You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@kudu.apache.org by Casey Ching <ca...@cloudera.com> on 2016/04/07 21:27:01 UTC

NTP really required for local mini-cluster?

For Impala development we setup a 3 node kudu "mini-cluster", with all process running locally. Is NTP really required in this case? Some people didn’t have NTP installed on their dev boxes. But if NTP isn’t needed maybe we can just set —max_clock_sync_error_usec to a high value?  

Re: NTP really required for local mini-cluster?

Posted by Casey Ching <ca...@cloudera.com>.
Thanks Dan. Can you give more details on which testing scenarios would be problematic? Or is there some place I can read more about this? Just trying to understand what to expect.


On April 7, 2016 at 12:48:25 PM, Dan Burkert (dan@cloudera.com) wrote:

Hey Casey,  

I think setting the 'use_hybrid_clock' flag to false on startup is your  
best bet. This is what we do on OS X, for instance. It's not a  
configuration suitable for production, but should be fine for most testing  
scenarios.  

- Dan  

On Thu, Apr 7, 2016 at 12:27 PM, Casey Ching <ca...@cloudera.com> wrote:  

> For Impala development we setup a 3 node kudu "mini-cluster", with all  
> process running locally. Is NTP really required in this case? Some people  
> didn’t have NTP installed on their dev boxes. But if NTP isn’t needed maybe  
> we can just set —max_clock_sync_error_usec to a high value?  
>  

Re: NTP really required for local mini-cluster?

Posted by Dan Burkert <da...@cloudera.com>.
Hey Casey,

I think setting the 'use_hybrid_clock' flag to false on startup is your
best bet.  This is what we do on OS X, for instance.  It's not a
configuration suitable for production, but should be fine for most testing
scenarios.

- Dan

On Thu, Apr 7, 2016 at 12:27 PM, Casey Ching <ca...@cloudera.com> wrote:

> For Impala development we setup a 3 node kudu "mini-cluster", with all
> process running locally. Is NTP really required in this case? Some people
> didn’t have NTP installed on their dev boxes. But if NTP isn’t needed maybe
> we can just set —max_clock_sync_error_usec to a high value?
>