Get started with Charmed Kubeflow

charm
,
2

I ran through the above tutorial, but had some issues with the MetalLB configuration. For me, I didn’t have the 10.64.140.0/24 network set up after installing MicroK8s. The documentation reads like this is automatically set up as a side effect of installing MicroK8s, but that’s not what I saw.

I needed to set up a Linux bridge for that network in order to land my MicroK8s node on that network so MetalLB could work. I created an LXD network as an easy way to set that up and to easily configure it so DHCP would avoid the MetalLB IP ranges - and that seemed to allow things to work for me.

Full disclosure: I had previously installed Microk8s on my system, but I did remove the snap (with --purge) prior to running through this tutorial, so my system should have been more-or-less “clean” before attempting the tutorial.

1 Like
3

Need to explain what MicroK8s is and why we install it.

We say that we need to wait for MK8s to complete config, what is the state that I need to wait for ?

4

does not work

5

[quote=“nohaihab, post:1, topic:7819”]

microk8s enable dns storage ingress metallb:10.64.140.43-10.64.140.49

[/quote]DEPRECIATION WARNING: ‘storage’ is deprecated and will soon be removed. Please use ‘hostpath-storage’ instead.

storage is deprecated, and should be hostpath-storage as per the warning when you execute the enable:

DEPRECIATION WARNING: 'storage' is deprecated and will soon be removed. Please use 'hostpath-storage' instead.

it’s also good to specify dns:{your upstream dns server} if google dns is not reachable otherwise it won’t work.

6

Any details about how to do this? I have the same issue. The istio-pilot waiting for ip address

7

Thanks - updated now and also in the MLflow one.

8

Hey! What do you get when you run this:

microk8s kubectl -n kubeflow get svc istio-ingressgateway-workload -o jsonpath='{.status.loadBalancer.ingress[0].ip}'
9

After installation I cannot access the dashboard at http://10.64.140.43.nip.io, ‘the site can’t be reached.’

microk8s kubectl get gateway -A NAMESPACE NAME AGE knative-serving knative-ingress-gateway 107m knative-serving knative-local-gateway 107m

juju config dex-auth public-url=http://10.64.140.43.nip.io WARNING the configuration setting “public-url” already has the value “http://10.64.140.43.nip.io”

The istio looks fine: istio-ingressgateway-workload LoadBalancer 10.152.183.197 10.64.140.43 80:30302/TCP,443:30118/TCP 97m

My laptop is running on 2 ips: the wired is on 10.64.140.38 and the Wifi is on 192.168.12.149

10

So I got this 10.64.140.43. Since my Wifi interface is on 192.168.12.149 I connected my wired LAN to another router. I assume since the laptop and the service IP are on the same block I should be able to access it?

11

Followed the instruction and reinstalled many times with clean microk8s and juju. It seems the following errors are common across all my installations:

> App                        Version                Status   Scale  Charm                    Channel         Rev  Address         Exposed  Message
> katib-controller           res:oci-image@111495a  waiting      1  katib-controller         0.15/stable     206  10.152.183.117  no
> kfp-api                    res:oci-image@e08e41d  waiting      1  kfp-api                  2.0/stable      298  10.152.183.4    no
> kfp-persistence            res:oci-image@516e6b8  waiting      1  kfp-persistence          2.0/stable      294                  no
> tensorboard-controller                            waiting      1  tensorboard-controller   1.7/stable      156                  no       Waiting for gateway info relation
> Unit                          Workload  Agent  Address       Ports              Message
> katib-controller/0*           error     idle   10.1.216.91   443/TCP,8080/TCP   crash loop backoff: back-off 5m0s restarting failed container=katib-controller pod=katib-controller-54846dbdbf-krk6z_...
> kfp-api/0*                    error     idle   10.1.216.92   8888/TCP,8887/TCP  crash loop backoff: back-off 5m0s restarting failed container=ml-pipeline-api-server pod=kfp-api-5cd4db4554-n76qf_kub...
> kfp-persistence/0*            error     idle   10.1.216.3                       crash loop backoff: back-off 5m0s restarting failed container=ml-pipeline-persistenceagent pod=kfp-persistence-5bbb9d...
> tensorboard-controller/0*     waiting   idle                                    Waiting for gateway info relation

Also the machine has wireless card connected to the internet that has 192.168.12.xxx address and a LAN card on 10.40.140.xxx block. I also added 10.40.140.43.nip.io to the host file after seeing ns lookup failure.

13

Hey did you try this:

An issue you might have is the tensorboard-controller component might be stuck with a status of waiting and a message “Waiting for gateway relation”. To fix this, run:

juju run --unit istio-pilot/0 -- "export JUJU_DISPATCH_PATH=hooks/config-changed; ./dispatch"

This is a known issue, see tensorboard-controller GitHub issue for more info.

