Install guide modifications based on recent build attempts

[dcardon]

I found that an install on a fresh VM didn't succeed without a few global npm installations and without telling git to used https instead of git (optional depending on your network).

[dcardon]

What is this PR for?

A minor (but hopefully useful to others) documentation change.

What type of PR is it?

Documentation

Todos

N/A

Is there a relevant Jira issue?

No

How should this be tested?

Someone else could test this on a fresh VM / server.

Screenshots (if appropriate)

N/A

Questions:

• Does the licenses files need update? No
• Is there breaking changes for older versions? No
• Does this needs documentation? No

[bzz]

Technically, neither node, nor npm or
grunt-cli https://github.com/apache/incubator-zeppelin/blob/master/zeppelin-web/package.json#L29
or bower https://github.com/apache/incubator-zeppelin/blob/master/zeppelin-web/package.json#L10
are required to be installed manually on the environment, as they all are beeing taken care of by mvn package in zeppelin-web through the frontend-maven-pluging and there is a ./grunt wrapper for manual run too

[dcardon]

Unfortunately, this was not my experience. I followed these instructions on a fresh CentOS 7 install and the zeppelin-web module build failed without the npm install steps. Are you saying that it should work if you remove the sudo apt-get install npm step?

[bzz]

Thank you for sharing the experience!

What I was trying to say is that mvn package was designed to build zeppelin on clean environment, including web application, without requirements to have node.js&npm pre-installed there.

If it does not work like that on some environments - it has to be a bug well worth firing a JIRA issue.
Having those things installed sounds like a reasonable workaround to me in such cases.

[dcardon]

Thanks! I totally agree it would be better to have these install automatically--I'll look and see if I can modify the pom.xml to get that working without the additional steps.

[jeffsteinmetz]

If you follow the virtual machine instructions here, it installs with all required dependencies.
You will see that grunt and bower are not required (since maven handles it).

https://zeppelin.incubator.apache.org/docs/0.6.0-incubating-SNAPSHOT/install/virtual_machine.html

[dcardon]

Thanks, @jeffsteinmetz! I didn't know that guide existed. I did run into issues running this on Windows, but eventually this does work (after manual Ansible installation within the VM itself).

I retried my previous install attempt, this time on a clean CentOS 7 install and it looks like the embedded bower and grunt installs contain hidden git pulls. So, this time I just did the git config --global url."https://".insteadOf git:// command and everything built just fine. It seems all of my build issues relate back to the corporate firewall.... So, maybe it's better just to keep the note about the git workaround in the install.

[jeffsteinmetz]

We could reference this VM guide earlier in the Zeppelin documentation. Let me know if you think there is a documentation location at the top of the project that you feel would be beneficial.
It could be added to:
https://zeppelin.incubator.apache.org/docs/0.6.0-incubating-SNAPSHOT/development/howtocontribute.html

Also, could you let me know what version of Windows you were running when you installed Vagrant, Virtual Box and Ansible?
You shouldn’t need to install Ansible in the VM itself. The base VM only requires the bundled python – which exists in Ubuntu 14.04 LTS.

If it warrants additional research, we could test the VM installation steps on the specific version of Windows and improve the documentation.

What version of Ansible does your Window host machine (not VM) report?

[felixcheung]

@dcardon@jeffsteinmetz where are we on this?

[dcardon]

Thanks for following up, @felixcheung!

I've modified the changes to only call out the possible need to run git through https:// instead of git://, which was the root cause of the issue I encountered.

I think it's worthwhile to keep this documented because:

• the maven build failures made it really unclear that this was causing the failure
• most users behind corporate firewalls are likely to run into this issue

The Ansible VM on Windows is a separate issue that isn't part of the PR.

[jeffsteinmetz]

I too confirmed, the build runs in the VM without NPM / Node installed.

The VM installs node and ruby (ruby for docs only) so that a developer can build and test updates to their code outside of the Maven workflow.

Let me know if you think there is a documentation location at the top of the project that you feel would be beneficial.
A reference to the VM could be added to:
https://zeppelin.incubator.apache.org/docs/0.6.0-incubating-SNAPSHOT/development/howtocontribute.html