NEVER use lastest tag

[unkcpz]

This is one thing I got from JupyterCon23 from the talk of https://github.com/yuvipanda who is one of core developers for jupyter docker-stack. "Never use latest" is one take-home message.
I can also find some relevant articles about it:

• What's Wrong With The Docker :latest Tag?
• DO NOT RUN A CONTAINER WITH THE :LATEST TAG
• ...

We actually encounter the issue when using it as the default image in aiidalab-launch, when there is a new release of aiida-core, it uses the new image automatically. This brings the issue that in @superstar54's live demo, not only when starting the container takes time to pull the new image, but also a critical bug happened.

I propose to use latest aiida-core version tag as the default tag of aiidalab-launch profile.
@yakutovicha @superstar54 @danielhollas comments?

[danielhollas]

I don't find the two linked articles that convincing in our use case, since we build and push images on the CI, and the latest tag therefore essentially means "the latest official version".

This brings the issue that in @superstar54's live demo, not only when starting the container takes time to pull the new image, but also a critical bug happened.

Could you please describe exactly what happened? What was the bug? To me, this is more of an issue that we need to test more the official released versions (and perhaps release them less often).

I propose to use latest aiida-core version tag as the default tag of aiidalab-launch profile.

I am not sure I understand your proposal here. The latest tag is equivalent to the latest aiida-core tag. They only start to differ when we bump the aiida-core version. What aiida-core version would you propose to be the default? And how often should the default be updated?

What would make more sense to me is to use the postgres version tag as the default, since supposedly, since updating PSQL is something that the user would typically want to avoid.

What I think is the bigger underlying issue is that in aiidalab-docker-stack the images are based on the latest jupyter-stack image, that is updated very often and often breaks us in unexpected ways (see many recent PRs where we needed to pin versions of various jupyter packages). Specifically, we're tracking Python3.9 tag, which is still updated regularly by the jupyter team. It would be more sane for us to track a tag that is not actively updated (e.g. some older Python version), and only update it e.g. once per six months or something. WDYT?

[danielhollas]

What I think is the bigger underlying issue is that in aiidalab-docker-stack the images are based on the latest jupyter-stack image, that is updated very often and often breaks us in unexpected ways (see many recent PRs where we needed to pin versions of various jupyter packages). Specifically, we're tracking Python3.9 tag, which is still updated regularly by the jupyter team. It would be more sane for us to track a tag that is not actively updated (e.g. some older Python version), and only update it e.g. once per six months or something. WDYT?

Hmm, looks like the above is not quite true, the last release of jupyter/minimal-notebook:python-3.19 was 7 months ago.
https://hub.docker.com/r/jupyter/minimal-notebook/tags?page=1&name=python-3.9.13

So I am not sure where for example issue from aiidalab/aiidalab-home#149 came from.

[unkcpz]

This brings the issue that in @superstar54's live demo, not only when starting the container takes time to pull the new image, but also a critical bug happened.

Could you please describe exactly what happened? What was the bug? To me, this is more of an issue that we need to test more the official released versions (and perhaps release them less often).

I was wrong about this, I thought aiida-2.3.0 is already released and full-stack:latest uses aiida-core==2.3.0. The issue I mentioned is after the container stopped it cannot start because the verdi daemon stop will throw the exception, as also reported by Carlo in the arm64 test and discussed in aiidalab/aiidalab-docker-stack#379.
@superstar54 show me the aiida-core version inside the container was 2.3.0 so it was the same issue but aiida-core is update not because of using a new docker stack but must be somewhere else. @superstar54 do you remember if you reinstall your customized QEapp or re-installed the AWB before you saw the issue?

But I still think using latest tag is not encouraged, it is true the aiida-2.2.2 is equal to latest but afterall the latest is less static than using aiida tag.

[unkcpz]

So I am not sure where for example issue from aiidalab/aiidalab-home#149 came from.

It does not come from the docker stack image but installing aiidalab-home will install the new widgetsnbextension (as the dependency of ipywidget). So we pinned the version inside aiidalab-home

[superstar54]

I got the error (Critical: The daemon could not be reached, ...) when I restart my computer.

