docker-stacks and jupyterhub #718

paskino · 2022-05-19T16:32:09Z

drop travis, update GHA docker CI

push to ghcr.io/synerbi/sirf

docker tag	SuperBuild branch/tag
`latest`, `latest-gpu`	latest tag `vM.m.p`
`M`, `M.m`, `M.m.p`, `M-gpu`, `M.m-gpu`, `M.m.p-gpu`	tag `vM.m.p`
`edge`, `edge-gpu`	`master`
only build & test (no tag)	CI (current commit)
`devel`, `devel-gpu`	`master` with `cmake -DDEVEL_BUILD=ON -DBUILD_CIL=ON`

build --pull optimisation

push to DockerHub
ccache builds with https://github.com/actions/cache

auto OMP_NUM_THREADS = cpu_count//2
use https://github.com/jupyter/docker-stacks submodule
- update entrypoint to launch gadgetron
drop old docker images & scripts
- merge /jupyterhub/ -> /docker/
add docker/compose.sh helper script
update Dockerfile & docker-compose*.yml
separate CPU & GPU images
ship CIL-Demos, CIL-Exercises, SIRF-contrib
- run download_data.sh
replace make => ninja builder
- auto num parallel threads
force ipywidgets>=8 to fix widget display (at the cost of figure duplication islicer support for ipywidgets>=8 TomographicImaging/CIL#1600)
rebase after update versions of many dependencies #862 (maybe expose more cmake options in Dockerfile)
update readmes
- docker permissions (--user, --group-add)
- docker compose override permissions (USER_ID, UID)
  - USER_ID: you could define this as a custom override
  - UID: current user
  - GID: not needed, vis. https://jupyter-docker-stacks.readthedocs.io/en/latest/using/common.html#user-related-configurations
- build options (# CMake options in Dockerfile)
- adding user deps
```
FROM synerbi/sirf:jupyter-gpu
USER root
RUN mamba install pytorch && fix-permissions "${CONDA_DIR}" /home/${NB_USER}
USER ${NB_UID}
```
- SIRF_DOWNLOAD_DATA_ARGS
update changelog
use current repo state when building rather than cloning from remote URL
- caveat: uncommitted changes are ignored for safety

fixes jupyterhub: apt finds unsigned package repositories #715
closes Docker: service.sh expects the cloned notebooks in the wrong directory #816
part of installation/testing of CIL on docker and VM #836
part of multiple requirements*.*/environment.* #834
fixes docker: make sure apt cache is cleared #828
closes Error when building Boost on docker #810
fixes GHA should push docker images to DockerHub #804
fixes GHA: docker always builds master #775
fixes docker: service-gpu too big #770
fixes GHA: docker run rebuilds images #763
fixes docker: to build from the local files rather than from a branch on GitHub #664, fixes Docker - build using current SuperBuild #597
fixes Add GitHub Action to build jupyterhub image #652, fixes Create Github Actions #514
fixes docker: remove need for 2 users and chown etc #596
closes docker: will need scripts for release images #592
fixes docker building: cache misses #563
fixes Dockerfile license is Mozilla MPL2.0 #559
~~This PR is currently used to run jupyterhub for the PSMRTBP2022 school~~
However this PR changes slightly the use of requirements.txt
- why?
Probably we should use requirements.txt if BUILD_CIL=OFF or BUILD_CIL_LITE=OFF, and switch to requirements installed from conda-forge in case we build CIL.
- why?
Some time ago, I also made so that CIL could be installed by conda, however maintaining this will be a problem because I see that we might need to have different requirements for the case we build CIL, we want CIL but not want to build it, etc. So I propose that CIL can only be built by the SuperBuild.
- opposite now (favour conda install)?

to move to follow-up issues/PRs

add devcontainer spec
- CPU-only default for gh codespaces
- GPU option (e.g. using https://containers.dev/implementors/templates/#folder-structure)?

