[DISCUSS] Making simple installs easier.

[David Nalley <da...@gnsa.us>]

Hi folks.

I've been thinking about our install process lately.

We currently require folks to muck about with firewall settings, NFS
settings, network configuration, etc.
This makes configuration painful, our docs VERY platform specific, and
easily prone to mistakes which result in failure to get things to
work. Even the 'install.sh' from the 3.0.x and earlier days doesn't do
enough. What I want to do is get rid of sections 2-4 of the quick
install guide, and replace it with - 'run this one or two lines worth
of commands' (http://s.apache.org/runbook)

My natural reaction was to reach for puppet - but I am not sure that's
the right answer. To do things right, I'd need several puppet modules
like stdlib, puppetlabs-firewall, etc, which is a fair bit of
overhread - and oh, yeah, need to install the puppet client. I think
Chef is probably in a similar problem space. I don't want to resort to
shell scripts of python - config management tools know the difference
between apt and yum, and can still get a package installed with one
declaration, same thing with firewall rules. Is something like Ansible
or SaltStack a better choice?? I don't see it right now if it is, but
I don't have much experience with either of those two.

The all-in-one installation process I'd like to see:

Install your host OS
Install an meta-RPM/Deb that either (installs everything, or
alternatively configures a repo - or just installs the repo and the
stuff I need to install with)
Run a command that activates one of these config tools - configures
the machine, installs the packages I need, and gets me to the point
where I'm ready to login and go through the beautiful new user gui
setup stuff.

I still want to keep the documentation around, it's invaluable for
experienced users and more complex deployments - but right now it's
far too much overhead (probably an hour or two) to get things
installed and setup to the point where you are ready to run the
'Welcome to CloudStack GUI' if you just want to try CloudStack out.

So why am I writing this email instead of diving in and solving this
problem? Well honestly, I'd like some external opinions. I want to
make sure that I am not seeing a 'nail' simply because I have a hammer
in my hand. How can we most easily do this? So - how do we make the
'brand-new' user experience much better? We develop a platform for
orchestration of complex systems, this should be a solved problem.

--David

Re: [DISCUSS] Making simple installs easier.

[David Nalley <da...@gnsa.us>]

On Thursday, April 25, 2013, Jörgen Maas wrote:

> As a new user (as of today) i can say the following from my personal
> experience:
>
>
Welcome - and thanks for pushing the feedback.


> The biggest hurdle (in an enterprise setting) is getting the packages for
> your platform.Building packages from source is nice and all, but in my case
> i don't have Internet access in the datacenter. So the whole maven stuff is
> pretty much a big time burner (there's also a bug somewhere wrt to calling
> /usr/bin/mvn when there's only /usr/bin/mvn3 as per the docs). I also had
> to downgrade tomcat to work around some bugs where the management service
> would just shutdown (Ubuntu 12.04 LTS).
>
>
Please file a bug for the tomcat issue.


> When the issues surrounding the packages were resolved and the management
> service was up and running it was pretty much smooth sailing from there on
> (except for the systemvm stuff).
>
> So from my point of view it would make much more sense to first align the
> releases more with the targeted distro's and provide repo's for those
> distro's.
>
>
The project officially doesn't release binaries at all - but we have some
enterprising contributors who have done just that.
Debs for Ubuntu are here:
http://cloudstack.apt-get.eu/ubuntu
RPMs for EL6 are here:
http://cloudstack.apt-get.eu/rhel/4.0/

The above site is maintained by Wido den Hollander who is a committer and
PMC member for CloudStack.

Re: [DISCUSS] Making simple installs easier.

[Jörgen Maas <jo...@gmail.com>]

As a new user (as of today) i can say the following from my personal
experience:

The biggest hurdle (in an enterprise setting) is getting the packages for
your platform.Building packages from source is nice and all, but in my case
i don't have Internet access in the datacenter. So the whole maven stuff is
pretty much a big time burner (there's also a bug somewhere wrt to calling
/usr/bin/mvn when there's only /usr/bin/mvn3 as per the docs). I also had
to downgrade tomcat to work around some bugs where the management service
would just shutdown (Ubuntu 12.04 LTS).