[unkcpz]

Thanks @superstar54. Yes, that is exactly the issue as reported aiidalab/aiidalab-docker-stack#379. Do you know why you have aiida-core==2.3.0 rather than v2.2.2 which is provided by the container? You installed it explicitly or it is updated by other packages (for example if you pin aiida-quantumespresso==4.3.0 it pines aiida-core to v2.3.0).

[superstar54]

The original version is v2.2.2. I didn't upgrade it manually. Other packages should update it.

[danielhollas]

Aiida-core cannot be updated, it is pinned in the image. So I would think you were not using the latest tag. Dne pá 19. 5. 2023 16:28 uživatel Xing Wang ***@***.***> napsal:

…

The original version is v2.2.2. I didn't upgrade it manually. Other packages should update it. — Reply to this email directly, view it on GitHub <#179 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACIY64NYYJY6E4EXSTWSUKDXG6GS3ANCNFSM6AAAAAAYFXIOJM> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[superstar54]

Aiida-core cannot be updated, it is pinned in the image. So I would think you were not using the latest tag.

I used a customized image, which uses FROM aiidalab/full-stack:aiida-2.2.2

[danielhollas]

I got the error (Critical: The daemon could not be reached, ...) when I restart my computer.

How exactly did you get this error? (i.e. which commands exactly and where do you see the error?) Daemon cannot be reached is a very general error, it could have been something else.

I used a customized image, which uses FROM aiidalab/full-stack:aiida-2.2.2

In that case I really don't understand how you could get to aiida-core 2.3.0. We don't have anything imo that would require this version, so unless you went to great lengths to reinstall it, I am puzzled as you could get it to you image.

[unkcpz]

Aiida-core cannot be updated, it is pinned in the image.

Not fully true, I just check the constrain set in docker stack and it is pinned to a range https://github.com/aiidalab/aiidalab-docker-stack/blob/5a1de2efdf86aab6ab2153eebf9cc728f31d9e68/stack/base/requirements.txt#L1

As for the version constrain set in aiidalab by "core-packages", first of all, it will check only through app manager. Meanwhile, there is still way to override the these core-packages by not pinning the aiida-core but having dependencies to install the incompatible version of aiida-core, I guess in @superstar54's case, he install QeApp or aiida-quantumespresso==4.3.0 (either through pip or aiidalab insntall or app manager) and it will then override the aiida-core version to 2.3.0.

I don't know the better solution for it, the dilemma happened to this case is app and package break the container which we didn't consider. Maybe we need to pinning aiida-core version in https://github.com/aiidalab/aiidalab-docker-stack/blob/5a1de2efdf86aab6ab2153eebf9cc728f31d9e68/stack/base/requirements.txt#L1?

[unkcpz]

In that case I really don't understand how you could get to aiida-core 2.3.0.

I just tested, I use full-stack:latest and install aiidalab-qe==23.10.0a0 from app store, the aiida-core==2.3.0 is installed.

[danielhollas]

Not fully true, I just check the constrain set in docker stack and it is pinned to a range https://github.com/aiidalab/aiidalab-docker-stack/blob/5a1de2efdf86aab6ab2153eebf9cc728f31d9e68/stack/base/requirements.txt#L1

I see, you are correct, apologies.

I just tested, I use full-stack:latest and install aiidalab-qe==23.10.0a0 from app store, the aiida-core==2.3.0 is installed.

How exactly did you install it? Also, what is the pip version?

It is still weird that the version got updated, because qeapp requires aiida-core~=2.2, and so it should be satisfied by the already installed aiida-core version, unless you instruct pip to upgrade all packages in the installation process.

In other words, we should find out what triggered the aiida-core reinstallation, and how to possibly avoid it.

[unkcpz]

It is still weird that the version got updated, because qeapp requires aiida-core~=2.2

Nope, aiidalab-qe==23.10.0a0 -> aiida-quantumespresso==4.3.0 -> aiida-core==2.3.0.

How exactly did you install it? Also, what is the pip version?

I installed it from app store, just tick the include pre-release box.

[danielhollas]

Ah, I see, in that case there's really nothing we could do in the docker stack (except if we exactly pin aiida-core version, but that seems problematic).