14

Hey,

So maybe there’s a conflict with your LAN.

The tutorial assumes the Kubeflow dashboard will be accessible at http://10.64.140.43.nip.io. Considering that your laptop is running on two IP addresses and one of them is mentioned as 10.64.140.38, there’s a possibility that you are already on a local area network (LAN) that uses the IP address range 10.64.140.0/24, which includes the IP address 10.64.140.43 mentioned in the tutorial.

To resolve this, you can try the following steps:

  1. Confirm the IP address range being used by your local network. Check if the range overlaps with the IP address 10.64.140.43. If there’s a conflict, it can prevent you from accessing the Kubeflow dashboard.
  2. If a conflict exists, you can modify the IP address range specified in the tutorial’s configuration. For example, let’s say you want to change the IP address to 10.64.141.43 to avoid conflicts. Run the following command instead of the original microk8s enable command:
microk8s enable dns hostpath-storage ingress metallb:10.64.141.43-10.64.141.49

This will set up the private IP address 10.64.141.43 as accessible within your VM environment.

Note: I haven’t tested this specific configuration myself, it’s just an idea

15 
> App                        Version                  Status   Scale  Charm                    Channel         Rev  Address         Exposed  Message
> kfp-api                    res:oci-image@e08e41d    waiting      1  kfp-api                  2.0/stable      298  10.152.183.106  no
> kfp-persistence            res:oci-image@516e6b8    waiting      1  kfp-persistence          2.0/stable      294                  no
> Unit                          Workload  Agent  Address       Ports              Message
> kfp-api/0*                    error     idle   10.1.216.122  8888/TCP,8887/TCP  crash loop backoff: back-off 5m0s restarting failed container=ml-pipeline-api-server pod=kfp-api-6658f6984b-dd8mp_kub...
> kfp-persistence/0*            error     idle   10.1.216.123                     crash loop backoff: back-off 5m0s restarting failed container=ml-pipeline-persistenceagent pod=kfp-persistence-5d7987...

The errors are pretty consistent across my installation attempts.

Also when I finally get to the Dashboard it stuck at ‘creating namespace’. After the namespace created it stays at the same page. When you back off and click the ‘Finish’ again it tells you the namespace already exists.

16

Could be an issue with microk8s and the default inotify limits. Did you run this?

17

Thanks!. That solved the problem.

1 Like
18

Error with juju deploy:

Located charm "tensorboards-web-app" in charm-hub, channel 1.7/stable
Located charm "training-operator" in charm-hub, channel 1.6/stable
ERROR lost connection to pod
ERROR lost connection to pod
ERROR cannot deploy bundle: cannot resolve charm or bundle "jupyter-ui": connection is shut down

Tried several times and it seems that the ERROR happened at different stages each time. I am doing it on a 3 node cluster on microk8s

19

Hi Andrew, can you run some commands to inspect / diagnose / debug what’s going on and let us know what’s happening?

Here are just some ideas:

  • What are the results of microk8s status and juju status
  • Use kubectl to get more info about what’s going on: get / describe / logs
  • Check your network connectivity between the microk8s nodes
  • Try sshing into the pods and see what happens

Also describe in as much detail steps to reproduce your exact setup, including the machine specs (RAM, OS etc.) you’re using, what cloud provider if in the cloud, how you setup the microk8s cluster etc.

Try to get as much info as you can for us, and then ping us back here with all the info. Then we’ll take it from there.

20

Tried on a clean new Ubuntu install (1 node only) following the documents with microk8s and this. Here is the screen.

andrew@G9:~$ juju bootstrap microk8s
Creating Juju controller "microk8s-localhost" on microk8s/localhost
Bootstrap to Kubernetes cluster identified as microk8s/localhost
Fetching Juju Dashboard 0.8.1
Creating k8s resources for controller "controller-microk8s-localhost"
Starting controller pod
Bootstrap agent now started
Contacting Juju controller at 10.152.183.148 to verify accessibility...
ERROR lost connection to pod

Bootstrap complete, controller "microk8s-localhost" is now available in namespace "controller-microk8s-localhost"

Now you can run juju add-model to create a new model to deploy k8s workloads.