When the issues surrounding the packages were resolved and the management
service was up and running it was pretty much smooth sailing from there on
(except for the systemvm stuff).

So from my point of view it would make much more sense to first align the
releases more with the targeted distro's and provide repo's for those
distro's.

Hope this helps :)




On Wed, Apr 24, 2013 at 1:58 PM, Noah Slater <ns...@apache.org> wrote:

> (Typo, but you can fill in any number you feel like...)
>
>
> On 24 April 2013 12:57, Noah Slater <ns...@apache.org> wrote:
>
> > Also + for the initiative!
> >
> >
> > On 23 April 2013 20:15, Chiradeep Vittal <Chiradeep.Vittal@citrix.com
> >wrote:
> >
> >>
> >>
> >> On 4/21/13 3:21 PM, "David Nalley" <da...@gnsa.us> wrote:
> >>
> >> >Hi folks.
> >> >
> >> >I've been thinking about our install process lately.
> >> >
> >> >We currently require folks to muck about with firewall settings, NFS
> >> >settings, network configuration, etc.
> >> >This makes configuration painful, our docs VERY platform specific, and
> >> >easily prone to mistakes which result in failure to get things to
> >> >work. Even the 'install.sh' from the 3.0.x and earlier days doesn't do
> >> >enough. What I want to do is get rid of sections 2-4 of the quick
> >> >install guide, and replace it with - 'run this one or two lines worth
> >> >of commands' (http://s.apache.org/runbook)
> >> >
> >> >My natural reaction was to reach for puppet - but I am not sure that's
> >> >the right answer. To do things right, I'd need several puppet modules
> >> >like stdlib, puppetlabs-firewall, etc, which is a fair bit of
> >> >overhread - and oh, yeah, need to install the puppet client. I think
> >> >Chef is probably in a similar problem space. I don't want to resort to
> >> >shell scripts of python - config management tools know the difference
> >> >between apt and yum, and can still get a package installed with one
> >> >declaration, same thing with firewall rules. Is something like Ansible
> >> >or SaltStack a better choice?? I don't see it right now if it is, but
> >> >I don't have much experience with either of those two.
> >> >
> >> >The all-in-one installation process I'd like to see:
> >> >
> >> >Install your host OS
> >> >Install an meta-RPM/Deb that either (installs everything, or
> >> >alternatively configures a repo - or just installs the repo and the
> >> >stuff I need to install with)
> >> >Run a command that activates one of these config tools - configures
> >> >the machine, installs the packages I need, and gets me to the point
> >> >where I'm ready to login and go through the beautiful new user gui
> >> >setup stuff.
> >> >
> >> >I still want to keep the documentation around, it's invaluable for
> >> >experienced users and more complex deployments - but right now it's
> >> >far too much overhead (probably an hour or two) to get things
> >> >installed and setup to the point where you are ready to run the
> >> >'Welcome to CloudStack GUI' if you just want to try CloudStack out.
> >> >
> >> >So why am I writing this email instead of diving in and solving this
> >> >problem? Well honestly, I'd like some external opinions. I want to
> >> >make sure that I am not seeing a 'nail' simply because I have a hammer
> >> >in my hand. How can we most easily do this? So - how do we make the
> >> >'brand-new' user experience much better? We develop a platform for
> >> >orchestration of complex systems, this should be a solved problem.
> >> >
> >> >--David
> >>
> >> +1 for the initiative.
> >> If I look at Apache Hadoop's single node operation documentation[1], it
> is
> >> considerably simpler.
> >> Apache Tomcat installation is also fairly trivial.
> >>
> >> [1] http://hadoop.apache.org/docs/stable/single_node_setup.html
> >>
> >>
> >
> >
> > --
> > NS
> >
>
>
>
> --
> NS
>



-- 
Grtz,
Jörgen Maas

Re: [DISCUSS] Making simple installs easier.

