[ 
https://issues.apache.org/jira/browse/CASSANDRA-9672?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14605477#comment-14605477
 ] 

Sylvain Lebresne commented on CASSANDRA-9672:
---------------------------------------------

I'm bikesheding, but one slightly different alternative could be to add a 
{{minimum_data_retention_time}} that would be a guarantee on the minimum time 
data is guarantee to live in the database. An advantage being that it would 
still allow different ttls (and in particular mixing data with ttl and data 
without it) but with the guarantee that you can't fat-finger a ttl too low, or 
delete data by mistake, which feels to me closer to what a DBA would intend.  
It also have a hunch that it's easier to explain why that kind of option forces 
us to refuse deletes, but well, that's just a hunch. It should also be fairly 
intuitive to reserve a special value (say negative ones) for that option to 
mean "keep all data forever".

Of course, such option would still allows us to drop sstables cheaply 
(technically as long as the "min retention" is lower than gcGrace, but you can 
lower gcGrace if needs be).


> Provide a per-table param that would force default ttl on all updates
> ---------------------------------------------------------------------
>
>                 Key: CASSANDRA-9672
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-9672
>             Project: Cassandra
>          Issue Type: Improvement
>            Reporter: Aleksey Yeschenko
>            Priority: Minor
>
> Many users have tables that rely on TTL entirely - no deletes, and only fixed 
> TTL value.
> The way that default ttl works now, we only apply it if none is specified.
> We should provide an option that would *enforce* the specified TTL. Not 
> allowing ttl-less {{INSERT}} or {{UPDATE}}, not allowing ttl that's lower or 
> higher than the default ttl, and not allowing deletes.
> That option when enabled ({{force_default_ttl}}) should allow us to drop more 
> tables during compaction and do so cheaper. Would also allow the DBAs to 
> enforce the constraint in a guaranteed manner.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Reply via email to