[Hawkular-dev] agent "new server" event

[John Sanda]

> On Mar 29, 2017, at 9:22 AM, Heiko W.Rupp <hrupp at redhat.com> wrote:
> 
> On 29 Mar 2017, at 13:29, John Sanda wrote:
> 
>> however, deletion does not result in immediate deletion of data 
>> points. We still let data points in in the data_compressed table 
>> expire. We do this because one of the guidelines for using 
>> TimeWindowCompactionStrategy (TWCS) is to use a global TTL and not 
>> delete data._______________________________________________
> 
> This needs to be well documented as this is not what you
> expect when you delete metrics in an urgency to free up space.

Yes, it does. The API docs mention this. It needs to be included in the user guide as well. I have read that deleting data when using TWCS can actually cause it to hang around longer than it would if you wait for it to expire, potentially resulting in more space being used. If we really want to delete data vs just letting it expire, then we should not be using TWCS, but I really think we want TWCS as it is designed specifically for time series data. We are using TWCS with compressed data. A week’s worth of compressed data for a metric is going to be quite small.

I suppose we need a section on deletes and how it works. The only ways in Cassandra to immediately purge data are:

* drop table
* truncate table
* manually purge sstables

TWCS basically deletes an entire sstable when it determines it is safe to do so.