Git clone hangs with large repo

[Abbyyan]

I've used Gitpython to clone some repos but it hangs with a specified repo with size of 17G. I've create a Pool to do git clone using Gitpython. There is a large git repo and needs more time than others to clone. Each process do a clone work for one repo. The Pool i used as follows:

  multi_res = [p.apply_async(runfunc, args=(
            incl_info, project_root, skip_dirs,)) for incl_info in incl_infos]
    LogInfo('Waiting for all subprocesses done...')
    for i in range(len(incl_infos)):
        while not multi_res[i].ready():
            LogInfo("Downloading now")
            time.sleep(5)
    p.close()
    p.join()

It works perfectly in most case. But will often hangs in the largest repo. It's wired that when i just clone the repo individually, It works fine. So i wonder if there is some block in python multiprocessing Pool at first.

I've strace the hanged git clone process . The git process output as follows:

Process 27649 attached
read(6, 0x7ffc36dae050, 4)              = ? ERESTARTSYS (To be restarted if SA_RESTART is set)
--- SIGALRM {si_signo=SIGALRM, si_code=SI_KERNEL, si_value={int=2895997, ptr=0x2c307d}} ---
rt_sigreturn()                          = 0
read(6, 0x7ffc36dae050, 4)              = ? ERESTARTSYS (To be restarted if SA_RESTART is set)
--- SIGALRM {si_signo=SIGALRM, si_code=SI_KERNEL, si_value={int=2895997, ptr=0x2c307d}} ---
rt_sigreturn()                          = 0
read(6, 0x7ffc36dae050, 4)              = ? ERESTARTSYS (To be restarted if SA_RESTART is set)
--- SIGALRM {si_signo=SIGALRM, si_code=SI_KERNEL, si_value={int=2895997, ptr=0x2c307d}} ---
rt_sigreturn()                          = 0
read(6, 0x7ffc36dae050, 4)              = ? ERESTARTSYS (To be restarted if SA_RESTART is set)
--- SIGALRM {si_signo=SIGALRM, si_code=SI_KERNEL, si_value={int=2895997, ptr=0x2c307d}} ---
rt_sigreturn()   

The git-lfs output as follows:

Process 28006 attached
[ Process PID=28006 runs in 32 bit mode. ]
futex(0x88b982c, FUTEX_WAIT_PRIVATE, 0, NULL

But when i replace the git.repo.clone_from with shell script git clone in a new subprocess, it works fine. So maybe there are some block in git.repo.clone_from, and i wonder whether it's solved. Thanks a lot.

[Byron]

Thanks for the detailed investigation! If memory serves, the way GitPython handles progress reporting on long-running clones can be prone to hanging. Even though it was thought to be fixed, apparently there is still a chance of it failing.

The workaround proposed here is certainly preferred over using GitPython at all, since it's doing what's needed much more directly. GitPython in the end just spawns a git process itself and fails to properly handle it's output on long-running process.

I am closing this issue as I don't think the underlying cause can clearly be determined or fixed, and due to the presence of a viable workaround. Please feel free to keep commenting here in case you would propose a different way of handling this - your opinion would be greatly appreciated.

[Abbyyan]

Thanks a lot. I enter the same problem today using checkout function and I wonder if there is a timeout when Gitpython start a new subprocess?

[Byron]

Unfortunately no, GitPython is inherently synchronous and blocking. Everyone is invited to have a look at the code, communicating to a subprocess should not ever hang :/.