paskino · 2022-05-20T08:44:40Z

Also I am building with the following command

 docker build --build-arg BASE_IMAGE=nvidia/cuda:10.0-cudnn7-devel-ubuntu18.04 \
--build-arg PYTHON_INSTALL_DIR=/opt/conda \
--build-arg EXTRA_BUILD_FLAGS="-DBUILD_CIL=ON -DIPP_LIBRARY=/opt/conda/lib -DIPP_INCLUDE=/opt/conda/include -DSTIR_URL=https://github.com/paskino/STIR.git -DSTIR_TAG=bump_parallelproj" \
--build-arg SIRF_SB_URL="https://github.com/paskino/SIRF-SuperBuild.git" \
--build-arg SIRF_SB_TAG="jupyterhub_env" --build-arg NUM_PARALLEL_BUILDS=6 --target sirf .

jupyterhub/Dockerfile

version_config.cmake

KrisThielemans

a few things that I spotted that might help

jupyterhub/README.md

KrisThielemans · 2023-01-23T15:14:53Z

are we building ASTRA? Best for the intro notebooks.

By the way, I'm not sure that CIL dependencies are alright when building ASTRA (i.e. does it need to be guaranteed that ASTRA is built before CIL? i.e.

SIRF-SuperBuild/SuperBuild/External_CIL.cmake

Line 24 in d50fce8

set(${proj}_DEPENDENCIES "")

might need an if on ASTRA

KrisThielemans · 2023-01-23T15:18:21Z

I'm a bit surprised that all the actions are running. we're excluding jupyterhub/**. I thought for the docker-one could be because the white-space change in Dockerfile, but the others shouldn't run. oh well. Not important now, but if you don't know why this is happening, please create an issue for it after the dust settles.

paskino · 2023-01-23T16:57:47Z

does it need to be guaranteed that ASTRA is built before CIL?

CIL does not depend on ASTRA. We will use ASTRA if present.

For SIRF usage it is not required. @ckolbPTB introductory notebook does not require ASTRA. The gradient_descent one does, see SyneRBI/SIRF-Exercises#192.

KrisThielemans · 2023-01-23T17:05:00Z

CIL does not depend on ASTRA. We will use ASTRA if present.

I agree, but if ASTRA is built after CIL, will it still be able to use it?

Why not build it on JupyterHub? too late I guess.

For SIRF usage it is not required.

sure. https://github.com/SyneRBI/SIRF-Exercises/blob/master/notebooks/Introductory/acquisition_model_mr_pet_ct.ipynb uses it as well, although at least now it no longer fails as you've seen.

paskino · 2023-01-23T17:10:21Z

I currently didn't add CIL to the jupyterhub image. I'll add it via conda if required. It is a 5 minute job. :D

paskino · 2023-01-23T17:12:23Z

if ASTRA is built after CIL, will it still be able to use it?

Yes, we only interact with ASTRA at the Python level so, as long as ASTRA and the python wrappers are built we can use it.

KrisThielemans · 2023-01-23T17:18:07Z

as long as ASTRA and the python wrappers are built we can use it.

yeah, but are the wrappers built if ASTRA isn't there?

KrisThielemans · 2023-01-23T17:20:29Z

I currently didn't add CIL to the jupyterhub image. I'll add it via conda if required. It is a 5 minute job. :D

well, if there's no CIL at all, even the intro stuff will fail, as we now just test on ASTRA.

I don't understand though. I thought you're grabbing the normal docker image, which has CIL. looks like i don't understand the jupyterhub stuff at all after all!

KrisThielemans · 2023-01-23T17:37:03Z

SIRF-SuperBuild/jupyterhub/Dockerfile

Line 1 in fe6806f

ARG BASE_IMAGE=ubuntu:18.04

is still 18.04

KrisThielemans · 2023-01-23T17:39:02Z

SIRF-SuperBuild/jupyterhub/Dockerfile

Line 50 in fe6806f

ENV SIRF_EXERCISES_DATA_PATH "/mnt/materials/SIRF/Fully3D/SIRF/"

says Fully3D. I don't care really, but just letting you know

KrisThielemans · 2023-01-23T18:04:23Z

SIRF-SuperBuild/jupyterhub/Dockerfile

Lines 39 to 40 in fe6806f

    
           COPY --from=synerbi/sirf:sirf-core /opt/SIRF-SuperBuild/INSTALL/ /opt/SIRF-SuperBuild/INSTALL 
        
           COPY --from=synerbi/sirf:sirf-core /opt/SIRF-SuperBuild/sources/SIRF/ /opt/SIRF-SuperBuild/sources/SIRF/

sirf-core? Doesn't make sense to me. core doesn't have an INSTALL? I suppose that's your local name when building as the one on dockerhub is sirf/core.

