Skip to content
This repository has been archived by the owner on Feb 16, 2019. It is now read-only.

.0.8 latest can't pull the docker image #387

Open
bfleming-ciena opened this issue Jun 9, 2018 · 21 comments
Open

.0.8 latest can't pull the docker image #387

bfleming-ciena opened this issue Jun 9, 2018 · 21 comments

Comments

@bfleming-ciena
Copy link

I'm sorry if I'm just missing something here but I pulled the 0.8.0 branch and trying to install with helm and getting.

Failed to pull image "docker.io/istio/proxyv2:0.8.latest": rpc error: code = Unknown desc = Error response from daemon: manifest for istio/proxyv2:0.8.latest not found

@ymesika
Copy link
Member

ymesika commented Jun 10, 2018

You probably already figured it out but you need to provide --set global.tag=latest to your helm install/template command.

It probably should've hold the latest nightly build tag but we still don't have such since 0.8.0 release.

@bfleming-ciena
Copy link
Author

Nope, had no clue I needed that. Thanks.

@bfleming-ciena
Copy link
Author

Hmm, I trried this for the master branch and got this error.

Warning Failed 10s (x2 over 31s) kubelet, realdev-kube-worker-2 Failed to pull image "docker.io/istio/proxyv2:latest": rpc error: code = Unknown desc = Error response from daemon: manifest for istio/proxyv2:latest not found

helm template install/kubernetes/helm/istio --set global.tag=latest --name istio --namespace istio-system --set tracing.enabled=true | kubectl create -f -

@bfleming-ciena
Copy link
Author

Didn't work. I can't remember how I installed htis originally. But when trying to use the github repo I can't install from master or 0.8 now.

helm template install/kubernetes/helm/istio --set global.tag=latest --name istio --namespace istio-system --set tracing.enabled=true | kubectl create -f -

Warning Failed 11s kubelet, realdev-kube-worker-2 Failed to pull image "docker.io/istio/proxyv2:latest": rpc error: code = Unknown desc = Error response from daemon: manifest for istio/proxyv2:latest not found

@bfleming-ciena
Copy link
Author

Ok, got it.

Did I miss something somewhere in the docs that indicated I had to set this global.tag option?

helm template install/kubernetes/helm/istio --set global.tag=0.8.0 --name istio --namespace istio-system | kubectl create -f -

@ymesika
Copy link
Member

ymesika commented Jun 11, 2018

Sorry, right, 0.8.0 tag.

The docs are referring to the released packages.
I believe that there the global.tag is set correctly. Isn't it?

@costinm
Copy link

costinm commented Jun 11, 2018 via email

@ymesika
Copy link
Member

ymesika commented Jun 11, 2018

@costinm I think it's the images in the docker hub that needs to be tagged with 0.8.latest.
I would do it but I don't think I have permissions.

@costinm
Copy link

costinm commented Jun 11, 2018 via email

@sakshigoel12
Copy link
Contributor

cc @hklai

@hklai
Copy link

hklai commented Jun 13, 2018

I doubt anyone had given much thought about installing from a release branch instead of using the release artifacts.

It is unclear to me if 0.8.latest is latest cutting edge build or latest stable build. If former, we currently do not even push to docker hub.

Similarly, we do not label "latest" in the daily release on master either.

@costinm
Copy link

costinm commented Jun 13, 2018 via email

@costinm
Copy link

costinm commented Jun 13, 2018 via email

@hklai
Copy link

hklai commented Jun 13, 2018

I can't say it is well thought of when something depends on circle nightly builds that is different from the regular daily releases maintained by the release team.

If we continue that approach circle should probably build nightly from release-0.8 (and other release branches) and label 0.8.latest (and newer releases later).

Otherwise we should kill circle nightly and make sure daily releases label accordingly as well.

And with Tiller, I suppose it can just take the release label (i.e. 0.8.0) without requiring the templates from the release, right?

@veggiemonk
Copy link

@costinm
Copy link

costinm commented Jun 18, 2018 via email

@costinm
Copy link

costinm commented Jun 18, 2018 via email

@sdake
Copy link
Member

sdake commented Jun 18, 2018

@stonefury typically the model for running master containers is to build them yourself. I guess some folks are using the nightly builds, and it sounds like some folks are using the circleci builds, but if you really want the manifest implementation to match the code in the container implementation, self-building is required.

@costinm
Copy link

costinm commented Jun 18, 2018 via email

@sdake
Copy link
Member

sdake commented Jun 18, 2018

@costin I find nothing all that controversial about it. I was simply pointing out the status quo. Agree that the default shouldn't be self-build. Without a constantly trailing functional container implementation, the container implementation and manifest implementation become de-synchronized (and assume developers are running on the latest master).

With appropriate caveats about using daily builds, even if tested, I don't see a problem.

@hklai
Copy link

hklai commented Jun 18, 2018

Why kill circle nightly ??? It is the one that works with lastest.
Kill the daily releases if you want to kill something, or fix them.

The official daily release (the one that we document in github/wiki/etc) also builds with latest on different branches. And we have a plan to have it push images to dockerhub with appropriate tags as well.

Having two systems is an overhead, even if you sign up to maintain/fix the circle nightly.

(To be clear my point is that we should have one system making daily/nightly releases. It can be prow, circle or anything else, but at the moment circle not the one supported by the release team.)

Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Projects
None yet
Development

No branches or pull requests

7 participants