Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Install of 2.6.2 from conda-forge fails when was previously working #210

Open
brsolomon-deloitte opened this issue Feb 14, 2022 · 16 comments

Comments

@brsolomon-deloitte
Copy link

brsolomon-deloitte commented Feb 14, 2022

The build output below is for a successful install of tensorflow-gpu==2.6.2 from 10 days ago.

As of today (2022-02-14), tensorflow-gpu 2.6.2 install now seems not to work despite nothing about the build parameters being changed.

What happened here?

+ mamba install -y --file requirements.txt
conda-forge/noarch        
conda-forge/linux-64      
Transaction

  Prefix: /opt/conda

  Updating specs:

   - bokeh==2.3.3
   - boto3==1.20.48
   - elasticsearch-dsl==7.4.0
   - elasticsearch==7.16.3
   - matplotlib-base==3.5.1
   - nltk==3.6.7
   - pandas==1.4.0
   - pyod==0.9.7
   - pyspark==3.2.1
   - scikit-learn==1.0.2
   - seaborn==0.11.2
   - statsmodels==0.13.1
   - tensorflow-gpu==2.6.2
   - ca-certificates
   - certifi
   - openssl


  Package                      Version  Build                   Channel                    Size
─────────────────────────────────────────────────────────────────────────────────────────────────
  Install:
─────────────────────────────────────────────────────────────────────────────────────────────────