I guess it'd have to be release-service-gpu, except that is always going to point backwards. We currently don't seem to build devel-service-gpu (nor devel-gpu) on Travis, so we cannot pull from there.

I suppose this works for you, so no need to fix this now.

KrisThielemans · 2023-01-25T04:07:52Z

deprecation isn't installed. Can't say I understand the installation of the dependencies anymore, but I'm sure @paskino does!

KrisThielemans · 2023-01-25T07:41:44Z

nbstripout is not available, and there for the instructions fail. (I've updated the HackMD page)

KrisThielemans · 2023-01-25T07:50:14Z

nibabel docopt also missing

KrisThielemans · 2023-01-25T08:25:43Z

parallelproj isn't installed, so STIR doesn't have it (and neither does SIRF of course)

paskino · 2023-01-25T11:01:54Z

deprecation isn't installed. Can't say I understand the installation of the dependencies anymore, but I'm sure @paskino does!

Installed

docker/.bashrc

docker/Dockerfile

docker/requirements-service.yml

jupyterhub/README.md

version_config.cmake

jupyterhub/Dockerfile

jupyterhub/build_docker_stacks.sh

casperdcl · 2023-11-28T17:09:59Z

Some time ago, I also made so that CIL could be installed by conda, however maintaining this will be a problem because I see that we might need to have different requirements for the case we build CIL, we want CIL but not want to build it, etc.

So I propose that CIL can only be built by the SuperBuild.

Is this still the plan @paskino?

Also btw what about replacing all the current SIRF jupyter image stuff with docker-stacks (i.e. no need for paskino#2 nor paskino#3, and simplify/merge a lot of things in ./docker/*)?

jupyterhub/build_docker_stacks.sh

casperdcl

just a bit concerned about which strategy is best:

copy SIRF installation dirs from synerbi/sirf:latest (CPU) image
copy SIRF installation dirs from synerbi/sirf:service-gpu (GPU) image
just build SIRF properly in the proposed Dockerfile

This PR currently does option (1), but I think (2) is safer, and (3) is best. btw option (3) also means we could perhaps deprecate the old SIRF images in favour of this new docker-stacks way?

CHANGES.md

docker/requirements.yml

jupyterhub/Dockerfile

casperdcl · 2023-11-28T22:27:59Z

jupyterhub/README.md

-### Start building SIRF
-
-Build the `sirf` target of the SIRF Dockerfile with the `nvidia/cuda:11.5.0-cudnn8-devel-ubuntu18.04` base image.
-
-```
-git clone [email protected]:SyneRBI/SIRF-SuperBuild.git
-cd SIRF-SuperBuild/docker
-
-# build standard SIRF docker
-docker build --build-arg BASE_IMAGE=nvidia/cuda:10.2-cudnn8-devel-ubuntu18.04 --build-arg PYTHON_INSTALL_DIR=/opt/conda --target sirf .
-
-```


I removed this section @paskino but I'm concerned it doesn't match.

The removed readme text implies you have to rebuild SIRF images using nvidia/cuda... but in the proposed Dockerfile we just COPY --from=synerbi/sirf:latest (i.e. the CPU SIRF image). Is this ok?

The thing is that when doing this I was building the GPU SIRF. And this doc was for me.