For the previous install I ignored this ERROR and continued with add-model. I think this is the root of the problem. The network settings:

> 1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue state UNKNOWN group default qlen 1000
> link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
> inet 127.0.0.1/8 scope host lo
>    valid_lft forever preferred_lft forever
> inet6 ::1/128 scope host 
>    valid_lft forever preferred_lft forever
> 2: enp3s0: <NO-CARRIER,BROADCAST,MULTICAST,UP> mtu 1500 qdisc mq state DOWN group default qlen 1000
> link/ether ec:aa:a0:18:45:bf brd ff:ff:ff:ff:ff:ff
> 3: wlp2s0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue state UP group default qlen 1000
> link/ether 9c:b6:d0:0c:eb:35 brd ff:ff:ff:ff:ff:ff
> inet 192.168.12.100/24 brd 192.168.12.255 scope global dynamic noprefixroute wlp2s0
>    valid_lft 41777sec preferred_lft 41777sec
> inet6 2607:fb91:87e:825a:856:6482:8976:6010/128 scope global dynamic noprefixroute 
>    valid_lft 2180sec preferred_lft 830sec
> inet6 2607:fb91:87e:825a:1ed6:460:ed44:1905/64 scope global temporary dynamic 
>    valid_lft 86149sec preferred_lft 14149sec
> inet6 2607:fb91:87e:825a:9b37:9d88:b462:b6fa/64 scope global dynamic mngtmpaddr noprefixroute 
>    valid_lft 86149sec preferred_lft 14149sec
> inet6 fe80::6c92:966f:6ff9:8f06/64 scope link noprefixroute 
>    valid_lft forever preferred_lft forever
> 6: vxlan.calico: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1450 qdisc noqueue state UNKNOWN group default 
> link/ether 66:c8:18:67:df:7a brd ff:ff:ff:ff:ff:ff
> inet 10.1.243.128/32 scope global vxlan.calico
>    valid_lft forever preferred_lft forever
> inet6 fe80::64c8:18ff:fe67:df7a/64 scope link 
>    valid_lft forever preferred_lft forever
> 7: calica2d35e61c3@if3: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue state UP group default 
> link/ether ee:ee:ee:ee:ee:ee brd ff:ff:ff:ff:ff:ff link-netns cni-5110dfe6-5021-3929-7c13-826d6ff19ee8
> inet6 fe80::ecee:eeff:feee:eeee/64 scope link 
>    valid_lft forever preferred_lft forever
> 8: cali32780c3cbff@if3: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue state UP group default 
> link/ether ee:ee:ee:ee:ee:ee brd ff:ff:ff:ff:ff:ff link-netns cni-dbaaddb1-bc74-3af9-4ecb-13bc54c630de
> inet6 fe80::ecee:eeff:feee:eeee/64 scope link 
>    valid_lft forever preferred_lft forever
> 9: calicb43c72bc0b@if3: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue state UP group default 
> link/ether ee:ee:ee:ee:ee:ee brd ff:ff:ff:ff:ff:ff link-netns cni-c2fb5bc9-a1a0-5f1c-b3c1-16e679ecc7f9
> inet6 fe80::ecee:eeff:feee:eeee/64 scope link 
>    valid_lft forever preferred_lft forever
> 10: calic40d2f9fe91@if3: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue state UP group default 
> link/ether ee:ee:ee:ee:ee:ee brd ff:ff:ff:ff:ff:ff link-netns cni-94b4a477-f3fd-f732-bfd1-e1f85c8ace54
> inet6 fe80::ecee:eeff:feee:eeee/64 scope link 
>    valid_lft forever preferred_lft forever
> 11: calicd215964915@if3: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue state UP group default 
> link/ether ee:ee:ee:ee:ee:ee brd ff:ff:ff:ff:ff:ff link-netns cni-dc847b39-d6d1-4672-a176-99b4fc226b48
> inet6 fe80::ecee:eeff:feee:eeee/64 scope link 
>    valid_lft forever preferred_lft forever
> 13: cali2866693e52b@if3: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue state UP group default 
> link/ether ee:ee:ee:ee:ee:ee brd ff:ff:ff:ff:ff:ff link-netns cni-84a8ae58-9fbe-be94-7ada-6c63ac16b0e8
> inet6 fe80::ecee:eeff:feee:eeee/64 scope link 
>    valid_lft forever preferred_lft forever
> 14: calibe02fed43bb@if3: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue state UP group default 
> link/ether ee:ee:ee:ee:ee:ee brd ff:ff:ff:ff:ff:ff link-netns cni-a1aa2784-d203-f334-443a-83d33fd7c3d1
> inet6 fe80::ecee:eeff:feee:eeee/64 scope link 
>    valid_lft forever preferred_lft forever
21

The only time I ever saw Juju hanging was because I didn’t have enough disk space.

What are your machine specs for the machine are you running Juju / microk8s on? Do they satisfy all the requirements from the tutorial?

  • Runs Ubuntu 20.04 (focal) or later.
  • Has at least 4 cores, 32GB RAM and 50GB of disk space available.
  • Is connected to the internet for downloading the required snaps and charms.
  • Has python3 installed.
22

ubuntu22.04

I just followed the tutorial.

All pods work normally, but the following error occurs.

스크린샷 2023-11-24 19-44-05
스크린샷 2023-11-24 19-44-051402×273 65.3 KB

1 Like