  + absl-py                     0.15.0  pyhd8ed1ab_0            conda-forge/noarch        98 KB
  + aiohttp                      3.8.1  py39h3811e60_0          conda-forge/linux-64     588 KB
  + aiosignal                    1.2.0  pyhd8ed1ab_0            conda-forge/noarch        12 KB
  + appdirs                      1.4.4  pyh9f0ad1d_0            conda-forge/noarch        13 KB
  + astunparse                   1.6.3  pyhd8ed1ab_0            conda-forge/noarch        15 KB
  + async-timeout                4.0.2  pyhd8ed1ab_0            conda-forge/noarch         9 KB
  + aws-c-auth                   0.6.7  hfef2836_0              conda-forge/linux-64      94 KB
  + aws-c-compression           0.2.14  h7c7754b_7              conda-forge/linux-64      17 KB
  + aws-c-http                  0.6.10  h58a30cf_2              conda-forge/linux-64     173 KB
  + aws-c-mqtt                   0.7.9  h042a236_0              conda-forge/linux-64      67 KB
  + aws-c-s3                    0.1.27  hae5f17b_11             conda-forge/linux-64      49 KB
  + aws-c-sdkutils               0.1.1  h7c7754b_4              conda-forge/linux-64      23 KB
  + aws-crt-cpp                 0.17.8  h82bac0c_1              conda-forge/linux-64     204 KB
  + boto3                      1.20.48  pyhd8ed1ab_0            conda-forge/noarch        71 KB
  + botocore                   1.23.48  pyhd8ed1ab_0            conda-forge/noarch         5 MB
  + cachetools                   4.2.4  pyhd8ed1ab_0            conda-forge/noarch        12 KB
  + cudatoolkit                 11.0.3  h15472ef_10             conda-forge/linux-64     953 MB
  + cudnn                     8.2.1.32  h86fa8c9_0              conda-forge/linux-64     674 MB
  + elasticsearch               7.16.3  pyhd8ed1ab_0            conda-forge/noarch       139 KB
  + elasticsearch-dsl            7.4.0  py39hf3d152e_1          conda-forge/linux-64      86 KB
  + frozenlist                   1.3.0  py39h3811e60_0          conda-forge/linux-64     161 KB
  + gast                         0.4.0  pyh9f0ad1d_0            conda-forge/noarch        12 KB
  + google-auth                 1.35.0  pyh6c4a22f_0            conda-forge/noarch        81 KB
  + google-auth-oauthlib         0.4.6  pyhd8ed1ab_0            conda-forge/noarch        19 KB
  + google-pasta                 0.2.0  pyh8c360ce_0            conda-forge/noarch        42 KB
  + grpcio                      1.41.1  py39hff7568b_1          conda-forge/linux-64       2 MB
  + jmespath                    0.10.0  pyh9f0ad1d_0            conda-forge/noarch        21 KB
  + keras                        2.6.0  pyhd8ed1ab_1            conda-forge/noarch       822 KB
  + keras-preprocessing          1.1.2  pyhd8ed1ab_0            conda-forge/noarch        34 KB
  + markdown                     3.3.6  pyhd8ed1ab_0            conda-forge/noarch        67 KB
  + mock                         4.0.3  py39hf3d152e_2          conda-forge/linux-64      51 KB
  + multidict                    6.0.2  py39h3811e60_0          conda-forge/linux-64     156 KB
  + nccl                      2.11.4.1  h17a0586_2              conda-forge/linux-64     151 MB
  + nltk                         3.6.7  pyhd8ed1ab_0            conda-forge/noarch         1 MB
  + opt_einsum                   3.3.0  pyhd8ed1ab_1            conda-forge/noarch        53 KB
  + py4j                      0.10.9.3  pyhd8ed1ab_1            conda-forge/noarch       180 KB
  + pyasn1                       0.4.8  py_0                    conda-forge/noarch        53 KB
  + pyasn1-modules               0.2.7  py_0                    conda-forge/noarch        60 KB
  + pyod                         0.9.7  pyhd8ed1ab_0            conda-forge/noarch        81 KB
  + pyspark                      3.2.1  pyhd8ed1ab_0            conda-forge/noarch       268 MB
  + python-flatbuffers            1.12  pyhd8ed1ab_1            conda-forge/noarch        19 KB
  + pyu2f                        0.1.5  pyhd8ed1ab_0            conda-forge/noarch        31 KB
  + regex                    2022.1.18  py39h3811e60_0          conda-forge/linux-64     845 KB
  + requests-oauthlib            1.3.1  pyhd8ed1ab_0            conda-forge/noarch        22 KB
  + rsa                            4.8  pyhd8ed1ab_0            conda-forge/noarch        31 KB
  + s3transfer                   0.5.1  pyhd8ed1ab_0            conda-forge/noarch        55 KB
  + tensorboard                  2.6.0  pyhd8ed1ab_1            conda-forge/noarch         5 MB
  + tensorboard-data-server      0.6.0  py39h95dcef6_1          conda-forge/linux-64       3 MB
  + tensorboard-plugin-wit       1.8.1  pyhd8ed1ab_0            conda-forge/noarch       668 KB
  + tensorflow                   2.6.2  cuda110py39h016931e_1   conda-forge/linux-64      29 KB
  + tensorflow-base              2.6.2  cuda110py39h405f49e_1   conda-forge/linux-64     286 MB
  + tensorflow-estimator         2.6.2  cuda110py39h016931e_1   conda-forge/linux-64     653 KB
  + tensorflow-gpu               2.6.2  cuda110py39h5b0ac8e_1   conda-forge/linux-64      29 KB
  + termcolor                    1.1.0  py_2                    conda-forge/noarch         6 KB
  + typing-extensions          3.7.4.3  0                       conda-forge/noarch         8 KB
  + werkzeug                     2.0.2  pyhd8ed1ab_0            conda-forge/noarch       221 KB
  + wrapt                       1.12.1  py39h3811e60_3          conda-forge/linux-64      47 KB
  + yarl                         1.7.2  py39h3811e60_1          conda-forge/linux-64     138 KB

  Change:
─────────────────────────────────────────────────────────────────────────────────────────────────

  - arrow-cpp                    6.0.1  py39h01fd06f_8_cpu      installed                      
  + arrow-cpp                    6.0.1  py39h1d68239_0_cpu      conda-forge/linux-64      25 MB
  - aws-c-event-stream           0.2.7  h3541f99_13             installed                      
  + aws-c-event-stream           0.2.7  hb80ed28_31             conda-forge/linux-64      47 KB
  - pyarrow                      6.0.1  py39hff6fa39_8_cpu      installed                      
  + pyarrow                      6.0.1  py39hff6fa39_0_cpu      conda-forge/linux-64       3 MB

