I've gotten complaints that cloning from github.com with "-q" will
sometimes abort before sending any data. The problem is that during a
quiet clone, upload-pack may be silent for a long time while preparing
the pack (i.e., the "counting objects" and "compressing" phases).

We have load balancers and reverse proxies sitting in front of the git
machines, and these machines may sometimes think the connection has hung
and drop it, even though if they had waited a few more seconds, the pack
would have started coming.

We mitigated it somewhat by bumping the timeouts on our side, but that's
only one piece of the puzzle. Clients may be going through http proxies
or stateful firewalls on their end, and neither we nor they have any

This series teaches upload-pack to periodically send an empty sideband
data packet when pack-objects is being quiet. That keeps a small amount
of data flowing, and should be backwards compatible. I hand-tested it
against JGit, dulwich (via the mercurial git plugin), libgit2, and old
versions of git, and all worked fine.  It has also been running on
github.com for about a week and a half, and nobody has reported any

  [1/2]: upload-pack: send keepalive packets during pack computation
  [2/2]: upload-pack: bump keepalive default to 5 seconds

To unsubscribe from this list: send the line "unsubscribe git" in
the body of a message to majord...@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Reply via email to