...
> > In aeolus, condor is what's monitoring running instances and I thought
> > it did that be GETing each instance individually every 90 seconds
> 
> No, condor does a batch update of every provider every 30 seconds.  We
can
> certainly tune that down, but the problem is that it is a tradeoff; the
longer
> the timeout, the worse the user experience becomes in terms of updating
the
> UI with the current state.  Even with the 30 second timeout, we are
getting
> complaints that we take "way longer" to show an instance going to
running
> than the EC2 front-end.

We use a 'sliding window' approach in rhev-m in some places from user
perspective
i.e., if the user performed an action (or clicked refresh), we'll refresh
more frequently for a few cycles.
But that's UI --> Backend.

Your use case is more close to backend --> vdsm, where we poll every 2
seconds actually.
But the polling every 2 seconds is a very light poll - only list of VM's
and their status.
The heavier polling for all detils/stats is not as frequent.

So you should probably define the minimal data you need to poll for more
frequently, and run the lighter query (the api has a 'detail' level,
though probably not one that light yet).

Reply via email to