  Upgrade:
─────────────────────────────────────────────────────────────────────────────────────────────────

  - aws-c-cal                   0.5.11  h95a6274_0              installed                      
  + aws-c-cal                   0.5.12  h70efedd_7              conda-forge/linux-64      36 KB
  - aws-c-common                 0.6.2  h7f98852_0              installed                      
  + aws-c-common                0.6.17  h7f98852_0              conda-forge/linux-64     179 KB
  - aws-c-io                    0.10.5  hfb6a706_0              installed                      
  + aws-c-io                   0.10.13  he836878_5              conda-forge/linux-64     137 KB
  - aws-checksums               0.1.11  ha31a3da_7              installed                      
  + aws-checksums               0.1.12  h7c7754b_6              conda-forge/linux-64      50 KB
  - aws-sdk-cpp                1.8.186  hb4091e7_3              installed                      
  + aws-sdk-cpp                1.9.145  hfe59705_2              conda-forge/linux-64       5 MB
  - s2n                         1.0.10  h9b69904_0              installed                      
  + s2n                          1.3.0  h9b69904_0              conda-forge/linux-64     483 KB

  Downgrade:
─────────────────────────────────────────────────────────────────────────────────────────────────

  - black                       22.1.0  pyhd8ed1ab_0            installed                      
  + black                       21.7b0  pyhd8ed1ab_0            conda-forge/noarch       110 KB
  - bokeh                        2.4.2  py39hf3d152e_0          installed                      
  + bokeh                        2.3.3  py39hf3d152e_0          conda-forge/linux-64       8 MB
  - grpc-cpp                    1.42.0  ha1441d3_1              installed                      
  + grpc-cpp                    1.41.1  h75e9d12_2              conda-forge/linux-64       4 MB
  - h5py                         3.6.0  nompi_py39h7e08c79_100  installed                      
  + h5py                         3.1.0  nompi_py39h25020de_100  conda-forge/linux-64       1 MB
  - hdf5                        1.12.1  nompi_h2750804_103      installed                      
  + hdf5                        1.10.6  nompi_h6a2412b_1114     conda-forge/linux-64       3 MB
  - libprotobuf                 3.19.4  h780b84a_0              installed                      
  + libprotobuf                 3.18.1  h780b84a_0              conda-forge/linux-64       3 MB
  - numpy                       1.21.5  py39haac66dc_0          installed                      
  + numpy                       1.19.5  py39hdbf815f_2          conda-forge/linux-64       5 MB
  - orc                          1.7.2  h1be678f_0              installed                      
  + orc                          1.7.1  h68e2c4e_0              conda-forge/linux-64       1 MB
  - protobuf                    3.19.4  py39he80948d_0          installed                      
  + protobuf                    3.18.1  py39he80948d_0          conda-forge/linux-64     347 KB
  - pytables                     3.7.0  py39h2669a42_0          installed                      
  + pytables                     3.6.1  py39hf6dc253_3          conda-forge/linux-64       2 MB
  - six                         1.16.0  pyh6c4a22f_0            installed                      
  + six                         1.15.0  pyh9f0ad1d_0            conda-forge/noarch        14 KB
  - tomli                        2.0.0  pyhd8ed1ab_1            installed                      
  + tomli                        1.2.2  pyhd8ed1ab_0            conda-forge/noarch        15 KB
  - typing_extensions            4.0.1  pyha770c72_0            installed                      
  + typing_extensions          3.7.4.3  py_0                    conda-forge/noarch        25 KB

  Summary:

  Install: 58 packages
  Change: 3 packages
  Upgrade: 6 packages
  Downgrade: 13 packages

  Total download: 2 GB

─────────────────────────────────────────────────────────────────────────────────────────────────