At the time I needed to build a specific branch/tag of SIRF(SuperBuild); these were the commands I used to build the SIRF image I required. Then I copied the files from such image to the docker-stack one.

casperdcl · 2023-11-28T22:30:33Z

jupyterhub/README.md

-#### Building CIL
-
-Please see [here](https://github.com/SyneRBI/SIRF-SuperBuild#building-ccpi-cil) for detailed info on the command below.
-
-
-```
-
-docker build --build-arg BASE_IMAGE=nvidia/cuda:10.2-cudnn8-devel-ubuntu18.04 --build-arg PYTHON_INSTALL_DIR=/opt/conda --build-arg EXTRA_BUILD_FLAGS="-DBUILD_CIL=ON -DIPP_LIBRARY=/opt/conda/lib -DIPP_INCLUDE=/opt/conda/include -DBUILD_ASTRA=ON" --target sirf . 


I also removed this section @paskino because in the proposed PR it seems CIL is just installed via mamba... is this correct?

I think that having this is important for when one wants to build a particular branch/tag of CIL which is not available as conda package.

jupyterhub/Dockerfile

KrisThielemans

I haven't tried this, nor looked at it in detail, but overall impression is good, and my remaining worries have been resolved, so let's merge!

KrisThielemans · 2024-02-21T13:29:03Z

surely this must close quite a few more issues than what's listed?

casperdcl · 2024-02-21T20:59:26Z

surely this must close quite a few more issues than what's listed?

I... just looked through the issues and added to the list.
This PR no longer looks daunting.
The length of the issues it fixes looks daunting!

KrisThielemans · 2024-02-21T22:33:10Z

Major update indeed! Let's assign it to 3.6 and merge :-)

epapoutsellis mentioned this pull request May 20, 2022

Remaining Jupyter notebooks issues SyneRBI/SIRF-Exercises#156

Closed

4 tasks

KrisThielemans reviewed May 25, 2022

View reviewed changes

jupyterhub/Dockerfile Outdated Show resolved Hide resolved

KrisThielemans reviewed May 26, 2022

View reviewed changes

version_config.cmake Outdated Show resolved Hide resolved

KrisThielemans requested changes Jan 23, 2023

View reviewed changes

jupyterhub/README.md Outdated Show resolved Hide resolved

jupyterhub/README.md Outdated Show resolved Hide resolved

jupyterhub/README.md Outdated Show resolved Hide resolved

jupyterhub/README.md Outdated Show resolved Hide resolved

jupyterhub/README.md Outdated Show resolved Hide resolved

KrisThielemans mentioned this pull request Mar 9, 2023

Error when building Boost on docker #810

Closed

paskino force-pushed the jupyterhub_env branch from 75e5c9f to 6d057c1 Compare November 23, 2023 20:35

casperdcl changed the title ~~Jupyterhub env~~ docker-stacks and jupyterhub Nov 28, 2023

casperdcl requested changes Nov 28, 2023

View reviewed changes

casperdcl reviewed Nov 28, 2023

View reviewed changes

jupyterhub/build_docker_stacks.sh Outdated Show resolved Hide resolved

casperdcl force-pushed the jupyterhub_env branch from d4f0793 to 410d7ac Compare November 28, 2023 22:04

casperdcl requested changes Nov 28, 2023

View reviewed changes

casperdcl reviewed Nov 28, 2023

View reviewed changes

jupyterhub/Dockerfile Outdated Show resolved Hide resolved

paskino added 9 commits February 20, 2024 16:37

added some docs in Readme

03c3206

update to python 3.10 and do not fail if Gadgetron tests cannot be run

188f8c1

install dxchange via conda

dfc9e20

temporarily disable CIL test 11

0b93393

reenable tests as fixed in CIL

56a0db1

set python version to 3.10 as string

056f186

test CIL PR

6823d86

revert to default volume mount

25f6a14

[ci skip] update readme

4a93cac

casperdcl force-pushed the jupyterhub_env branch from 0fa2a03 to 164445e Compare February 20, 2024 16:50

casperdcl added 6 commits February 20, 2024 17:41

cleanup post-rebase

7e2573e

customisable OMP_NUM_THREADS

d8a5249

fix executable permissions

7cb3734

CI: optimise devel build

ea73e05

docker docs

ac7f72f

compose.sh: better help output

04e266d

casperdcl force-pushed the jupyterhub_env branch from b1340a8 to 04e266d Compare February 20, 2024 20:49

