Re: RFC: Design and code of partial clones (now, missing commits and trees OK)

On Fri, 22 Sep 2017 17:02:11 -0400
Jeff Hostetler [off-list ref] wrote:

quoted

I was struggling a bit with the terminology, true.

Right now I'm thinking of:
  - promisor remote (as you defined)
  - promisor packfile (as you defined)
  - promisor object is an object known to belong to the promisor (whether
    because we have it in a promisor packfile or because it is referenced
    by an object in a promisor packfile)

This might eliminate "promise(d)", and thus eliminate the confusion
between "promised" and "promisor", but I haven't done an exhaustive
search.

maybe just call the "promised" ones "missing".

They are not the same, though - "missing" usually just means that the
local repo does not have it, without regard to whether another repo has
it.

quoted

quoted

I guess it depends on how many missing-objects you expect the client
to have. My concern here is that we're limiting the design to the
"occasional" big file problem, rather than the more general scale
problem.

Do you have a specific situation in mind?

I have would like to be able do sparse-enlistments in the Windows
source tree. (3.5M files at HEAD.)  Most developers only need a small
feature area (a device driver or file system or whatever) and not the
whole tree.  A typical Windows developer may have only 30-50K files
populated.  If we can synchronize on a sparse-checkout spec and use
that for both the checkout and the clone/fetch, then we can bulk fetch
the blobs that they'll actually need.  GVFS can hydrate the files as
they touch them, but I can use this to pre-fetch the needed blobs in
bulk, rather than faulting and fetching them one-by-one.

So, my usage may have >95% of the ODB be missing blobs.  Contrast that
with the occasional large blob / LFS usage where you may have <5% of
the ODB as large objects (by count of OIDs not disk usage).

I don't think the current design precludes a more intelligent bulk
fetching (e.g. being allowed to specify a "want" tree and several "have"
trees, although we will have to figure out a design for that, including
how to select the "have" trees to inform the server about).

In the meantime, yes, this will be more useful for occasional large blob
repos, and (if/when the hook support is added) a GVFS situation where
the missing objects are available network-topologically close by.