Finished s2n                                  (00m:00s)             483 KB      5 MB/s
Finished aws-c-common                         (00m:00s)             179 KB      1 MB/s
Finished aws-c-compression                    (00m:00s)              17 KB    104 KB/s
Finished aws-c-cal                            (00m:00s)              36 KB    189 KB/s
Finished libprotobuf                          (00m:00s)               3 MB     12 MB/s
Finished hdf5                                 (00m:00s)               3 MB     13 MB/s
Finished aws-c-http                           (00m:00s)             173 KB    671 KB/s
Finished aws-c-auth                           (00m:00s)              94 KB    344 KB/s
Finished tensorboard-data-server              (00m:00s)               3 MB     12 MB/s
Finished termcolor                            (00m:00s)               6 KB     18 KB/s
Finished numpy                                (00m:00s)               5 MB     16 MB/s
Finished py4j                                 (00m:00s)             180 KB    490 KB/s
Finished python-flatbuffers                   (00m:00s)              19 KB     53 KB/s
Finished cachetools                           (00m:00s)              12 KB     31 KB/s
Finished tensorboard-plugin-wit               (00m:00s)             668 KB      2 MB/s
Finished aiosignal                            (00m:00s)              12 KB     29 KB/s
Finished keras-preprocessing                  (00m:00s)              34 KB     75 KB/s
Finished pyod                                 (00m:00s)              81 KB    163 KB/s
Finished aws-sdk-cpp                          (00m:00s)               5 MB      9 MB/s
Finished rsa                                  (00m:00s)              31 KB     61 KB/s
Finished boto3                                (00m:00s)              71 KB    134 KB/s
Finished protobuf                             (00m:00s)             347 KB    620 KB/s
Finished google-auth-oauthlib                 (00m:00s)              19 KB     32 KB/s
Finished pytables                             (00m:00s)               2 MB      2 MB/s
Finished tensorflow-gpu                       (00m:00s)              29 KB     44 KB/s
Finished botocore                             (00m:00s)               5 MB      8 MB/s
Finished grpc-cpp                             (00m:00s)               4 MB      4 MB/s
Finished aws-checksums                        (00m:00s)              50 KB     54 KB/s
Finished yarl                                 (00m:00s)             138 KB    143 KB/s
Finished aws-c-mqtt                           (00m:00s)              67 KB     67 KB/s
Finished six                                  (00m:00s)              14 KB     11 KB/s
Finished keras                                (00m:00s)             822 KB    626 KB/s
Finished jmespath                             (00m:00s)              21 KB     15 KB/s
Finished werkzeug                             (00m:00s)             221 KB    159 KB/s
Finished opt_einsum                           (00m:00s)              53 KB     38 KB/s
Finished absl-py                              (00m:00s)              98 KB     68 KB/s
Finished typing-extensions                    (00m:00s)               8 KB      5 KB/s
Finished pyasn1-modules                       (00m:00s)              60 KB     39 KB/s
Finished mock                                 (00m:00s)              51 KB     33 KB/s
Finished aiohttp                              (00m:00s)             588 KB    369 KB/s
Finished tensorflow                           (00m:00s)              29 KB     16 KB/s
Finished tensorboard                          (00m:00s)               5 MB      3 MB/s
Finished wrapt                                (00m:00s)              47 KB     25 KB/s
Finished aws-c-sdkutils                       (00m:00s)              23 KB     12 KB/s
Finished aws-c-io                             (00m:00s)             137 KB     71 KB/s
Finished aws-crt-cpp                          (00m:00s)             204 KB    104 KB/s
Finished tomli                                (00m:00s)              15 KB      8 KB/s
Finishe
                  __    __    __    __
                 /  \  /  \  /  \  /  \
                /    \/    \/    \/    \
███████████████/  /██/  /██/  /██/  /████████████████████████
              /  / \   / \   / \   / \  \____
             /  /   \_/   \_/   \_/   \    o \__,
            / _/                       \_____/  `
            |/
        ███╗   ███╗ █████╗ ███╗   ███╗██████╗  █████╗
        ████╗ ████║██╔══██╗████╗ ████║██╔══██╗██╔══██╗
        ██╔████╔██║███████║██╔████╔██║██████╔╝███████║
        ██║╚██╔╝██║██╔══██║██║╚██╔╝██║██╔══██╗██╔══██║
        ██║ ╚═╝ ██║██║  ██║██║ ╚═╝ ██║██████╔╝██║  ██║
        ╚═╝     ╚═╝╚═╝  ╚═╝╚═╝     ╚═╝╚═════╝ ╚═╝  ╚═╝

        mamba (0.20.0) supported by @QuantStack

        GitHub:  https://github.com/mamba-org/mamba
        Twitter: https://twitter.com/QuantStack

