Set up personal Anaconda mirror
  • Nov 16, 2018
  • 6 minute read

Predicaments

If you encounter any problems as follows, you should try to set up a personal Python(Anaconda) mirror:

  1. Deploy a python environment on servers that have no access to the Internet;
  2. Deploy a python environment which can be used directly by other users on the server;
  3. Internet speed or connection to cloud mirror is slow.

Steps

Installation

Just install Anaconda according to this guide.

Of course, you can download Anaconda installer archive from here.

Download(mirror)

You can use wget to mirror the packages from official Anaconda websites or other mirror websites (TUNA and USTC).

If your server is located in China, I recommend the later one.

I’ll take the later one as example.

Create your directory for saving packages:

$ mkdir anaconda

Create two files for downloading packages:

$ cd anaconda
$ touch download.sh links.txt
$ chmod +x download.sh

Save all urls you wanted to links.txt:

https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/free/linux-64/
https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/free/noarch/
https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/linux-64/
https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/noarch/
https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/pro/linux-64/
https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/pro/noarch/
https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/conda-forge/linux-64/
https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/conda-forge/noarch/

Then, edit the download.sh file to download all files in each directory:

#!/bin/bash
for link in `cat links.txt`; do
    wget -m -np -nH --cut-dirs=1 -e robots=off -R "index.html*" $link
done

Explanations of options:

-m,  --mirror
    Turn on options suitable for mirroring.  This option turns on
    recursion and time-stamping, sets infinite recursion depth and
    keeps FTP directory listings.  It is currently equivalent to 
    -r -N -l inf --no-remove-listing.
-np, --no-parent
    don't ascend to the parent directory.
-nH, --no-host-directories
    Disable generation of host-prefixed directories.
--cut-dirs=number
    Ignore number directory components. This is useful for getting a fine-grained control over     the directory where recursive retrieval will be saved.
-e robots=off
    ignore robots.txt.
-R,  --reject=LIST
    comma-separated list of rejected extensions.

PS : conda-forge is much larger than others. Of course, you can just download specific version and packages:

# Reject other versions except Python2.7
# Beware, not to use something like "only py27" packages.
# Indeed, many packages in the repo #don't have version string in their name;
# and you'll miss them as dependency.
# All package information is written in a repodata.json and repodata.json.bz2.
$ wget -m -np -nH --cut-dirs=2 -e robots=off -R --regex-type pcre --reject-regex '(.*py26.*)|(.*py3[3456].*)' https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/free/linux-64/

Don’t forget to download noarch, otherwise conda won’t work correctly.

Anyway, if your links.txt and download.sh are as same sa mine, you will get this tree of directory:

└── anaconda
    ├── cloud
    │   └── conda-forge
    │       ├── linux-64
    │       └── noarch
    └── pkgs
        ├── free
        │   ├── linux-64
        │   └── noarch
        ├── main
        │   ├── linux-64
        │   └── noarch
        └── pro
            ├── linux-64
            └── noarch

Settings

It’s more convenient to save all packages in one directory and just using channel_alias for channels.

Let’s move cloud/conda-forge to anaconda/pkgs.

└── anaconda
    └── pkgs
        ├── conda-forge
        │   ├── linux-64
        │   └── noarch
        ├── free
        │   ├── linux-64
        │   └── noarch
        ├── main
        │   ├── linux-64
        │   └── noarch
        └── pro
            ├── linux-64
            └── noarch

Now, you can modify the channel_alias and add channels:

$ conda config --set channel_alias file://Path_to_your_channel/anaconda/pkgs
$ conda config --add channels free
$ conda config --add channels pro
$ conda config --add channels main
$ conda config --add channels conda-forge
$ conda config --remove channels defaults
$ conda config --set offline True

Then, check ~/.condarc :

channels:
  - conda-forge
  - main
  - pro
  - free
channel_alias: file://Path_to_your_channel/anaconda/pkgs

If you prefer command line, you can just use conda config --get :

$ conda config --get
--set channel_alias file://Path_to_your_channel/anaconda/pkgs
--add channels 'free'   # lowest priority
--add channels 'pro'
--add channels 'main'
--add channels 'conda-forge'   # highest priority
--set offline True

If you make a mistake during adding channel, you can remove it:

$ conda config --remove channels <your_mistake>

By default, conda now prefers packages from a higher priority channel over any version from a lower priority channel. Therefore, you can now safely put channels at the bottom of your channel list to provide additional packages that are not in the default channels, and still be confident that these channels will not override the core package set.

