Re: git-clone causes out of memory

From: Mike Hommey <mh@glandium.org>
Date: Oct 13, 2017, 10:37 UTC
Message-ID: <20171013103722.rvr7536mu2hoo4wb@glandium.org>
In-Reply-To: <CAP8UFD1KuBdUCo=x_q4__s1kW15CWMH1jJkKzXqmf3=T3jcrng@mail.gmail.com>

On Fri, Oct 13, 2017 at 12:26:46PM +0200, Christian Couder wrote:

Show 69 quoted lines

> On Fri, Oct 13, 2017 at 12:06 PM, Mike Hommey <mh@glandium.org> wrote:
> > On Fri, Oct 13, 2017 at 12:51:58PM +0300, Constantine wrote:
> >> There's a gitbomb on github. It is undoubtedly creative and funny, but since
> >> this is a bug in git, I thought it'd be nice to report. The command:
> >>
> >>       $ git clone https://github.com/x0rz/ShadowBrokersFiles
> >
> > What fills memory is actually the checkout part of the command. git
> > clone -n doesn't fail.
> >
> > Credit should go where it's due: https://kate.io/blog/git-bomb/
> > (with the bonus that it comes with explanations)
> 
> Yeah, there is a thread on Hacker News about this too:
> 
> https://news.ycombinator.com/item?id=15457076
> 
> The original repo on GitHub is:
> 
> https://github.com/Katee/git-bomb.git
> 
> After cloning it with -n, there is the following "funny" situation:
> 
> $ time git rev-list HEAD
> 7af99c9e7d4768fa681f4fe4ff61259794cf719b
> 18ed56cbc5012117e24a603e7c072cf65d36d469
> 45546f17e5801791d4bc5968b91253a2f4b0db72
> 
> real    0m0.004s
> user    0m0.000s
> sys     0m0.004s
> $ time git rev-list HEAD -- d0/d0/d0/d0/d0/d0/d0/d0/d0/d0/f0
> 
> real    0m0.004s
> user    0m0.000s
> sys     0m0.000s
> $ time git rev-list HEAD -- d0/d0/d0/d0/d0/d0/d0/d0/d0/d0
> 
> real    0m0.004s
> user    0m0.000s
> sys     0m0.000s
> $ time git rev-list HEAD -- d0/d0/d0/d0/d0/d0/d0/d0/
> 45546f17e5801791d4bc5968b91253a2f4b0db72
> 
> real    0m0.005s
> user    0m0.008s
> sys     0m0.000s
> $ time git rev-list HEAD -- d0/d0/d0/d0/d0/
> 45546f17e5801791d4bc5968b91253a2f4b0db72
> 
> real    0m0.203s
> user    0m0.112s
> sys     0m0.088s
> $ time git rev-list HEAD -- d0/d0/d0/d0/
> 45546f17e5801791d4bc5968b91253a2f4b0db72
> 
> real    0m1.305s
> user    0m0.720s
> sys     0m0.580s
> $ time git rev-list HEAD -- d0/d0/d0/
> 45546f17e5801791d4bc5968b91253a2f4b0db72
> 
> real    0m12.135s
> user    0m6.700s
> sys     0m5.412s
> 
> So `git rev-list` becomes exponentially more expensive when you run it
> on a shorter directory path, though it is fast if you run it without a
> path.

That's because there are 10^7 files under d0/d0/d0, 10^6 under d0/d0/d0/d0/, 10^5 under d0/d0/d0/d0/d0/ etc.

So really, this is all about things being slower when there's a crazy number of files. Picture me surprised.

What makes it kind of special is that the repository contains a lot of paths/files, but very few objects, because it's duplicating everything.

All the 10^10 blobs have the same content, all the 10^9 trees that point to them have the same content, all the 10^8 trees that point to those trees have the same content, etc.

If git wasn't effectively deduplicating identical content, the repository would be multiple gigabytes large.

Mike