█████████████████████████████████████████████████████████████


Looking for: ['bokeh==2.3.3', 'boto3==1.20.48', 'elasticsearch-dsl==7.4.0', 'elasticsearch==7.16.3', 'matplotlib-base==3.5.1', 'nltk==3.6.7', 'pandas==1.4.0', 'pyod==0.9.7', 'pyspark==3.2.1', 'scikit-learn==1.0.2', 'seaborn==0.11.2', 'statsmodels==0.13.1', 'tensorflow-gpu==2.6.2']


Pinned packages:
  - python 3.9.*
  - python 3.9.7


Preparing transaction: ...working... done
Verifying transaction: ...working... done
Executing transaction: ...working... By downloading and using the CUDA Toolkit conda packages, you accept the terms and conditions of the CUDA End User License Agreement (EULA): https://docs.nvidia.com/cuda/eula/index.html

By downloading and using the cuDNN conda packages, you accept the terms and conditions of the NVIDIA cuDNN EULA -
  https://docs.nvidia.com/deeplearning/cudnn/sla/index.html

done
d markdown                             (00m:00s)              67 KB     33 KB/s
Finished astunparse                           (00m:00s)              15 KB      7 KB/s
Finished nltk                                 (00m:00s)               1 MB    534 KB/s
Finished async-timeout                        (00m:00s)               9 KB      4 KB/s
Finished elasticsearch-dsl                    (00m:00s)              86 KB     40 KB/s
Finished frozenlist                           (00m:00s)             161 KB     73 KB/s
Finished arrow-cpp                            (00m:01s)              25 MB     12 MB/s
Finished h5py                                 (00m:00s)               1 MB    546 KB/s
Finished aws-c-s3                             (00m:00s)              49 KB     21 KB/s
Finished typing_extensions                    (00m:00s)              25 KB     11 KB/s
Finished pyasn1                               (00m:00s)              53 KB     22 KB/s
Finished bokeh                                (00m:00s)               8 MB      3 MB/s
Finished grpcio                               (00m:00s)               2 MB    888 KB/s
Finished tensorflow-estimator                 (00m:00s)             653 KB    246 KB/s
Finished orc                                  (00m:00s)               1 MB    407 KB/s
Finished pyarrow                              (00m:00s)               3 MB      1 MB/s
Finished requests-oauthlib                    (00m:00s)              22 KB      7 KB/s
Finished black                                (00m:00s)             110 KB     38 KB/s
Finished elasticsearch                        (00m:00s)             139 KB     47 KB/s
Finished appdirs                              (00m:00s)              13 KB      2 KB/s
Finished s3transfer                           (00m:00s)              55 KB      7 KB/s
Finished multidict                            (00m:00s)             156 KB     19 KB/s
Finished pyu2f                                (00m:00s)              31 KB      4 KB/s
Finished regex                                (00m:00s)             845 KB    102 KB/s
Finished google-pasta                         (00m:00s)              42 KB      5 KB/s
Finished gast                                 (00m:00s)              12 KB      1 KB/s
Finished google-auth                          (00m:00s)              81 KB     10 KB/s
Finished aws-c-event-stream                   (00m:00s)              47 KB      6 KB/s
Finished nccl                                 (00m:06s)             151 MB     19 MB/s
Finished tensorflow-base                      (00m:11s)             286 MB     28 MB/s
Finished pyspark                              (00m:10s)             268 MB     24 MB/s
Finished cudnn                                (00m:21s)             674 MB     41 MB/s
Finished cudatoolkit                          (00m:23s)             953 MB     49 MB/s
@izahn
Copy link

izahn commented Feb 14, 2022

If you are trying to install tensorflow-gpu on a computer that doesn't have a gpu you need to set CONDA_OVERRIDE_CUDA as documented in https://conda-forge.org/docs/maintainer/knowledge_base.html#cuda-builds

@brsolomon-deloitte
Copy link
Author

If you are trying to install tensorflow-gpu on a computer that doesn't have a gpu you need to set CONDA_OVERRIDE_CUDA as documented in https://conda-forge.org/docs/maintainer/knowledge_base.html#cuda-builds

As shown in my original post this build worked barely over a week ago and now does not. I'm hoping to get clarification on why. The requirement pinned to tensorflow-gpu==2.6.2 is unchanged.

