kill-controller, unregister, destroy-controller

[Casey Marshall]

My main use case for killing controllers is development & testing. For
this, I have a script that force deletes all the juju-* LXC containers, and
then unregisters all controllers with cloud: lxd. It's *much* faster than
waiting for each juju controller to tear itself down. It's also nothing I'd
provide casually to end users.

I think it should be possible, but not trivial, to destroy controllers and
everything in them. It's not a bad thing to have to type a long command or
flag name to do something so destructive -- or write a script to do
precisely what I want. Feels like my use case for destroying controllers
isn't really as a normal Juju user -- I'm (ab)using Juju with a developer &
QA tester's mindset.

-Casey

On Fri, Sep 2, 2016 at 6:15 AM, roger peppe <roger.peppe at canonical.com>
wrote:

> It seems to me that this kind of thing is exactly what "blocks" are
> designed for. An explicit unblock command seems better to me than either an
> explicit flag or an extra prompt, both of which are vulnerable to typing
> without thinking. Particularly if "throwaway" controllers created for
> testing purposes are not blocked by default, so you don't get used to to
> typing "unblock" all the time.
>
> On 1 Sep 2016 16:14, "Mark Ramm-Christensen (Canonical.com)" <
> mark.ramm-christensen at canonical.com> wrote:
>
> I get the desire to remove friction everywhere we can, but unless
> destroying controllers is a regular activity, I actually think SOME
> friction is valuable here.
>
> If controllers are constantly getting into a wedged state where they must
> be killed, that's likely a product of a fast moving set of betas, and
> should be addressed *directly* rather than as they say "applying lipstick
> to a pig"
>
> On Thu, Sep 1, 2016 at 4:04 PM, Marco Ceppi <marco.ceppi at canonical.com>
> wrote:
>
>> On Thu, Sep 1, 2016 at 9:59 AM Mark Ramm-Christensen (Canonical.com) <
>> mark.ramm-christensen at canonical.com> wrote:
>>
>>> I believe keeping the --destroy-all-models flag is helpful in keeping
>>> you from accidentally destroying a controller that is hosting important
>>> models for someone without thinking.
>>>
>>
>> What happens if I destroy-controller without that flag? Do I have to go
>> into my cloud portal to kill those instances? Is there any way to recover
>> from that to get juju reconnected? If not, it's just a slower death.
>>
>>
>>> On Thu, Sep 1, 2016 at 3:40 PM, Marco Ceppi <marco.ceppi at canonical.com>
>>> wrote:
>>>
>>>> Hey everyone,
>>>>
>>>> I know we've had discussions about this over the past few months, but
>>>> it seems we have three commands that overlap pretty aggressively.
>>>>
>>>> Using Juju beta16, and trying to 'destroy' a controller it looks like
>>>> this now:
>>>>
>>>> ```
>>>> root at ubuntu:~# juju help destroy-controller
>>>> Usage: juju destroy-controller [options] <controller name>
>>>>
>>>> ...
>>>>
>>>> Details:
>>>> All models (initial model plus all workload/hosted) associated with the
>>>> controller will first need to be destroyed, either in advance, or by
>>>> specifying `--destroy-all-models`.
>>>>
>>>> Examples:
>>>>     juju destroy-controller --destroy-all-models mycontroller
>>>>
>>>> See also:
>>>>     kill-controller
>>>>     unregister
>>>> ```
>>>>
>>>> When would you ever want to destroy-controller and not
>>>> destroy-all-models? I have to specify that flag everytime, it seems it
>>>> should just be the default behavior. Kill-controller seems to do what
>>>> destroy-controller --destroy-all-models does but more aggressively?
>>>>
>>>> Finally, unregister and destroy-controller (without
>>>> --destroy-all-models) does the same thing. Can we consider dropping the -
>>>> very long winded almost always required - flag for destroy-controller?
>>>>
>>>> Finally, there used to be a pretty good amount of feedback during
>>>> destroy-controller, while it was rolling text, I at least knew what was
>>>> happening. Now it's virtually silent. Given it runs for quite a long time,
>>>> can we get some form of feedback to the user back into the command?
>>>>
>>>> ```
>>>> root at ubuntu:~# juju destroy-controller --destroy-all-models cabs
>>>> WARNING! This command will destroy the "cabs" controller.
>>>> This includes all machines, applications, data and other resources.
>>>>
>>>> Continue? (y/N):y
>>>> ERROR failed to destroy controller "cabs"
>>>>
>>>> If the controller is unusable, then you may run
>>>>
>>>>     juju kill-controller
>>>>
>>>> to forcibly destroy the controller. Upon doing so, review
>>>> your cloud provider console for any resources that need
>>>> to be cleaned up.
>>>>
>>>> ERROR cannot connect to API: unable to connect to API: websocket.Dial
>>>> wss://10.0.0.4:17070/api: dial tcp 10.0.0.4:17070: getsockopt: no
>>>> route to host
>>>> root at ubuntu:~# juju kill-controller cabs
>>>> WARNING! This command will destroy the "cabs" controller.
>>>> This includes all machines, applications, data and other resources.
>>>>
>>>> Continue? (y/N):y
>>>> Unable to open API: open connection timed out
>>>> Unable to connect to the API server. Destroying through provider.
>>>> ERROR listing resource groups: azure.ServicePrincipalToken#Refresh:
>>>> Failure sending request for Service Principal 83d638b0-841c-4bd1-9e7c-868cae3393f4:
>>>> StatusCode=0 -- Original Error: http: nil Request.URL
>>>> root at ubuntu:~# juju bootstrap cabs azure
>>>> ERROR controller "cabs" already exists
>>>> ```
>>>>
>>>> Marco
>>>>
>>>> --
>>>> Juju-dev mailing list
>>>> Juju-dev at lists.ubuntu.com
>>>> Modify settings or unsubscribe at: https://lists.ubuntu.com/mailm
>>>> an/listinfo/juju-dev
>>>>
>>>>
>
> --
> Juju-dev mailing list
> Juju-dev at lists.ubuntu.com
> Modify settings or unsubscribe at: https://lists.ubuntu.com/mailm
> an/listinfo/juju-dev
>
>
>
> --
> Juju-dev mailing list
> Juju-dev at lists.ubuntu.com
> Modify settings or unsubscribe at: https://lists.ubuntu.com/
> mailman/listinfo/juju-dev
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.ubuntu.com/archives/juju-dev/attachments/20160902/ce308954/attachment-0001.html>