Open Stack

On 7/25/2018 1:46 AM, Ghanshyam Mann wrote:
> yeah, there are many tests taking too long time. I do not know the reason this time but last time we did audit for slow tests was mainly due to ssh failure.
> I have created the similar ethercalc [3] to collect time taking tests and then round figure of their avg time taken since last 14 days from health dashboard. Yes, there is no calculated avg time on o-h so I did not take exact avg time its round figure.
> 
> May be 14 days  is too less to take decision to mark them slow but i think their avg time since 3 months will be same. should we consider 3 month time period for those ?
> 
> As per avg time, I have voted (currently based on 14 days avg) on ethercalc which all test to mark as slow. I taken the criteria of >120 sec avg time.  Once we have more and more people votes there we can mark them slow.
> 
> [3]https://ethercalc.openstack.org/dorupfz6s9qt

Thanks for this. I haven't gone through all of the tests in there yet, 
but noticed (yesterday) a couple of them were personality file compute 
API tests, which I thought was strange. Do we have any idea where the 
time is being spent there? I assume it must be something with ssh 
validation to try and read injected files off the guest. I need to dig 
into this one a bit more because by default, file injection is disabled 
in the libvirt driver so I'm not even sure how these are running (or 
really doing anything useful). Given we have deprecated personality 
files in the compute API [1] I would definitely mark those as slow tests 
so we can still run them but don't care about them as much.

[1] 
https://docs.openstack.org/nova/latest/reference/api-microversion-history.html#id52

-- 

Thanks,

Matt