@brsolomon-deloitte
Copy link
Author

brsolomon-deloitte commented Feb 14, 2022

From docker container run -it --rm --entrypoint=bash jupyter/all-spark-notebook:spark-3.2.1:

(base) jovyan@e62f15dc56ec:~$ CONDA_OVERRIDE_CUDA='11.2' mamba install tensorflow-gpu==2.6.2

                  __    __    __    __
                 /  \  /  \  /  \  /  \
                /    \/    \/    \/    \
███████████████/  /██/  /██/  /██/  /████████████████████████
              /  / \   / \   / \   / \  \____
             /  /   \_/   \_/   \_/   \    o \__,
            / _/                       \_____/  `
            |/
        ███╗   ███╗ █████╗ ███╗   ███╗██████╗  █████╗
        ████╗ ████║██╔══██╗████╗ ████║██╔══██╗██╔══██╗
        ██╔████╔██║███████║██╔████╔██║██████╔╝███████║
        ██║╚██╔╝██║██╔══██║██║╚██╔╝██║██╔══██╗██╔══██║
        ██║ ╚═╝ ██║██║  ██║██║ ╚═╝ ██║██████╔╝██║  ██║
        ╚═╝     ╚═╝╚═╝  ╚═╝╚═╝     ╚═╝╚═════╝ ╚═╝  ╚═╝

        mamba (0.21.1) supported by @QuantStack

        GitHub:  https://github.com/mamba-org/mamba
        Twitter: https://twitter.com/QuantStack

█████████████████████████████████████████████████████████████


Looking for: ['tensorflow-gpu==2.6.2']

conda-forge/linux-64                                        Using cache
conda-forge/noarch                                          Using cache

Pinned packages:
  - python 3.9.*
  - python 3.9.10


Encountered problems while solving:
  - package tensorflow-gpu-2.6.2-cuda102py39hf05f184_0 requires tensorflow 2.6.2 cuda102py39h87695c4_0, but none of the providers can be installed

(base) jovyan@e62f15dc56ec:~$ mamba info

                  __    __    __    __
                 /  \  /  \  /  \  /  \
                /    \/    \/    \/    \
███████████████/  /██/  /██/  /██/  /████████████████████████
              /  / \   / \   / \   / \  \____
             /  /   \_/   \_/   \_/   \    o \__,
            / _/                       \_____/  `
            |/
        ███╗   ███╗ █████╗ ███╗   ███╗██████╗  █████╗
        ████╗ ████║██╔══██╗████╗ ████║██╔══██╗██╔══██╗
        ██╔████╔██║███████║██╔████╔██║██████╔╝███████║
        ██║╚██╔╝██║██╔══██║██║╚██╔╝██║██╔══██╗██╔══██║
        ██║ ╚═╝ ██║██║  ██║██║ ╚═╝ ██║██████╔╝██║  ██║
        ╚═╝     ╚═╝╚═╝  ╚═╝╚═╝     ╚═╝╚═════╝ ╚═╝  ╚═╝

        mamba (0.21.1) supported by @QuantStack

        GitHub:  https://github.com/mamba-org/mamba
        Twitter: https://twitter.com/QuantStack

