Re: [RFC PATCH 01/13] serve: add command to advertise bundle URIs

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Tue, Aug 10 2021, Derrick Stolee wrote:

> On 8/5/2021 11:07 AM, Ævar Arnfjörð Bjarmason wrote:
> ...
>> +bundle-uri CLIENT AND SERVER EXPECTATIONS
>> +^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
>> +
>> +The advertised bundles MUST contain one or more reference tips for use
>> +by the client. Bundles that are not self-contained MUST use the
>> +standard "-" prefixes in the bundle format to indicate their
>> +prerequisites. I.e. they must be in the standard format "git bundle
>> +create" would create.
>> +
>> +If after an `ls-refs` the client finds that the ref tips it wants can
>> +be retrieved entirety from advertised bundle(s), it MAY
>> +disconnect. The results of such a "clone" or "fetch" should be
>> +indistinguishable from the state attained without using bundle-uri.
>> +
>> +The client MAY also keep the connection open pending download of the
>> +bundle-uris, e.g. should on or more downloads (or their validation)
>> +fail.
>
> The only technical thought I had (so far) about this proposal was that
> leaving the connection open while downloading the bundle would leave
> unnecessary load on the servers when no communication is happening.
> There is a cost to keeping an open SSH connection, so here it would be
> good to at least have the Git client close the connection after
> getting a 200 response from the bundle (but not waiting for all of its
> contents).

Thanks. Yes it's something I'll have to fix. I was hoping that I'd get
away with it for an initial implementation, but e.g. using
transfer.injectBundleURI to bootstrap chromium.git's repo from a bundle
will take so long that Google's server will give up and hang up on you.

I wonder if it's something the transport layer should be doing in
general to resume connections if they go stale if it's at a point of
clean separation in the dialog, but in any case I'll need it for
bundle-uri.

Closing the connection is also going to be more expensive in some cases,
e.g. if the bundle takes 1s we'll open/close/download
bundle-uri/open/close the connection, instead of of open/download
bundle-uri/close. I wonder if anyone cares though, we can always apply
some heuristic later I guess...




[Index of Archives]     [Linux Kernel Development]     [Gcc Help]     [IETF Annouce]     [DCCP]     [Netdev]     [Networking]     [Security]     [V4L]     [Bugtraq]     [Yosemite]     [MIPS Linux]     [ARM Linux]     [Linux Security]     [Linux RAID]     [Linux SCSI]     [Fedora Users]

  Powered by Linux