Pulsar: Helm deployment fails to deploy properly

[ckdarby]

Describe the bug
Unable to get Pulsar working correctly following the helm deployments.

To Reproduce
Steps to reproduce the behavior:

1. Follow instructions on helm deployment
2. Try to add a new environment from Pulsar admin manager

Expected behavior
Expect Pulsar to come up & be able to use pulsar admin manager to add new environment.

For what I can see in K8 the proxy isn't being exposed properly on the network. Unable to use the minikube ip address for port 30001.

[ckdarby]

Since at least commit: 23ae5ec

Everything related to Zookeeper coming up properly is broken in the helm deployment.

I went to the latest stable release & was able to get up & running successfully at least.

[merlimat]

@ckdarby Could you verify just before that commit if it works?

[ckdarby]

Works in 811dd45 and it doesn't work with 23ae5ec

No idea where it broke between those commits but I know when I pulled master at 23ae5ec the helm deployment stopped working.

[jiazhai]

@ckdarby as @merlimat mentioned, would you please help try the commits before 23ae5ec ? , such as
f95c71f21aae9e8299597295ccb2bc6954d82ae5 or 88fc454173af94d33986e608e94037db556a9226

commit 23ae5eca871bde2599f86cebfe9971cb08584bf9
Author: Ali Ahmed <[email protected]>
Date:   Sat Jan 4 18:30:34 2020 -0800

    Make docker files consistent. (#5949)

    Allow volume override for docker commands
    with env variable, this allows the scripts to
    work in dind environments.

commit f95c71f21aae9e8299597295ccb2bc6954d82ae5
Author: Matteo Merli <[email protected]>
Date:   Sat Jan 4 04:25:04 2020 -0800

    Make standalone to advertise "localhost" to avoid failure when hostname is not resolvable (#5856)

    ### Motivation

    The standalone service is failing to startup when the hostname is not resolvable to an IP. This is common in laptops.

    The reason is that the function worker in standalone is trying to connect to the advertised address rather than "localhost".

    To fix this, we should make the standalone to always advertise "localhost" as the the broker address. This is ok because in any case the standalone treats the advertisement as the pulsar proxy, in that it forces the clients to always go through the same service url, so a client connecting from a different host won't have problems.

commit 88fc454173af94d33986e608e94037db556a9226
Author: Jia Zhai <[email protected]>
Date:   Fri Jan 3 23:47:21 2020 +0800

    fix reader builder clone error (#5923)

    Motivation
    In reader builder, clone() method does not clone the config, and will cause error once we call a builder.clone() concurrently to create readers. This PR mainly try to fix this issue.

    Modifications
    use conf.clone(), to make sure conf is cloned.
    add test to verify it.
    fix other small issues, like get() methods, and access level, make it align with consumer/producer builder.
    Verifying this change
    ut passed.

[merlimat]

cc/ @aahmed-se

[aahmed-se]

@merlimat I don't my changes to docker build files can affect this,I did not change the runtime docker images.

[sijie]

@ckdarby are you building your image or do you just use the helm chart?

[ckdarby]

@sijie Just using the helm charts

[sijie]

if you are just using the helm charts, I don't think the changes to images will impact you. What are the errors you have encountered?

[jack-newbury]

@sijie I can confirm the issues that are @ckdarby is having as I'm only using the helm charts and was running into the same issue as described above using the latest release. However, going back to the commit https://github.com/apache/pulsar/commit/811dd45e87600d7163c8f74b40d377831f64bbfb Helm charts work again.

[sijie]

@jack-newbury and @ckdarby we have just updated the helm chart. Currently the helm chart is verified by the chart testing framework. You can:

• follow the instructions (http://pulsar.apache.org/docs/en/kubernetes-helm/) to get started
• follow the instructions to configure (http://pulsar.apache.org/docs/en/helm-deploy/) the helm chart to install a production setup

[ckdarby]

@sijie Wow, very impressed with the documentation & helm updates!

I am a bit confused though, a lot of these changes are in the 2.5.0 docs but I looked at 2.5.0 rc4 and a bunch of this isn't in there :(

[sijie]

@ckdarby Currently the helm chart is maintained in the main source repo while it shouldn't be. Because the changes to the helm chart should be decoupled from the changes to the main code. We are working on moving the helm chart into a separate repo maintained by the PMC - https://github.com/apache/pulsar-helm-chart as part of PIP-62.

[ckdarby]

@sijie Sounds good! Was able to get everything running without any issue.

Excellent contributions to whomever did all the changes!