.jira.plugin.system.issuetabpanels%3Acomment-tabpanel#comment-13427314]
>  so we should revisit and potentially address these concerns.
> I would like to explore supporting dynamic TTL, which would reflect any 
> updates to the table-level {{default_time_to_live}} immediately to all table 
> data.



--
This message was sent by Atlassian Jira
(v8.20.1#820001)

---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org
For additional commands, e-mail: commits-h...@cassandra.apache.org

Reply via email to