Docker Failed (RDKit)

[amelie-iska]

Issue: Docker fails at installing rdkit.
Attempted Workarounds: Tried removing the version of rdkit, also tried using pip install rdkit.

docker build -t molecule_stm .
[+] Building 40.7s (13/27)                                 docker:default
 => [internal] load build definition from Dockerfile                 0.0s
 => => transferring dockerfile: 1.39kB                               0.0s 
 => [internal] load metadata for nvcr.io/nvidia/pytorch:22.01-py3    1.6s 
 => [internal] load .dockerignore                                    0.0s
 => => transferring context: 2B                                      0.0s 
 => [ 1/24] FROM nvcr.io/nvidia/pytorch:22.01-py3@sha256:06f27ba669  0.0s 
 => CACHED [ 2/24] RUN useradd -ms /bin/bash shengchaol              0.0s 
 => CACHED [ 3/24] WORKDIR /home/shengchaol                          0.0s 
 => CACHED [ 4/24] RUN chmod -R 777 /home/shengchaol                 0.0s 
 => CACHED [ 5/24] RUN chmod -R 777 /usr/bin                         0.0s 
 => CACHED [ 6/24] RUN chmod -R 777 /bin                             0.0s 
 => CACHED [ 7/24] RUN chmod -R 777 /usr/local                       0.0s 
 => CACHED [ 8/24] RUN chmod -R 777 /opt/conda                       0.0s 
 => CACHED [ 9/24] RUN conda install -y python=3.7                   0.0s 
 => ERROR [10/24] RUN conda install -y -c rdkit rdkit=2020.09.1.0   39.1s 
------
 > [10/24] RUN conda install -y -c rdkit rdkit=2020.09.1.0:
1.165 Collecting package metadata (current_repodata.json): ...working... done
14.62 Solving environment: ...working... failed with initial frozen solve. Retrying with flexible solve.
30.82 Solving environment: ...working... failed with repodata from current_repodata.json, will retry with next repodata source.
38.38
38.38 ResolvePackageNotFound:
38.38   - conda==4.11.0
38.38
------
Dockerfile:20
--------------------
  18 |     RUN conda install -y python=3.7
  19 |
  20 | >>> RUN conda install -y -c rdkit rdkit=2020.09.1.0
  21 |     # RUN pip install rdkit
  22 |     RUN conda install -y -c conda-forge -c pytorch pytorch=1.9.1   
--------------------
ERROR: failed to solve: process "/bin/sh -c conda install -y -c rdkit rdkit=2020.09.1.0" did not complete successfully: exit code: 1

[chao1224]

Hi @amelie-iska,

So this exception is about wrong version of rdkit. One solution is as follow:

1. First you use this command to find the available rdkit versions: conda search rdkit -c rdkit. Here is what I got (my current runnable env, not the one from Nvidia):

rdkit                      2014.03.1      np18py27_3  rdkit               
rdkit                      2014.09.1      np19py26_1  rdkit               
rdkit                      2014.09.1      np19py27_1  rdkit
...
...
...          
rdkit                    2020.03.3.0  py37hc20afe1_1  rdkit               
rdkit                    2020.09.1.0  py36hd50e099_1  rdkit               
rdkit                    2020.09.1.0  py37hd50e099_1  rdkit 

2. Then you try other versions instead of 2020.09.1.0.

[kosonocky]

@amelie-iska I had the same issue. Changing most conda installs to pip installs fixed this for me. This is the docker file that worked:

FROM nvcr.io/nvidia/pytorch:22.01-py3 as base

#create a new new user
RUN useradd -ms /bin/bash shengchaol

# #change to this user
# USER shengchaol

#set working directory
WORKDIR /home/shengchaol

RUN chmod -R 777 /home/shengchaol
RUN chmod -R 777 /usr/bin
RUN chmod -R 777 /bin
RUN chmod -R 777 /usr/local
RUN chmod -R 777 /opt/conda

RUN conda install -y python=3.7

RUN pip install rdkit
RUN conda install -y -c conda-forge -c pytorch pytorch=1.9.1

RUN conda install -y -c pyg -c conda-forge pyg

RUN pip install requests
RUN pip install tqdm
RUN pip install matplotlib
RUN pip install spacy

# for SciBert
RUN pip install boto3
RUN pip install transformers

# for MoleculeNet
RUN pip install ogb

# install pysmilesutils
RUN python -m pip install git+https://github.com/MolecularAI/pysmilesutils.git

RUN pip install deepspeed

# install Megatron
RUN cd /tmp && git clone https://github.com/MolecularAI/MolBART.git --branch megatron-molbart-with-zinc && cd /tmp/MolBART/megatron_molbart/Megatron-LM-v1.1.5-3D_parallelism && pip install .

# install apex
RUN cd /tmp && git clone https://github.com/chao1224/apex.git
RUN cd /tmp/apex/ && pip install -v --disable-pip-version-check --no-cache-dir --global-option="--cpp_ext" --global-option="--cuda_ext" ./


#expose port for Jupyter
EXPOSE 8888

[amelie-iska]

Thanks @kosonocky any chance you've had any luck with ProteinDT (the other CLIP based model by @chao1224) as well?