[Noah Slater <ns...@apache.org>]

(Typo, but you can fill in any number you feel like...)


On 24 April 2013 12:57, Noah Slater <ns...@apache.org> wrote:

> Also + for the initiative!
>
>
> On 23 April 2013 20:15, Chiradeep Vittal <Ch...@citrix.com>wrote:
>
>>
>>
>> On 4/21/13 3:21 PM, "David Nalley" <da...@gnsa.us> wrote:
>>
>> >Hi folks.
>> >
>> >I've been thinking about our install process lately.
>> >
>> >We currently require folks to muck about with firewall settings, NFS
>> >settings, network configuration, etc.
>> >This makes configuration painful, our docs VERY platform specific, and
>> >easily prone to mistakes which result in failure to get things to
>> >work. Even the 'install.sh' from the 3.0.x and earlier days doesn't do
>> >enough. What I want to do is get rid of sections 2-4 of the quick
>> >install guide, and replace it with - 'run this one or two lines worth
>> >of commands' (http://s.apache.org/runbook)
>> >
>> >My natural reaction was to reach for puppet - but I am not sure that's
>> >the right answer. To do things right, I'd need several puppet modules
>> >like stdlib, puppetlabs-firewall, etc, which is a fair bit of
>> >overhread - and oh, yeah, need to install the puppet client. I think
>> >Chef is probably in a similar problem space. I don't want to resort to
>> >shell scripts of python - config management tools know the difference
>> >between apt and yum, and can still get a package installed with one
>> >declaration, same thing with firewall rules. Is something like Ansible
>> >or SaltStack a better choice?? I don't see it right now if it is, but
>> >I don't have much experience with either of those two.
>> >
>> >The all-in-one installation process I'd like to see:
>> >
>> >Install your host OS
>> >Install an meta-RPM/Deb that either (installs everything, or
>> >alternatively configures a repo - or just installs the repo and the
>> >stuff I need to install with)
>> >Run a command that activates one of these config tools - configures
>> >the machine, installs the packages I need, and gets me to the point
>> >where I'm ready to login and go through the beautiful new user gui
>> >setup stuff.
>> >
>> >I still want to keep the documentation around, it's invaluable for
>> >experienced users and more complex deployments - but right now it's
>> >far too much overhead (probably an hour or two) to get things
>> >installed and setup to the point where you are ready to run the
>> >'Welcome to CloudStack GUI' if you just want to try CloudStack out.
>> >
>> >So why am I writing this email instead of diving in and solving this
>> >problem? Well honestly, I'd like some external opinions. I want to
>> >make sure that I am not seeing a 'nail' simply because I have a hammer
>> >in my hand. How can we most easily do this? So - how do we make the
>> >'brand-new' user experience much better? We develop a platform for
>> >orchestration of complex systems, this should be a solved problem.
>> >
>> >--David
>>
>> +1 for the initiative.
>> If I look at Apache Hadoop's single node operation documentation[1], it is
>> considerably simpler.
>> Apache Tomcat installation is also fairly trivial.
>>
>> [1] http://hadoop.apache.org/docs/stable/single_node_setup.html
>>
>>
>
>
> --
> NS
>



-- 
NS

Re: [DISCUSS] Making simple installs easier.

[Noah Slater <ns...@apache.org>]

Also + for the initiative!