█████████████████████████████████████████████████████████████


     active environment : base
    active env location : /opt/conda
            shell level : 1
       user config file : /home/jovyan/.condarc
 populated config files : /opt/conda/.condarc
          conda version : 4.11.0
    conda-build version : not installed
         python version : 3.9.10.final.0
       virtual packages : __linux=4.14.262=0
                          __glibc=2.31=0
                          __unix=0=0
                          __archspec=1=x86_64
       base environment : /opt/conda  (writable)
      conda av data dir : /opt/conda/etc/conda
  conda av metadata url : None
           channel URLs : https://conda.anaconda.org/conda-forge/linux-64
                          https://conda.anaconda.org/conda-forge/noarch
          package cache : /opt/conda/pkgs
                          /home/jovyan/.conda/pkgs
       envs directories : /opt/conda/envs
                          /home/jovyan/.conda/envs
               platform : linux-64
             user-agent : conda/4.11.0 requests/2.27.1 CPython/3.9.10 Linux/4.14.262-200.489.amzn2.x86_64 ubuntu/20.04.3 glibc/2.31
                UID:GID : 1000:100
             netrc file : None
           offline mode : False

Am I misinterpreting your suggestion @izahn ?

@izahn
Copy link

izahn commented Feb 14, 2022

No, you have the right idea. See how it no longer says nothing provides __cuda needed by tensorflow-base-2.6.2-cuda102py37h55054dc_2? But now you have another problem which is presumably that tensorflow 2.6.2 cuda102py39h87695c4_0 conflicts with something already installed in your environment. What happens if you start over with

CONDA_OVERRIDE_CUDA='11.2' mamba install -y --file requirements.txt

?

@ngam
Copy link
Contributor

ngam commented Feb 14, 2022

How about even starting with something simpler: CONDA_OVERRIDE_CUDA='11.2' mamba create -n test tensorflow-gpu==2.6.2? This would make it clear if there are dependency problems. All builds from 2.7.0 shouldn't be affected by the __cuda constraint, I think... wrong it started with 2.6.2: https://github.com/conda-forge/tensorflow-feedstock/pull/188/files but not sure why it is happening to you only now...

@brsolomon-deloitte
Copy link
Author

brsolomon-deloitte commented Feb 14, 2022

What happens if you start over

In this case requirements.txt is a red herring. It's added on top of the base image which is jupyter/all-spark-notebook:spark-3.2.1 and not relevant to creating a minimum reproducible example.

At the risk of going slightly off topic here, ultimately it looks like there is a difference between 2.6.2 and 2.7.0.

Does not work:

CONDA_OVERRIDE_CUDA='11.2' mamba install tensorflow-gpu==2.6.2

(Errors with package tensorflow-gpu-2.6.2-cuda102py39hf05f184_0 requires tensorflow 2.6.2 cuda102py39h87695c4_0, but none of the providers can be installed.)

Works:

(base) jovyan@1d87aae3f816:~$ CONDA_OVERRIDE_CUDA='11.2' mamba install tensorflow-gpu==2.7.0
...
(base) jovyan@1d87aae3f816:~$ mamba list | egrep 'tensorflow|cuda'
cudatoolkit               10.2.89             h8f6ccaa_10    conda-forge
tensorflow                2.7.0           cuda102py39h30a2e9f_0    conda-forge
tensorflow-base           2.7.0           cuda102py39h15c874f_0    conda-forge
tensorflow-estimator      2.7.0           cuda102py39h87695c4_0    conda-forge
tensorflow-gpu            2.7.0           cuda102py39hf05f184_0    conda-forge

@ngam
Copy link
Contributor

ngam commented Feb 14, 2022

Also, fyi, there is a known bug (see #208) that CONDA_OVERRIDE_CUDA='11.2' mamba install tensorflow-gpu==2.7.0 will almost always get you cuda102 (if you want cuda112+, then you should do something like

CONDA_OVERRIDE_CUDA='11.2' mamba install tensorflow-gpu==2.7.0=cuda112*

or

CONDA_OVERRIDE_CUDA='11.2' mamba install tensorflow-gpu==2.7.0 cudatoolkit>=11.2

@brsolomon-deloitte
Copy link
Author

Also, I do realize the title of this issue is incorrect and that is my mistake. 2.6.2 is indeed available on conda-forge as is 2.7.0. I was looking mistakenly at the anaconda channel, which only contained 2.6.0.

