[Gluster-infra] [Gluster-devel] Announcing Softserve- serve yourself a VM

[Sankarshan Mukhopadhyay]

On Tue, Mar 20, 2018 at 1:57 PM, Sanju Rakonde <srakonde at redhat.com> wrote:

> I have a suggestion here. It will be good if we have a option like request
> for extension of the VM duration, and the option will be automatically
> activated after 3 hours of usage of VM. If somebody is using the VM after 3
> hours and they feel like they need it for 2 more hours they will request to
> extend the duration by 1 more hour. It will save the time of engineering
> since if a machine is expired, one has to configure the machine and all
> other stuff from the beginning.
>

As Nigel states later on, please file an issue. I fear that very soon
we will require an enhancement to the issue which might have to handle
how many times an extension can be sought. Supplementing it would be
another enhancement around how often can a particular individual
request for extension on a particular machine instance within the
space of a work week. And thus, at some point, a simple machine
instance allocation system would need to be backed by a decision
making system.

This was supposed to be an easy-to-use infrastructure, not where the
Infra team members chase down individuals who are not freeing up
machines for as high as 5 weeks (I am not going to publicly cite this
instance).

That being said, I think it is worth a moment to think about what is
happening with Softserve.

The system is intended to provide a set of finite metered resources
available to developers in order to debug issues. The number of
available machine instances are not high (at this point) and the
resource allocation during this initial period has been designed to be
this way so as to arrive at a better understanding of usage patterns.

A 4 hour allocation is half of a working day. If we do think that this
is not enough, we want to also think deeply about the following (a)
all our issues require more than a day’s worth of debugging (b)
perhaps we are unable to allocate a dedicated 4 hour block on one
topic ie. there are distractions (c) there will not be enough machine
instances to make available to a backlog of developers waiting to
start off with their work

Any one or, a combination of the above possibilities are larger
challenges which need to be addressed. I understand why the
very-short-term approach of “just extend the darned time slice and be
done” is conducive to getting things done. I’d rather ask ourselves -
why do we have all issues that consume multiple days and what can we
do to improve that. Because I am convinced that it surely isn’t
enjoyable to always budget for multiple days when something comes up
to debug, even if it is a race condition (as Atin mentions in his
response).




-- 
sankarshan mukhopadhyay
<https://about.me/sankarshan.mukhopadhyay>