So, if you want to set the order to free → main → conda-forge, you can rearrange them by terminal or edit the .condarc file directly.

Anyway, you will get this .condarc file:

allow_other_channels : false
channel_alias: file://datadir2/anaconda/pkgs
channels:
  - free
  - main
  - pro 
  - conda-forge
offline: true

Content between these dividing lines are wrong according to this answer.

Let’s create a new env named test to prepare for next step (Aggregation).

$ conda create -n test python=3.7
$ source activate test
$ conda install conda-build

Aggregation

What about the ‘anaconda’ channel on anaconda.org at https://anaconda.org/anaconda? The Anaconda channel on anaconda.org is an aggregated mirror of the packages in https://repo.continuum.io/pkgs/main, https://repo.continuum.io/pkgs/free, and https://repo.continuum.io/pkgs/pro. If you are using conda’s ‘defaults’ channel, you do not need the ‘anaconda’ channel.

Although you’ve downloaded packages, but if you want to install as usual like conda install -c anaconda netcdf4, you have to aggregate /main, /free and /pro.

So, you should create a directory named anaconda under anaconda_mirror.

Then, link all files under each linux-64 and noarch of /main, /free and /pro to linux-64 and noarch under anaconda directory.

Now, you can use conda-build to generates a file repodata.json.

$ conda index Path_to_your_channel/anaconda/pkgs/anaconda/linux-64/ Path_to_your_channel/anaconda/pkgs/anaconda/noarch/

Update/Install

Now, you can use conda update/install :

$ conda install <package>
$ conda install --channel conda-forge <package>

Don’t use anaconda channel anymore, because we just have free, main, pro and conda-forge channels.

For example, if you want to install netcdf4 from anaconda channel, don’t specify channel name:

$ conda install netcdf4

If you have many packaged to install, you can create requirements.txt and install all of them by one command.

$ while read requirement; do conda install --yes $requirement; done < requirements.txt 2>error.log

Here’s one example of requirements.txt :

netcdf4
h5py
matplotlib
statsmodels

Advanced

If you’re the administrator of the server, you can use system .condarc to set a environment for all users.

You can make conda and any number of packages available to a group of 1 or more users, while preventing these users from installing unwanted packages with conda:

  1. Install conda and the allowed packages, if any, in a location that is under administrator control and accessible to users.
  2. Create a .condarc system configuration file in the root directory of the installation. This system-level configuration file will override any user-level configuration files installed by the user.

Each user accesses the central conda installation, which reads settings from the user .condarcconfiguration file located in their home directory. The path to the user file is the same as the root environment prefix displayed by conda info, as shown in User configuration file below. The user.condarc file is limited by the system .condarc file.

System configuration settings are commonly used in a system .condarc file but may also be used in a user .condarc file. All user configuration settings may also be used in a system .condarc file.

You can just check the official document about administrator-controlled installation.

Here’s the example of my system .condarc :

allow_other_channels : false
channel_alias: file://Path_to_your_channel/anaconda/pkgs
channels:
  - free
  - main
  - pro
  - conda-forge
offline: true

References

  1. https://stackoverflow.com/questions/37391824/simply-use-python-anaconda-without-internet-connection
  2. http://www.meteoboy.com/conda-without-internet.html#
  3. https://conda.io/docs/user-guide/configuration/use-condarc.html
  4. https://conda.io/docs/user-guide/configuration/admin-multi-user-install.html
  5. https://conda.io/docs/user-guide/tasks/create-custom-channels.html
  • Python
    • Previous
    Trick of pcolormesh
    Next
    Send e-mail when file modified

    Say something

    Thank you

    Your comment has been submitted and will be published once it has been approved.

    OOPS!

    Your comment has not been submitted. Please go back and try again. Thank You!

    If this error persists, please open an issue by clicking here.