Re: What's cooking in git.git (Jul 2016, #05; Wed, 13)

[Stefan Beller <sbeller@google.com>]

On Wed, Jul 13, 2016 at 10:32 AM, Junio C Hamano <gitster@pobox.com> wrote:
> Stefan Beller <sbeller@google.com> writes:
>
>>> * sb/push-options (2016-07-12) 5 commits
>>>  - add a test for push options
>>>  - push: accept push options
>>>  - SQUASH???
>>
>> Squash? I do not find a squashable commit in what you pushed,
>> do you intend to squash the first 2 patches instead?

Oh I pulled a few minutes before you sent this email, and forgot
that you likely have pushed again when sending this email. :/
Thanks!

>> Yeah there were some late comments, so I did not reroll right away.
>> I think Shawns proposal to have a receive.maxCommandBytes is a
>> good way for an overall upper bound, but how does it stop us from
>> going forward with this series?
>
> If we were to do maxcommandbytes, then max_options would become
> irrelevant, no?

Maybe?

I do not know what kind of safety measures we want in place here, and
if we want to go for overlapping things?

Currently there are none at all in your upstream code, although you cannot
push arbitrary large things to either Shawns or Peffs $Dayjob servers, so
I wonder if we want to either agree on one format or on many overlapping
things, as some different hosts may perceive different things as DoS threats,
so they can fine tune as they want?

In the Gerrit world, you have a ref per code review, such that it is easy to
have 50k refs or more, similar to the repo Jeff pointed out [1], that has 40k
tags (and getting a new tag every 2 hours apparently).

So I could understand if different services care about the different loads
(refs vs push options) differently (one would want to allow unlimited refs
pushing for mirroring such repos as pointed out above, while another
one might care about the total load of the server for a single rogue user)

Thanks,
Stefan

[1] https://github.com/JetBrains/intellij-community