[doc] Decrease number of retries for puppet deployments

[OpenStack Infra]

https://review.openstack.org/131696
commit 3c374c9f7bfbdbcd7ce2f716cd704e3044e6fb41
Author: Vladimir Sharshov <email address hidden>
Date: Wed Oct 29 13:36:06 2014 +0300

    Decrease number of retries for puppet deployments

    Default behavior before: 3 puppet runs for the
    current role on the current node: 1 run + 2 retries

    Now we will run only 2 times: 1 run + 1 retries

    In 99 out of 100 cases, if the problem persists
    when we restart puppet, it will not disappear after
    next restart. That's why it makes no sense to run
    more than 2 times. Thus we save time, follow the
    ideology of 'fail fast', speed up deployment
    process, show user more detailed message about
    problem instead of 'deployment timeout'.

    Change-Id: Idd4b2a2e7ecdeff9ec298a2c5ceaea91f74cee4c
    Related-Bug: #1384332
    DocImpact

[Mike Scherbakov (mihgen)]

Few questions:
1) where do we document how many retries we do?
2) It might be fine to run 2 times instead of 3, tnough it has nothing to do with fail fast during testing. Our tests must fail after even first pass, so we should not wait 2 runs actually at all. Let's figure our how we can make it happen. May be additional deployment param passed via REST API, which would go thru to Astute?

[Vladimir Sharshov (vsharshov)]

> 1) where do we document how many retries we do?
As i know, we do not notice about such detail in documentation.

> 2) Our tests must fail after even first pass, so we should not wait 2 runs actually at all
If we setup zero retries in test, we should also specify such behavior for production. In other case we will get tests that run with different condition and show different result. 1 retry in other case as Dmitry Illyin explain me, sometimes can solve some problem with puppet graph.

Sure, decrease number of retries to zero looks like desirable goal. We can create patchset with zero retries and run all tests in parallel with master iso. If result will be the same, apply it to master. What do you think?

[Vladimir Sharshov (vsharshov)]

After discussion with Mike Scherbakov (mihgen) original idea is to use debug mode for fast CI test which can save for us a lot of time.

System tests and BVT still should use puppet runs with retries.

Related with: https://bugs.launchpad.net/fuel/+bug/1343136

[Sheena Conant (sheena-conant)]

Is this just a change to the number of times puppet scripts are run before failing to the customer? If so, is it necessary to expose this to the customer? I see this as targeted to 6.1 - is anyone really reading the 6.1 release notes to see this kind of information now?