RE: [PATCH] add http.maxReceiveSpeed to limit git-receive-pack receiving speed

From: "Randall S. Becker" <rsbecker@nexbridge.com>
To: "'Jeff King'" <peff@peff.net>, "'Junio C Hamano'" <gitster@pobox.com>
Cc: "'Xia XiaoWen'" <haoyurenzhuxia@gmail.com>, <git@vger.kernel.org>,
	<worldhello.net@gmail.com>,
	"'Xia XiaoWen'" <chenan.xxw@alibaba-inc.com>
Subject: RE: [PATCH] add http.maxReceiveSpeed to limit git-receive-pack receiving speed
Date: Fri, 20 Aug 2021 14:21:28 -0400	[thread overview]
Message-ID: <018c01d795f0$350a5c00$9f1f1400$@nexbridge.com> (raw)
In-Reply-To: <YR/wbWhei0PPLSgX@coredump.intra.peff.net>

On August 20, 2021 2:12 PM, Jeff King wrote:
>On Thu, Aug 19, 2021 at 11:28:53AM -0700, Junio C Hamano wrote:
>
>> > https protocol, and only supports libcurl 7.15.5 and above.
>>
>> We are likely be raising the floor versions of libcURL to 7.16.0 or
>> even 7.19.4 soonish.  It probably would make it easier to allow it
>> unconditionally (otherwise you'd probably need to implement error or
>> warning messages when configuration is given but the libcURL version
>> used is too old, etc.).
>
>Yeah, I agree that if we can drop the conditional totally, we should.
>
>If we do need to make it conditional, I think there was a preference for shifting to checking the actual option constants, like:
>
>  #ifdef CURLOPT_MAX_RECV_SPEED_LARGE
>  ...ok, we have it...
>  #endif
>
>The rest of your comments all seemed quite reasonable to me.
>
>As a general feature, IMHO speed-limiting is best done outside of application-level tools like Git (and instead done via proxies or kernel
>network config). But since we are not building the feature ourselves, but rather just plugging our config to curl's feature, I don't have any
>problem with it here.

I second that. This should be in the realm of QoS at a managed switch/router in the consumer's network (or the upstream repo in some cases).
-Randall