On Tue, Apr 19, 2016 at 11:58 AM, Malmee Weerasinghe <[email protected]>
wrote:

> Hi All,
>
> As there is a default database 'InfluxDB' that comes with 'Heapster' using
> the same database would be a better approach than an external database.
>

Yes IMO it is better to use InfluxDB for this.

Thanks

>
> Default Heapster gives only up to 15 minutes of historical data for any
> Container. Would using a custom Heapster docker image be a better approach
> on this?
>
> In Heapster Repository, Heapster.go file contains these configurations.
>
>    - stats_resolution (Deafult 1*time.Second)  The resolution at which
>    heapster will retain stats
>    - sink_frequency (Default 10*time.Second) Frequency at which data will
>    be pushed to sinks
>    - cache_duration (Default 4*time.Minute) The total duration of the
>    historical data that will be cached by heapster
>    - use_model (Default true) When true, the internal model
>    representation will be used
>    - model_resolution (Default 1*time.Minute)  The resolution of the time
>    series stored in the model.
>    - model_frequency (Default 145*time.Second) Frequency at which model
>    will be updated.
>
>
>
> *      1 second < stats_resolution < model_resolution < model_frequency       
> sink_frequency < cache_duration*
>
> If these configuration are changed as below, build a custom docker image
> and use that docker image instead of default Heapster image
> Heapster:canary, would that be a better approach on this? Or is there any
> other approaches for this?
>
>    - stats_resolution    1*time.Hour
>    - sink_frequency     2*time.Hour
>    - cache_duration     3*time.Hour
>    - use_model            true
>    - model_resolution   2*time.Hour
>    - model_frequency   3*time.Hour
>
> I highly appreciate your suggestions regarding this.
>
> Thank you.
>
> On Fri, Apr 8, 2016 at 5:05 PM, Malmee Weerasinghe <[email protected]>
> wrote:
>
>> Hi All,
>> I have been working on detecting the idle containers in appcloud. I have
>> installed Heapster on Kubernetes and Heapster Metric Model gives only up to
>> 15 minutes of historical data for any Container.
>>
>> I have tried adding the flag "-model_resolution" and making it more than
>> 30s, but the flag can not be set as it gives errors when installing the
>> replication controller of heapster.
>>
>> Then I have tried adding "start=X&end=Y" to the end of the API endpoint,
>> but it gives the same response that gives without adding that part.
>>
>> AFAIU if the database of heapster, influxdb stores all the data from 30s
>> to 30s, it needs a huge storage. Therefore it stores only up to 15 minutes
>> of historical data at any moment.
>>
>> To find the idle containers of appcloud, the resource usage details of
>> the containers have to be stored for a longer period than 15 minutes. One
>> of the approaches would be to use another database and store the resource
>> usage details of containers checking whether the container has been idle
>> for a period of 5 days.
>>
>> I highly appreciate your suggestions on the approaches that can be taken
>> regarding this.
>>
>> Thank you.
>>
>> --
>> Malmee Weerasinghe
>> WSO2 Intern
>> mobile : (+94)* 71 7601905* |   email :   <[email protected]>
>> [email protected]
>>
>
>
>
> --
> Malmee Weerasinghe
> WSO2 Intern
> mobile : (+94)* 71 7601905* |   email :   <[email protected]>
> [email protected]
>



-- 
*Imesh Gunaratne*
Senior Technical Lead
WSO2 Inc: http://wso2.com
T: +94 11 214 5345 M: +94 77 374 2057
W: http://imesh.io
Lean . Enterprise . Middleware
_______________________________________________
Dev mailing list
[email protected]
http://wso2.org/cgi-bin/mailman/listinfo/dev

Reply via email to