Git submodule init async

Question

When I run git submodule update --init first time on a projects which have a lot of submodules, this usually take a lot of time, because most of submodules are stored on slow public servers.

Is there a possibility to initialize submodules asynchronously?

Answer 1

Linux:

cat .gitmodules | grep -Po '".*"' | sed 's/.\(.\+\).$/\1/' | while sleep 0.1 && read line; do git submodule update --init $line & done

Mac:

cat .gitmodules | grep -o '".*"' | cut -d '"' -f 2 | while sleep 0.1 && read line; do git submodule update --init $line & done

Answer 2

Update January 2016:

With Git 2.8 (Q1 2016), you will be able to fetch submodules in parallel (!) with git fetch --recurse-submodules -j2.
See "How to speed up / parallelize downloads of git submodules using git clone --recursive?"

---

Original answer mid-2013

You could try:

• to initialize first all submodules:

  git submodule init

Then, the foreach syntax:

git submodule foreach git submodule update --recursive -- $path &

If the '&' applies to the all line (instead of just the 'git submodule update --recursive -- $path' part), then you could call a script which would make the update in the background.

git submodule foreach git_submodule_update

Answer 3

This can also be done in Python. In Python 3 (because we're in 2015...), we can use something like this:

#!/usr/bin/env python3

import os
import re
import subprocess
import sys
from functools import partial
from multiprocessing import Pool

def list_submodules(path):
    gitmodules = open(os.path.join(path, ".gitmodules"), 'r')
    matches = re.findall("path = ([\w\-_\/]+)", gitmodules.read())
    gitmodules.close()
    return matches


def update_submodule(name, path):
    cmd = ["git", "-C", path, "submodule", "update", "--init", name]
    return subprocess.call(cmd, shell=False)


if __name__ == '__main__':
    if len(sys.argv) != 2:
        sys.exit(2)
    root_path = sys.argv[1]

    p = Pool()
    p.map(partial(update_submodule, path=root_path), list_submodules(root_path))

This may be safer than the one-liner given by @Karmazzin (since that one just keeps spawning processes without any control on the number of processes spawned), still it follows the same logic: read .gitmodules, then spawn multiple processes running the proper git command, but here using a process pool (the maximum number of processes can be set too). The path to the cloned repository needs to be passed as an argument. This was tested extensively on a repository with around 700 submodules.

Note that in the case of a submodule initialization, each process will try to write to .git/config, and locking issues may happen:

error: could not lock config file .git/config: File exists

Failed to register url for submodule path '...'

This can be caught with subprocess.check_output and a try/except subprocess.CalledProcessError block, which is cleaner than the sleep added to @Karmazzin's method. An updated method could look like:

def update_submodule(name, path):
    cmd = ["git", "-C", path, "submodule", "update", "--init", name]
    while True:
        try:
            subprocess.check_output(cmd, stderr=subprocess.PIPE, shell=False)
            return
        except subprocess.CalledProcessError as e:
            if b"could not lock config file .git/config: File exists" in e.stderr:
                continue
            else:
                raise e

With this, I managed to run the init/update of 700 submodules during a Travis build without the need to limit the size of the process pool. I often see a few locks caught that way (~3 max).

Answer 4

As of Git 2.8 you can do this:

git submodule update --init --jobs 4

where 4 is the number of submodules to download in parallel.