@brsolomon-deloitte brsolomon-deloitte changed the title What happened to 2.6.2 on conda-forge? Install of 2.6.2 from conda-forge fails when was previously working Feb 14, 2022
@izahn
Copy link

izahn commented Feb 14, 2022

In this case requirements.txt is a red herring. It's added on top of the base image which is jupyter/all-spark-notebook:spark-3.2.1 and not relevant to creating a minimum reproducible example.

Right, well then everything has to be compatible with what is already installed in that environment or it won't work.

At the risk of going slightly off topic here, ultimately it looks like there is a difference between 2.6.2 and 2.7.0.

Of course, wouldn't be much point in having both version is they were the same :-)

Does not work:

CONDA_OVERRIDE_CUDA='11.2' mamba install tensorflow-gpu==2.6.2

(Errors with package tensorflow-gpu-2.6.2-cuda102py39hf05f184_0 requires tensorflow 2.6.2 cuda102py39h87695c4_0, but none of the providers can be installed.)

Works:

(base) jovyan@1d87aae3f816:~CONDA_OVERRIDE_CUDA='11.2' mamba install tensorflow-gpu==2.7.0

But this is now about compatibility between tensorflow 2.6.2 cuda102 and jupyter/all-spark-notebook:spark-3.2.1 rather than a problem with any of the tenseflow builds per se.

@brsolomon-deloitte
Copy link
Author

brsolomon-deloitte commented Feb 14, 2022

But this is now about compatibility between tensorflow 2.6.2 cuda102 and jupyter/all-spark-notebook:spark-3.2.1 rather than a problem with any of the tenseflow builds per se.

That's a fair point ... since all-spark-notebook is derived from several intermediate base images I can try to drill down and see where a conflict is introduced. Wish mamba would give a more informative error message here :(

@izahn
Copy link

izahn commented Feb 14, 2022

I've found it helpful to try to install the thing it says it can't and then keep going until it mentions something already in the environment. So here you could start with

CONDA_OVERRIDE_CUDA='10.2' mamba install tensorflow=2.6.2cuda102py39h87695c4_0

and go from there.

@ngam
Copy link
Contributor

ngam commented Feb 14, 2022

But this is now about compatibility between tensorflow 2.6.2 cuda102 and jupyter/all-spark-notebook:spark-3.2.1 rather than a problem with any of the tenseflow builds per se.

That's a fair point ... since all-spark-notebook is derived from several intermediate base images I can try to drill down and see where a conflict is introduced. Wish mamba would give a more informative error message here :(

recent (global pinnings) changes include grpc_cpp and protobuf, I would start with these two --- the all-spark-notebook might be late (in case of protobuf) or early (in case of grpc_cpp)

@h-vetinari
Copy link
Member

h-vetinari commented Feb 14, 2022

recent (global pinnings) changes include grpc_cpp and protobuf

For context: the global pinning sets a universal version for certain dependencies (like protobuf) that many packages are compiled against. To increase this version, conda-forge runs a migration that builds all packages depending on (e.g.) protobuf against the new version (i.e. a transition period where there are packages against both the old and the new version), and once all packages are done, the migration is closed, and the global pin updated.

The reason this is relevant and perhaps somewhat non-obvious is that eventually, after things have moved to a new global pin, all package builds following from then on will be compiled against the new version, and old package builds become impossible to install side-by-side with the newest packages that depend on the same shared libraries, but are compiled against a different ABI.

@bjarthur
Copy link

bjarthur commented Jul 26, 2022

thanks for the CONDA_OVERRIDE_CUDA suggestion. that works for me. however, in the link to the docs you provide (https://conda-forge.org/docs/maintainer/knowledge_base.html#cuda-builds), it actually says to put {{ compiler('cuda') }} in the build sections of your requirements. no mention of the override. this does not work for me. it fails with conda.exceptions.ResolvePackageNotFound: - cuda_linux-64. here is the relevant section of my meta.yaml file:

requirements:
  build:
    - {{ compiler('cuda') }}
  run:
    - tensorflow-gpu >=2.8

@izahn
Copy link

izahn commented Jul 26, 2022

The link is correct, scroll down to the Note about "How is cudatoolkit selected at install time"

@izahn
Copy link

izahn commented Aug 3, 2022

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

No branches or pull requests

5 participants