On 23 April 2013 20:15, Chiradeep Vittal <Ch...@citrix.com>wrote:

>
>
> On 4/21/13 3:21 PM, "David Nalley" <da...@gnsa.us> wrote:
>
> >Hi folks.
> >
> >I've been thinking about our install process lately.
> >
> >We currently require folks to muck about with firewall settings, NFS
> >settings, network configuration, etc.
> >This makes configuration painful, our docs VERY platform specific, and
> >easily prone to mistakes which result in failure to get things to
> >work. Even the 'install.sh' from the 3.0.x and earlier days doesn't do
> >enough. What I want to do is get rid of sections 2-4 of the quick
> >install guide, and replace it with - 'run this one or two lines worth
> >of commands' (http://s.apache.org/runbook)
> >
> >My natural reaction was to reach for puppet - but I am not sure that's
> >the right answer. To do things right, I'd need several puppet modules
> >like stdlib, puppetlabs-firewall, etc, which is a fair bit of
> >overhread - and oh, yeah, need to install the puppet client. I think
> >Chef is probably in a similar problem space. I don't want to resort to
> >shell scripts of python - config management tools know the difference
> >between apt and yum, and can still get a package installed with one
> >declaration, same thing with firewall rules. Is something like Ansible
> >or SaltStack a better choice?? I don't see it right now if it is, but
> >I don't have much experience with either of those two.
> >
> >The all-in-one installation process I'd like to see:
> >
> >Install your host OS
> >Install an meta-RPM/Deb that either (installs everything, or
> >alternatively configures a repo - or just installs the repo and the
> >stuff I need to install with)
> >Run a command that activates one of these config tools - configures
> >the machine, installs the packages I need, and gets me to the point
> >where I'm ready to login and go through the beautiful new user gui
> >setup stuff.
> >
> >I still want to keep the documentation around, it's invaluable for
> >experienced users and more complex deployments - but right now it's
> >far too much overhead (probably an hour or two) to get things
> >installed and setup to the point where you are ready to run the
> >'Welcome to CloudStack GUI' if you just want to try CloudStack out.
> >
> >So why am I writing this email instead of diving in and solving this
> >problem? Well honestly, I'd like some external opinions. I want to
> >make sure that I am not seeing a 'nail' simply because I have a hammer
> >in my hand. How can we most easily do this? So - how do we make the
> >'brand-new' user experience much better? We develop a platform for
> >orchestration of complex systems, this should be a solved problem.
> >
> >--David
>
> +1 for the initiative.
> If I look at Apache Hadoop's single node operation documentation[1], it is
> considerably simpler.
> Apache Tomcat installation is also fairly trivial.
>
> [1] http://hadoop.apache.org/docs/stable/single_node_setup.html
>
>


-- 
NS

Re: [DISCUSS] Making simple installs easier.

[Chiradeep Vittal <Ch...@citrix.com>]

On 4/23/13 12:15 PM, "Chiradeep Vittal" <Ch...@citrix.com>
wrote:

>
>
>On 4/21/13 3:21 PM, "David Nalley" <da...@gnsa.us> wrote:
>
>>Hi folks.
>>
>>I've been thinking about our install process lately.
>>
>>We currently require folks to muck about with firewall settings, NFS
>>settings, network configuration, etc.
>>This makes configuration painful, our docs VERY platform specific, and
>>easily prone to mistakes which result in failure to get things to
>>work. Even the 'install.sh' from the 3.0.x and earlier days doesn't do
>>enough. What I want to do is get rid of sections 2-4 of the quick
>>install guide, and replace it with - 'run this one or two lines worth
>>of commands' (http://s.apache.org/runbook)
>>
>>My natural reaction was to reach for puppet - but I am not sure that's
>>the right answer. To do things right, I'd need several puppet modules
>>like stdlib, puppetlabs-firewall, etc, which is a fair bit of
>>overhread - and oh, yeah, need to install the puppet client. I think
>>Chef is probably in a similar problem space. I don't want to resort to
>>shell scripts of python - config management tools know the difference
>>between apt and yum, and can still get a package installed with one
>>declaration, same thing with firewall rules. Is something like Ansible
>>or SaltStack a better choice?? I don't see it right now if it is, but
>>I don't have much experience with either of those two.
>>
>>The all-in-one installation process I'd like to see:
>>
>>Install your host OS
>>Install an meta-RPM/Deb that either (installs everything, or
>>alternatively configures a repo - or just installs the repo and the
>>stuff I need to install with)
>>Run a command that activates one of these config tools - configures
>>the machine, installs the packages I need, and gets me to the point
>>where I'm ready to login and go through the beautiful new user gui
>>setup stuff.
>>
>>I still want to keep the documentation around, it's invaluable for
>>experienced users and more complex deployments - but right now it's
>>far too much overhead (probably an hour or two) to get things
>>installed and setup to the point where you are ready to run the
>>'Welcome to CloudStack GUI' if you just want to try CloudStack out.
>>
>>So why am I writing this email instead of diving in and solving this
>>problem? Well honestly, I'd like some external opinions. I want to
>>make sure that I am not seeing a 'nail' simply because I have a hammer
>>in my hand. How can we most easily do this? So - how do we make the
>>'brand-new' user experience much better? We develop a platform for
>>orchestration of complex systems, this should be a solved problem.
>>
>>--David
>
>+1 for the initiative.
>If I look at Apache Hadoop's single node operation documentation[1], it is
>considerably simpler.
>Apache Tomcat installation is also fairly trivial.
>
>[1] http://hadoop.apache.org/docs/stable/single_node_setup.html

