On Tue, Feb 4, 2020 at 9:57 PM Daniel Walsh <dwalsh@redhat.com> wrote:
On 2/4/20 11:02 AM, Laurent Pellegrino wrote:
OK, I better understand. 

For sure, if the purpose is just to recreate the pod/containers with a new image then the command suggested previously has low interest, the main value would be to keep the connections (i.e. updating the service/pod without an outage).

> In your case the admin needs to realize that one of the containers in the Pod uses an image that has been updated, and now triggers the service.

In my case, the admin is a continuous deployment script. The update is triggered for instance by a Github action on a new commit in a Git repository.

Valentin WDYT?

TL;DR : My 2 cents are that a podman-update command is possible (maybe restricted to K8s yaml) but no-downtime and rollbacks are IMHO out of scope.

The two scenarios seem quite related. If we had a podman-update command that could do that, then both scenarios would be covered.

However, we did not plan for such a command but to implement updates by restarting systemd services that recreate containers. This certainly includes some downtime of the service. Doing that in Podman without a downtime is at least incredibly complex, if not impossible for some cases (e.g., duplicate ports etc.).

Use cases that require a downtime-free update, maybe even rollbacks, are IMHO leaving the single-node use case that Podman targets and enter the orchestration & Kubernetes world. A concrete technical challenge for Podman is recreating the same Pod/Container we want to update. In the Kubernetes world, that's fairly easy since everything is created via yaml files where bumping the image to a newer version is trivial. We don't have this guarantee in Podman as there are many different ways to create a Pod or container (e..g, via the CLI, the remote API, the native golang API, or K8s yaml). This makes the cloning of a pod/container much more complex and prone to errors. On the other hand, we could enforce that we only support rolling updates for Pods created via podman-play-kube (i.e., via a K8s yaml) but the downtime issue would still remain. IMHO, no downtime requires a cluster and that's out of Podman's scope.


Le mar. 4 févr. 2020 à 16:13, Daniel Walsh <dwalsh@redhat.com> a écrit :
On 2/4/20 9:04 AM, Laurent Pellegrino wrote:
I don't know about internals so it may not apply but regarding the use case I described briefly, I know what pod to update and what new image to use. Also, the image is available at the time the update must be performed. 

Taking Kubernetes as an example [1], from a CLI point of view, I would see something like this:

> podman rolling-update PODNAME --image=IMAGE:TAG

Sure but you could script up the same thing.

podman pull images related to pod.

Then stop/destroy the pod/containers.  Recreate the pod/containers.

I am not against your idea, but it brings in the Human element, I want to automate this process which is why I suggested a cron job.

In your case the admin needs to realize that one of the containers in the Pod uses an image that has been updated, and now triggers the service.

I would prefer to just have the system figure out whether an update is needed.



Le mar. 4 févr. 2020 à 14:46, Daniel Walsh <dwalsh@redhat.com> a écrit :
On 2/4/20 7:28 AM, Laurent Pellegrino wrote:
Thanks for the information.

Just to understand, why are you considering a periodic job?

We don't know when the image that the container is based on will be updated.  So checking once a day seems like a reasonable thing to do.  How else would you suggest we do this? 
> This however would not maintain the the connections.

OK, so for now, the only solution seems to deploy a proxy that supports dynamic configurations with no restarts?

Le lun. 3 févr. 2020 à 19:28, Daniel Walsh <dwalsh@redhat.com> a écrit :
On 2/1/20 3:28 AM, laurent.pellegrino@gmail.com wrote:
> Hi there!
>
> I discovered Podman recently and started to use it as a great alternative to Docker.
>
> My question is about application updates. Let's say you have a pod running and you want to update the code for the app in that pod. Is there a simple manner to manage something similar to a Kubernetes rolling update? By rolling update, I mean performing an update with no service interruptions (i.e. starting a new pod running the new app version, switch the network traffic to the new pod and terminate the old pod).
>
> Kind Regards,
>
> Laurent Pellegrino
> _______________________________________________
> Podman mailing list -- podman@lists.podman.io
> To unsubscribe send an email to podman-leave@lists.podman.io

We are actually investigating this workflow. But have not taken as far
as you want.  The idea would be to be running a container, and then have
a job start periodically that would see if their was an update to the
image that the container was running on.  It would pull the image down
and restart the container service, which would destroy the old container
and create a new container with the same params.  This however would not
maintain the the connections.
_______________________________________________
Podman mailing list -- podman@lists.podman.io
To unsubscribe send an email to podman-leave@lists.podman.io

_______________________________________________
Podman mailing list -- podman@lists.podman.io
To unsubscribe send an email to podman-leave@lists.podman.io


_______________________________________________
Podman mailing list -- podman@lists.podman.io
To unsubscribe send an email to podman-leave@lists.podman.io



_______________________________________________
Podman mailing list -- podman@lists.podman.io
To unsubscribe send an email to podman-leave@lists.podman.io