casperdcl added 2 commits February 21, 2024 01:41

merge dev docs

ca0e4c3

fix compose.sh -U

69b0963

KrisThielemans approved these changes Feb 21, 2024

View reviewed changes

CI: separate pull/build, don't cache devel

aa0478b

casperdcl merged commit bfb5f9f into SyneRBI:master Feb 21, 2024
8 checks passed

This was referenced Feb 21, 2024

add devcontainer spec #865

Open

fix docker deploy #866

Closed

casperdcl mentioned this pull request Feb 29, 2024

Dockerfile license is Mozilla MPL2.0 #559

Closed

casperdcl mentioned this pull request Mar 18, 2024

available docker tags (no devel?) #890

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

docker-stacks and jupyterhub #718

docker-stacks and jupyterhub #718

paskino commented May 19, 2022 •

edited by casperdcl

Loading

paskino commented May 20, 2022

KrisThielemans left a comment

KrisThielemans commented Jan 23, 2023

KrisThielemans commented Jan 23, 2023

paskino commented Jan 23, 2023

KrisThielemans commented Jan 23, 2023

paskino commented Jan 23, 2023

paskino commented Jan 23, 2023

KrisThielemans commented Jan 23, 2023 •

edited

Loading

KrisThielemans commented Jan 23, 2023

KrisThielemans commented Jan 23, 2023

KrisThielemans commented Jan 23, 2023

KrisThielemans commented Jan 23, 2023

KrisThielemans commented Jan 25, 2023

KrisThielemans commented Jan 25, 2023

KrisThielemans commented Jan 25, 2023 •

edited by paskino

Loading

KrisThielemans commented Jan 25, 2023

paskino commented Jan 25, 2023

casperdcl commented Nov 28, 2023

casperdcl left a comment

casperdcl Nov 28, 2023

paskino Nov 29, 2023

casperdcl Nov 28, 2023

paskino Nov 29, 2023

KrisThielemans left a comment •

edited

Loading

KrisThielemans commented Feb 21, 2024

casperdcl commented Feb 21, 2024

KrisThielemans commented Feb 21, 2024

docker-stacks and jupyterhub #718

docker-stacks and jupyterhub #718

Conversation

paskino commented May 19, 2022 • edited by casperdcl Loading

paskino commented May 20, 2022

KrisThielemans left a comment

Choose a reason for hiding this comment

KrisThielemans commented Jan 23, 2023

KrisThielemans commented Jan 23, 2023

paskino commented Jan 23, 2023

KrisThielemans commented Jan 23, 2023

paskino commented Jan 23, 2023

paskino commented Jan 23, 2023

KrisThielemans commented Jan 23, 2023 • edited Loading

KrisThielemans commented Jan 23, 2023

KrisThielemans commented Jan 23, 2023

KrisThielemans commented Jan 23, 2023

KrisThielemans commented Jan 23, 2023

KrisThielemans commented Jan 25, 2023

KrisThielemans commented Jan 25, 2023

KrisThielemans commented Jan 25, 2023 • edited by paskino Loading

KrisThielemans commented Jan 25, 2023

paskino commented Jan 25, 2023

casperdcl commented Nov 28, 2023

casperdcl left a comment

Choose a reason for hiding this comment

casperdcl Nov 28, 2023

Choose a reason for hiding this comment

paskino Nov 29, 2023

Choose a reason for hiding this comment

casperdcl Nov 28, 2023

Choose a reason for hiding this comment

paskino Nov 29, 2023

Choose a reason for hiding this comment

KrisThielemans left a comment • edited Loading

Choose a reason for hiding this comment

KrisThielemans commented Feb 21, 2024

casperdcl commented Feb 21, 2024

KrisThielemans commented Feb 21, 2024

paskino commented May 19, 2022 •

edited by casperdcl

Loading

KrisThielemans commented Jan 23, 2023 •

edited

Loading

KrisThielemans commented Jan 25, 2023 •

edited by paskino

Loading

KrisThielemans left a comment •

edited

Loading