Also, I will put in a plug for QuickCloud which should help getting the
first vm up and running even quicker.

Re: [DISCUSS] Making simple installs easier.

[Chiradeep Vittal <Ch...@citrix.com>]

On 4/21/13 3:21 PM, "David Nalley" <da...@gnsa.us> wrote:

>Hi folks.
>
>I've been thinking about our install process lately.
>
>We currently require folks to muck about with firewall settings, NFS
>settings, network configuration, etc.
>This makes configuration painful, our docs VERY platform specific, and
>easily prone to mistakes which result in failure to get things to
>work. Even the 'install.sh' from the 3.0.x and earlier days doesn't do
>enough. What I want to do is get rid of sections 2-4 of the quick
>install guide, and replace it with - 'run this one or two lines worth
>of commands' (http://s.apache.org/runbook)
>
>My natural reaction was to reach for puppet - but I am not sure that's
>the right answer. To do things right, I'd need several puppet modules
>like stdlib, puppetlabs-firewall, etc, which is a fair bit of
>overhread - and oh, yeah, need to install the puppet client. I think
>Chef is probably in a similar problem space. I don't want to resort to
>shell scripts of python - config management tools know the difference
>between apt and yum, and can still get a package installed with one
>declaration, same thing with firewall rules. Is something like Ansible
>or SaltStack a better choice?? I don't see it right now if it is, but
>I don't have much experience with either of those two.
>
>The all-in-one installation process I'd like to see:
>
>Install your host OS
>Install an meta-RPM/Deb that either (installs everything, or
>alternatively configures a repo - or just installs the repo and the
>stuff I need to install with)
>Run a command that activates one of these config tools - configures
>the machine, installs the packages I need, and gets me to the point
>where I'm ready to login and go through the beautiful new user gui
>setup stuff.
>
>I still want to keep the documentation around, it's invaluable for
>experienced users and more complex deployments - but right now it's
>far too much overhead (probably an hour or two) to get things
>installed and setup to the point where you are ready to run the
>'Welcome to CloudStack GUI' if you just want to try CloudStack out.
>
>So why am I writing this email instead of diving in and solving this
>problem? Well honestly, I'd like some external opinions. I want to
>make sure that I am not seeing a 'nail' simply because I have a hammer
>in my hand. How can we most easily do this? So - how do we make the
>'brand-new' user experience much better? We develop a platform for
>orchestration of complex systems, this should be a solved problem.
>
>--David

+1 for the initiative.
If I look at Apache Hadoop's single node operation documentation[1], it is
considerably simpler.
Apache Tomcat installation is also fairly trivial.

[1] http://hadoop.apache.org/docs/stable/single_node_setup.html